Answered from these docs only, by a model that cannot see your account. Check the pages it cites.

Metric registry

Every metric key Roiva ships: what it measures, its type and unit, and which kind of connection feeds it.

View as Markdown View as JSON

Answered from these docs only, by a model that cannot see your account. Check the pages it cites.

Generated from what Roiva ships, on every deploy.

Every key a library formula or an initiative template can read, grouped by the domain its key opens with. A key with a Fed by entry arrives from a connected system, or from a query on your own data warehouse where none of those is connected. The rest are recorded by a person, imported, read by a warehouse query, or produced by a formula. See Metrics from your own data warehouse.

What the key's shape means, and where a metric's baseline comes from, is in Metric keys, and where a value comes from.

Financial services

Key What it measures Type Unit Fed by
financial_services.advisors.accounts_per_advisor Client accounts managed per advisor.

Watch out for

  • Count accounts an advisor is responsible for, not households, unless you use households on both sides. Divide by the same advisors as Advisors.
  • The formula books each extra account per advisor as new revenue, so accounts moved over from another advisor overstate it.

Good for

  • Valuing advisor tools by the revenue of the extra accounts each advisor carries, at your contribution margin.

Guide: Revenue is booked at your margin

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.advisors.aum_per_advisor Average assets under management per advisor.

Watch out for

  • Use the period-end valuation and say so: markets move it between two dates whatever advisors do.
  • Formulas read it only as a baseline, the assets a change in organic growth is applied to.

Good for

  • With advisors, the assets a rise in organic growth is applied to.

Guide: What a formula reads

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.advisors.count Advisors using the client intelligence platform.

Watch out for

  • Advisors actually using the platform, from its usage report. Licenses bought is a different number.

Good for

  • The advisors a rise in accounts per advisor, or in organic growth, is multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.advisors.organic_growth_rate Net new assets in the period, inflows less outflows with market movement left out, annualized, as a percentage of assets under management at the period's start, from your custodian or portfolio system.

Watch out for

  • Net new assets with market movement left out, annualized: a month's figure times 12. Take it from your custodian's or portfolio system's flows report.
  • Markets, referrals, hiring and pricing move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing proactive client outreach by the advisory fees on the extra assets, at your contribution margin.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.advisors.proactive_outreach_rate Percentage of clients contacted proactively per quarter. percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.client_intake.completed_count New client intakes completed in the period.

Watch out for

  • Intakes that reached a plan or a quote, not documents received: an abandoned intake took time but finished nothing.

Good for

  • The intakes staff hours saved are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.client_intake.staff_minutes_per_intake Staff minutes of hands-on work to take in one new client's documents.

Watch out for

  • Minutes worked, not time waiting, including confirming what the AI extracted. Leave out KYC review if another initiative measures it.
  • Compare intakes of similar complexity: a household with trusts and several returns isn't the typical one.
  • The minutes freed are freed capacity, not cash: the staff are paid the same afterwards.

Good for

  • Valuing document extraction by the staff time it frees on each intake.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.commercial_credit.analyst_hours_per_request Credit analyst and underwriter hours of hands-on spreading, analysis and memo writing per commercial credit request.

Watch out for

  • Hours worked, not the days a request waits for documents or a committee, including checking the AI's spreads and drafts.
  • Compare requests of similar size and structure: a multi-entity relationship takes far longer than a small renewal.
  • It is part of the review hours an underwriting initiative may count on the same requests, so if one does, time only this work here.

Good for

  • Valuing AI spreading and credit memo drafting by the analyst time it frees on each request.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.commercial_credit.requests_analyzed_count Commercial credit requests spread and written up in the period, new loans, renewals and periodic loan reviews included.

Watch out for

  • Requests that reached a finished spread and memo, not applications received: a request withdrawn before analysis took little analyst time.
  • Count a request once however many borrowers or guarantors it spreads, the same way before and after.

Good for

  • The requests analyst hours saved are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.contact_center.agent_minutes_per_month Agent minutes in the month on customer or member contacts across calls, chats and secure messages, talk, hold and after-contact work included.

Watch out for

  • Every assisted channel, the same ones before and after. A phone report that leaves out after-call work or chat reads low.
  • It's compared per customer or member, not per contact, so easy questions the assistant brings in don't read as time saved. Record customers over the same months.
  • Outbound calls and back-office work agents do between contacts aren't contacts. Leave them out on both sides.

Good for

  • Valuing a virtual assistant and agent assist together by the agent time they free.

Guide: When the count moves with how busy you are

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.customers.count Customers or members with at least one open account at the end of the month.

Watch out for

  • People or households, not accounts, and the same choice before and after.
  • A merger or a conversion adds customers in one month. Leave it out of the comparison or start the baseline after it.

Good for

  • The volume contact center minutes are compared per, so a growing base isn't read as worse service.

Guide: When the count moves with how busy you are

count customers Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.digital_banking.avg_deposit_balance_per_user Average deposit balance held with you by the customers or members who use insights and coaching.

Watch out for

  • Average daily balance over the month, checking, savings and certificates, the same accounts before and after.
  • People who open insights already keep more with you, so the baseline is the same people before they started, or a holdout group that doesn't see insights.
  • Rates, tax refunds and promotions move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing financial insights and coaching by the deposits they keep with you.

Guide: When the outcome has other causes

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.digital_banking.insight_users_count Customers or members who opened an insight or a coaching session in digital banking in the period.

Watch out for

  • People who opened one, not everyone it was shown to, counted the same way every month.

Good for

  • The people a rise in average deposit balance is multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.fraud.false_positive_rate Percentage of flagged transactions that were legitimate.

Watch out for

  • A share of the same flags you record as Transactions Flagged, using resolved flags only: one still under review is neither.
  • Fewer false positives can also mean fewer flags of any kind, real fraud included. Read it beside the fraud loss rate.

Good for

  • Valuing ML fraud scoring by the analyst time no longer spent on legitimate transactions.

Guide: Cash value and freed capacity

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.fraud.flagged_count Transactions flagged by the fraud system in the period.

Watch out for

  • Flags resolved in the period, from the fraud system's logs, not cases opened: the false positive rate is a share of resolved flags, so count the same ones.

Good for

  • The flags a fall in the false positive rate is applied to.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.fraud.loss_rate_bps Fraud losses as basis points of transaction volume.

Watch out for

  • Losses over transaction volume × 10,000, on the same volume you record. Say whether it's net of recoveries: recoveries arrive months later and move a past period.
  • Fraud trends, merchant and channel mix, network rules and the economy move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing ML fraud scoring by the fraud losses avoided on this period's volume.

Guide: When the outcome has other causes

count bps Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.fraud.minutes_per_investigation Analyst minutes to investigate and resolve one flagged transaction, customer contact included, from a time study or your case system's time tracking.

Watch out for

  • Hands-on minutes from opening an alert to closing it, customer contact included.
  • Until it's measured, your own cost per investigation stands in and the value reads as estimated.

Good for

  • Pricing the analyst time fewer false positives free, at your loaded hourly rate.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.kyc.review_hours_per_account Staff hours to review and approve one account.

Watch out for

  • Compare accounts of the same risk tier: an enhanced-diligence case isn't the typical one.
  • The hours freed are freed capacity, not cash: reviewers are paid the same afterwards.

Good for

  • Valuing KYC automation by the review time it frees on each account.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.onboarding.accounts_count New accounts onboarded through KYC/AML in the period.

Watch out for

  • Accounts that completed onboarding, not applications received.

Good for

  • The accounts review hours and onboarding days saved are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.onboarding.days_to_active Average days from application to active account.

Watch out for

  • Calendar days from application to a usable account, waiting included, between the same two events before and after.
  • An account active sooner brings its revenue sooner, not more of it: the formula values the days at your cost of capital, on your margin.

Good for

  • Valuing faster onboarding by what having each account's revenue sooner is worth.

Guide: Revenue that arrives sooner

duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.operations.staff_minutes_per_task Staff minutes of hands-on work per deposit or loan operations task, averaged over every task in scope.

Watch out for

  • Minutes worked, not time a task waits in a queue, including reviewing and fixing what the AI agent did.
  • A task the agent finishes untouched counts only the minutes spent checking it.
  • The minutes freed are freed capacity, not cash: the staff are paid the same afterwards.

Good for

  • Valuing AI agents in back-office operations by the staff time they free on each task.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.operations.tasks_completed_count Deposit or loan operations tasks in scope completed in the period, by an AI agent or by staff.

Watch out for

  • Every task of the kinds in scope, whoever finished it, not only the ones the agent did.
  • Keep the same kinds of task in scope before and after; adding a new kind changes what the average minutes mean.

Good for

  • The tasks staff minutes saved are multiplied by.

Guide: What a formula reads

count tasks Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.regulatory_reporting.error_rate Percentage of regulatory report line items requiring correction before submission.

Watch out for

  • Line items, not filings: one amended schedule with twelve bad figures is twelve.
  • Errors caught before submission, from your validation log or sign-off record, counted the same way before and after.

Good for

  • With line items per cycle, the remediation work fewer reporting errors free.

Guide: Cash value and freed capacity

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.regulatory_reporting.line_items_per_cycle Line items across the regulatory reports filed in one reporting cycle.

Watch out for

  • Every figure the schedules in one cycle ask for counts as one item, across all the reports filed.
  • Per cycle, not per month: the formula spreads a year's cycles evenly over its months.

Good for

  • The line items a fall in the error rate is applied to.

Guide: What a formula reads

average — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.regulatory_reporting.prep_hours_per_report Staff hours to prepare and validate one regulatory report.

Watch out for

  • Hours for one report, from gathering to filing. Use a recent filing of the kind the initiative changes rather than averaging report types.
  • Per report cycle: the formula spreads a year's cycles evenly over its months.

Good for

  • Valuing reporting automation by the preparation time it frees.

Guide: What a formula reads

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.transactions.volume Total transaction volume processed in the period.

Watch out for

  • Use the same volume the fraud loss rate is measured over.

Good for

  • The volume a fall in the fraud loss rate is applied to.

Guide: When the outcome has other causes

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.underwriting.decision_hours Average hours from application submission to decision. duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.underwriting.manual_review_rate Percentage of applications requiring manual underwriter review.

Watch out for

  • Count a case decided automatically and then reopened as manual: the review happened.
  • The formula prices each application the AI now decides that the old process declined at its revenue less expected loss, spread over the loan term.

Good for

  • With the overturned decline rate, the extra loans AI decisioning approves.

Guide: Revenue is booked at your margin

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.underwriting.overturned_decline_rate Of past applications the AI would decide without manual review, the percentage it approves that the old process declined, from a back-test against known decisions.

Watch out for

  • From a back-test over past applications with known decisions, not from live results. Your model validation team or vendor runs it.
  • A loan the old process declined may have been declined for a reason: the formula takes expected loss off each one, so set that figure carefully.

Good for

  • The share of newly auto-decided applications that become extra approvals.

Guide: What a formula reads

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
financial_services.underwriting.review_hours_per_application Average underwriter and analyst hours of hands-on review per application decided, an automatically decided one counting as none — not the elapsed time to a decision.

Watch out for

  • Hours of work per application decided, an automatically decided one counting as none. Not the elapsed time: a file that sat three days and took forty minutes is 0.67.
  • The hours freed are freed capacity, not cash: underwriters are paid the same afterwards.

Good for

  • Valuing AI decisioning by the underwriter and analyst time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

HubSpot

Key What it measures Type Unit Fed by
hubspot.contacts.lead_to_mql_rate HubSpot's lead-to-MQL rate for the month. Written to crm.contacts.lead_to_mql_rate instead when this connection is the primary source. percentage % HubSpot
hubspot.contacts.mql_count HubSpot's marketing qualified leads for the month. Written to crm.contacts.mql_count instead when this connection is the primary source. count leads HubSpot
hubspot.contacts.mql_to_sql_rate HubSpot's MQL-to-SQL rate for the month. Written to crm.contacts.mql_to_sql_rate instead when this connection is the primary source. percentage % HubSpot
hubspot.contacts.new_count HubSpot's new contacts for the month. Written to crm.contacts.new_count instead when this connection is the primary source. count contacts HubSpot
hubspot.contacts.sql_count HubSpot's sales qualified leads for the month. Written to crm.contacts.sql_count instead when this connection is the primary source. count leads HubSpot
hubspot.deals.avg_demo_to_proposal_days Average days from the stage chosen as demo to the stage chosen as proposal, over the HubSpot deals first entering the proposal stage in the month. Written to crm.deals.avg_demo_to_proposal_days instead when this connection is the primary source. duration days HubSpot
hubspot.deals.avg_sales_cycle_days HubSpot's average sales cycle for the month. Written to crm.deals.avg_sales_cycle_days instead when this connection is the primary source. duration days HubSpot
hubspot.deals.avg_value HubSpot's average won deal size for the month. Written to crm.deals.avg_value instead when this connection is the primary source. currency USD HubSpot
hubspot.deals.lost_count HubSpot's lost deals for the month. Written to crm.deals.lost_count instead when this connection is the primary source. count deals HubSpot
hubspot.deals.open_count Deals open in HubSpot at sync time, a snapshot rather than a period total, written once per month and rewritten on every sync. count — HubSpot
hubspot.deals.open_value Value of the deals open in HubSpot at sync time, a snapshot rather than a period total, written once per month and rewritten on every sync. currency USD HubSpot
hubspot.deals.proposals_sent_count HubSpot deals in pipelines that count as sales that first entered the stage chosen as proposal in the month. Written to crm.proposals.sent_count instead when this connection is the primary source. count proposals HubSpot
hubspot.deals.renewal_arr_due The value of the HubSpot deals chosen as renewals that fell due in the month, whatever became of them. Written to customer_success.renewals.arr_up_for_renewal instead when this connection is the primary source. currency USD HubSpot
hubspot.deals.renewal_rate Of the HubSpot deals chosen as renewals that fell due in the month, the share won, by number or by value. Written to customer_success.renewals.rate instead when this connection is the primary source. percentage % HubSpot
hubspot.deals.renewal_rate_base The HubSpot deals chosen as renewals that fell due in the month, by number or by value as the renewal rate counts them. Written to customer_success.renewals.rate_base instead when this connection is the primary source. count — HubSpot
hubspot.deals.services_proposals_count HubSpot deals in pipelines ticked as services that first entered the stage chosen as proposal in the month. Written to professional_services.proposals.submitted_count instead when this connection is the primary source. count proposals HubSpot
hubspot.deals.win_rate HubSpot's deal win rate for the month. Written to crm.deals.win_rate instead when this connection is the primary source. percentage % HubSpot
hubspot.deals.won_count HubSpot's won deals for the month. Written to crm.deals.won_count instead when this connection is the primary source. count deals HubSpot
hubspot.deals.won_value HubSpot's won deal value for the month. Written to crm.deals.won_value instead when this connection is the primary source. currency USD HubSpot
hubspot.marketing.campaigns_launched HubSpot's campaigns launched for the month. Written to marketing.campaigns.launched_count instead when this connection is the primary source. count campaigns HubSpot
hubspot.marketing.email_open_rate HubSpot's marketing email open rate for the month. Written to marketing.emails.open_rate instead when this connection is the primary source. percentage % HubSpot
hubspot.marketing.organic_sessions HubSpot's organic search sessions for the month. Written to marketing.web.organic_sessions instead when this connection is the primary source. count sessions HubSpot
hubspot.marketing.posts_published HubSpot's content published for the month. Written to marketing.content.published_count instead when this connection is the primary source. count posts HubSpot
hubspot.tickets.avg_resolution_days HubSpot's average ticket resolution time for the month. Written to support.tickets.avg_resolution_days instead when this connection is the primary source. duration days HubSpot
hubspot.tickets.closed_count HubSpot's support tickets closed for the month. Written to support.tickets.closed_count instead when this connection is the primary source. count tickets HubSpot
hubspot.tickets.new_count HubSpot's support tickets opened for the month. Written to support.tickets.volume instead when this connection is the primary source. count tickets HubSpot
hubspot.tickets.open_count Tickets open in HubSpot at sync time, a snapshot rather than a period total, written once per month and rewritten on every sync. count — HubSpot
hubspot.tickets.resolution_rate HubSpot's ticket resolution rate for the month. Written to support.tickets.resolution_rate instead when this connection is the primary source. percentage % HubSpot

CRM

Key What it measures Type Unit Fed by
crm.contacts.lead_to_mql_rate Percentage of new contacts that became Marketing Qualified Leads. percentage % HubSpot
crm.contacts.mql_count Contacts that reached MQL (Marketing Qualified Lead) status in the period.

Watch out for

  • From HubSpot it's the contacts created in the month whose lifecycle stage is now MQL or later, so a recent month reads low, and earlier months keep changing as contacts move on.
  • Changing what makes a contact an MQL changes the count without anything happening to demand. Keep the definition the same before and after.

Good for

  • The MQLs a change in the MQL-to-SQL rate is multiplied by.

Guide: When the outcome has other causes

count leads HubSpot
crm.contacts.mql_to_sql_rate Percentage of the period's marketing-qualified contacts that also became sales-qualified.

Watch out for

  • From HubSpot it's the share of a month's new MQLs that are sales-qualified now. That share grows as contacts move on, so older months read higher than recent ones.
  • SDR capacity, how an MQL is defined, lead mix and the market move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing lead nurturing by the extra sales-qualified leads from the same MQLs.

Guide: When the outcome has other causes

percentage % HubSpot
crm.contacts.new_count New contacts created in the CRM in the period.

Watch out for

  • From HubSpot it counts contacts by the date they were created, so a list import or an event upload reads as a month of new leads.
  • The lead-scoring formulas take it as the leads the AI scores, which assumes it scores every new contact. Where it scores only some, it overstates the volume.

Good for

  • The leads that rep minutes saved per lead are multiplied by.
  • The volume nurture hours are compared per, so a campaign that floods the funnel isn't read as slower follow-up.

Guide: When the count moves with how busy you are

count contacts HubSpot
crm.contacts.sql_count Contacts that reached SQL (Sales Qualified Lead) status in the period.

Watch out for

  • From HubSpot it's the contacts created in the month whose lifecycle stage is now SQL or later, not the contacts that became sales-qualified in it. A recent month reads low.
  • Changing what makes a lead sales-qualified changes the count without any change in conversion. Keep the definition the same before and after.

Good for

  • Over new contacts, the month's lead-to-SQL rate; over MQLs, its MQL-to-SQL rate. The conversion formulas work both out from it.

Guide: When the outcome has other causes

count leads HubSpot
crm.contracts.avg_cycle_days Average days from contract request to signature.

Watch out for

  • A contract signed sooner brings the deal's money sooner, not more of it: the formula values the days at your cost of capital, on the margin of the deals won.
  • The formula applies the days saved to every deal won in the period, which assumes each one went through this contract process.

Good for

  • Valuing contract automation by the deals that sign sooner.

Guide: Revenue that arrives sooner

duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
crm.contracts.hours_per_contract Combined legal, sales and operations hours to take one contract from request to signature.

Watch out for

  • Hands-on hours across legal, sales and operations, not the days the contract was open. Count every group's time before and after, or the difference shifts with who was left out.
  • The hours freed are freed capacity, not cash: the same people are paid afterwards.

Good for

  • Valuing contract drafting, redlining and approval routing by the staff time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
crm.contracts.processed_count Contracts drafted, redlined and executed in the period.

Watch out for

  • Decide which contracts count, such as customer agreements only or NDAs and vendor contracts too, and keep that the same before and after. Average the hours per contract over the same contracts.

Good for

  • The contracts hours saved per contract are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
crm.deals.at_risk_share Percentage of open pipeline deals slipping or forecast to be lost.

Watch out for

  • Decide once what makes a deal at risk, such as a slipped close date or a forecast of lost, and keep it the same before and after.
  • It moves with the market as well as the AI, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing pipeline forecasting by the at-risk deals it helps recover.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
crm.deals.avg_demo_to_proposal_days Average days from demo to proposal sent, over the deals first sent a proposal in the period after a demo.

Watch out for

  • From a CRM, it runs from a deal's first entry into the stage chosen as demo to its first entry into the one chosen as proposal; one with no demo first is left out. By hand, start and stop the clock the same way before and after.
  • A faster proposal brings revenue sooner, not more of it: it's valued at your cost of capital.

Good for

  • Valuing proposal and quote automation.

Guide: Revenue that arrives sooner

duration days HubSpot, Pipedrive, Salesforce
crm.deals.avg_sales_cycle_days Average days from deal creation to close across the deals won in the period.

Watch out for

  • It's measured on the deals won in the period, so a batch of old deals closing at once lengthens it.
  • A shorter cycle brings revenue sooner, not more of it: formulas value the days at your cost of capital, and only the share of the change you credit to the initiative.

Good for

  • Valuing deals closing sooner after sales coaching or call intelligence.

Guide: Revenue that arrives sooner

duration days HubSpot, Pipedrive, Salesforce
crm.deals.avg_value Average closed-won deal size in the period. currency USD HubSpot, Pipedrive, Salesforce
crm.deals.lost_count Number of deals/opportunities closed-lost in the period.

Watch out for

  • A pipeline cleanup that closes stale deals as lost in one go raises the count, and drops that month's win rate, without anything happening to sales.

Good for

  • With deals won, the decided deals a change in win rate is applied to.

Guide: When the outcome has other causes

count deals HubSpot, Pipedrive, Salesforce
crm.deals.pipeline_value Value of the deals open in the period, taken as a snapshot when the CRM syncs.

Watch out for

  • A snapshot of the deals open when the CRM synced, not a total for the month.
  • It's the amounts reps entered, which run optimistic, and it includes deals that will never close.

Good for

  • Sizing what slipping deals put at risk, with win rate and sales cycle.

Guide: What a formula reads

currency USD HubSpot, Pipedrive, Salesforce
crm.deals.win_rate Percentage of closed deals that were won. won_count / (won_count + lost_count).

Watch out for

  • It moves with the market, pricing and the mix of your pipeline, so a formula valuing its change takes only the share of it you credit to the initiative.
  • Only closed deals count. Deals left open don't lower it, so it can rise while fewer deals close.

Good for

  • Valuing sales coaching or deal intelligence by the change in deals won.
  • Turning more qualified leads into an expected number of won deals.

Guide: When the outcome has other causes

percentage % HubSpot, Pipedrive, Salesforce
crm.deals.won_count Number of deals/opportunities closed-won in the period.

Watch out for

  • A deal counts in the month it closes, so a quarter-end push makes some months lumpy. Compare like periods.
  • A count says nothing about deal size. Read it beside the value of deals won.

Good for

  • The deals a change in conversion, cycle time or proposal speed is multiplied by.

Guide: What a formula reads

count deals HubSpot, Pipedrive, Salesforce
crm.deals.won_value Total value of all closed-won deals in the period, in the deals' currency when they share one and in US dollars at the European Central Bank's rate when they mix.

Watch out for

  • It's revenue, not return: a formula books what you keep of it at your contribution margin, never the full amount.
  • A month whose deals mix currencies is converted to US dollars at the European Central Bank's rate, so its total moves with exchange rates.

Good for

  • With the count of deals won, your average deal value.

Guide: Revenue is booked at your margin

currency USD HubSpot, Pipedrive, Salesforce
crm.leads.avg_lead_to_won_days Average days from lead creation to closed-won. duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
crm.leads.avg_qualify_minutes Average staff minutes to qualify one inbound lead.

Watch out for

  • Time the same task before and after, from the same start to the same finish, including the minutes spent checking the AI's score.
  • The minutes freed are freed capacity, not cash: reps are paid the same afterwards.

Good for

  • Valuing the rep time AI lead scoring frees, at your loaded rep rate.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
crm.leads.qualification_rate Percentage of scored leads that pass the qualification threshold. percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
crm.leads.qualified_count Leads that met the qualification threshold and were routed to a rep.

Watch out for

  • Lowering the score a lead needs to reach a rep raises it without the leads getting any better, and the formula prices each extra lead at this month's win rate. Keep the threshold the same before and after.
  • It's compared per lead scored, so record both over the same periods.

Good for

  • Valuing AI lead scoring by the extra deals expected from more qualified leads, at your contribution margin.

Guide: When the count moves with how busy you are

count leads Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
crm.leads.scored_count Inbound leads scored in the period.

Watch out for

  • Before the AI there was no score, so the baseline is the inbound leads that arrived, over the same months as the baseline for qualified leads.

Good for

  • The volume qualified leads are compared per, so a campaign that doubles inbound leads isn't read as better qualification.

Guide: When the count moves with how busy you are

count leads Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
crm.leads.sql_rate Sales-qualified leads in the period as a percentage of the new leads created in it.

Watch out for

  • Formulas read it only as the baseline. The figure now is worked out from your CRM's SQLs ÷ new contacts, so count the baseline the same way, from the same CRM.
  • Lead mix, SDR capacity, how an SQL is defined and the market move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing AI lead scoring by the extra sales-qualified leads from the same new leads.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
crm.proposals.hours_per_proposal Staff hours to assemble and send one proposal.

Watch out for

  • Count everyone's hands-on hours, pre-sales included, the same way before and after, including the time spent checking what the AI assembled.
  • The hours freed are freed capacity, not cash: reps and pre-sales are paid the same afterwards.

Good for

  • Valuing proposal and quote automation by the rep and pre-sales time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
crm.proposals.sent_count Proposals sent to prospects in the period.

Watch out for

  • From a CRM, each deal counts once, in the month it first entered the stage chosen as proposal, so a revised proposal isn't a new one. By hand, count the same way before and after, and average the hours per proposal over the same proposals.

Good for

  • The proposals hours saved per proposal are multiplied by.

Guide: What a formula reads

count — HubSpot, Pipedrive, Salesforce
crm.reps.admin_hours_per_week Sales rep hours per week on administrative work rather than selling. duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Finance

Key What it measures Type Unit Fed by
finance.ap.data_entry_minutes_per_invoice Staff minutes to key the data from one supplier invoice. duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
finance.ap.days_payable_outstanding Average days between invoice receipt and payment. duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
finance.ap.invoices_processed Supplier invoices processed through the extraction pipeline in the period. count invoices Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
finance.ap.minutes_per_invoice Staff minutes to process one supplier invoice end to end.

Watch out for

  • Hands-on minutes for entry, matching, coding and routing, not the time an invoice waits on an approver.
  • The formula multiplies it by every vendor bill received, so time it over your usual mix of bills, not only the easy ones.

Good for

  • Valuing invoice extraction and matching by the AP time it frees, at your loaded hourly rate.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
finance.bills.count Total vendor bills received in the period.

Watch out for

  • From your accounting platform it's counted by the date on each bill, not the day AP received or entered it.
  • A bill counts once however many lines it has, so time the minutes per bill over your usual mix of bills.

Good for

  • The volume AP minutes saved per bill, and bill errors avoided, are multiplied by.
  • The volume matching hours are compared per, so a heavy month of bills isn't read as slower matching.

Guide: When the count moves with how busy you are

count bills NetSuite, QuickBooks, Xero
finance.bills.error_rate Percentage of vendor bills entered with an error that needed correcting — amount, coding, vendor or a duplicate.

Watch out for

  • It's AP's error rate on the vendor bills it enters, not the billing error rate on invoices you send, which is Invoice Error Rate.
  • Decide what counts as an error, such as a wrong amount, coding, vendor or a duplicate, and keep that the same before and after.

Good for

  • With bills received, the bill errors avoided: the AP time they took is freed capacity and the money they let out is cash, in separate formulas.

Guide: Cash value and freed capacity

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
finance.bills.minutes_per_error Staff minutes to catch, investigate and correct one vendor bill error, vendor follow-up included, from a time study or your own timing.

Watch out for

  • Staff time only. Money an error cost, such as a duplicate payment never recovered, goes in the cash lost per bill error, or it's counted as freed capacity and again as cash.
  • Time it from spotting the error to its correction, vendor follow-up included. Until it's measured, your own cost per error stands in and the value reads as estimated.

Good for

  • Pricing the AP time fewer bill errors free, at your loaded hourly rate.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
finance.bills.volume Total value of all vendor bills received in the period. currency USD NetSuite, QuickBooks, Xero
finance.close.cycle_days Calendar days from period end to final sign-off of the close.

Watch out for

  • A faster close isn't cash: the formula prices each day saved at a daily cost of delayed reporting that you estimate, and books it as freed capacity.
  • Calendar days from period end to final sign-off. Compare like closes, a year-end close against year-end closes.

Good for

  • Valuing close automation by the days the books are final sooner.

Guide: Cash value and freed capacity

duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
finance.disputes.avg_resolution_days Average days to resolve a billing dispute.

Watch out for

  • Days a dispute stays open, not staff time. The formula values the days saved at your cost of capital on the disputed invoices, not as revenue.
  • The formula sizes disputed invoices from the billing error rate before the initiative, so it values disputes resolved sooner, not fewer disputes.

Good for

  • Valuing faster dispute handling by the cash it brings in sooner.

Guide: Revenue that arrives sooner

duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
finance.invoices.count Total outbound invoices created in the period.

Watch out for

  • Invoices you send customers, not the vendor bills AP receives, which are Bills Received.
  • From your accounting platform it's counted by the date on each invoice. An invoice counts once however many lines it has.

Good for

  • The volume billing errors and revenue recognition work are compared per, so a heavy billing month isn't read as more errors.

Guide: When the count moves with how busy you are

count invoices NetSuite, QuickBooks, Xero
finance.invoices.error_rate Percentage of invoices issued with an error requiring correction.

Watch out for

  • It's the error rate on invoices you send, not AP's on the bills it enters, which is Vendor Bill Error Rate.
  • Decide what counts as an error and keep that the same before and after.
  • Fixing an overcharge isn't value, since that money was never owed. Only the credits and write-offs disputes end in, and the staff time, are counted.

Good for

  • With invoices sent, the billing errors avoided: the staff time is freed capacity and the credits and write-offs are cash, in separate formulas.
  • Its baseline sizes how many invoices were disputed, when valuing disputes resolved sooner.

Guide: Cash value and freed capacity

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
finance.invoices.minutes_per_dispute Finance staff minutes to resolve one billing dispute, customer communication and credit memo processing included, from a time study or your own timing.

Watch out for

  • Staff time only. Credits and write-offs a dispute ends in go in the credits and write-offs per billing error, or they're counted as freed capacity and again as cash.
  • Include customer communication and credit memo processing, the same way before and after. Until it's measured, your own cost per dispute stands in and the value reads as estimated.

Good for

  • Pricing the finance staff time fewer billing disputes free, at your loaded hourly rate.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
finance.invoices.paid_count Number of invoices marked paid in the period. count invoices NetSuite, QuickBooks, Xero
finance.invoices.volume Total billed on invoices created in the period, tax included.

Watch out for

  • What the invoices billed, tax included: what customers owe, not the revenue earned. Billed isn't collected, either.
  • In your books' home currency: the connection converts each invoice to it, and an entry calculated from it counts in US dollars at the European Central Bank's average rate for its month.

Good for

  • With invoices sent, the average invoice value that days of dispute are priced at, at your cost of capital.

Guide: What a formula reads

currency USD NetSuite, QuickBooks, Xero
finance.pnl.expenses Total operating expenses for the period from the accounting system. currency USD NetSuite, QuickBooks, Xero
finance.pnl.net_income Net income for the period, the books' bottom line, other income included. currency USD NetSuite, QuickBooks, Xero
finance.pnl.revenue Revenue from sales recognized in the period, from the accounting system. Other income, such as interest and gains, isn't counted.

Watch out for

  • Recognized revenue from your books: it's only complete once the month is closed, so a month still open reads low.
  • It moves with everything the business does. Roiva's formulas use it as the base a rate's change is applied to, not as the AI's result. A formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • The base a change in stockouts, markdowns, shrink or overruns is applied to.

Guide: When the outcome has other causes

currency USD NetSuite, QuickBooks, Xero
finance.reconciliation.hours Staff hours on manual reconciliations per close cycle.

Watch out for

  • Hours per close cycle, not per month. The formula spreads a year's close cycles evenly over its months, twelve a year unless you set otherwise.
  • Compare like closes, a year-end close against year-end closes, or average over a full year.

Good for

  • Valuing reconciliation automation by the accounting time it frees each close.

Guide: What a formula reads

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
finance.revenue_recognition.adjustment_count Revenue recognition adjustments or corrections required per period.

Watch out for

  • Compared per invoice issued, so a heavier billing month isn't read as more errors. Record it over the same periods as the invoices.
  • Decide what counts as one adjustment, such as one correcting entry or one contract corrected, and keep that the same before and after.

Good for

  • With the minutes per adjustment, the correction work fewer recognition errors free.

Guide: When the count moves with how busy you are

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
finance.revenue_recognition.manual_hours Staff hours per period building revenue recognition schedules by hand.

Watch out for

  • Hours building recognition schedules by hand, not all revenue accounting. It's compared per invoice issued, so record it over the same periods as the invoices.
  • The hours freed are freed capacity, not cash: the accounting team is paid the same afterwards.

Good for

  • Valuing automated revenue recognition by the accounting time it frees.

Guide: When the count moves with how busy you are

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
finance.revenue_recognition.minutes_per_adjustment Accounting staff minutes to identify, correct, document and sign off one revenue recognition adjustment, from a time study or your own timing.

Watch out for

  • Hands-on minutes from identifying an adjustment to its sign-off, its documentation and review included.
  • Until it's measured, your own cost per adjustment stands in and the value reads as estimated.

Good for

  • Pricing the accounting time fewer recognition adjustments free, at your loaded hourly rate.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
finance.treasury.avg_cash_balance Average daily cash balance in the period.

Watch out for

  • An average of daily balances over the period, not the month-end balance.
  • It's the pool the gain in forecast accuracy is applied to, so include only cash treasury can move. Cash that's restricted or can't be redeployed overstates the value.

Good for

  • With forecast accuracy and your cost of capital, the value of holding less idle cash.

Guide: What a formula reads

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
finance.treasury.manual_hours_per_cycle Staff hours to build and update the cash flow model per forecast cycle.

Watch out for

  • Hours per forecast cycle, not per month: the formula multiplies it by the cycles run in the period, so count cycles the same way before and after.
  • The hours freed are freed capacity, not cash: the treasury team is paid the same afterwards.

Good for

  • Valuing AI cash forecasting by the treasury time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Logistics

Key What it measures Type Unit Fed by
logistics.brokerage.buy_rate_vs_market Carrier pay as a percentage of a market benchmark rate, such as DAT or Greenscreens, on loads paid to outside carriers and owner-operators; under 100 means paying below market.

Watch out for

  • Loads paid to outside carriers and owner-operators only. A load your own company drivers haul has no carrier pay and would read as free.
  • The same benchmark, lanes and equipment before and after, with fuel treated the same way on both sides.
  • Lane mix, lead time and carrier relationships move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing AI buy-rate targets and wider carrier reach without reading a falling spot market as a saving.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.brokerage.fraud_theft_losses Losses you bore from carrier fraud, double brokering and cargo theft in the period, after insurance and recoveries.

Watch out for

  • Book each incident in the month it happened, net of insurance and recoveries, and update it as the claim settles.
  • Incidents are rare, so set the baseline over a year or more of loads covered.
  • Theft and fraud waves by region, commodity and season move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing carrier vetting and load monitoring by the losses they stop.

Guide: When the outcome has other causes

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.brokerage.loads_covered_count Loads booked with a carrier, owner-operator or your own driver in the period.

Watch out for

  • Loads booked and picked up, not loads tendered: a load that fell off and was re-covered counts once.

Good for

  • The loads rep minutes saved are multiplied by.
  • The volume fraud and theft losses are compared per, so more freight isn't read as more fraud.

Guide: When the count moves with how busy you are

count loads Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.brokerage.market_cost_of_loads_covered The market benchmark rate summed over the period's loads paid to outside carriers and owner-operators, from the same benchmark the buy rate is compared with.

Watch out for

  • Use the benchmark's rate on each load's lane and equipment on the day it was booked, not a national average.
  • The same loads as buy rate vs market: leave out loads your own company drivers haul.

Good for

  • The carrier cost a change in buy rate against market is applied to.

Guide: What a formula reads

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.brokerage.quote_requests_count Spot-quote requests received from shippers by email, portal or phone in the period, answered or not.

Watch out for

  • Every request received, not only quotes sent: requests nobody answered before are what an agent adds.
  • Count a request for several loads on one lane once, the same way before and after.

Good for

  • The volume spot net revenue is compared per, so both win rate and margin per load count.

Guide: When the count moves with how busy you are

count requests Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.brokerage.rep_minutes_per_load Carrier sales and capacity rep minutes per load covered, averaged over every load, from finding a carrier to the last check call.

Watch out for

  • Minutes worked, not the hours a load sat open, including reviewing what the AI offered, agreed or logged.
  • A load a carrier books on its own in the app counts only the minutes spent checking it.
  • Booking and check calls are timed here, so an operations agents initiative on the same loads should leave them out of its minutes per shipment.

Good for

  • Valuing load matching and carrier voice agents by the rep time they free on each load.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.brokerage.spot_net_revenue Shipper revenue less carrier pay on spot loads won from quote requests in the period.

Watch out for

  • Spot loads won from quote requests only, credited to the month the request came in, with contract freight left out on both sides.
  • It's compared per quote request received, so a busy month isn't read as better pricing. Record requests over the same months.
  • Spot market rates, capacity and season move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing AI quoting and pricing by the net revenue each quote request brings, win rate and margin together.

Guide: When the outcome has other causes

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.customer_service.minutes_per_wismo_contact Minutes of agent work one "Where Is My Order" contact takes, from a time study or your help desk's time tracking on those contacts alone. Not every ticket's average.

Watch out for

  • Minutes on where-is-my-order contacts alone, not every ticket's average, which mixes in harder contacts.
  • Until it's measured, your own cost per contact stands in and the value reads as estimated.

Good for

  • Pricing the agent time fewer where-is-my-order contacts free, at your loaded hourly rate.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.customer_service.wismo_contacts_per_month Where-is-my-order inbound contacts per month.

Watch out for

  • Use the same tag or intent rule before and after: a broader rule reads as a rise in contacts.
  • It's compared per shipment, so peak season isn't read as a loss. Record shipments over the same months.

Good for

  • Valuing real-time ETAs and delivery notifications by the contacts customers no longer make.

Guide: When the count moves with how busy you are

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.deliveries.on_time_rate Percentage of deliveries completed within the promised window.

Watch out for

  • Measure against the window the customer was given. A window widened after dispatch was still missed.
  • Weather, traffic, order volume and staffing move it too, so a formula valuing its change takes only the share of it you credit to the initiative.
  • It counts deliveries against a window. On-Time Order Delivery Rate counts orders against a promised date, so the two don't compare.

Good for

  • Valuing route optimization by the penalties, chargebacks and lost orders late deliveries no longer cost, at your contribution margin.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.drivers.count Last-mile drivers in the measured operation. count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.drivers.operating_hours Driver operating hours in the period.

Watch out for

  • Hours driving and delivering, not depot time, which no route optimization changes.
  • Use the same hours as stops per driver hour is measured over, so the two figures agree.

Good for

  • The driver hours a change in stops per hour is multiplied by.

Guide: What a formula reads

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.drivers.stops_per_hour Completed delivery stops per driver per labor hour.

Watch out for

  • Completed stops, not attempted: a failed delivery costs the hour and delivers nothing.
  • The formula books added stops as revenue per stop, which assumes the time freed carries more paid stops. If volume didn't grow, that revenue didn't happen: credit less of the change, or don't use the formula.
  • Route density, order volume, weather and traffic move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing dispatch and stop sequencing by the extra stops the same drivers complete.

Guide: When the outcome has other causes

count stops Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.fleet.annual_insurance_premium Annual commercial auto and cargo insurance premium across the fleet.

Watch out for

  • The premium only, from each policy's declarations page, not the total cost of risk: deductibles and claims are a different figure.
  • Nothing books until a renewal lowers the premium per vehicle. Record it again at each renewal.
  • Market rates, miles driven, limits, deductibles and a change of carrier move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing a driver safety program by the premium it saves at renewal, per vehicle insured.

Guide: When the outcome has other causes

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.fleet.maintenance_hours_per_month Technician hours per month on scheduled and reactive fleet maintenance.

Watch out for

  • Scheduled and reactive work together. Where maintenance is outsourced, use the hours billed, not the invoice value.
  • It's compared per vehicle in service, so a larger fleet isn't read as less efficient maintenance. Record vehicles over the same months.

Good for

  • Valuing predictive maintenance by the technician time it frees.

Guide: When the count moves with how busy you are

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.fleet.miles Total fleet miles in the period.

Watch out for

  • Route savings compare miles per stop completed, so record stops over the same days as the miles.
  • Use the same miles for preventable accidents per million miles, so the rate and the miles it's applied to agree.

Good for

  • With stops completed, the miles per stop route optimization brings down.
  • The miles a change in the preventable accident rate is applied to.

Guide: When the count moves with how busy you are

count miles Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.fleet.roadside_events_per_year Unplanned roadside breakdown events per year.

Watch out for

  • Count breakdowns that stopped a vehicle on the road, not every defect an inspection found.
  • A year's count, compared per vehicle in service. The formula spreads it evenly over the year's months.

Good for

  • Valuing predictive maintenance by the towing, recovery and missed loads breakdowns no longer cost.

Guide: When the count moves with how busy you are

count events Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.fleet.stop_count Delivery and pickup stops the fleet completed in the period.

Watch out for

  • Stops made, not orders: two orders at one door are one stop.
  • Record it over the same days as the fleet miles, since route savings compare miles per stop.

Good for

  • The volume fleet miles are compared per, so a peak-season month isn't read as worse routing.

Guide: When the count moves with how busy you are

count stops Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.fleet.vehicle_count Vehicles in service across the fleet.

Watch out for

  • Vehicles that ran in the period, not units parked or awaiting sale.
  • For the insurance premium, use the vehicles on the policy's schedule, recorded with the premium at each renewal.

Good for

  • The base maintenance hours, breakdowns and premium are compared per, so a growing fleet isn't read as worse.

Guide: When the count moves with how busy you are

count vehicles Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.operations.shipments_handled_count Shipments or loads your freight operations team handled in the period, brokered, own-fleet or forwarded, by AI agents or by staff.

Watch out for

  • Every shipment the team handled, whoever did the work, from the TMS, not a warehouse's outbound count.

Good for

  • The shipments operations staff minutes saved are multiplied by.

Guide: What a formula reads

count shipments Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.operations.staff_minutes_per_shipment Staff minutes of hands-on freight operations work per shipment on the tasks in scope, averaged over every shipment.

Watch out for

  • Minutes worked, not time a shipment waits, including reviewing and fixing what the AI agents did.
  • Keep the same tasks in scope before and after; adding a task changes what the average means.
  • Carrier booking and check calls belong to carrier rep minutes per load; if a load matching initiative times them there, leave them out here.

Good for

  • Valuing AI agents in freight operations by the staff time they free on each shipment.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
logistics.safety.preventable_accidents_per_million_miles Preventable accident frequency per million miles driven.

Watch out for

  • Preventable is your safety review's judgment. Keep whoever makes that call, and how, the same before and after.
  • On a small fleet, one accident moves the rate a lot. Read it over several months.
  • Weather, traffic, driver mix and other safety programs move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing driver coaching by the accident costs the fleet itself no longer pays, with what insurers pay left out.

Guide: When the outcome has other causes

count accidents Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Professional services

Key What it measures Type Unit Fed by
professional_services.billing.realization_rate Billed fees as a percentage of standard-rate value of hours worked.

Watch out for

  • Fees billed over the standard value of hours worked, from your time and billing system.
  • Client pushback, pricing, write-off policy and matter mix move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing AI time capture and billing review by the fees kept on hours already worked, booked in full.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.billing.unbilled_hours_per_practitioner_weekly Extra billable hours per practitioner per week that AI time capture records and that would otherwise have gone unrecorded.

Watch out for

  • Only the extra hours captured since the AI, not every unbilled hour: the formula books each one as revenue.
  • Per practitioner per week: hours over billable practitioners and over the weeks in the period.

Good for

  • Valuing AI time suggestions by the billable hours they recover.

Guide: Revenue is booked at your margin

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.knowledge.active_users Practitioners actively using the knowledge system in the period.

Watch out for

  • Practitioners who used the system in the period, from its usage report. Licenses bought is a different number.

Good for

  • The users search minutes saved are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.knowledge.search_minutes Average minutes per search for prior work and precedent.

Watch out for

  • From starting to look to having something usable, or giving up. A search that found nothing, or ended in asking a colleague, still counts.

Good for

  • Pricing the practitioner time a knowledge system frees, at your loaded hourly rate.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.knowledge.searches_per_practitioner Times per period a practitioner searches for prior work, methods or precedent.

Watch out for

  • Where the system logs searches, use that per active user. Otherwise count a normal week for a few practitioners and scale it to the period.

Good for

  • With active users, the searches minutes saved per search are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.practitioners.available_hours Working hours available per practitioner in the period.

Watch out for

  • Contracted hours less holidays, before any is sold. It's the base of both utilization figures, so set it once and reuse it.
  • Billable utilization and bench time formulas both read it, and value the same hours, so an initiative gets one of the two.

Good for

  • The hours a change in billable utilization or bench time is applied to.

Guide: Formulas that count the same money

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.practitioners.billable_count Billable practitioners in the period.

Watch out for

  • People expected to bill time, from the roster, with a half-time practitioner counted as a half.

Good for

  • The practitioners a change in utilization, bench time or captured hours is multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.practitioners.billable_hours_worked Hours worked by billable practitioners in the period.

Watch out for

  • Billable hours recorded in your time and billing system, the same hours realization is measured on.

Good for

  • The hours a rise in realization is applied to.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.projects.at_risk_detection_lead_weeks Average weeks before delivery failure that an at-risk project is identified. duration weeks Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.projects.at_risk_identified_count At-risk projects identified in the period.

Watch out for

  • Each project once, at its first flag, however many reviews it stays on.

Good for

  • The at-risk projects recovery hours saved per project are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.projects.avg_overrun_depth Of projects closed in the period that exceeded the approved budget, their cost over budget as a percentage of their contract value.

Watch out for

  • Cost over budget as a percentage of contract value, for projects closed over budget. Before the AI, take the last twelve months of closed projects.
  • Formulas read only the depth before the AI, so overruns made smaller aren't counted and the value errs low.

Good for

  • Pricing each project that no longer overruns, for the overrun formula.

Guide: What a formula reads

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.projects.budget_overrun_rate Percentage of projects that exceeded the approved budget.

Watch out for

  • Projects, not dollars, against the budget approved at kickoff: a re-baselined project overran.
  • Scope changes, clients, staffing and pricing move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing delivery-risk tracking by the overrun costs it avoids.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.projects.pm_count_using_automation Project managers using the automated reporting system. count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.projects.recovery_hours_per_project Senior PM hours spent recovering a late-detected at-risk project.

Watch out for

  • Senior hours pulling one late-caught project back. No system records it, so ask the people who did the last few recoveries.
  • The hours freed are freed capacity, not cash: senior staff are paid the same afterwards.

Good for

  • Valuing early risk detection by the escalation time it saves.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.proposals.prep_hours Staff hours to assemble one proposal.

Watch out for

  • Every contributor's hours on one proposal of the kind the initiative changes, the same way before and after.
  • The hours freed are freed capacity, not cash: the team is paid the same afterwards.

Good for

  • Valuing AI knowledge retrieval by the proposal time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.proposals.submitted_count Proposals or statement-of-work responses submitted in the period.

Watch out for

  • Submissions, not opportunities: two bids for the same client are two.
  • From a CRM, it counts deals first entering the stage chosen as proposal in the pipelines ticked as services, once each, so two bids are two only when each is its own deal.

Good for

  • The proposals prep hours saved per proposal are multiplied by.

Guide: What a formula reads

count — HubSpot, Pipedrive, Salesforce
professional_services.reporting.client_reports_delivered Client status reports delivered in the period.

Watch out for

  • Reports that went to the client, one per client per cycle, not drafts.
  • PM capacity expansion on the same template values the same hours at a billing rate, so an initiative gets one of the two.

Good for

  • The reports hours saved per report are multiplied by.

Guide: Formulas that count the same money

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.reporting.hours_per_client_report Staff hours to compile, format and deliver one client report.

Watch out for

  • Everyone who touches one report, not only the person who sends it, for reports of the same kind.
  • PM capacity expansion on the same template values the same hours at a billing rate, so an initiative gets one of the two.

Good for

  • Valuing automated report assembly by the staff time it frees.

Guide: Formulas that count the same money

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.reporting.reports_per_pm Client reports each project manager prepares in the period. count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.utilization.bench_share Percentage of available practitioner hours spent on the bench.

Watch out for

  • Available hours on neither client work nor agreed internal work. Without a bench code, take available hours less billable and internal hours.
  • A bench hour filled with billed work counts as revenue in billable utilization uplift, so an initiative gets that formula or bench time freed capacity, not both.
  • Client demand, the pipeline and hiring move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing AI capacity planning by the paid hours no longer idle.

Guide: Formulas that count the same money

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
professional_services.utilization.billable_rate Billable hours as a percentage of available practitioner hours.

Watch out for

  • Hours booked to client work, not hours invoiced: write-offs are realization, a different figure.
  • Bench time freed capacity on the same template values the same hours, so an initiative gets one of the two.
  • Client demand, the pipeline, staffing and holidays move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing AI staffing by the extra billed hours, at your contribution margin.

Guide: Formulas that count the same money

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Construction

Key What it measures Type Unit Fed by
construction.bids.awarded_value Contract value of projects awarded in the period.

Watch out for

  • Contract value at signing, before change orders, so the before and after measure the same thing.

Good for

  • The awarded work a fall in estimate variance is applied to, at your project margin.

Guide: What a formula reads

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.bids.submitted_count Bids submitted in the period.

Watch out for

  • Submissions, not jobs: two bids for the same job after a re-scope are two.

Good for

  • The bids estimating hours saved per bid are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.documents.search_hours_per_week Staff hours per week locating current drawing versions and document history.

Watch out for

  • Finding the current drawing or a document's history, not reviewing it, across field and office staff.
  • The hours freed are freed capacity, not cash: the team is paid the same afterwards.

Good for

  • Valuing a common data environment by the search time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.estimating.hours_per_bid Estimator hours per bid.

Watch out for

  • Compare bids of comparable size: a small tender and a major works bid aren't one figure.
  • The hours freed are freed capacity, not cash: estimators are paid the same afterwards.

Good for

  • Valuing AI-assisted takeoffs by the estimating time they free.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.estimating.variance_rate Average estimate-to-actual variance as a percentage of the estimate.

Watch out for

  • Use the absolute difference between estimate and final cost, so overs and unders don't cancel to a flattering zero.

Good for

  • Valuing bid intelligence by the margin more accurate estimates protect on awarded work.

Guide: What a formula reads

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.job_costing.review_cycle_days Average days between formal job cost reviews on active projects. duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.job_costing.review_cycles Formal job cost review cycles in the period.

Watch out for

  • Review cycles run in the period, such as one for a monthly review, not one per project: the formula already multiplies by active projects.

Good for

  • With active projects, the reviews hours saved per review are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.job_costing.review_hours_per_project PM and controller hours per project per job cost review cycle.

Watch out for

  • PM and controller hours for one project's review, from the people who run it.
  • The hours freed are freed capacity, not cash: the team is paid the same afterwards.

Good for

  • Valuing real-time cost analytics by the review time they free.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.labor.hours_worked Field labor hours worked in the period.

Watch out for

  • Field hours from certified payroll or timecards. Use the same hours for the recordable incident rate, which is per 200,000 of them.

Good for

  • The hours a fall in the recordable incident rate is applied to.
  • The volume safety admin hours are compared per, so a busy month on site isn't read as slower safety work.

Guide: When the count moves with how busy you are

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.payroll.minutes_per_timecard_correction Payroll admin minutes to find and correct one timecard error, from a time study or your own timing.

Watch out for

  • Hands-on minutes from finding one timecard error to its correction, confirming it with the employee or supervisor included.
  • Staff time only: payroll reprocessing fees a correction triggers are cash and aren't counted. Until it's measured, your own cost per correction stands in.

Good for

  • Pricing the payroll time fewer timecard errors free, at your loaded hourly rate.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.payroll.processing_hours_per_week Staff hours per week collecting, reconciling and processing timecards.

Watch out for

  • Collecting, chasing, reconciling and processing timecards. Include the chasing: on most sites it's most of the work.
  • It's compared per timecard, so record timecards over the same months.

Good for

  • Valuing digital time capture by the payroll time it frees.

Guide: When the count moves with how busy you are

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.payroll.timecard_error_rate Percentage of timecards requiring correction.

Watch out for

  • Timecards needing a correction before payroll runs, such as missing hours or a wrong cost code or job, from your payroll exception log.

Good for

  • With timecards processed, the corrections digital time capture avoids.

Guide: Cash value and freed capacity

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.payroll.timecards_processed Timecards processed in the period.

Watch out for

  • Cards, not workers or weeks, with a corrected card counted once.

Good for

  • The timecards a fall in the error rate is applied to.
  • The volume payroll hours are compared per, so a busy month isn't read as slower processing.

Guide: When the count moves with how busy you are

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.projects.active_count Active projects in the portfolio during the period.

Watch out for

  • Projects under way, counted on the same status each month.

Good for

  • The projects review hours saved per project are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.projects.avg_overrun_depth Of projects closed in the period that exceeded the approved budget, their cost over budget as a percentage of their contract value.

Watch out for

  • Cost over budget as a percentage of contract value, for projects closed over budget. Before the AI, take twelve months of closed projects so a few jobs don't decide it.
  • Formulas read only the depth before the AI, so overruns made smaller aren't counted and the value errs low.

Good for

  • Pricing each project that no longer overruns, for the overrun formula.

Guide: What a formula reads

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.projects.cost_overrun_rate Percentage of projects that exceed the approved budget.

Watch out for

  • The share of projects closed over the approved budget, by count. How far over is Average Cost Overrun Depth, a separate figure.
  • Material and labor prices, subcontractors, weather and scope changes move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing cost analytics and variance alerts by the overruns avoided.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.rfis.response_days Average calendar days from RFI submission to formal response.

Watch out for

  • Calendar days from submission to a formal answer: a week of waiting costs the same whichever days it falls on.
  • The formula prices each day saved at a daily delay cost you estimate, booked as freed capacity, not cash.

Good for

  • Valuing AI-assisted RFI routing by the delay days it saves.

Guide: Cash value and freed capacity

duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.rfis.submitted_count Requests for information submitted in the period.

Watch out for

  • RFIs raised, not answered, so the count reflects the questions the job produced.

Good for

  • The RFIs response days saved are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.safety.observation_hours_per_week Staff hours per week on safety observation reporting and job hazard analysis distribution.

Watch out for

  • The walking, writing up and getting hazard analyses to crews, in a normal week.
  • It's compared per field labor hour, so record field hours over the same months.

Good for

  • Valuing digital safety forms by the admin time they free.

Guide: When the count moves with how busy you are

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
construction.safety.recordable_incident_rate OSHA recordable incident rate (incidents × 200,000 ÷ hours worked).

Watch out for

  • Recordable incidents × 200,000 ÷ hours worked, on the same hours you record. A year of hours is the usual base; say so if you use less.
  • Site mix, crews, weather and other safety programs move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing AI safety observation by the incident costs it avoids.

Guide: When the outcome has other causes

ratio — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Healthcare

Key What it measures Type Unit Fed by
healthcare.appointments.avg_length Scheduled minutes of a patient visit of the type recovered clinician time would be booked as. duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.appointments.no_show_rate Percentage of scheduled appointments the patient did not attend.

Watch out for

  • Cancellations with notice aren't no-shows. A share of the same appointments you record as scheduled.
  • Weather, flu season, patient mix, transport and scheduling policy move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing automated reminders by the visits patients now keep, at your contribution margin.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.appointments.scheduled_count Appointments scheduled in the period.

Watch out for

  • Scheduled, not attended, from the practice management system.

Good for

  • The appointments a fall in no-shows is applied to.
  • The volume outreach hours are compared per, so a busy month isn't read as slower outreach.

Guide: When the count moves with how busy you are

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.audits.prep_hours_per_cycle Staff hours preparing documentation for one quality audit or regulatory review.

Watch out for

  • Hours for one audit or review, every contributing department included.
  • Per cycle: the formula spreads a year's audit cycles evenly over its months.

Good for

  • Valuing always-ready documentation by the audit preparation time it frees.

Guide: What a formula reads

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.claims.denial_rate Percentage of submitted claims denied on first submission. percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.claims.denied_count Claims denied on first submission in the period.

Watch out for

  • Denials received in the month, whatever month the claim was submitted.
  • The formula applies the change in appeal success to every denial, which assumes each one is appealed. Where many aren't, count only the denials you appeal.

Good for

  • The denials a rise in appeal success is applied to.

Guide: Revenue is booked at your margin

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.coding.charts_coded Charts coded in the period.

Watch out for

  • Charts, not codes or claims: one chart with six codes is one.

Good for

  • The charts a fall in the coding error rate is applied to.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.coding.charts_per_coder_per_day Charts coded per medical coder per day.

Watch out for

  • Charts over coder working days, so leave and holidays don't read as a fall. Use the same coders as Medical Coders.
  • The formula books each extra chart as revenue, which holds only while charts are waiting to be coded. Without a backlog they'd have been coded anyway, and it overstates.

Good for

  • Valuing AI-assisted coding by the extra charts coded, at your contribution margin.

Guide: Revenue is booked at your margin

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.coding.coder_count Medical coders using AI-assisted coding.

Watch out for

  • Coders using AI-assisted coding, from the tool's usage report rather than the department roster.

Good for

  • The coders a rise in charts per coder per day is multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.coding.error_rate Percentage of coded charts with an error.

Watch out for

  • Say which source, an internal audit sample or payer denials: they measure different failures. Use the same one before and after.

Good for

  • Valuing AI-assisted coding by the revenue coding errors no longer lose.

Guide: What a formula reads

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.denials.appeal_success_rate Share of appealed denials overturned.

Watch out for

  • A ratio between 0 and 1, not a percentage: 0.62, not 62.
  • Of the appeals resolved in the period, the share overturned. Appeals still open aren't in either count.

Good for

  • Valuing denial management by the claims recovered on appeal, booked in full since the care was already delivered.

Guide: Revenue is booked at your margin

ratio 0–1 Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.documentation.after_hours_charting_hours_per_week Hours per week clinicians spend on documentation outside clinical hours.

Watch out for

  • Define outside clinical hours once, such as after the last scheduled patient, and measure both sides the same way.
  • Added visit revenue on the same template values the same time as extra visits, so an initiative gets one of the two.
  • It's freed capacity. Locum or overtime spend that falls because of it is cash: record it as capacity realized as cash, with the document that shows it.

Good for

  • Valuing ambient documentation by the clinician time it gives back.

Guide: Formulas that count the same money

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.documentation.clinician_count Clinicians using ambient documentation.

Watch out for

  • Clinicians actually using ambient documentation, from the tool's usage report. Licenses assigned is a different number.

Good for

  • The clinicians hours saved, or extra encounters, are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.documentation.encounters_per_clinician_per_day Patient encounters per clinician per clinical day.

Watch out for

  • Clinical days only: an admin day has no encounters and drags the average down on whichever side has more of them.
  • Clinician documentation freed capacity on the same template values the same time at clinicians' cost, so an initiative gets one of the two.

Good for

  • Valuing ambient documentation by the extra visits clinicians see, at your contribution margin.

Guide: Formulas that count the same money

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.documentation.minutes_per_encounter Clinician documentation minutes per patient encounter. duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.outreach.hours_per_week Staff hours per week on appointment reminders, care gap calls and other patient outreach done by hand.

Watch out for

  • Reminders, care gap calls and other outreach done by hand, in a normal week.
  • It's compared per appointment scheduled, so record appointments over the same months.

Good for

  • Valuing automated outreach by the staff time it frees.

Guide: When the count moves with how busy you are

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.quality.care_gap_closure_rate Percentage of identified care gaps closed.

Watch out for

  • Take both halves from the same registry, so a change in what it identifies doesn't read as a change in what was closed.
  • Patients, providers, payer programs and how gaps are measured move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing quality monitoring by the value-based payments a higher closure rate earns.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.quality.value_based_revenue_at_risk Revenue at risk or bonus-eligible under value-based contracts, for a performance year.

Watch out for

  • The amount the contracts put on performance for a performance year, not total contract revenue.
  • It's for a year, so the formula books a month's share of it at a time.

Good for

  • The revenue a rise in care gap closure is applied to.

Guide: What a formula reads

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
healthcare.revenue_cycle.days_in_ar Accounts receivable over average daily charges, in days.

Watch out for

  • Take it the same way before and after, as your revenue cycle system reports it, or month-end AR over the period's daily charges.
  • Cash collected sooner isn't more cash: the formula values the days at your cost of capital, not as revenue.
  • Payer mix, payer policy and billing staff levels move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing revenue cycle automation by what collecting sooner is worth.

Guide: Revenue that arrives sooner

duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

HR

Key What it measures Type Unit Fed by
hr.hiring.days_to_first_interview Average days from job posting to first interview.

Watch out for

  • Calendar days from the job going live to the first interview with a hiring manager, not a screening call.
  • The formula takes each day saved before the first interview as a day the role is filled sooner, and books it as freed capacity, not cash.

Good for

  • Valuing AI screening by the vacancy days it saves on each hire.

Guide: Cash value and freed capacity

duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.hiring.hires_count Positions filled in the period.

Watch out for

  • Count the offer accepted, not the start date, so a long notice period doesn't move a hire into the wrong month.
  • It isn't New Hires Onboarded: someone hired in one month often onboards in the next.

Good for

  • The hires recruiter hours and vacancy days saved per hire are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.hiring.recruiter_hours_per_hire Recruiter hours on resume review and screening per hire.

Watch out for

  • Compare roles of the same kind: a volume role and an executive search aren't the same figure.
  • The hours freed are freed capacity, not cash: recruiters are paid the same afterwards.

Good for

  • Valuing AI screening by the recruiter time it frees on each hire.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.inquiries.avg_response_hours Average hours from employee inquiry to HR response for routine questions. duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.inquiries.chatbot_deflection_rate Share of employee inquiries fully resolved by the chatbot without HR involvement.

Watch out for

  • A decimal between 0 and 1, not a percentage: 0.6, not 60.
  • Before the chatbot it's 0, so the formula counts the whole rate. Count only questions closed without a person; one the employee then took to HR wasn't deflected.

Good for

  • Valuing an HR chatbot by the HR time spent on the questions it answers.

Guide: Cash value and freed capacity

ratio 0–1 Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.inquiries.minutes_per_inquiry HR staff minutes to handle one tier-1 inquiry by hand, answer, follow-up and documentation included, from a time study or your case tool's time tracking.

Watch out for

  • Hands-on minutes to answer, follow up and document one routine question by hand.
  • Until it's measured, your own cost per inquiry stands in and the value reads as estimated.

Good for

  • Pricing the HR time a chatbot frees, at your loaded hourly rate.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.inquiries.tier1_per_month Routine HR inquiries received per month (policy, PTO, benefits, payroll).

Watch out for

  • Count questions, not replies, and leave out anything that needed a case opened.
  • Count every routine question that arrived, chatbot included: it's the total the deflection rate is a share of.

Good for

  • The questions a chatbot's deflection rate is applied to.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.onboarding.admin_hours_per_hire HR and IT staff hours on onboarding tasks per new hire.

Watch out for

  • Every team's hours, HR and IT, not the new starter's own time.
  • The hours freed are freed capacity, not cash: the same people are paid afterwards.

Good for

  • Valuing onboarding automation by the HR and IT time it frees on each hire.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.onboarding.completion_rate Percentage of required onboarding tasks completed within the target window. percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.onboarding.days_to_productivity Days from start date until a new hire is independently productive.

Watch out for

  • Nothing reports it. Agree what productive means once, and ask managers the same way before and after.
  • Role mix, managers, hiring quality and training programs move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing faster onboarding by the output new hires add sooner, booked as freed capacity.

Guide: When the outcome has other causes

duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.onboarding.hires_count New employees who completed onboarding in the period.

Watch out for

  • New starters who finished onboarding in the period. It isn't Hires Completed, which counts offers accepted.

Good for

  • The new hires onboarding hours and ramp-up days saved are multiplied by.

Guide: What a formula reads

count hires Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.onboarding.tasks_automated Onboarding checklist items (provisioning, welcome communications, acknowledgments) now handled automatically. count tasks Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.reporting.hours_per_month Staff hours per month building workforce reports.

Watch out for

  • Count the assembly, pulling data and building the pack, not the reading, and include everyone who touches it.
  • The hours freed are freed capacity, not cash: the same people are paid afterwards.

Good for

  • Valuing workforce analytics by the reporting time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.reviews.admin_hours_per_cycle HR and manager hours coordinating, collecting and calibrating one performance review cycle. duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.reviews.admin_hours_per_employee HR and manager admin hours per employee per review cycle.

Watch out for

  • Hours per person per review cycle: the cycle's coordination hours divided by the people reviewed, the same way before and after.
  • The formula spreads a year's review cycles evenly over its months, so value books every month, not only in review season.

Good for

  • Valuing review-cycle automation by the HR and manager time it frees.

Guide: What a formula reads

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.reviews.completion_rate Percentage of required performance reviews completed on time. percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.reviews.employees_reviewed Employees included in the performance review cycle.

Watch out for

  • People the cycle covered, from its own roster, not those who completed it.

Good for

  • The employees review admin hours saved per person are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.workforce.headcount Total employee headcount in the period.

Watch out for

  • Take it at the same point in each month, and count contractors only if the initiative touches their work.

Good for

  • The base a change in voluntary attrition is applied to.

Guide: When the outcome has other causes

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hr.workforce.voluntary_attrition_rate Annualized voluntary attrition rate.

Watch out for

  • Voluntary leavers only, annualized. Redundancies and dismissals hide what a retention program moved.
  • The job market, pay, managers and the economy move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing retention analytics by the replacement cost of the people who stay.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Insurance

Key What it measures Type Unit Fed by
insurance.claims.adjuster_hours_per_claim Average adjuster hours of hands-on work per claim closed, a straight-through claim counting as none — not the days from first notice of loss to close.

Watch out for

  • Hands-on adjuster hours averaged over every claim closed, a straight-through claim counting as none. Not the days from first notice of loss to close.
  • Adjuster time is part of loss adjustment expense, so an initiative gets either this or LAE reduction, never both.
  • Measure it the same way before and after, over the same mix of claims.

Good for

  • Valuing AI triage and straight-through adjudication by the adjuster time it frees, at your loaded hourly rate.

Guide: Formulas that count the same money

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.claims.auto_adjudication_rate Share of claims processed straight-through without adjuster review.

Good for

  • Taking straight-through claims out of the volume adjuster file-work minutes are multiplied by, since they have no file work.
ratio 0–1 Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.claims.avg_cycle_days Average days from first notice of loss to claim closure. duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.claims.file_work_minutes_per_claim Adjuster minutes per claim spent reading claim files and records and writing correspondence, averaged over the claims an adjuster worked, not those closed straight-through.

Watch out for

  • Reading and writing only, including reviewing what the AI summarized or drafted. Not investigation, negotiation or the days a claim stays open.
  • It is part of adjuster hours per claim, so if another initiative values those hours on the same claims, time only this work here.
  • Compare claims of similar complexity: a bodily injury file with medical records takes far longer than a windshield claim.

Good for

  • Valuing claim file summaries and drafted letters by the adjuster time they free.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.claims.incurred_losses Incurred losses on the affected book in the period.

Watch out for

  • It includes reserves, so it moves as reserves on open claims develop, and with catastrophe events.
  • It's the base a change in a ratio is applied to, not the AI's result. Use the losses on the book the AI works on, not the whole company's.

Good for

  • The base a change in the LAE ratio or in claims leakage is applied to.

Guide: When the outcome has other causes

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.claims.lae_ratio Loss adjustment expense as a percentage of incurred losses.

Watch out for

  • Here it's loss adjustment expense as a percentage of incurred losses, not of premium. Check the base of any figure you bring in.
  • Claim mix and severity, catastrophes, litigation and staffing move it too, so a formula valuing its change takes only the share of it you credit to the initiative.
  • Adjuster time is inside it, so an initiative gets LAE reduction or adjudicator freed capacity, never both.

Good for

  • Valuing AI claims triage by the loss adjustment expense saved on your incurred losses.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.claims.leakage_rate Paid loss attributable to fraud or excess adjustments as a percentage of claims paid.

Watch out for

  • It's a percentage of claims paid, but the formula applies its change to incurred losses, which include reserves. Measure it the same way before and after.
  • Claim mix, catastrophe surges, SIU staffing and fraud trends move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing fraud detection: the formula applies the fall in leakage to the incurred losses on the book exposed to fraud, an estimate of the losses kept from going out.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.claims.processed_count Claims submitted and processed in the period.

Watch out for

  • Count claims submitted and worked, not transactions on them, and keep that the same before and after.
  • Adjuster hours are averaged over claims closed, so in a month when far more claims arrive than close, the adjuster time freed reads high.

Good for

  • The claims adjuster hours saved per claim are multiplied by.
  • The volume false-positive SIU referrals are compared per, so a catastrophe month's surge in claims isn't read as worse triage.

Guide: When the count moves with how busy you are

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.fraud.false_positive_siu_referrals Special investigations unit referrals per month that do not substantiate fraud.

Watch out for

  • Compared per claim processed, so a catastrophe month's surge in claims isn't read as worse triage. Record both over the same months.
  • Fewer false positives can also mean fewer referrals of any kind, real fraud included. Read it beside the fraud the unit substantiates.
  • A referral is known to be a false positive only once its case closes. Count by the month referred or the month closed, the same way before and after.

Good for

  • Valuing AI fraud triage by the investigator time no longer spent on referrals that find nothing.

Guide: When the count moves with how busy you are

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.fraud.hours_per_siu_investigation Investigator hours one SIU investigation takes, from your SIU case system's time tracking or your own timing. Not the days the case stays open.

Watch out for

  • Investigator hours on one investigation, not the days the case stays open.
  • Staff time only. Vendor costs an investigation runs up are cash, and aren't counted here.
  • Until it's measured, your own cost per investigation stands in and the value reads as estimated.

Good for

  • Pricing the investigator time fewer false-positive referrals free, at your loaded hourly rate.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.policy_servicing.calls_per_month Inbound policy-servicing call volume per month.

Watch out for

  • Compared per servicing request, so a renewal-season surge isn't read as a loss. Record both over the same months.
  • Two formulas value the same calls no longer made: call deflection at their full cost, and agent redeployment at agent time. An initiative gets one of the two.

Good for

  • Valuing self-service by the calls it keeps from reaching an agent.

Guide: When the count moves with how busy you are

count calls Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.policy_servicing.requests_per_month Policy-servicing requests per month across every channel (phone, portal, app and chat), as the policy administration system counts them.

Watch out for

  • Every channel, as your policy administration system counts them. A request that moves from the phone to the portal still counts, which is what lets calls per request fall.
  • Leave out a channel, such as chat, and calls per request reads high.

Good for

  • The volume calls are compared per, so a renewal-season surge isn't read as worse self-service.

Guide: When the count moves with how busy you are

count requests Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.pricing.on_level_loss_ratio Incurred losses as a percentage of earned premium restated at today's rate level, so an overall rate change doesn't move it.

Watch out for

  • Restate premium at today's rate level on both sides, with the rate history your actuaries use for indications. A plain calendar-year loss ratio counts a rate increase as the model's.
  • Weather and catastrophe losses, reserve development and the book's mix still move it, so a formula valuing its change takes only the share of it you credit to the initiative.
  • Losses develop for years on long-tail lines. Use losses developed to the same age on both sides.

Good for

  • Valuing a new rating plan by the losses not paid on the same premium, without counting the rate level.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.submissions.intake_minutes_per_submission Staff minutes of hands-on work to read, clear and set up one submission, up to the point an underwriter picks up the file.

Watch out for

  • Minutes worked, not time the submission waits in the inbox, including checking what the AI extracted.
  • Stop the clock when an underwriter picks up the file. Underwriter review is measured separately, as underwriter hours per quote.
  • The minutes freed are freed capacity, not cash: the staff are paid the same afterwards.

Good for

  • Valuing AI submission intake by the staff time it frees on each submission.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.submissions.received_count Broker submissions received in the period, declined and duplicate ones included.

Watch out for

  • Every submission that reached intake, not only those an underwriter reviewed or quoted: intake works the declines too.
  • Count a resubmission of the same risk once, the same way before and after.

Good for

  • The submissions intake minutes saved are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.underwriting.earned_premium Earned premium on the targeted book in the period.

Watch out for

  • Earned, not written: premium written in a period is earned over the policy term.
  • Use the premium on the book the AI selects risks for, the same book the loss ratio is measured on, not the whole company's.

Good for

  • The base a change in the loss ratio is applied to.

Guide: When the outcome has other causes

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.underwriting.loss_ratio Incurred losses as a percentage of earned premium.

Watch out for

  • Weather and catastrophe losses, rate changes, reserve development and the book's mix move it too, so a formula valuing its change takes only the share of it you credit to the initiative.
  • Incurred losses include reserves, so a recent period's ratio can still change as claims develop. Measure it on the same book, the same way, before and after.

Good for

  • Valuing AI risk selection by the losses not paid on the same earned premium.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.underwriting.quote_hours Average underwriter hours worked on one submission, from application to bindable quote.

Watch out for

  • Hours worked on the file, not the elapsed time from application to quote, which includes waiting on brokers and data. The formula prices them as underwriter time.
  • Standard risks speed up more than complex ones, so a shift in the mix of submissions moves it.

Good for

  • Valuing AI pre-screening by the underwriter time it frees, at your loaded hourly rate.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
insurance.underwriting.submission_count New business submissions reviewed in the period.

Watch out for

  • New business submissions reviewed, not quotes issued or policies bound. Keep the count the same before and after.

Good for

  • The submissions underwriter hours saved per submission are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Salesforce

Key What it measures Type Unit Fed by
salesforce.cases.avg_resolution_days Salesforce's average case resolution time for the month. Written to support.tickets.avg_resolution_days instead when this connection is the primary source. duration days Salesforce
salesforce.cases.closed_count Salesforce's support tickets closed for the month. Written to support.tickets.closed_count instead when this connection is the primary source. count tickets Salesforce
salesforce.cases.open_count Cases open in Salesforce at sync time, a snapshot rather than a period total, written once per month and rewritten on every sync. count — Salesforce
salesforce.cases.resolution_rate Salesforce's case resolution rate for the month. Written to support.tickets.resolution_rate instead when this connection is the primary source. percentage % Salesforce
salesforce.cases.volume Salesforce's support tickets opened for the month. Written to support.tickets.volume instead when this connection is the primary source. count tickets Salesforce
salesforce.opportunities.avg_demo_to_proposal_days Average days from the stage chosen as demo to the stage chosen as proposal, over the Salesforce opportunities first entering the proposal stage in the month. Written to crm.deals.avg_demo_to_proposal_days instead when this connection is the primary source. duration days Salesforce
salesforce.opportunities.avg_sales_cycle_days Salesforce's average sales cycle for the month. Written to crm.deals.avg_sales_cycle_days instead when this connection is the primary source. duration days Salesforce
salesforce.opportunities.avg_value Salesforce's average won opportunity size for the month. Written to crm.deals.avg_value instead when this connection is the primary source. currency USD Salesforce
salesforce.opportunities.lost_count Salesforce's lost deals for the month. Written to crm.deals.lost_count instead when this connection is the primary source. count deals Salesforce
salesforce.opportunities.open_count Opportunities open in Salesforce at sync time, a snapshot rather than a period total, written once per month and rewritten on every sync. count — Salesforce
salesforce.opportunities.open_value Value of the opportunities open in Salesforce at sync time, a snapshot rather than a period total, written once per month and rewritten on every sync. currency USD Salesforce
salesforce.opportunities.proposals_sent_count Salesforce opportunities in record types that count as sales that first entered the stage chosen as proposal in the month. Written to crm.proposals.sent_count instead when this connection is the primary source. count proposals Salesforce
salesforce.opportunities.renewal_arr_due The value of the Salesforce opportunities chosen as renewals that fell due in the month, whatever became of them. Written to customer_success.renewals.arr_up_for_renewal instead when this connection is the primary source. currency USD Salesforce
salesforce.opportunities.renewal_rate Of the Salesforce opportunities chosen as renewals that fell due in the month, the share won, by number or by value. Written to customer_success.renewals.rate instead when this connection is the primary source. percentage % Salesforce
salesforce.opportunities.renewal_rate_base The Salesforce opportunities chosen as renewals that fell due in the month, by number or by value as the renewal rate counts them. Written to customer_success.renewals.rate_base instead when this connection is the primary source. count — Salesforce
salesforce.opportunities.services_proposals_count Salesforce opportunities in record types ticked as services that first entered the stage chosen as proposal in the month. Written to professional_services.proposals.submitted_count instead when this connection is the primary source. count proposals Salesforce
salesforce.opportunities.win_rate Salesforce's deal win rate for the month. Written to crm.deals.win_rate instead when this connection is the primary source. percentage % Salesforce
salesforce.opportunities.won_count Salesforce's won deals for the month. Written to crm.deals.won_count instead when this connection is the primary source. count deals Salesforce
salesforce.opportunities.won_value Salesforce's won deal value for the month. Written to crm.deals.won_value instead when this connection is the primary source. currency USD Salesforce

Supply chain

Key What it measures Type Unit Fed by
supply_chain.deliveries.on_time_rate Percentage of orders delivered on or before the promised date.

Watch out for

  • Measure against the date the customer was first given. An order re-promised and then delivered on the new date was still late.
  • Suppliers, carriers, weather, demand swings and capacity move it too, so a formula valuing its change takes only the share of it you credit to the initiative.
  • It counts orders against a promised date. On-Time Delivery Rate counts deliveries against a window, so the two don't compare.

Good for

  • Valuing production scheduling by the revenue late orders no longer put at risk, at your contribution margin.
  • Valuing route planning by the penalties, credits and expedites late shipments no longer cost.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.deliveries.revenue_at_risk Revenue exposed to late deliveries in the period through penalties, cancellations and credits.

Watch out for

  • Count what lateness cost or could have, such as penalties, cancellations and credits, not the value of every late order.
  • It's revenue, not return: a formula books what you keep of it at your contribution margin.

Good for

  • The amount a change in the on-time delivery rate is applied to.

Guide: Revenue is booked at your margin

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.forecasting.accuracy Demand forecast accuracy against actual demand. percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.inventory.avg_value Average inventory value on hand during the period.

Watch out for

  • An average over the period, not the month-end balance on the balance sheet. Record the same one before and after.

Good for

  • The base the share of excess inventory is applied to, so stocking up for a peak season isn't read as worse forecasting.
  • With revenue, inventory per dollar of sales: the measure of holding less stock for the same sales.

Guide: When the count moves with how busy you are

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.inventory.excess_value Value of inventory held above target levels.

Watch out for

  • Excess means above your target levels. Raising or lowering a target changes it without anything changing on the shelf, so keep the targets the same before and after.
  • It's compared as a share of inventory on hand, so record both over the same months.
  • It isn't all written off: the formula prices it at your annual holding cost rate, a share of its value a year, not its full value.

Good for

  • Valuing better demand forecasting by the carrying cost of excess stock no longer held.

Guide: When the count moves with how busy you are

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.inventory.stockout_events Stockout incidents in the period. count events Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.inventory.stockout_rate Percentage of SKUs or order lines that were out of stock.

Watch out for

  • Count it by SKU, by SKU-location or by order line, and keep that the same before and after: the three give very different rates.
  • Counted by SKU, a stockout on a slow seller counts the same as one on a best seller, while each point of change is priced at your whole revenue.
  • Supplier delays, demand spikes and buying decisions move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing demand forecasting or omnichannel inventory by the sales stockouts no longer lose, at your contribution margin.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.procurement.hours_per_po Staff hours per purchase order, from raising it through approval to sending it to the supplier, not matching its invoice.

Watch out for

  • Stop before the invoice. The same template values invoice matching hours on their own, so matching time recorded here as well would count twice.
  • Time ordinary orders, not the exception that took a week.

Good for

  • Valuing procurement automation by the staff time it frees on each purchase order.

Guide: Formulas that count the same money

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.procurement.invoice_matching_hours_per_week AP staff hours per week matching purchase orders to invoices.

Watch out for

  • Hours a week, including chasing invoices that don't match. Use a normal week, not a month-end week.
  • Count all the matching time here, and leave it out of hours per purchase order on the same template, or it counts twice.

Good for

  • Valuing automated three-way matching by the AP time it frees, compared per vendor bill so a heavy month isn't read as slower matching.

Guide: When the count moves with how busy you are

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.procurement.po_cycle_days Average calendar days from purchase requisition to approved purchase order. duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.procurement.purchase_orders_processed Purchase orders processed in the period.

Watch out for

  • Count orders, not lines, and count a changed order once.

Good for

  • The purchase orders hours saved per order are multiplied by.

Guide: What a formula reads

count — NetSuite, QuickBooks, Xero
supply_chain.shipments.cost_per_shipment Average total cost per shipment.

Watch out for

  • Include freight, fuel surcharges and accessorials, so a fall in surcharges shows rather than hides.
  • Fuel surcharges, carrier rates and the weight and lane mix of shipments move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing carrier selection and route planning by the shipping cost saved on this period's shipments.

Guide: When the outcome has other causes

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.shipments.count Outbound shipments in the period.

Watch out for

  • Count shipments, not parcels: a five-carton shipment is one.

Good for

  • The shipments a fall in cost per shipment, or a rise in on-time delivery, is multiplied by.
  • The volume where-is-my-order contacts are compared per, so peak season isn't read as a loss.

Guide: When the count moves with how busy you are

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.suppliers.active_count Suppliers the business is actively buying from.

Watch out for

  • Suppliers you've ordered from recently, such as in the past twelve months, not every vendor ever set up.

Good for

  • The base disruptions and review hours are compared per, so a wider supplier base isn't read as more risk or slower reviews.

Guide: When the count moves with how busy you are

count suppliers NetSuite, QuickBooks, Xero
supply_chain.suppliers.disruptions_per_year Supply disruptions per year.

Watch out for

  • Count disruptions that stopped or delayed production or fulfillment, as events, not the orders each one affected.
  • Port closures, shortages, weather and market events move it too, so the library's formula valuing its fall takes only the share of it you credit to the initiative.
  • Set it over several years, so one rough or quiet year isn't the baseline.

Good for

  • Valuing supplier risk monitoring by the stoppages, emergency sourcing and expedites it heads off.

Guide: When the outcome has other causes

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.suppliers.review_hours_per_month Staff hours per month on supplier health reviews and risk assessments.

Watch out for

  • Count the reviewing, gathering the data, reading it and writing it up, not the remediation after a bad review.
  • It's compared per active supplier. Record active suppliers over the same months.

Good for

  • Valuing continuous supplier monitoring by the procurement time it frees.

Guide: When the count moves with how busy you are

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.warehouse.pick_error_rate Percentage of picks resulting in an error.

Watch out for

  • Take it from the same source before and after, audits or downstream returns.
  • Count wrong item, wrong quantity and wrong place, the same way before and after.

Good for

  • With units picked, the returns, reshipping and corrections fewer mispicks avoid.

Guide: What a formula reads

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.warehouse.picks_count Units picked in the period.

Watch out for

  • Units, not orders or lines, since picks per labor hour counts units.

Good for

  • The units a change in picks per labor hour, or in the pick error rate, is applied to.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
supply_chain.warehouse.picks_per_labor_hour Units picked per warehouse labor hour.

Watch out for

  • Units over picking hours only. Packing or receiving hours understate it and hide what changed.
  • The hours freed are freed capacity, not cash: pickers are paid the same unless hours are actually cut.

Good for

  • Valuing directed picking and slotting by the labor hours the same picks no longer take.

Guide: Cash value and freed capacity

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Support

Key What it measures Type Unit Fed by
support.calls.avg_handle_time Average minutes an agent spends on one inbound phone call — talk, hold and wrap-up.

Watch out for

  • Talk, hold and wrap-up together. A phone system report that leaves out wrap-up reads low.
  • It moves with the mix of calls: once an AI takes the simple ones, the calls left for agents run longer, so the average can rise while total agent time falls.

Good for

  • Pricing the calls an AI or self-service channel keeps from reaching an agent, with your labor rate.

Guide: Cash value and freed capacity

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
support.knowledge_base.article_count Knowledge base articles being maintained.

Watch out for

  • Published articles customers can see at month end, not drafts or internal notes.

Good for

  • The articles update minutes saved per article are multiplied by.

Guide: What a formula reads

count — Freshdesk, Zendesk
support.knowledge_base.avg_article_age_days Average days since knowledge base articles were last reviewed or updated. duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
support.knowledge_base.update_minutes_per_article Minutes to review and update one knowledge base article per cycle.

Watch out for

  • One article from opening it for review to publishing the update, including checking what the AI suggested.
  • The formula spreads a year's update cycles evenly over its months, so time one cycle's update, not a year's.

Good for

  • Valuing AI freshness checks by the documentation time they free.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
support.quality.review_minutes_per_interaction Minutes for a QA analyst to review and score one interaction.

Watch out for

  • Minutes for one review: reading or listening, scoring and notes.
  • The formula frees every manual review at the old sample rate, which assumes people no longer review any. If they still review some, it overstates.

Good for

  • Pricing the QA analyst time AI scoring frees, at your loaded hourly rate.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
support.quality.sample_rate Share of support interactions reviewed for quality.

Watch out for

  • Reviews completed over conversations handled in the same month. Formulas read only the rate before the AI.

Good for

  • With interactions and minutes per review, the manual QA workload AI scoring takes on, as freed capacity.

Guide: Cash value and freed capacity

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
support.tickets.agent_minutes_per_ticket Minutes of agent work one resolved ticket takes, from a time study or your help desk's time tracking. Not how long the ticket was open.

Watch out for

  • It's hands-on minutes per resolved ticket, not how long the ticket was open, which is mostly waiting.
  • Time the same task before and after, from the same start to the same finish, including the minutes spent checking what the AI drafted.

Good for

  • Valuing the agent time an AI routing or drafting tool saves on the tickets agents still resolve.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
support.tickets.agent_resolved_count Tickets resolved by an agent: every resolved ticket that no AI deflection rule on the help desk matched, plus the ones a rule matched whose customer came back.

Watch out for

  • On Zendesk, Freshdesk and Intercom it leaves out tickets an AI deflection rule matched, unless the customer came back, so a ticket is agent work or deflection value, never both.
  • HubSpot tickets and Salesforce cases don't show whether an AI resolved them, so there it's counted by hand. Leave out what the AI handled alone.

Good for

  • The tickets the agent minutes saved per ticket are multiplied by.

Guide: Counting a ticket once

count tickets Freshdesk, Intercom, Zendesk
support.tickets.ai_resolution_rate Percentage of the tickets resolved in the period that AI resolved, without human escalation. percentage % Freshdesk, Intercom, Zendesk
support.tickets.ai_resolved_count Support tickets fully resolved by AI without human escalation in the period. count tickets Freshdesk, Intercom, Zendesk
support.tickets.avg_first_response_hours Average hours from ticket creation to the first agent reply, across tickets opened in the period. duration hours Freshdesk, Intercom, Zendesk
support.tickets.avg_resolution_days Average days from ticket creation to resolution, waiting included: how long a ticket was open, not how long anyone worked it. Zendesk, Freshdesk and Intercom report hours, converted ÷ 24.

Watch out for

  • It counts waiting as well as work: a ticket left open over a weekend gains two days. Don't price it as agent time.
  • It's an average, so a few tickets that ran for weeks can move it more than everything else in the month.

Good for

  • Valuing the time people wait less for an IT ticket to be resolved, when AI triage speeds it up.
duration days Freshdesk, HubSpot, Intercom, Salesforce, Zendesk
support.tickets.closed_count Total support tickets closed/resolved in the period. count tickets Freshdesk, HubSpot, Intercom, Salesforce, Zendesk
support.tickets.cost_per_ticket Blended cost to resolve one support ticket (agent labor plus tooling).

Watch out for

  • It's mostly agent time, so value priced from it is freed capacity, not cash.
  • Price a ticket an agent handles: agent minutes × loaded agent cost, plus overhead. Total support spend ÷ every ticket includes the tickets the AI already resolves, and reads low.

Good for

  • Pricing the tickets a knowledge base or self-service channel keeps from reaching an agent.

Guide: The price of a deflected ticket

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
support.tickets.csat_score Average customer satisfaction on closed tickets, normalized to 0–1 (Zendesk good/bad → 0/1; Intercom 1–5 → (rating − 1) ÷ 4). score 0–1 Intercom, Zendesk
support.tickets.deflected_count Tickets resolved by self-service or AI without an agent in the period. count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
support.tickets.resolution_rate Percentage of tickets opened in the period that were also closed. percentage % HubSpot, Salesforce
support.tickets.self_serve_deflection_rate Percentage of support inquiries resolved without agent involvement.

Watch out for

  • Count it the same way before and after: the same channels, and the same test for resolved without an agent.
  • A customer who gave up looks resolved, so the rate can rise while customers are served worse.

Good for

  • Valuing a knowledge base or self-service launch by the change in the share of inquiries resolved without an agent, at this period's ticket volume.

Guide: What counts as deflected

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
support.tickets.volume Total support tickets opened in the period (help-desk tickets, Intercom conversations, Salesforce cases).

Watch out for

  • It rises and falls with your customer base, launches and incidents, whatever the AI does. Compare a count per ticket rather than raw totals.
  • Each source counts its own unit: a help desk counts tickets, Intercom conversations, Salesforce cases.

Good for

  • The volume a change in a rate is multiplied by, so a busy month isn't read as a loss.

Guide: When the count moves with how busy you are

count tickets Freshdesk, HubSpot, Intercom, Salesforce, Zendesk

GitHub

Key What it measures Type Unit Fed by
github.copilot.acceptance_rate Accepted code activities as a share of code generation activities in GitHub's Copilot organization report for the month. Read as engineering.coding_assistant.acceptance_rate when GitHub is the primary coding-assistant source. percentage % GitHub
github.copilot.active_users_avg Daily active Copilot users in GitHub's organization report, averaged over the month. Read as engineering.coding_assistant.active_users when GitHub is the primary coding-assistant source. average — GitHub
github.copilot.lines_accepted Lines of code accepted in GitHub's Copilot organization report for the month. Read as engineering.coding_assistant.lines_accepted when GitHub is the primary coding-assistant source. count — GitHub
github.copilot.suggestions_accepted Code acceptance activities in GitHub's Copilot organization report for the month. Read as engineering.coding_assistant.acceptances when GitHub is the primary coding-assistant source.

Watch out for

  • One acceptance activity can bundle several completions, so it isn't the same unit as a Claude Code accepted edit. The two are priced at separate minutes saved, never added up.
  • It counts code a developer took, not code that shipped: an accepted suggestion can be rewritten or deleted later.

Good for

  • Valuing Copilot's share of the time saved where a team runs it beside Claude Code, each at its own minutes saved per acceptance.

Guide: The units aren't the same across tools

count — GitHub
github.copilot.suggestions_shown Code generation activities in GitHub's Copilot organization report for the month. Read as engineering.coding_assistant.suggestions when GitHub is the primary coding-assistant source. count suggestions GitHub
github.copilot_review.active_users_avg People active with Copilot code review on an average day in the month, from GitHub's Copilot organization report. They're counted in GitHub Avg Active Copilot Users too, which counts anyone active with Copilot in any way. average users GitHub
github.copilot_review.applied_suggestions Copilot code review suggestions committed with GitHub's Commit suggestion button in the month, from GitHub's Copilot organization report.

Watch out for

  • It counts only suggestions committed with GitHub's button. A comment fixed by hand, or by another coding agent, counts as not applied, so a low figure doesn't mean the reviews were ignored.
count suggestions GitHub
github.copilot_review.prs_reviewed Pull requests Copilot code review reviewed in the month, from GitHub's Copilot organization report. count pull requests GitHub
github.copilot_review.suggestions Suggestions Copilot code review left on pull requests in the month, from GitHub's Copilot organization report. Not the same as Copilot's editor suggestions (GitHub Copilot Suggestions Shown), and never added to them. count suggestions GitHub
github.copilot_seats.active_30d Of the Copilot seats assigned, those whose holder GitHub saw use Copilot in the 30 days to the month's latest read: suggestions or Chat in an IDE, pull request summaries, Chat on github.com or in the CLI. GitHub sees IDE use only where the IDE's telemetry is on, and its date can be up to a day behind. count users GitHub
github.copilot_seats.assigned Copilot seats assigned across the connection's organizations, as of the month's latest read of each organization's seat list, seats pending cancellation included. GitHub lists only who holds a seat now, so months before the first read have none. count users GitHub
github.cost.copilot_spend GitHub's Copilot and AI-credit net spend for the month. Written as ai.cost.spend when this connection is the primary AI-cost source. currency USD GitHub
github.cost.product_spend GitHub spend on one product (Actions, Copilot, Packages, …) for the month; the product is in the observation's dimensions. currency USD GitHub
github.cost.total_spend GitHub's total spend for the month, every product included. currency USD GitHub
github.deployments.count Deployments GitHub recorded to a production environment in the month. Written to it.deployments.count instead when this connection is the primary source for deployments. count deployments GitHub
github.pull_requests.additions_avg GitHub's average lines added per pull request for the month. average — GitHub
github.pull_requests.cycle_time_avg GitHub's average pull request cycle time for the month. duration — GitHub
github.pull_requests.merged_count GitHub's pull requests merged count for the month. count — GitHub

Real estate

Key What it measures Type Unit Fed by
real_estate.leasing.annual_renewal_rate Percentage of expiring leases renewed.

Watch out for

  • Renewals at signature over units whose leases expired, from the same rent roll. Decide once whether a month-to-month rollover counts.
  • Market rents, rent increases, relocations and building condition move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing tenant experience programs by the turnover costs renewals avoid.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.leasing.days_to_lease Average days from inquiry to signed lease.

Watch out for

  • Calendar days from first inquiry, not from the tour, to signed lease.
  • The formula books each day saved as rent, which holds only for a unit standing vacant. Measure leases of vacant units: a lead on one still occupied earns nothing sooner.
  • The rental market, seasonality, concessions and unit condition move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing leasing automation by the vacant days it saves.

Guide: Revenue that arrives sooner

duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.leasing.staff_hours_per_lease Leasing staff hours per signed lease (lead response, tours, application processing).

Watch out for

  • Include hours on leads that never signed, spread across the leases that did, or the figure flatters itself.
  • The hours freed are freed capacity, not cash: leasing staff are paid the same afterwards.

Good for

  • Valuing leasing automation by the staff time it frees on each lease.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.leasing.units_leased Units signed to new leases in the period.

Watch out for

  • Signed leases, not applications or reservations.

Good for

  • The leases vacant days and staff hours saved are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.leasing.units_up_for_renewal Units with lease expirations in the period.

Watch out for

  • Units whose leases expire in the period, from the same rent roll as the renewal rate.

Good for

  • The units a rise in the renewal rate is applied to.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.maintenance.emergency_repairs_per_month Unplanned emergency repair incidents per month.

Watch out for

  • Emergency as your maintenance system defines it, without changing the definition between the two sides.
  • It's compared per unit in the portfolio, so record units over the same months.

Good for

  • Valuing predictive building monitoring by the emergency repairs it avoids.

Guide: When the count moves with how busy you are

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.portfolio.annual_yield Net operating income over a year as a percentage of assets under management.

Watch out for

  • Net operating income over twelve months as a percentage of assets under management, with assets valued the same way on both sides.
  • Rents, interest rates, cap rates and the property market move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing portfolio intelligence by the extra income better capital allocation earns.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.portfolio.aum Portfolio value under management at the time of measurement.

Watch out for

  • Say the valuation date: a revaluation between two periods moves it more than anything an initiative does.

Good for

  • The assets a rise in yield is applied to.

Guide: When the outcome has other causes

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.portfolio.property_count Properties in the portfolio.

Watch out for

  • Properties, meaning buildings, not units: the formulas multiply it by reporting hours per property. Units in the portfolio is its own figure.

Good for

  • The properties reporting hours saved per property are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.portfolio.unit_count Residential or commercial units in the portfolio.

Watch out for

  • Units in the portfolio the initiative covers, not the whole company's. Several figures divide by it, so set it once and reuse it.

Good for

  • The base emergency repairs and tenant contacts are compared per.

Guide: When the count moves with how busy you are

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.reporting.hours_per_property_monthly Staff hours per property per month compiling performance and financial reports.

Watch out for

  • Hours to produce one property's monthly reporting pack, from the accountants who prepare it.
  • The hours freed are freed capacity, not cash: the team is paid the same afterwards.

Good for

  • Valuing automated reporting by the preparation time it frees per property.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.reporting.review_cycle_days Calendar days between formal portfolio performance reviews. duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.tenants.contacts_per_unit_monthly Inbound phone and email contacts per unit per month requiring staff handling.

Watch out for

  • Contacts that reached staff, divided by the same unit count you record. An automated acknowledgement isn't one.

Good for

  • Valuing a self-serve tenant app by the staff contacts it removes.

Guide: When the count moves with how busy you are

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.tenants.minutes_per_contact Staff minutes to handle one inbound tenant contact.

Watch out for

  • Hands-on minutes for an ordinary contact end to end, the follow-up it creates included, not the memorable ones.

Good for

  • Pricing the staff time fewer tenant contacts free, at your loaded hourly rate.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.tenants.satisfaction_score Average tenant satisfaction score from periodic surveys. score — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.work_orders.admin_minutes_per_order Staff minutes per work order on creation, vendor calls and status follow-ups.

Watch out for

  • Raising, vendor calls, chasing and closing: not the repair, and not the trade's own time.

Good for

  • Valuing automated routing and status tracking by the admin time they free.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.work_orders.created_count Work orders created in the period.

Watch out for

  • Orders raised, not completed, so a backlog doesn't read as fewer.

Good for

  • The work orders admin minutes saved are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
real_estate.work_orders.resolution_days Average days from work order submission to resolution. duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Marketing

Key What it measures Type Unit Fed by
marketing.budget.destination_pipeline_per_dollar Pipeline created per dollar of spend in the channels budget was moved to, from the attribution model.

Watch out for

  • Use the same attribution model and window as for the channels the money came from, or the difference is between two models.
  • It's the pipeline your attribution model assigns to the spend, not proof the spend caused it.
  • The formula credits the whole difference to the move, with no share for other causes, so a seasonal swing between channels books as value.

Good for

  • Less pipeline per dollar in the channels cut, the extra pipeline each dollar moved creates, at your win rate and contribution margin.

Guide: Revenue is booked at your margin

ratio — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
marketing.budget.reallocated The period's marketing spend that went to the channels attribution favored instead of the channels it was taken from.

Watch out for

  • Only spend moved to the favored channels, counted every month it stays moved: not the whole budget, and not a plan to move it later.

Good for

  • The dollars a difference in pipeline per dollar between channels is applied to.

Guide: Revenue is booked at your margin

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
marketing.budget.source_pipeline_per_dollar Pipeline created per dollar of spend in the channels budget was taken from, from the attribution model.

Watch out for

  • Use the same attribution model and window as for the channels the money went to, or the difference is between two models.
  • It's the pipeline your attribution model assigns to the spend, not proof the spend caused it.

Good for

  • Subtracted from pipeline per dollar in the channels funded, it gives the extra pipeline each dollar moved creates.

Guide: Revenue is booked at your margin

ratio — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
marketing.campaigns.avg_list_size Average recipients per campaign send.

Watch out for

  • Recipients delivered per send, not the list you targeted.
  • The formula multiplies it by campaigns launched, which counts every campaign, not only email sends.

Good for

  • With campaigns launched, the emails a change in open rate is applied to.

Guide: When the outcome has other causes

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
marketing.campaigns.conversion_rate Percentage of campaign recipients who completed the campaign's goal. percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
marketing.campaigns.hours_per_launch Staff hours to configure and launch one campaign.

Watch out for

  • Hours from brief to send for the people who built it: build, list, creative, QA and scheduling, not the meetings.
  • The hours freed are freed capacity, not cash: the team is paid the same afterwards.

Good for

  • Valuing campaign automation by the setup time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
marketing.campaigns.launched_count Marketing campaigns with a start date in the period.

Watch out for

  • From HubSpot it counts campaigns with a start date in the period, whatever kind of campaign they are.

Good for

  • The campaigns setup hours saved per launch are multiplied by.

Guide: What a formula reads

count campaigns HubSpot
marketing.content.hours_per_piece Staff hours from brief to published asset.

Watch out for

  • Compare pieces of the same kind: a blog post and a white paper aren't the same figure.
  • The hours freed are freed capacity, not cash: writers and editors are paid the same afterwards.

Good for

  • Valuing AI-assisted drafting by the writing and editing time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
marketing.content.pieces_per_month Content pieces produced per month. count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
marketing.content.published_count Blog posts published in the period on the connected CMS.

Watch out for

  • From HubSpot it counts blog posts published in the period, not other kinds of content.
  • The outsourcing formula counts each extra post as one you'd otherwise have paid an agency for. That holds only if you would have.

Good for

  • The posts content hours saved per piece are multiplied by.

Guide: What a formula reads

count posts HubSpot
marketing.emails.open_rate Opens as a percentage of deliveries across marketing emails published in the period.

Watch out for

  • Mail privacy features record opens nobody made, so the rate runs high and can jump with no change in readers.
  • Deliverability, list changes and seasonality move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing AI-assisted email by the extra conversions from more opens, at your contribution margin.

Guide: When the outcome has other causes

percentage % HubSpot
marketing.emails.open_to_conversion_rate Conversions from marketing email as a percentage of its opens in the period, from your email platform or CRM.

Watch out for

  • Conversions attributed to marketing email over that email's opens. Count conversions the same way each period.
  • Inflated opens make it read low, since the same conversions are divided by more opens.

Good for

  • Turning extra opens into expected conversions.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
marketing.nurture.manual_hours_per_week Staff hours per week on lead follow-up and segmentation done by hand.

Watch out for

  • Follow-up and segmentation done by hand, in a normal week rather than a launch week.
  • It's compared per new contact, so a campaign that floods the funnel isn't read as slower follow-up. Record new contacts over the same weeks.

Good for

  • Valuing nurture automation by the demand gen time it frees.

Guide: When the count moves with how busy you are

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
marketing.pipeline.sourced_share Percentage of total sales pipeline attributed to marketing-sourced leads. percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
marketing.reporting.hours_per_week Staff hours per week on marketing performance reports.

Watch out for

  • Count the assembly, exports, spreadsheets and the deck, not the reading, and include everyone who touches the pack.
  • The hours freed are freed capacity, not cash: the team is paid the same afterwards.

Good for

  • Valuing automated dashboards by the reporting time they free.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
marketing.seo.research_hours_per_month Staff hours per month on keyword research and on-page optimization.

Watch out for

  • Where an agency does the work, use the hours on its retainer for those tasks, not the whole retainer.
  • Hours freed in-house are freed capacity, not cash: the team is paid the same afterwards.

Good for

  • Valuing SEO automation by the research and optimization time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
marketing.web.organic_sessions Website sessions from organic search in the period.

Watch out for

  • From HubSpot it's HubSpot's own figure from its Analytics sources report, so it counts only what HubSpot's tracking sees.
  • Search demand, seasonality, brand campaigns, press and algorithm changes move it too, so a formula valuing its change takes only the share of it you credit to the initiative.
  • The formula prices each extra visit at an equivalent paid click, which assumes you'd otherwise have bought it.

Good for

  • Valuing SEO work by the paid traffic it saves buying.

Guide: When the outcome has other causes

count sessions HubSpot

Manufacturing

Key What it measures Type Unit Fed by
manufacturing.data.reconciliation_hours_per_week Staff hours per week reconciling data between shop floor systems and the ERP.

Watch out for

  • Everyone who re-keys, chases and corrects shop-floor numbers into the ERP, not only the planner.
  • The hours freed are freed capacity, not cash: staff are paid the same afterwards.

Good for

  • Valuing shop-floor integration by the reconciliation time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
manufacturing.equipment.oee Overall equipment effectiveness.

Watch out for

  • Availability × performance × quality. Record how each part was measured: OEE quoted without its parts isn't comparable to itself a year later.
  • Demand and scheduling, maintenance, product mix and staffing move it too, so a formula valuing its change takes only the share of it you credit to the initiative.
  • The formula books the extra output as revenue, which holds only if you can sell it.

Good for

  • Valuing a loss-reduction program by the extra good output, at your contribution margin.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
manufacturing.equipment.oee_reporting_hours_per_week Staff hours per week collecting and compiling OEE data.

Watch out for

  • Gathering and compiling the numbers only, not analyzing them.
  • The hours freed are freed capacity, not cash: staff are paid the same afterwards.

Good for

  • Valuing automated OEE dashboards by the collection time they free.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
manufacturing.equipment.scheduled_hours_per_month Hours per month the monitored equipment was scheduled to run.

Watch out for

  • Planned production time before any downtime is taken off, on the same equipment as the downtime hours.

Good for

  • The base unplanned downtime is compared per, so a fuller schedule isn't read as worse reliability.

Guide: When the count moves with how busy you are

count hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
manufacturing.equipment.unplanned_downtime_hours_per_month Unplanned equipment downtime hours per month.

Watch out for

  • Unplanned only: scheduled maintenance and changeovers hide what condition monitoring changed.
  • It's compared per scheduled hour, so record scheduled hours on the same equipment over the same months.

Good for

  • Valuing predictive maintenance by the lost production it avoids.

Guide: When the count moves with how busy you are

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
manufacturing.maintenance.annual_cost_per_asset Average annual maintenance cost per asset including reactive repairs and planned servicing. currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
manufacturing.maintenance.annual_spend Annual maintenance spend across monitored assets, planned and reactive.

Watch out for

  • A year's spend, parts, contractors and labor, planned and reactive. The formula books a month's share of the saving at a time.
  • If in-house labor is included, part of the saving is staff time, booked here as cash. Say whether it's included and keep to it.

Good for

  • The spend a maintenance cost reduction is applied to.

Guide: What a formula reads

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
manufacturing.maintenance.cost_reduction_rate Percentage reduction in total maintenance spend from condition-based servicing.

Watch out for

  • The fall itself, as a percentage of spend before condition-based servicing, not a before-and-after pair. Measure it on the same spend you record as Annual Maintenance Spend.
  • No credited share is taken, so leave out falls from other causes, such as fewer assets or cheaper parts.

Good for

  • Valuing condition-based maintenance by the maintenance spend it saves.

Guide: What a formula reads

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
manufacturing.materials.shortage_incidents_per_month Production stoppages caused by material shortages per month.

Watch out for

  • Stoppages, not the shortages behind them: one missing part that stops a line twice is two.
  • It's compared per unit produced, so record units over the same months.

Good for

  • Valuing supply visibility by the stoppage costs it avoids.

Guide: When the count moves with how busy you are

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
manufacturing.production.theoretical_capacity_units Units producible in the period at nameplate capacity.

Watch out for

  • Units at nameplate rate over the period's scheduled hours, with no stoppages. Say which shift pattern you assumed.

Good for

  • The capacity a rise in OEE is applied to.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
manufacturing.production.units_produced Units produced in the period.

Watch out for

  • Good units, not units started, so scrap doesn't count twice.

Good for

  • The volume scrap cost and shortage stoppages are compared per, so a busy month isn't read as worse quality.
  • The units a fall in the defect escape rate is applied to.

Guide: When the count moves with how busy you are

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
manufacturing.quality.defect_escape_rate Percentage of produced units with a defect that reaches the customer.

Watch out for

  • Defects that reached the customer, from returns, warranty claims and complaints. Defects caught in-house aren't escapes.

Good for

  • Valuing computer vision inspection by the warranty costs fewer escapes avoid.

Guide: What a formula reads

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
manufacturing.quality.first_pass_yield Percentage of units completing production without defect, rework or scrap. percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
manufacturing.quality.scrap_cost_per_month Monthly scrap and rework cost.

Watch out for

  • Material, the labor already spent and the rework labor. Material alone understates it and the saving.
  • It's compared per unit produced, so a busy month isn't read as worse quality. Record units over the same months.

Good for

  • Valuing real-time defect detection by the scrap and rework cost it saves.

Guide: When the count moves with how busy you are

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
manufacturing.scheduling.hours_per_week Staff hours per week building and adjusting production schedules.

Watch out for

  • Building schedules and reworking them after breakdowns or late deliveries: the reworking is usually where the hours go.
  • The hours freed are freed capacity, not cash: planners are paid the same afterwards.

Good for

  • Valuing AI scheduling by the planner time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Retail

Key What it measures Type Unit Fed by
retail.customers.active_count Active customers in the period.

Watch out for

  • Say what makes a customer active, such as a purchase in the past twelve months, and keep it the same before and after.
  • Both the repeat purchase and the retention formula multiply by it. An initiative gets one of the two, since second purchases are inside retention.

Good for

  • The customers a change in repeat purchase or retention is applied to.

Guide: Formulas that count the same money

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
retail.customers.repeat_purchase_rate Percentage of customers making a second purchase within twelve months.

Watch out for

  • Follow the same cohort forward before and after, or a growing count of new customers reads as a fall.
  • Promotions, seasonality, loyalty programs and the economy move it too, so a formula valuing its change takes only the share of it you credit to the initiative.
  • Second purchases are inside customer retention, so an initiative gets the repeat purchase formula or the retention one, not both.

Good for

  • Valuing lifecycle marketing by the second purchases it adds, at your contribution margin.

Guide: Formulas that count the same money

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
retail.customers.retention_rate Twelve-month customer retention rate.

Watch out for

  • Customers active twelve months ago who are still active now. It isn't the repeat purchase rate; keep the two apart.
  • Prices, competitors, loyalty programs and the economy move it too, so a formula valuing its change takes only the share of it you credit to the initiative.
  • It already includes second purchases, so an initiative gets the retention formula or the repeat purchase one, not both.

Good for

  • Valuing customer-lifetime-value programs by the revenue of customers kept, at your contribution margin.

Guide: Formulas that count the same money

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
retail.inventory.accuracy Percentage of inventory records matching physical stock. percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
retail.inventory.markdown_eligible_sold_value Full retail value of the seasonal or clearance-eligible stock that sold or cleared in the period, at target price or marked down.

Watch out for

  • Full retail value of the seasonal or clearance stock that left, not what it sold for. Say the age or season rule and keep it.
  • The sell-through formula and gross margin improvement on the same template count the same money, so an initiative gets one of the two.

Good for

  • The stock a change in sell-through is applied to, at the markdown depth it avoids.

Guide: Formulas that count the same money

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
retail.inventory.sell_through_rate Of the seasonal or clearance-eligible stock that sold or cleared in the period, the percentage by retail value that sold at or above target price.

Watch out for

  • By full retail value on both sides. Units will do where the system can't value them, if both sides count units.
  • Weather, trends, promotions and buying decisions move it too, so a formula valuing its change takes only the share of it you credit to the initiative.
  • Gross margin improvement on the same template counts this too, so an initiative gets one of the two.

Good for

  • Valuing AI markdown optimization by the markdowns avoided on stock that sold at target price.

Guide: Formulas that count the same money

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
retail.inventory.shrink_rate Book inventory less what a physical count finds, as a percentage of sales over the same span, from your stock counts.

Watch out for

  • From your stock counts, so it moves only when you count. Use the same counting cycle before and after.
  • Theft, staffing, store mix and loss-prevention programs move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing inventory visibility by the shrink it avoids on this period's sales.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
retail.orders.avg_value Average order value.

Watch out for

  • Say whether it's before or after returns and keep to it: returns move it more than most changes will.
  • Inflation, pricing, promotions and product mix move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing personalization by the extra revenue per order, at your contribution margin.
  • The value each extra order is priced at when valuing a rise in conversion.

Guide: When the outcome has other causes

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
retail.orders.count Orders placed in the period.

Watch out for

  • Orders, not units, with canceled orders left out. Keep it to the same orders as average order value.

Good for

  • The orders a change in average order value is multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
retail.orders.fulfillment_cost_per_order Outbound shipping spend on customer orders in the period, divided by the period's orders as Orders counts them, canceled orders left out.

Watch out for

  • Per order, not per shipment: an order sent in three parcels costs three labels, which cost per shipment hides.
  • Include freight, fuel surcharges and accessorials from every location that ships, stores and 3PLs as well as DCs.
  • Carrier rates, fuel surcharges, order size and where customers live move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing ship-from-store and order routing by the shipping cost saved on this period's orders, split parcels included.

Guide: When the outcome has other causes

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
retail.pricing.avg_markdown_depth Markdown dollars as a percentage of the full retail value of the sales made at a markdown price in the period, or how far below full price a marked-down sale went on average.

Watch out for

  • Markdown dollars over the full retail value of the marked-down sales, for the same stores and categories as the markdown rate.
  • Formulas read only the depth before the AI, so markdowns made shallower aren't counted and the value errs low.

Good for

  • Pricing each sale no longer marked down, for the markdown and sell-through formulas.

Guide: What a formula reads

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
retail.pricing.gross_margin Gross margin as a percentage of revenue.

Watch out for

  • Gross profit over revenue from the books, synced for each month with a cost of goods sold. A month still open reads high until its cost of sales is booked at close.
  • Cost of goods, freight, product mix and promotions move it too, so a formula valuing its change takes only the share of it you credit to the initiative.
  • The sell-through formula on the same template counts part of the same money, so an initiative gets one of the two.

Good for

  • Valuing AI pricing by the margin it adds on this period's revenue.

Guide: Formulas that count the same money

percentage % NetSuite, QuickBooks, Xero
retail.pricing.markdown_rate Percentage of the period's sales, by full retail value, made at a markdown price.

Watch out for

  • Sales below full price over all sales, both at full retail value, or both in units.
  • Weather, trends, competitors' promotions and buying decisions move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing demand forecasting by the markdowns it avoids on this period's revenue.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
retail.web.conversion_rate Orders as a percentage of website sessions.

Watch out for

  • Orders over sessions, from the same figures you record for each. Where your analytics reports its own rate, use that and say so.
  • Traffic mix, promotions, seasonality and pricing move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing site personalization by the extra orders from the same sessions, at your contribution margin.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
retail.web.session_count Website sessions in the period.

Watch out for

  • Use one analytics tool's definition throughout: session counts differ between tools by more than most initiatives move them.

Good for

  • The sessions a change in conversion rate is applied to.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Anthropic

Key What it measures Type Unit Fed by
anthropic.claude_code.acceptance_rate Accepted tool actions as a share of those proposed in Anthropic's Claude Code report for the month. Read as engineering.coding_assistant.acceptance_rate when Anthropic is the primary coding-assistant source. percentage % Anthropic
anthropic.claude_code.acceptances Tool actions developers accepted in Anthropic's Claude Code report for the month — one accepted edit or write each, not a bundled activity as GitHub counts them. Read as engineering.coding_assistant.acceptances when Anthropic is the primary coding-assistant source.

Watch out for

  • One accepted edit or write each, a smaller unit than a Copilot acceptance, so the two are priced at separate minutes saved, never added up.
  • It counts edits a developer took, not edits that shipped.
  • It covers the Claude Code your Anthropic Console organization reports. Claude Code used on Claude Enterprise seats reports separately.

Good for

  • Valuing Claude Code's share of the time saved where a team runs it beside GitHub Copilot, each at its own minutes saved per acceptance.

Guide: The units aren't the same across tools

count acceptances Anthropic
anthropic.claude_code.active_users_avg Developers who ran Claude Code on an average day in the month. Read as engineering.coding_assistant.active_users when Anthropic is the primary coding-assistant source. average users Anthropic
anthropic.claude_code.lines_accepted Lines of code Claude Code added in the month. Read as engineering.coding_assistant.lines_accepted when Anthropic is the primary coding-assistant source. count lines Anthropic
anthropic.claude_code.proposals Tool actions Claude Code proposed in the month, accepted and rejected together. Read as engineering.coding_assistant.suggestions when Anthropic is the primary coding-assistant source. count suggestions Anthropic
anthropic.usage.cost_per_million_output_tokens Anthropic's spend for the month over the output tokens it generated, per million. Written to ai.cost.per_million_output_tokens instead when this connection is the primary source for AI usage. currency USD Anthropic
anthropic.usage.input_tokens Anthropic's input tokens for the month. Written to ai.usage.input_tokens instead when this connection is the primary source. count tokens Anthropic
anthropic.usage.model_input_tokens Input tokens sent to one Claude model in the month; the model is in the observation's dimensions. count tokens Anthropic
anthropic.usage.model_output_tokens Output tokens generated by one Claude model in the month; the model is in the observation's dimensions. count tokens Anthropic
anthropic.usage.model_spend Anthropic API spend on one model for the month; the model is in the observation's dimensions. currency USD Anthropic
anthropic.usage.model_total_tokens Input plus output tokens for one Claude model in the month; the model is in the observation's dimensions. count tokens Anthropic
anthropic.usage.output_tokens Anthropic's output tokens for the month. Written to ai.usage.output_tokens instead when this connection is the primary source. count tokens Anthropic
anthropic.usage.spend Anthropic's usage spend for the month. currency USD Anthropic
anthropic.usage.total_tokens Anthropic's total tokens for the month. Written to ai.usage.total_tokens instead when this connection is the primary source. count tokens Anthropic

IT

Key What it measures Type Unit Fed by
it.audits.hours_per_cycle IT staff hours per software inventory and license reconciliation audit cycle.

Watch out for

  • One whole inventory and license reconciliation cycle, not a year of them. The formula spreads a year's cycles evenly over its months.

Good for

  • Valuing automated discovery by the audit time it frees.

Guide: What a formula reads

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
it.deployments.change_failure_rate Percentage of deployments causing a rollback or incident.

Watch out for

  • Production deployments rolled back, hotfixed or that opened an incident, over the same deployment count. Not GitHub's failed deployment status: those never landed.

Good for

  • Valuing CI/CD improvements by the incident costs fewer failed changes avoid.

Guide: What a formula reads

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
it.deployments.count Production deployments in the period.

Watch out for

  • From GitHub, production deployments only, by GitHub's production flag or the environment's name. Staging and preview deployments don't count.

Good for

  • The deployments engineer hours saved, or a fall in change failures, are multiplied by.

Guide: What a formula reads

count — GitHub
it.deployments.hours_per_deployment Engineer hours per production deployment, including manual steps, monitoring and validation.

Watch out for

  • Manual steps, watching afterwards and validation, for releases of the kind the initiative changes: a hotfix and a quarterly release aren't one figure.
  • The hours freed are freed capacity, not cash: engineers are paid the same afterwards.

Good for

  • Valuing automated pipelines by the engineer time they free on each release.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
it.deployments.per_week Production deployments per week. count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
it.helpdesk.avg_users_per_incident End-users a typical IT incident affects, whether or not it stops them working.

Watch out for

  • People a typical incident affects, the median, not the worst one, whether or not it stops them working.
  • The formula takes only the blocked share of their time, so it's the users touched, not the users stopped.

Good for

  • The users faster resolution gives working time back to.

Guide: Cash value and freed capacity

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
it.helpdesk.mean_time_to_resolution_hours Mean hours to resolve a tier-1 IT issue. duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
it.helpdesk.tickets_per_month IT helpdesk tickets per month requiring agent handling. count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
it.security.incident_count Security incidents investigated in the period.

Watch out for

  • Incidents investigated, not every alert that fired, counted the same way before and after: a change in what counts reads as a change in security.

Good for

  • The incidents faster containment or detection is multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
it.security.incidents_per_year Security incidents per year. count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
it.security.mean_time_to_contain_hours Mean hours from detection of a security incident to its containment.

Watch out for

  • From detection to the spread stopped, not the root cause fixed or the ticket closed.
  • The formula prices each hour saved as one analyst hour per incident. If several people work an incident, that understates it.

Good for

  • Valuing automated playbooks by the response time they free.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
it.security.mean_time_to_detect_hours Mean hours from a security incident starting to its detection.

Watch out for

  • From the start the post-incident review establishes, not the first alert: an incident found late started before it was found.
  • When incidents are few, one late find moves the average a lot.

Good for

  • Valuing AI threat detection by the breach exposure hours it cuts.

Guide: What a formula reads

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
it.software.annual_spend Total annual software license spend.

Watch out for

  • Where Roiva already books your software bills as costs, add up the same vendors, so the saving and the spend agree.
  • Formulas read it as the spend before the initiative, and book a month's share of the saving at a time.

Good for

  • The spend a fall in the unused license share is applied to.

Guide: What a formula reads

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
it.software.unused_license_share Percentage of purchased licenses that are unassigned or unused.

Watch out for

  • Say whether unused means unassigned or idle, and over how long: a 30-day and a 90-day idle test give very different numbers.
  • The share also falls when idle seats are reassigned, which saves money only if it avoids buying new ones. A seat saves nothing until it's canceled or not renewed.

Good for

  • Valuing license discovery by the seats you stop paying for.

Guide: What a formula reads

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Pipedrive

Key What it measures Type Unit Fed by
pipedrive.deals.avg_demo_to_proposal_days Average days from the stage chosen as demo to the stage chosen as proposal, over the Pipedrive deals first entering the proposal stage in the month. Written to crm.deals.avg_demo_to_proposal_days instead when this connection is the primary source. duration days Pipedrive
pipedrive.deals.avg_sales_cycle_days Pipedrive's average sales cycle for the month. Written to crm.deals.avg_sales_cycle_days instead when this connection is the primary source. duration days Pipedrive
pipedrive.deals.avg_value Pipedrive's average won deal size for the month. Written to crm.deals.avg_value instead when this connection is the primary source. currency USD Pipedrive
pipedrive.deals.lost_count Pipedrive's lost deals for the month. Written to crm.deals.lost_count instead when this connection is the primary source. count deals Pipedrive
pipedrive.deals.open_count Deals open in Pipedrive at sync time, a snapshot rather than a period total, written once per month and rewritten on every sync. count — Pipedrive
pipedrive.deals.open_value Value of the deals open in Pipedrive at sync time, a snapshot rather than a period total, written once per month and rewritten on every sync. currency USD Pipedrive
pipedrive.deals.proposals_sent_count Pipedrive deals in pipelines that count as sales that first entered the stage chosen as proposal in the month, from each deal's changelog. Written to crm.proposals.sent_count instead when this connection is the primary source. count proposals Pipedrive
pipedrive.deals.renewal_arr_due The value of the Pipedrive deals chosen as renewals that fell due in the month, whatever became of them. Written to customer_success.renewals.arr_up_for_renewal instead when this connection is the primary source. currency USD Pipedrive
pipedrive.deals.renewal_rate Of the Pipedrive deals chosen as renewals that fell due in the month, the share won, by number or by value. Written to customer_success.renewals.rate instead when this connection is the primary source. percentage % Pipedrive
pipedrive.deals.renewal_rate_base The Pipedrive deals chosen as renewals that fell due in the month, by number or by value as the renewal rate counts them. Written to customer_success.renewals.rate_base instead when this connection is the primary source. count — Pipedrive
pipedrive.deals.services_proposals_count Pipedrive deals in pipelines ticked as services that first entered the stage chosen as proposal in the month. Written to professional_services.proposals.submitted_count instead when this connection is the primary source. count proposals Pipedrive
pipedrive.deals.win_rate Pipedrive's deal win rate for the month. Written to crm.deals.win_rate instead when this connection is the primary source. percentage % Pipedrive
pipedrive.deals.won_count Pipedrive's won deals for the month. Written to crm.deals.won_count instead when this connection is the primary source. count deals Pipedrive
pipedrive.deals.won_value Pipedrive's won deal value for the month. Written to crm.deals.won_value instead when this connection is the primary source. currency USD Pipedrive

Energy

Key What it measures Type Unit Fed by
energy.assets.critical_count Critical assets in service: the transformers, breakers, lines and generating units whose failures count as unplanned outages.

Watch out for

  • The assets whose failures you count as unplanned outages, on the same definition each time, so a reclassification doesn't read as a change in reliability.

Good for

  • The base unplanned outages are compared per, so a larger asset base isn't read as worse reliability.

Guide: When the count moves with how busy you are

count assets Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
energy.assets.deferred_capex Asset replacement spend deferred by extending asset life.

Watch out for

  • The replacement spend actually pushed out, not the asset's whole replacement value.
  • Deferring spend isn't saving it: the formula values it at your cost of capital, for each month it stays deferred.

Good for

  • Valuing predictive maintenance by what deferring replacement spend is worth.

Guide: What a formula reads

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
energy.assets.unplanned_outages_per_year Unplanned critical-asset outage events per year.

Watch out for

  • Unplanned outages on critical assets only: planned work is what the initiative is trying to shift them into.
  • A year's count, compared per critical asset. The formula spreads it evenly over the year's months.

Good for

  • Valuing predictive maintenance by the outage costs it avoids.

Guide: When the count moves with how busy you are

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
energy.grid.peak_demand_mw Peak system demand the reserve is held against, in megawatts.

Watch out for

  • Forecast or metered peak, from the same operating plan as the reserve, the same one on both sides.

Good for

  • The base reserve capacity is compared per, so a higher peak isn't read as worse forecasting.

Guide: When the count moves with how busy you are

count MW Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
energy.grid.reserve_capacity_mw Reserve generation capacity held, in megawatts.

Watch out for

  • Say whether it's the regulatory requirement or what you actually hold, and use the same one on both sides.
  • It's compared per megawatt of peak demand, so record peak demand over the same periods.

Good for

  • Valuing AI demand forecasting by the reserve capacity it lets you stop holding.

Guide: When the count moves with how busy you are

count MW Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
energy.load_forecasting.mape Mean absolute percentage error of the load forecast. percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
energy.rate_cases.count Rate cases and major regulatory filings prepared in the period.

Watch out for

  • Filings prepared, not hearings or supplements to a filing already counted.

Good for

  • The filings internal and outside counsel hours saved are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
energy.rate_cases.external_counsel_hours Outside counsel hours billed for a rate case or major regulatory filing.

Watch out for

  • Hours outside counsel billed for one filing, from their invoices. Internal hours are Staff Prep Hours per Rate Case.

Good for

  • Valuing AI-assisted drafting by the outside counsel fees it saves, in cash.

Guide: What a formula reads

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
energy.rate_cases.prep_hours Internal staff hours to prepare a rate case or major regulatory filing.

Watch out for

  • Internal hours for one filing only: outside counsel and consultants are their own cost.
  • The hours freed are freed capacity, not cash: staff are paid the same afterwards.

Good for

  • Valuing AI drafting by the internal preparation time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
energy.reliability.saidi_minutes System Average Interruption Duration Index, in outage minutes per customer per year. duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
energy.storms.call_volume_index Inbound call volume during a typical major storm event.

Watch out for

  • Calls in a typical major storm: take the median event, not the worst.
  • Storm severity, how many customers lose power and how long restoration takes move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing proactive outage notifications by the storm calls they keep from reaching agents.

Guide: When the outcome has other causes

count index Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
energy.storms.events_per_year Major storm or high-call-volume outage events per year.

Watch out for

  • Use your own threshold for major and say what it is: it decides both this count and the storm call figures.
  • A year's count: the formulas spread it evenly over the year's months.

Good for

  • The storms calls and notification hours saved per event are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
energy.storms.notification_hours_per_event Staff hours spent communicating outage status to customers during one major storm event.

Watch out for

  • Hours across every team that sent updates in one major storm, taking the median event.
  • The hours freed are freed capacity, not cash: staff are paid the same afterwards.

Good for

  • Valuing automated outage notifications by the staff time they free.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Hospitality

Key What it measures Type Unit Fed by
hospitality.bookings.direct_share Share of bookings made through direct channels.

Watch out for

  • The formula applies its change to room revenue, so measure it as a share of room revenue. A share of bookings holds only where direct and OTA stays are alike in rate and length.
  • Loyalty programs, OTA ranking and promotions, and guest mix move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing direct booking tools by the OTA commission saved on room revenue.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hospitality.food_beverage.cost_of_goods Food and beverage cost of goods in the period.

Watch out for

  • The same food and beverage cost of goods the waste share is measured against.

Good for

  • The cost a fall in the waste share is applied to.

Guide: When the outcome has other causes

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hospitality.food_beverage.covers_served Covers served in the period.

Watch out for

  • Covers for the same outlets and meal periods the prep hours cover.

Good for

  • The volume prep hours are compared per, so a busy month isn't read as slower prep.

Guide: When the count moves with how busy you are

count covers Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hospitality.food_beverage.prep_hours Kitchen prep hours in the period.

Watch out for

  • Rostered prep hours, not scheduled ones, and prep only, not service.
  • It's compared per cover served, so record covers over the same periods.

Good for

  • Valuing AI demand forecasting by the prep time it frees.

Guide: When the count moves with how busy you are

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hospitality.food_beverage.waste_share Food cost lost to waste and spoilage as a percentage of food and beverage cost of goods.

Watch out for

  • Say whether it comes from waste logs or from theoretical against actual usage: the two rarely agree. Use the same before and after.
  • Demand swings, menu changes, events and supplier quality move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing AI par and prep forecasting by the food cost no longer wasted.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hospitality.guests.ancillary_spend_per_guest Average ancillary spend per guest.

Watch out for

  • Non-room spend over the same guest count you record, guests or stays, not one of each.
  • Guest mix, seasonality, prices and what the property offers move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing personalized offers by the extra spend per guest, at your contribution margin.

Guide: When the outcome has other causes

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hospitality.guests.count Guests in the period.

Watch out for

  • Say whether it's guests or stays: a family of four on one reservation is one stay and four guests.

Good for

  • The guests a rise in ancillary spend per guest is applied to.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hospitality.housekeeping.minutes_per_room_turn Housekeeping labor minutes per room turn.

Watch out for

  • Include walking between rooms and trolley set-up: they're real minutes, and dispatch changes the walking most.
  • The minutes freed are freed capacity, not cash. If rostered hours fall, record that it became cash.

Good for

  • Valuing AI housekeeping dispatch by the labor time it frees per turn.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hospitality.housekeeping.rooms_turned Room turns (checkouts and cleans) in the period.

Watch out for

  • Checkouts cleaned and readied. Count stayover services separately or not at all, the same way before and after.

Good for

  • The turns minutes saved per turn are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hospitality.revenue_management.hours_per_week Revenue manager hours per week on rate setting and reporting.

Watch out for

  • Rate setting, competitor checks and reporting, not the meeting where rates are presented.
  • The hours freed are freed capacity, not cash: revenue managers are paid the same afterwards.

Good for

  • Valuing automated rate decisions by the revenue manager time they free.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hospitality.rooms.available_room_nights Available room-nights in the period.

Watch out for

  • Rooms available times nights, with rooms out of order taken out, or RevPAR reads low for a reason no initiative fixes.

Good for

  • The room nights a rise in RevPAR is applied to.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hospitality.rooms.revenue Total room revenue across all channels in the period.

Watch out for

  • Rooms only, across every channel: food, beverage and ancillary are their own figures.

Good for

  • The revenue a rise in the direct booking share is applied to, at the OTA commission rate.

Guide: When the outcome has other causes

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
hospitality.rooms.revpar Revenue per available room.

Watch out for

  • Room revenue over available room nights, from the same figures you record, or the PMS's own figure with both taken from the same report.
  • Seasonality, local events, market demand and new supply move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing AI revenue management by the extra room revenue, at your contribution margin.

Guide: When the outcome has other causes

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Education

Key What it measures Type Unit Fed by
education.admissions.admitted_count Applicants offered admission in the period.

Watch out for

  • Offers made, not deposits or enrollments, since the yield rate is a share of the offers.

Good for

  • The admitted students a change in yield is applied to.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
education.admissions.applicant_count Applicants in the enrollment cycle. count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
education.admissions.decision_days Average days from application received to decision. duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
education.admissions.review_minutes_per_application Average staff minutes of hands-on work to review and process one application — not the days it waits for a decision.

Watch out for

  • Hands-on minutes reviewing and processing one application, not the days it waits for a decision.
  • Time complete applications of the usual kind, including the minutes spent checking what the AI summarized.

Good for

  • Valuing application screening by the admissions staff time it frees.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
education.admissions.yield_rate Percentage of admitted students who enroll.

Watch out for

  • Measure enrollment at the same point each cycle, census day rather than deposits, since students who deposit don't all arrive.
  • Financial aid, competing institutions, demographics and the economy move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing personalized admissions outreach by the extra students who enroll, at your contribution margin.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
education.courses.dfw_rate Percentage of students earning a D or F or withdrawing.

Watch out for

  • Keep withdrawals in, and take both halves from the same census.
  • The formula prices only the share of D, F and W students who then leave, at your attrition figure, not every one of them.
  • Instructors, grading policy, course design and each cohort's preparation move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing AI tutoring by the tuition kept from students who would have failed or withdrawn and left.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
education.enrollment.retention_rate Year-over-year student retention rate.

Watch out for

  • Say which cohort: first-year to second-year retention isn't the same as all students. Keep it the same before and after.
  • Students' finances and aid, cohort preparation, advising staff and the job market move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing student outreach by the tuition kept from students who return, at your contribution margin.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
education.enrollment.student_count Students enrolled in the affected courses or programs.

Watch out for

  • Students in the courses or programs the initiative covers, not the institution, taken at census day.
  • Retention, outreach, DFW and tutoring formulas all read it, so an initiative records one count for all of them: the students it covers.

Good for

  • The students a change in retention, DFW rate or tutoring cost is applied to.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
education.faculty.admin_hours_per_week Faculty hours per week on course administration.

Watch out for

  • Course administration per faculty member, such as posting materials and chasing submissions, not teaching or preparation.
  • The hours freed are freed capacity, not cash: faculty are paid the same afterwards.

Good for

  • Valuing course-operations tools by the faculty time they free.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
education.faculty.count Faculty using AI course-operations tools.

Watch out for

  • Faculty actually using the tools, from the tool's own usage. The department roster includes people the hours saved don't apply to.

Good for

  • The faculty admin hours saved per person are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
education.outreach.hours_per_week Staff hours per week spent identifying at-risk students and reaching out to them.

Watch out for

  • Finding at-risk students as well as contacting them, in a normal week.
  • It's compared per enrolled student, so record students in the same programs over the same months.

Good for

  • Valuing AI-assisted advising by the outreach time it frees.

Guide: When the count moves with how busy you are

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
education.tutoring.cost_per_student Cost of tutoring services per enrolled student in the period.

Watch out for

  • Include staffed and contracted tutoring both.
  • The formula multiplies it by the students the tutoring program covers, so divide by that same count.

Good for

  • Valuing AI tutoring by the tutoring spend it saves.

Guide: What a formula reads

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Operations

Key What it measures Type Unit Fed by
operations.applications.processed_count Applications reviewed and decided in the period.

Watch out for

  • Decisions reached, not applications received.
  • Admissions and credit underwriting formulas both read it. An initiative records one count: the applications it covers.

Good for

  • The applications review time saved, or extra approvals, are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
operations.documents.data_entry_minutes_per_document Staff minutes to key the data from one document.

Watch out for

  • Keying and checking one document of the kind the initiative handles, including correcting what the AI extracted, and no waiting.

Good for

  • Valuing AI extraction by the data entry time it frees.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
operations.documents.extraction_accuracy Percentage of extracted fields that are correct without human correction. percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
operations.documents.processed_count Documents (PDFs, forms, emails) processed by the extraction pipeline in the period.

Watch out for

  • Documents, not pages or fields, with a re-submission counted again if it was handled again.

Good for

  • The documents data entry minutes saved, or fewer errors, are multiplied by.

Guide: What a formula reads

count documents Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
operations.forecasting.accuracy Accuracy of the operating forecast (cash flow, demand or pipeline) against actuals.

Watch out for

  • 100 minus the average absolute error against actuals, at the same horizon every time: a month-ahead and a quarter-ahead forecast are different measures.
  • Volatile cash flows, one-off payments, slipping deals and team changes move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing AI forecasting by the idle cash it frees or the forecast misses it avoids.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
operations.forecasting.cycles Forecast cycles run in the period.

Watch out for

  • The versions your planning process actually produced in the period, monthly, weekly or ad hoc.

Good for

  • The cycles hours saved per cycle are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
operations.forecasting.hours_per_cycle Staff hours to produce one complete forecast cycle.

Watch out for

  • Everyone's hours on one complete cycle, gathering, building and reviewing, from the cycle's owner.
  • The hours freed are freed capacity, not cash: staff are paid the same afterwards.

Good for

  • Valuing forecast automation by the time it frees each cycle.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
operations.processing.error_rate Percentage of processed items that contained an error or needed manual correction.

Watch out for

  • From your QA sample or rework log. Where only a sample is checked, keep the sample the same size before and after.

Good for

  • With documents processed, the rework fewer errors free.

Guide: Cash value and freed capacity

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
operations.reporting.delivery_lag_days Days from period close to delivery of the period's report.

Watch out for

  • Days from period end to the pack reaching its readers, by the date it was distributed, not created.
  • The formula prices each day saved at a daily cost of investor inquiries you set, booked as freed capacity.

Good for

  • Valuing automated reporting by the days investors wait less.

Guide: Cash value and freed capacity

duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
operations.reporting.distribution_lag_hours Hours between data availability and report delivery to stakeholders. duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
operations.reporting.hours_per_week Staff hours per week spent compiling and distributing reports by hand. duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
operations.reporting.reports_automated Distinct report types fully automated and removed from manual production. count reports Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

NetSuite

Key What it measures Type Unit Fed by
netsuite.financials.expenses NetSuite's expenses for the month. Written to finance.pnl.expenses instead when this connection is the primary source. currency USD NetSuite
netsuite.financials.gross_margin NetSuite's gross margin for the month: Income accounts less COGS accounts, over Income accounts, for a month with a cost of goods sold. Written to retail.pricing.gross_margin instead when this connection is the primary source. percentage % NetSuite
netsuite.financials.net_income NetSuite's net income for the month: revenue and other income less expenses. Written to finance.pnl.net_income instead when this connection is the primary source. currency USD NetSuite
netsuite.financials.revenue NetSuite's revenue for the month: Income accounts, other income left out. Written to finance.pnl.revenue instead when this connection is the primary source. currency USD NetSuite
netsuite.invoices.count NetSuite's invoices created for the month. Written to finance.invoices.count instead when this connection is the primary source. count invoices NetSuite
netsuite.invoices.paid_count NetSuite's invoices paid for the month. Written to finance.invoices.paid_count instead when this connection is the primary source. count invoices NetSuite
netsuite.invoices.volume NetSuite's invoice volume for the month. Written to finance.invoices.volume instead when this connection is the primary source. currency USD NetSuite
netsuite.purchase_orders.count NetSuite purchase orders dated in the month, each counted once, other than canceled (Closed) ones. Read as supply_chain.procurement.purchase_orders_processed when this connection is the primary source for bills. count orders NetSuite
netsuite.suppliers.active_count NetSuite vendors on vendor bills and card charges in the trailing window to the month's end, less those the connection leaves out. Read as supply_chain.suppliers.active_count when this connection is the primary source for bills. count suppliers NetSuite
netsuite.vendor_bills.count NetSuite's bills received for the month. Written to finance.bills.count instead when this connection is the primary source. count bills NetSuite
netsuite.vendor_bills.volume NetSuite's bill volume for the month. Written to finance.bills.volume instead when this connection is the primary source. currency USD NetSuite

OpenAI

Key What it measures Type Unit Fed by
openai.usage.cost_per_million_output_tokens OpenAI's spend for the month over the output tokens it generated, per million. Written to ai.cost.per_million_output_tokens instead when this connection is the primary source for AI usage. currency USD OpenAI
openai.usage.cost_per_request OpenAI's spend for the month over the requests it served. Written to ai.cost.per_request instead when this connection is the primary source for AI usage. currency USD OpenAI
openai.usage.input_tokens OpenAI's input tokens for the month. Written to ai.usage.input_tokens instead when this connection is the primary source. count tokens OpenAI
openai.usage.model_input_tokens Input tokens sent to one OpenAI model in the month; the model is in the observation's dimensions. count tokens OpenAI
openai.usage.model_output_tokens Output tokens generated by one OpenAI model in the month; the model is in the observation's dimensions. count tokens OpenAI
openai.usage.model_request_count API requests to one OpenAI model in the month; the model is in the observation's dimensions. count requests OpenAI
openai.usage.model_total_tokens Input plus output tokens for one OpenAI model in the month; the model is in the observation's dimensions. count tokens OpenAI
openai.usage.output_tokens OpenAI's output tokens for the month. Written to ai.usage.output_tokens instead when this connection is the primary source. count tokens OpenAI
openai.usage.request_count OpenAI's API requests for the month. Written to ai.usage.request_count instead when this connection is the primary source. count requests OpenAI
openai.usage.spend OpenAI's usage spend for the month. currency USD OpenAI
openai.usage.total_tokens OpenAI's total tokens for the month. Written to ai.usage.total_tokens instead when this connection is the primary source. count tokens OpenAI

QuickBooks

Key What it measures Type Unit Fed by
quickbooks.bills.count QuickBooks's bills received for the month. Written to finance.bills.count instead when this connection is the primary source. count bills QuickBooks
quickbooks.bills.volume QuickBooks's bill volume for the month. Written to finance.bills.volume instead when this connection is the primary source. currency USD QuickBooks
quickbooks.financials.expenses QuickBooks's expenses for the month. Written to finance.pnl.expenses instead when this connection is the primary source. currency USD QuickBooks
quickbooks.financials.gross_margin QuickBooks's gross margin for the month: income less cost of goods sold, over income, for a month with a cost of goods sold. Written to retail.pricing.gross_margin instead when this connection is the primary source. percentage % QuickBooks
quickbooks.financials.net_income QuickBooks's net income for the month. Written to finance.pnl.net_income instead when this connection is the primary source. currency USD QuickBooks
quickbooks.financials.revenue QuickBooks's revenue for the month. Written to finance.pnl.revenue instead when this connection is the primary source. currency USD QuickBooks
quickbooks.invoices.count QuickBooks's invoices created for the month. Written to finance.invoices.count instead when this connection is the primary source. count invoices QuickBooks
quickbooks.invoices.paid_count QuickBooks's invoices paid for the month. Written to finance.invoices.paid_count instead when this connection is the primary source. count invoices QuickBooks
quickbooks.invoices.volume QuickBooks's invoice volume for the month. Written to finance.invoices.volume instead when this connection is the primary source. currency USD QuickBooks
quickbooks.purchase_orders.count QuickBooks purchase orders dated in the month, each counted once, open or closed. Read as supply_chain.procurement.purchase_orders_processed when this connection is the primary source for bills. count orders QuickBooks
quickbooks.suppliers.active_count QuickBooks vendors on bills and purchases in the trailing window to the month's end, less those the connection leaves out. Read as supply_chain.suppliers.active_count when this connection is the primary source for bills. count suppliers QuickBooks

Xero

Key What it measures Type Unit Fed by
xero.bills.count Xero's bills received for the month. Written to finance.bills.count instead when this connection is the primary source. count bills Xero
xero.bills.volume Xero's bill volume for the month. Written to finance.bills.volume instead when this connection is the primary source. currency USD Xero
xero.financials.expenses Xero's expenses for the month. Written to finance.pnl.expenses instead when this connection is the primary source. currency USD Xero
xero.financials.gross_margin Xero's gross margin for the month: revenue less cost of sales, over revenue, for a month with a cost of sales. Written to retail.pricing.gross_margin instead when this connection is the primary source. percentage % Xero
xero.financials.net_income Xero's net income for the month. Written to finance.pnl.net_income instead when this connection is the primary source. currency USD Xero
xero.financials.revenue Xero's revenue for the month. Written to finance.pnl.revenue instead when this connection is the primary source. currency USD Xero
xero.invoices.count Xero's invoices created for the month. Written to finance.invoices.count instead when this connection is the primary source. count invoices Xero
xero.invoices.paid_count Xero's invoices paid for the month. Written to finance.invoices.paid_count instead when this connection is the primary source. count invoices Xero
xero.invoices.volume Xero's invoice volume for the month. Written to finance.invoices.volume instead when this connection is the primary source. currency USD Xero
xero.purchase_orders.count Xero purchase orders dated in the month, each counted once, other than drafts and deleted ones. Read as supply_chain.procurement.purchase_orders_processed when this connection is the primary source for bills. count orders Xero
xero.suppliers.active_count Xero contacts on bills and spend money in the trailing window to the month's end, less those the connection leaves out. Read as supply_chain.suppliers.active_count when this connection is the primary source for bills. count suppliers Xero

Claude Enterprise

Key What it measures Type Unit Fed by
claude_enterprise.chat.monthly_active_users Members who sent at least one message in Claude chat in the month, in claude.ai chat or Chat and Cowork unified, each counted once however many days they chatted. Members who only used Claude Code or Cowork sessions aren't counted. Read as ai_assistant.users.monthly_active when Claude Enterprise is the primary AI-assistant source. count users Claude Enterprise
claude_enterprise.claude_code.acceptance_rate Accepted tool actions as a share of those proposed across the Claude Enterprise organization for the month. Read as engineering.coding_assistant.acceptance_rate when Claude Enterprise is the primary coding-assistant source. percentage % Claude Enterprise
claude_enterprise.claude_code.acceptances Edits and writes developers accepted in Claude Code across the Claude Enterprise organization for the month — one accepted tool action each, the same unit as Anthropic's Claude Code report. Read as engineering.coding_assistant.acceptances when Claude Enterprise is the primary coding-assistant source.

Watch out for

  • One accepted edit or write each, a smaller unit than a Copilot acceptance, so the two are priced at separate minutes saved, never added up.
  • It covers Claude Code on Claude Enterprise seats only. If your developers also use Claude Code through an Anthropic Console organization, Roiva reads one of the two, not both.

Good for

  • Valuing Claude Code's share of the time saved where a team on Claude Enterprise runs it beside GitHub Copilot, each at its own minutes saved per acceptance.

Guide: The units aren't the same across tools

count acceptances Claude Enterprise
claude_enterprise.claude_code.active_users_avg Members who used Claude Code on an average day in the month — members active only in other Claude products are not counted. Read as engineering.coding_assistant.active_users when Claude Enterprise is the primary coding-assistant source. average users Claude Enterprise
claude_enterprise.claude_code.lines_accepted Lines of code Claude Code added across the Claude Enterprise organization in the month. Read as engineering.coding_assistant.lines_accepted when Claude Enterprise is the primary coding-assistant source. count lines Claude Enterprise
claude_enterprise.claude_code.proposals Tool actions Claude Code proposed across the Claude Enterprise organization in the month, accepted and rejected together. Read as engineering.coding_assistant.suggestions when Claude Enterprise is the primary coding-assistant source. count suggestions Claude Enterprise
claude_enterprise.seats.assigned Seats assigned to members of the Claude Enterprise organization, as of the month's last day in Anthropic's analytics. Pending invitations aren't counted. Read as ai_assistant.seats.licensed when Claude Enterprise is the primary AI-assistant source. count users Claude Enterprise
claude_enterprise.usage.spend What a Claude Enterprise organization's usage cost for the month, from Anthropic's cost report. Seat fees aren't in it. currency USD Claude Enterprise
claude_enterprise.users.active_30d Members Anthropic's organization summary counts as monthly active users: those with token use in the rolling 30 days to the month's last available day in Anthropic's analytics, the same day as the month's assigned seats. Not only chat, so a member who only used Claude Code counts; a rolling 30 days, not the calendar month, so it isn't monthly chat users. count users Claude Enterprise

Engineering

Key What it measures Type Unit Fed by
engineering.coding_assistant.acceptance_rate Accepted suggestions as a share of the suggestions offered in the period. percentage % Anthropic, Claude Enterprise, Gemini Code Assist, GitHub
engineering.coding_assistant.acceptances Suggestions developers accepted from an AI coding assistant in the period.

Watch out for

  • Copilot counts acceptance activities, which can bundle several completions, Claude Code single accepted edits, and Gemini Code Assist suggestions as Google records them. Totals from two tools aren't like for like.
  • It counts code a developer took, not code that shipped: an accepted suggestion can be rewritten or deleted later.
  • It rises with the number of developers using the tool. Read it beside active users before calling a rise better use.

Good for

  • Valuing the developer time an assistant saves: accepted suggestions × minutes saved per acceptance × your developer hourly rate.
  • Following adoption after a rollout, beside active users.

Guide: The units aren't the same across tools

count acceptances Anthropic, Claude Enterprise, Gemini Code Assist, GitHub
engineering.coding_assistant.active_users Developers using an AI coding assistant on an average day in the period. Summed across assistants when read combined. average users Anthropic, Claude Enterprise, GitHub
engineering.coding_assistant.lines_accepted Lines of code developers accepted from an AI coding assistant in the period.

Watch out for

  • A line is a poor measure of effort: a hard line and an easy one count the same, and a line the assistant wrote isn't always one anyone would have written.
  • Each tool counts its own lines: Copilot and Gemini Code Assist the lines accepted, Claude Code the lines it added, Codex the lines it added that developers kept. Don't add them up or compare them across tools.

Good for

  • Valuing the time saved by an assistant that reports no count of accepted suggestions, such as OpenAI's Codex. Where acceptances are reported, they're the better measure.

Guide: The units aren't the same across tools

count lines Anthropic, ChatGPT Enterprise, Claude Enterprise, Gemini Code Assist, GitHub
engineering.coding_assistant.self_reported_hours_saved_per_week The average hours a week developers who answered the month's survey say an AI coding assistant saved them. What developers report, not a measurement.

Watch out for

  • It's what developers report, not a measurement. In METR's 2025 trial, experienced developers felt 20% faster with AI tools while taking 19% longer.
  • Ask the same question the same way each month, and average the answers of those who replied. A changed question changes the figure.

Good for

  • Valuing the developer time an assistant frees where it reports no acceptances or lines, as freed capacity a person approves each month.

Guide: When the assistant reports nothing

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
engineering.coding_assistant.self_reported_users Developers who used an AI coding assistant in the month and answered the survey asking how many hours a week it saved them.

Watch out for

  • Count only developers who answered this month's survey, not every seat or everyone invited to it.
  • The developers who answer are often the keenest users, so the hours they report may run high for the team as a whole.

Good for

  • With hours saved per developer per week, what developers report an assistant saves them, where the assistant itself reports nothing Roiva can read.

Guide: When the assistant reports nothing

count developers Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
engineering.coding_assistant.suggestions Code suggestions an AI coding assistant offered developers in the period. count suggestions Anthropic, Claude Enterprise, Gemini Code Assist, GitHub
engineering.pull_requests.cycle_time_avg Average hours from a pull request being opened to being merged, over the pull requests merged in the period. duration hours GitHub
engineering.pull_requests.merged_count Pull requests merged in the period.

Watch out for

  • From GitHub, pull requests merged in the repositories the connection can read. Adding repositories later raises it with no change in output.
  • Where the assistant reports accepted suggestions or lines of code, an initiative gets a formula on those counts instead, because they come from the assistant itself.
  • Many things move it besides the assistant, so the formula takes only the share of the change you credit to the initiative.

Good for

  • Valuing an assistant that reports no usage figures, by the extra pull requests merged.

Guide: Best practices

count pull requests GitHub

Media

Key What it measures Type Unit Fed by
media.audience.avg_active_users Average monthly active users on the service.

Watch out for

  • Average the months rather than taking the best one, and say what makes a user active.

Good for

  • The users a rise in watch hours per user is applied to.

Guide: When the outcome has other causes

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
media.audience.monthly_watch_hours_per_user Average monthly watch hours per active user.

Watch out for

  • Watch hours over the same active users you record. A figure per viewing user is a smaller denominator and a larger number.
  • New releases, seasonality, live sports and competing services move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing AI recommendations by the extra ad revenue from more watching, at your contribution margin.

Guide: When the outcome has other causes

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
media.localization.cost_per_episode Average cost to localize one episode (dubbing, subtitling and QC).

Watch out for

  • Dubbing or subtitling, QC and project management, per the same unit as episodes localized, per episode or per episode-language.

Good for

  • Valuing AI-assisted dubbing and subtitling by the localization cost saved.

Guide: What a formula reads

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
media.localization.episodes_localized Episodes localized into additional languages in the period.

Watch out for

  • Count episode-language pairs when you localize into several languages: one episode into five languages is five pieces of work.

Good for

  • The episodes a fall in localization cost is multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
media.post_production.days_per_episode Days from picture lock to delivery per episode.

Watch out for

  • Calendar days from picture lock to delivery, waiting on notes and QC included.
  • The formula prices each day saved at a daily post-production cost you set, which holds only if spend falls with the days.

Good for

  • Valuing AI-assisted post by the production days it saves.

Guide: What a formula reads

duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
media.post_production.episode_count Episodes entering post-production in the period.

Watch out for

  • Episodes entering post, not projects or seasons.

Good for

  • The episodes days saved per episode are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
media.rights.admin_hours_per_week Staff hours per week on rights tracking, metadata correction and catalog reconciliation.

Watch out for

  • Tracking, metadata correction and catalog reconciliation, not negotiating rights.
  • The hours freed are freed capacity, not cash: staff are paid the same afterwards.

Good for

  • Valuing AI rights management by the admin time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
media.royalties.dispute_rate Share of royalty statements requiring dispute or correction.

Watch out for

  • A statement counts once however many lines were wrong.

Good for

  • With statements issued, the disputes AI-verified rights data avoids.

Guide: Cash value and freed capacity

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
media.royalties.statement_count Royalty statements issued in the period.

Watch out for

  • Statements, not payees or line items.

Good for

  • The statements a fall in the dispute rate is applied to.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

AI

Key What it measures Type Unit Fed by
ai.cost.per_million_output_tokens All-in AI spend per million output tokens: everything the provider billed in the period (input, cache and output charges) divided by the output tokens it generated.

Watch out for

  • One provider's whole bill, input and cache charges included, over its output tokens only. Longer prompts with the same answers raise it with no change in price.
  • Take spend and tokens from the same provider and month: never one provider's spend over another's tokens.

Good for

  • Valuing AI spend optimization for a provider that reports tokens but no request count, such as Anthropic.

Guide: What moved an AI spend saving

currency USD Anthropic, OpenAI
ai.cost.per_request All-in AI spend per API request: everything the provider billed in the period divided by the requests made to it.

Watch out for

  • One provider's whole bill over the requests made to it. A change in what one request does, such as longer prompts or batching, moves it with no change in price.
  • Take spend and requests from the same provider and month.

Good for

  • Valuing AI spend optimization for a provider that reports request counts, such as OpenAI.

Guide: What moved an AI spend saving

currency USD OpenAI
ai.cost.spend Total AI provider spend in the period — the AI-cost source's, or every one's added up when the account reads them combined. currency USD Anthropic, GitHub, OpenAI
ai.outputs.human_review_rate Percentage of AI outputs routed to a human reviewer before being acted on. percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
ai.usage.input_tokens Total prompt/input tokens consumed across the primary AI provider in the period. count tokens Anthropic, OpenAI
ai.usage.output_tokens Total completion/output tokens generated by the primary AI provider in the period.

Watch out for

  • Output tokens only, from the provider's usage report: input and cache tokens aren't in it, though their charges are in the cost per million.

Good for

  • The tokens a fall in cost per million output tokens is applied to.

Guide: What moved an AI spend saving

count tokens Anthropic, OpenAI
ai.usage.request_count Number of API requests made to the primary AI provider in the period. OpenAI and Gemini report it; Anthropic reports tokens but no request count.

Watch out for

  • OpenAI and Gemini report it. Anthropic reports tokens but no requests, so for Claude count the Messages API calls from your own logs.

Good for

  • The requests a fall in cost per request is applied to.

Guide: What moved an AI spend saving

count requests OpenAI
ai.usage.total_tokens Total tokens (input + output) used through the primary AI provider in the period. count tokens Anthropic, OpenAI

Government

Key What it measures Type Unit Fed by
government.cases.decision_days Average days from case intake to eligibility decision. duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
government.cases.processed_count Benefit applications and cases processed in the period.

Watch out for

  • Decisions reached, not cases received, so a growing backlog doesn't read as throughput.

Good for

  • The cases caseworker hours saved per case are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
government.cases.staff_hours_per_case Average caseworker hours of hands-on work per case decided — not the days it waits for a decision.

Watch out for

  • Hours of eligibility work per case decided, not the days a case waits.
  • The hours freed are freed capacity, not cash: caseworkers are paid the same afterwards.

Good for

  • Valuing AI-assisted extraction and routing by the caseworker time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
government.constituent_services.calls_per_month Inbound constituent contacts per month.

Watch out for

  • Contacts that reached staff; abandoned calls are a separate measure.
  • Call deflection and staff redeployment value the same calls no longer made, so an initiative gets one of the two.

Good for

  • Valuing AI self-service by the calls it keeps from reaching staff, compared per service request.

Guide: Formulas that count the same money

count calls Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
government.constituent_services.requests_per_month Constituent service requests per month across every channel (phone, web, app and chat), as the 311 or case system counts them.

Watch out for

  • Every channel, as the 311 or case system counts them, for the same request types the calls cover.
  • A request that moves from the phone to the web still counts, which is what lets calls per request fall.

Good for

  • The volume calls are compared per, so a busy month isn't read as worse self-service.

Guide: When the count moves with how busy you are

count requests Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
government.permits.decision_days Average days from permit intake to decision. duration days Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
government.permits.processed_count Permit applications processed in the period.

Watch out for

  • Decisions, not submissions.

Good for

  • The permits staff hours saved per permit are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
government.permits.staff_hours_per_permit Average staff hours of hands-on work per permit decided — not the days it waits for a decision.

Watch out for

  • Hours worked per permit decided, intake, plan review and scheduling, not the days a permit waits.
  • The hours freed are freed capacity, not cash: staff are paid the same afterwards.

Good for

  • Valuing AI-assisted plan review by the staff time it frees.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Telecom

Key What it measures Type Unit Fed by
telecom.customer_care.calls_per_month Inbound customer care calls per month.

Watch out for

  • It's compared per contact across every channel, so a surge after an outage isn't read as a loss. Record contacts over the same months.
  • Care cost per subscriber on the same template already falls by the agent time deflection saves, so an initiative gets one of the two.

Good for

  • Valuing AI deflection bots by the calls they keep from reaching agents.

Guide: Formulas that count the same money

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
telecom.customer_care.contacts_per_month Customer care contacts per month across every channel: calls, chat, app and self-service.

Watch out for

  • Every channel, calls, chat, app and self-service sessions that ended a question. Say which sessions count and keep that rule.

Good for

  • The volume calls are compared per, so a surge in contacts isn't read as worse deflection.

Guide: When the count moves with how busy you are

count contacts Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
telecom.customer_care.cost_per_subscriber Customer care cost per subscriber per period.

Watch out for

  • One month's fully loaded care cost over the subscribers served.
  • Outages, billing changes, device launches and subscriber mix move it too, so a formula valuing its change takes only the share of it you credit to the initiative.
  • Care call deflection on the same template values part of the same fall, so an initiative gets one of the two.

Good for

  • Valuing deflection and agent assist by the care cost saved per subscriber.

Guide: Formulas that count the same money

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
telecom.field_service.truck_rolls_per_month Field dispatch events per month.

Watch out for

  • Dispatches, not jobs: a repeat visit to the same fault is another truck roll.
  • It's compared per active subscriber, so a growing base isn't read as more failures. Record subscribers over the same months.

Good for

  • Valuing remote diagnostics by the truck rolls they avoid.

Guide: When the count moves with how busy you are

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
telecom.network.incident_count Network incidents in the period.

Watch out for

  • Incidents, not alarms: one fault raising forty alarms is one incident.

Good for

  • The incidents minutes of recovery saved are multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
telecom.network.mttr_minutes Mean minutes to recover from a network incident.

Watch out for

  • From an incident starting to service restored, not the ticket closed or the root cause fixed, using the same two events on both sides.

Good for

  • Valuing AIOps by the cost of the incident hours it saves.

Guide: What a formula reads

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
telecom.retention.cost_per_retained_subscriber Average cost of retention offers per subscriber retained.

Watch out for

  • The save team's cost plus credits and discounts given, divided by the subscribers retained.
  • A lower cost per save can come from saving easier customers. Read it beside how many were retained.

Good for

  • Valuing AI-targeted save offers by the retention cost they save.

Guide: What a formula reads

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
telecom.retention.retained_subscribers Subscribers retained through save offers in the period.

Watch out for

  • Subscribers who were at risk, in a save queue or having given notice, and stayed, counted in the month saved.

Good for

  • The saves a fall in cost per retained subscriber is multiplied by.

Guide: What a formula reads

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Zendesk

Key What it measures Type Unit Fed by
zendesk.knowledge_base.article_count Zendesk Help Center articles that are published and that customers can see, as they stand at the month's last sync, from the month the connection was made. Read as support.knowledge_base.article_count when this connection is the primary source for support tickets. count articles Zendesk
zendesk.tickets.agent_resolved_count Zendesk's solved and closed tickets for the month that people handled and that no AI deflection rule on the connection matched, plus the ones a rule matched whose customer came back. Read as support.tickets.agent_resolved_count when this connection is the primary source. count tickets Zendesk
zendesk.tickets.ai_agent_resolved_count Zendesk's solved and closed tickets for the month that an AI agent or bot handled alone in Zendesk messaging, which Zendesk calls AI agent tickets. A ticket a person was brought into isn't one, and one that only got a default messaging response or that the customer left is, since Zendesk solves those four days after the last message. Zendesk's other ticket counts here leave these out, as its own views and reports do. count tickets Zendesk
zendesk.tickets.avg_first_response_hours Zendesk's average ticket first response time for the month, over the tickets people handled. Written to support.tickets.avg_first_response_hours instead when this connection is the primary source. duration hours Zendesk
zendesk.tickets.avg_resolution_days Zendesk's average time from a ticket's creation to its solve for the month, over the tickets people handled, in days (the computer converts from hours). Read as support.tickets.avg_resolution_days when this connection is the primary source. duration days Zendesk
zendesk.tickets.csat_score Zendesk's ticket CSAT score for the month, over the tickets people handled: an AI agent ticket's rating isn't in it. Written to support.tickets.csat_score instead when this connection is the primary source. score 0–1 Zendesk
zendesk.tickets.solved_count Zendesk's support tickets closed for the month, of those people handled: a ticket a messaging AI agent handled alone is counted in zendesk.tickets.ai_agent_resolved_count instead. Written to support.tickets.closed_count instead when this connection is the primary source. count tickets Zendesk
zendesk.tickets.volume Zendesk's support tickets opened for the month, of those people handled: a ticket a messaging AI agent handled alone isn't counted here, and once solved is counted in zendesk.tickets.ai_agent_resolved_count. Written to support.tickets.volume instead when this connection is the primary source. count tickets Zendesk

Billing

Key What it measures Type Unit Fed by
billing.charges.count Charges in the period that kept revenue, in the currency the account settles in.

Watch out for

  • From Stripe, charges that succeeded and kept something, in the account's own currency only: charges in other currencies are left out.

Good for

  • The transactions cloud spend is compared per, for a payments business.

Guide: When Stripe already counts it

count charges Stripe
billing.charges.volume What the period's charges kept, captured less refunded, tax included, in the currency the account settles in. currency USD Stripe
billing.subscriptions.active_count Subscriptions active at the end of the period.

Watch out for

  • A count at the end of the period, not an average over it.
  • Check whether your billing system counts trials, paused or past-due subscriptions as active, and keep that the same before and after.

Good for

  • The base a change in churn, or in a cost per subscriber, is multiplied by.

Guide: What a formula reads

count subscriptions Stripe
billing.subscriptions.churn_rate Subscriptions canceled in the period as a percentage of those active at its start. Absent for a period with nothing active at the start.

Watch out for

  • It moves with price, product and competitors, so a formula valuing its change takes only the share of it you credit to the initiative.
  • It's monthly. Compare it with a monthly baseline, never an annual rate.
  • On a small base a few cancellations move it a lot. Read it over several months.

Good for

  • Valuing churn prediction and save plays by the subscribers kept.

Guide: When the outcome has other causes

percentage % Stripe
billing.subscriptions.churned_count Subscriptions canceled in the period. count subscriptions Stripe
billing.subscriptions.mrr Monthly recurring revenue from subscriptions active at the end of the period.

Watch out for

  • A level at the end of the period, what active subscriptions bill each month, not the revenue the month earned.
  • It's revenue, not return: a formula books what you keep of it at your contribution margin.

Good for

  • Pricing a change in churn or net retention in recurring revenue kept.

Guide: Revenue is booked at your margin

currency USD Stripe
billing.subscriptions.new_count Subscriptions created in the period. count subscriptions Stripe

Customer success

Key What it measures Type Unit Fed by
customer_success.accounts.csm_count Customer success managers covering the accounts.

Watch out for

  • Count people carrying a book of accounts, not open roles, and a half-time CSM as a half.
  • The formula values more accounts per CSM as the CSM cost it saves, booked as freed capacity. If a hire is avoided or headcount falls, record that it became cash.

Good for

  • The CSMs a rise in accounts per CSM is applied to.

Guide: Cash value and freed capacity

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
customer_success.accounts.per_csm Customer accounts managed per customer success manager.

Watch out for

  • Divide by the same CSMs you record as CSM count, so the two figures agree.
  • It also rises when accounts get smaller or get less attention. Read it beside the renewal rate.

Good for

  • Valuing customer success automation by the extra accounts each CSM carries.

Guide: Cash value and freed capacity

count — Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
customer_success.renewals.arr_up_for_renewal Annual recurring revenue from accounts up for renewal in the period.

Watch out for

  • The contract value of renewals falling due in the period, not what has renewed yet.
  • It's revenue, not return: the formula books what you keep of it at your contribution margin, a month's share of the year at a time.

Good for

  • The base a change in the renewal rate is applied to.

Guide: Revenue is booked at your margin

currency USD HubSpot, Pipedrive, Salesforce
customer_success.renewals.rate Percentage of customers up for renewal who renewed.

Watch out for

  • Count contracts the same way on both sides, by number or by value, not one of each.
  • From a CRM, a renewal counts in the month it fell due, by its close date as the month began or a date field: early ones count then, late ones raise their month's rate, and open ones count as not renewed yet.
  • The product, pricing, customers' budgets and competitors move it too, so a formula valuing its change takes only the share of it you credit to the initiative.

Good for

  • Valuing proactive customer success by the recurring revenue kept, at your contribution margin.

Guide: When the outcome has other causes

percentage % HubSpot, Pipedrive, Salesforce
customer_success.renewals.rate_base The renewals due in the period as the renewal rate counts them, by number or by their value.

Watch out for

  • It's what the renewal rate is divided by, so it counts the same way the rate does: renewals, or their value.

Good for

  • Weighting renewal rates averaged over several months, so a busy month counts for more than a quiet one.

Guide: What a formula reads

count — HubSpot, Pipedrive, Salesforce
customer_success.revenue.cohort_starting_mrr The monthly recurring revenue the customers you had twelve months before the period were paying then, net revenue retention's denominator.

Watch out for

  • What those customers paid twelve months before the period, not what they pay now and not today's total MRR, which counts customers won since and is already lifted by the retention it measures.
  • Take it from the same cohort as the net revenue retention you record for the period, so the two agree.

Good for

  • The base a change in net revenue retention is applied to.

Guide: Revenue is booked at your margin

currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
customer_success.revenue.net_retention_rate Revenue retained from existing customers including expansion, as a percentage of the starting base.

Watch out for

  • This period's recurring revenue from the customers you had a year ago, over what they paid then. Keep the same cohort rule before and after.
  • Price increases, seat and usage growth, launches and the economy move it too, so a formula valuing its change takes only the share of it you credit to the initiative.
  • Revenue kept by lower churn is inside it. Don't value churn and net retention for the same customers.

Good for

  • Valuing health scoring and intervention by the recurring revenue kept and expanded, at your contribution margin.

Guide: When the outcome has other causes

percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

Stripe

Key What it measures Type Unit Fed by
stripe.charges.count Stripe's charges for the month that kept revenue, in the currency the account settles in. count — Stripe
stripe.charges.volume What Stripe's charges kept for the month, captured less refunded, tax included, in the currency the account settles in. Charges in other currencies aren't counted. currency USD Stripe
stripe.subscriptions.active_count Stripe's active subscriptions for the month. Written to billing.subscriptions.active_count instead when this connection is the primary source. count subscriptions Stripe
stripe.subscriptions.churn_rate Stripe's subscription churn rate for the month. Written to billing.subscriptions.churn_rate instead when this connection is the primary source. percentage % Stripe
stripe.subscriptions.churned_count Stripe's churned subscriptions for the month. Written to billing.subscriptions.churned_count instead when this connection is the primary source. count subscriptions Stripe
stripe.subscriptions.mrr Stripe's monthly recurring revenue for the month. Written to billing.subscriptions.mrr instead when this connection is the primary source. currency USD Stripe
stripe.subscriptions.new_count Stripe's new subscriptions for the month. Written to billing.subscriptions.new_count instead when this connection is the primary source. count subscriptions Stripe

AWS

Key What it measures Type Unit Fed by
aws.cost.service_spend One AWS service's unblended cost for the month; the service is in the observation's dimensions. Feeds cost entries through the initiative's service mappings. currency USD AWS
aws.cost.total_spend AWS's total spend for the month. currency USD AWS
aws.quick.enterprise_users People on Amazon Quick's Enterprise subscription (the Author Pro and Admin Pro roles), as of the month's latest sync. count users AWS
aws.quick.group_licensed_users One Amazon Quick group's members on a Professional or Enterprise subscription, as of the month's latest sync; the group is in the observation's dimensions. A person in two groups counts in both. count users AWS
aws.quick.licensed_users People on an Amazon Quick Professional or Enterprise subscription, as of the month's latest sync, signed in yet or not. Quick Sight's own Reader and Author subscriptions, without the AI assistant, aren't counted. Read as ai_assistant.seats.licensed when the AWS connection is the primary AI-assistant source. count users AWS
aws.quick.professional_users People on Amazon Quick's Professional subscription (the Reader Pro role), as of the month's latest sync. count users AWS

Freshdesk

Key What it measures Type Unit Fed by
freshdesk.knowledge_base.article_count Freshdesk solution articles that are published and that customers can see, as they stand at the month's last sync, from the month the connection was made. Read as support.knowledge_base.article_count when this connection is the primary source for support tickets. count articles Freshdesk
freshdesk.tickets.agent_resolved_count Freshdesk's resolved and closed tickets for the month that no AI deflection rule on the connection matched, plus the ones a rule matched whose customer came back. Read as support.tickets.agent_resolved_count when this connection is the primary source. count tickets Freshdesk
freshdesk.tickets.avg_first_response_hours Freshdesk's average ticket first response time for the month. Written to support.tickets.avg_first_response_hours instead when this connection is the primary source. duration hours Freshdesk
freshdesk.tickets.avg_resolution_days Freshdesk's average time from a ticket's creation to its resolution for the month, in days (the computer converts from hours). Read as support.tickets.avg_resolution_days when this connection is the primary source. duration days Freshdesk
freshdesk.tickets.resolved_count Freshdesk's support tickets closed for the month. Written to support.tickets.closed_count instead when this connection is the primary source. count tickets Freshdesk
freshdesk.tickets.volume Freshdesk's support tickets opened for the month. Written to support.tickets.volume instead when this connection is the primary source. count tickets Freshdesk

Intercom

Key What it measures Type Unit Fed by
intercom.conversations.agent_resolved_count Intercom's closed conversations for the month that no AI deflection rule on the connection matched, plus the ones a rule matched whose customer came back. Read as support.tickets.agent_resolved_count when this connection is the primary source. count tickets Intercom
intercom.conversations.avg_first_response_hours Intercom's average ticket first response time for the month. Written to support.tickets.avg_first_response_hours instead when this connection is the primary source. duration hours Intercom
intercom.conversations.avg_resolution_days Intercom's average time from a conversation opening to closing for the month, in days (Intercom reports hours; the computer converts). Read as support.tickets.avg_resolution_days when this connection is the primary source. duration days Intercom
intercom.conversations.closed_count Intercom's support tickets closed for the month. Written to support.tickets.closed_count instead when this connection is the primary source. count tickets Intercom
intercom.conversations.csat_score Intercom's average conversation rating for the month on the 0–1 scale the support keys use (Intercom rates 1–5; the computer converts). Read as support.tickets.csat_score when this connection is the primary source. score 0–1 Intercom
intercom.conversations.volume Intercom's support tickets opened for the month. Written to support.tickets.volume instead when this connection is the primary source. count tickets Intercom

Cloud

Key What it measures Type Unit Fed by
cloud.cost.budget_amount Cloud budget ceiling for the period, across providers. currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
cloud.cost.forecasted_spend Forecast end-of-period cloud spend based on the current usage trajectory. currency USD Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
cloud.cost.total_spend Total cloud infrastructure spend for the period on the primary provider.

Watch out for

  • Each provider is taken its own way: AWS as unblended cost with credits netted in, GCP net of credits, Databricks at list price, Snowflake as billed. Take a hand-entered baseline the same way.
  • It's compared per unit of Cloud Cost Driver Volume, so growth isn't read as a dearer bill. Record the volume for the same months.

Good for

  • Valuing FinOps work by the cloud spend saved per unit of demand.

Guide: Pick the count your business grows by

currency USD AWS, Azure, Databricks, Google Cloud, Snowflake
cloud.resources.idle_share Percentage of provisioned cloud resources running at low or zero utilization. percentage % Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
cloud.usage.unit_count What your cloud bill grows with (customers, orders, transactions or requests) in the period, counted the same way every period.

Watch out for

  • Pick the unit once, such as customers, orders, transactions or requests: a reading counting a different unit is held for review.
  • Count it for the same month as the spend, the same way every month.

Good for

  • The volume cloud spend is compared per, so a growing business isn't read as a dearer bill.

Guide: Keep it the same every month

count units Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

ChatGPT Enterprise

Key What it measures Type Unit Fed by
openai_codex.chat.monthly_active_users Members who sent at least one ChatGPT message in the month, each counted once however many days they used it. Members who only used Codex aren't counted. Read from OpenAI's daily usage per member, kept as a digest of each member's ID; members OpenAI has deleted since aren't in it. Read as ai_assistant.users.monthly_active when ChatGPT Enterprise is the primary AI-assistant source. count users ChatGPT Enterprise
openai_codex.seats.chatgpt Members holding a ChatGPT seat in the workspace, as of the month's latest sync. Codex-only seats aren't counted. Months before the connection's first sync have none, since OpenAI lists only who holds a seat now. Read as ai_assistant.seats.licensed when ChatGPT Enterprise is the primary AI-assistant source. count users ChatGPT Enterprise
openai_codex.usage.lines_accepted Lines of code Codex added that developers kept, for the month. Read as engineering.coding_assistant.lines_accepted when Codex is the primary coding-assistant source. Codex reports no count of accepted suggestions, so it feeds none of the other coding-assistant keys. count lines ChatGPT Enterprise
openai_codex.usage.lines_committed Lines attributed to Codex that reached a commit, for the month — the stronger of the two line counts, but one the other assistants do not report, so it stays a platform key. count lines ChatGPT Enterprise
openai_codex.usage.turns Codex turns in the workspace for the month. Stands in for an active-developer count: Roiva reads members' rows only to count who used ChatGPT, and keeps nothing from them about Codex. count turns ChatGPT Enterprise

Snowflake

Key What it measures Type Unit Fed by
snowflake.compute.credits_used Snowflake warehouse credits consumed in the month. count credits Snowflake
snowflake.cost.service_spend The month's Snowflake bill split by usage type — compute, storage, cloud services and the rest — with Cortex AI under the service type it bills by (AI_FUNCTIONS, CORTEX_AGENTS and so on), the line in the observation's dimensions. What a cost source's service mappings match. currency USD Snowflake
snowflake.cost.total_spend What Snowflake billed the account for the month, at its own contract rates, excluding usage Snowflake did not charge for. Read as cloud.cost.total_spend when this connection is the primary cloud cost source. currency USD Snowflake
snowflake.query.avg_execution_seconds Mean execution time of Snowflake queries in the month. duration seconds Snowflake
snowflake.query.count Successful Snowflake queries in the month. count queries Snowflake

AI assistants

Key What it measures Type Unit Fed by
ai_assistant.seats.licensed People holding a license for a workplace AI assistant, as of the period's latest report. Shown beside active users for adoption. Never priced as value; the seats nobody used are priced only as a cost, at the seat price a person enters.

Watch out for

  • People holding a license, not people using it. Cost follows licenses, value follows use.

Good for

  • Beside monthly active users, the licenses nobody used before a renewal.

Guide: Pitfalls

count users AWS, ChatGPT Enterprise, Claude Enterprise, Microsoft 365 Copilot
ai_assistant.tasks.minutes_per_task Hands-on minutes a person spends on one task of the kind the AI assistant is for. The baseline is without the assistant and the reading is with it; a time study on this key records both.

Watch out for

  • Hands-on minutes on the kind of task the assistant is for, including prompting it and checking what it wrote. Say which tasks count.
  • A time study on this metric records both: the without-AI mean as the baseline and the with-AI mean as the reading.

Good for

  • With monthly active users and tasks per user, the time an assistant frees.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
ai_assistant.tasks.timed_per_month How many times in the period the people an initiative covers did its timed task with the AI assistant, such as tickets researched or commentaries drafted.

Watch out for

  • Count the task the time study timed, done with the assistant, by the people this initiative covers. Record it on the initiative: a figure for the whole account is never read.
  • Use the time study's definition of one task, or the count and the minutes describe different work.

Good for

  • With the minutes a time study saved on one task, the time one use of an assistant frees.

Guide: Setting one up

count tasks Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
ai_assistant.users.monthly_active People who used a workplace AI assistant at least once in the period, each counted once however often they used it.

Watch out for

  • People who used the assistant at least once in the month, each counted once, however little. A seat assigned isn't a user.
  • The formula multiplies it by tasks per user, so average tasks over these same users, light ones included.

Good for

  • The users minutes saved per task are multiplied by.

Guide: Where the active users come from

count users ChatGPT Enterprise, Claude Enterprise, Gemini in Google Workspace, Glean, Microsoft 365 Copilot

Workflow automation

Key What it measures Type Unit Fed by
automation.runs.count Runs of automated workflows that finished in the period, successful and failed. count runs Make, Zapier, n8n
automation.runs.failed_count Runs of automated workflows that failed in the period. count runs Make, Zapier, n8n
automation.runs.manual_minutes_per_run Minutes a person took to do the task one successful automated run now does, from a time study or your own timing. One figure for the initiative's workflows, so a study samples the tasks as often as each one runs.

Watch out for

  • One figure for the initiative's workflows, so time the tasks in the mix they run in: the common ones most.
  • Until it's measured, Roiva's convention stands in and the value reads as estimated.

Good for

  • Pricing the manual work each successful run replaces, at your ops hourly rate.

Guide: Setting one up

duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
automation.runs.successful_count Runs of automated workflows that succeeded in the period.

Watch out for

  • From Zapier, Make or n8n, successful runs summed across the workflows the initiative covers.
  • The formula counts every successful run as a task a person would otherwise have done. Runs that replaced nothing, such as tests or notifications, overstate it.

Good for

  • The runs manual minutes per run are multiplied by.

Guide: What a formula reads

count runs Make, Zapier, n8n

Calendar

Key What it measures Type Unit Fed by
calendar.meetings.avg_attendees Average attendees per meeting in the period on the calendars this connection syncs.

Watch out for

  • From Google Workspace or Microsoft 365 it counts everyone on the invite, accepted or not, booked rooms included, so inviting widely reads as large meetings.
  • Only confirmed, timed meetings with two or more attendees count. All-day events and solo blocks are left out.

Good for

  • With meeting hours, the person-hours in meetings an assistant brings down.

Guide: Cash value and freed capacity

average people Google Workspace, Microsoft 365
calendar.meetings.avg_duration_mins Average length in minutes of a meeting on the calendars this connection syncs. duration minutes Google Workspace, Microsoft 365
calendar.meetings.count Meetings in the period on the calendars this connection syncs — confirmed, timed events with two or more attendees. A meeting on more than one of those calendars counts once. count meetings Google Workspace, Microsoft 365
calendar.meetings.hours Hours of meetings in the period on the calendars this connection syncs, each meeting counted once however many attend and however many of those calendars it sits on.

Watch out for

  • Each meeting counts once, however many attend. The formula multiplies by average attendees to get person-hours.
  • It covers the calendars the connection syncs, so connecting more calendars adds their meetings. Keep the same calendars before and after.

Good for

  • Valuing an assistant that shortens or cancels meetings by the person-hours it frees.

Guide: Cash value and freed capacity

duration hours Google Workspace, Microsoft 365

Gemini Code Assist

Key What it measures Type Unit Fed by
gemini_code_assist.code_assist.acceptance_rate Accepted code suggestions as a share of those offered in the project for the month. Read as engineering.coding_assistant.acceptance_rate when Gemini Code Assist is the primary coding-assistant source. percentage % Gemini Code Assist
gemini_code_assist.code_assist.lines_accepted Lines of code developers accepted from Gemini Code Assist in the project for the month, from Cloud Monitoring's code_lines_accepted_count. Read as engineering.coding_assistant.lines_accepted when Gemini Code Assist is the primary coding-assistant source. count lines Gemini Code Assist
gemini_code_assist.code_assist.suggestions_accepted Code suggestions developers accepted from Gemini Code Assist in the project for the month, from Cloud Monitoring's code_suggestions_accepted_count. Read as engineering.coding_assistant.acceptances when Gemini Code Assist is the primary coding-assistant source. count acceptances Gemini Code Assist
gemini_code_assist.code_assist.suggestions_shown Code suggestions Gemini Code Assist offered developers in the project for the month, from Cloud Monitoring's code_suggestions_count. Read as engineering.coding_assistant.suggestions when Gemini Code Assist is the primary coding-assistant source. count suggestions Gemini Code Assist

Google Workspace

Key What it measures Type Unit Fed by
google_workspace.calendar.avg_attendees Average attendees per meeting on the Google calendars this connection syncs, for the month. Read as calendar.meetings.avg_attendees when this connection is the primary calendar source. average people Google Workspace
google_workspace.calendar.avg_meeting_duration_mins Average meeting length in minutes on the Google calendars this connection syncs, for the month. Read as calendar.meetings.avg_duration_mins when this connection is the primary calendar source. duration minutes Google Workspace
google_workspace.calendar.meeting_count Meetings on the Google calendars this connection syncs, for the month, a meeting on more than one of them counted once. Read as calendar.meetings.count when this connection is the primary calendar source. count meetings Google Workspace
google_workspace.calendar.meeting_hours Hours of meetings on the Google calendars this connection syncs, for the month, each meeting counted once. Read as calendar.meetings.hours when this connection is the primary calendar source. duration hours Google Workspace

Make

Key What it measures Type Unit Fed by
make.scenarios.credits Make credits one scenario used in the month, which is what Make bills. Standard modules use 1 credit per operation; Make's built-in AI features add credits for tokens. The scenario is in the observation's dimensions. count credits Make
make.scenarios.executions Total runs of one Make scenario in the month; the scenario is in the observation's dimensions. Read as automation.runs.count, every scenario's row added, when this connection is the primary workflow-automation source. count runs Make
make.scenarios.failed_executions Failed runs of one Make scenario in the month; the scenario is in the observation's dimensions. Read as automation.runs.failed_count, every scenario's row added, when this connection is the primary workflow-automation source. count runs Make
make.scenarios.successful_executions Successful runs of one Make scenario in the month; the scenario is in the observation's dimensions. Read as automation.runs.successful_count, every scenario's row added, when this connection is the primary workflow-automation source. count runs Make

Microsoft 365

Key What it measures Type Unit Fed by
microsoft_365.calendar.avg_attendees Average attendees per meeting on the Outlook calendars this connection syncs, for the month. Read as calendar.meetings.avg_attendees when this connection is the primary calendar source. average people Microsoft 365
microsoft_365.calendar.avg_meeting_duration_mins Average meeting length in minutes on the Outlook calendars this connection syncs, for the month. Read as calendar.meetings.avg_duration_mins when this connection is the primary calendar source. duration minutes Microsoft 365
microsoft_365.calendar.meeting_count Meetings on the Outlook calendars this connection syncs, for the month, a meeting on more than one of them counted once. Read as calendar.meetings.count when this connection is the primary calendar source. count meetings Microsoft 365
microsoft_365.calendar.meeting_hours Hours of meetings on the Outlook calendars this connection syncs, for the month, each meeting counted once. Read as calendar.meetings.hours when this connection is the primary calendar source. duration hours Microsoft 365

Work

Key What it measures Type Unit Fed by
work.issues.cycle_time_avg Average hours from an issue's start to its resolution, over the issues resolved in the period. duration hours Jira
work.issues.minutes_per_issue Minutes of triage, coordination and hand-off one resolved issue takes, from a time study or your own timing. Not how long the issue was open. duration minutes Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
work.issues.resolved_count Issues resolved in the period in the team's issue tracker. count issues Jira
work.issues.story_points_completed Story points on the issues resolved in the period. count points Jira

BigQuery

Key What it measures Type Unit Fed by
bigquery.jobs.bytes_processed Bytes processed by BigQuery jobs in the month. count bytes BigQuery
bigquery.jobs.count BigQuery jobs run in the month. count jobs BigQuery
bigquery.jobs.slot_ms Slot milliseconds consumed by BigQuery jobs in the month. count slot-ms BigQuery

Google Cloud

Key What it measures Type Unit Fed by
gcp.billing.budget_amount The amount of one GCP budget for the month; the budget is in the row's raw payload. currency USD Google Cloud
gcp.billing.budget_spend Spend against one GCP budget for the month, before tax; the budget is in the row's raw payload. Feeds cost entries through the initiative's budget mappings. currency USD Google Cloud
gcp.billing.total_spend Google Cloud's total billed spend for the month. currency USD Google Cloud

Jira

Key What it measures Type Unit Fed by
jira.issues.cycle_time_avg Jira's average issue cycle time for the month. Read as work.issues.cycle_time_avg when Jira is the primary issue-tracking source. duration — Jira
jira.issues.resolved_count Jira's issues resolved in the month. Read as work.issues.resolved_count when Jira is the primary issue-tracking source. count — Jira
jira.issues.story_points_completed Jira's issues story points completed for the month. Read as work.issues.story_points_completed when Jira is the primary issue-tracking source. count — Jira

n8n

Key What it measures Type Unit Fed by
n8n.workflows.executions Production executions of one n8n workflow that started in the month and finished, successful plus failed. Counted as n8n's Insights counts them, except evaluation runs: no manual test runs or sub-workflow runs, and a retry counts as an execution of its own. The workflow is in the observation's dimensions. Read as automation.runs.count, every workflow's row added, when this connection is the primary workflow-automation source. count executions n8n
n8n.workflows.failed_executions Production executions of one n8n workflow in the month that ended in error or crashed, including runs a later retry fixed; the workflow is in the observation's dimensions. Read as automation.runs.failed_count, every workflow's row added, when this connection is the primary workflow-automation source. count executions n8n
n8n.workflows.successful_executions Production executions of one n8n workflow in the month that succeeded, on the first try or a retry; the workflow is in the observation's dimensions. Read as automation.runs.successful_count, every workflow's row added, when this connection is the primary workflow-automation source. count executions n8n

Zapier

Key What it measures Type Unit Fed by
zapier.zaps.failed_runs Failed Zap runs reported to Roiva's Zapier webhook in the month. Read as automation.runs.failed_count when this connection is the primary workflow-automation source. count runs Zapier
zapier.zaps.runs Total Zap runs reported to Roiva's Zapier webhook in the month. Read as automation.runs.count when this connection is the primary workflow-automation source. count runs Zapier
zapier.zaps.successful_runs Successful Zap runs reported to Roiva's Zapier webhook in the month. Read as automation.runs.successful_count when this connection is the primary workflow-automation source. count runs Zapier

Azure

Key What it measures Type Unit Fed by
azure.cost.service_spend One Azure service's cost for the month; the service is in the observation's dimensions. Feeds cost entries through the initiative's service mappings. currency USD Azure
azure.cost.total_spend Azure's total spend for the month. currency USD Azure

Databricks

Key What it measures Type Unit Fed by
databricks.cost.product_spend Databricks list-price spend for the month on one product (billing_origin_product: MODEL_SERVING, JOBS, SQL, ...). currency USD Databricks
databricks.cost.total_spend Databricks list-price spend for the month, from system.billing.usage priced at system.billing.list_prices. Read as cloud.cost.total_spend when this connection is the primary cloud cost source. currency USD Databricks

Microsoft 365 Copilot

Key What it measures Type Unit Fed by
microsoft_365_copilot.seats.enabled_users People holding a Microsoft 365 Copilot license, as of the month's latest usage report. Read as ai_assistant.seats.licensed when Microsoft 365 Copilot is the primary AI-assistant source. count users Microsoft 365 Copilot
microsoft_365_copilot.users.monthly_active People who used Microsoft 365 Copilot in any app in the month, each counted once however many days they used it. Counted from each user's latest activity date in Microsoft's usage report, read nightly from the connection's first sync. Read as ai_assistant.users.monthly_active when Microsoft 365 Copilot is the primary AI-assistant source. count users Microsoft 365 Copilot

Workforce

Key What it measures Type Unit Fed by
workforce.tasks.manual_per_week Discrete manual tasks (data pulls, formatting, distribution) performed each week. count tasks Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)
workforce.time.hours_saved_per_week Net staff hours recovered per week as a direct result of the initiative.

Watch out for

  • Measured after the initiative is live, not forecast before it, and from the people doing the work rather than the tool's own report.
  • Count only work that stopped, not work that moved to someone else. The hours freed are freed capacity, not cash.

Good for

  • Valuing a reporting or automation initiative directly by the hours it saves each week.

Guide: Cash value and freed capacity

duration hours Recorded by hand, imported, or read from your warehouse (BigQuery, Databricks, Snowflake or SQL Server)

American Express

Key What it measures Type Unit Fed by
amex.card.spend What the American Express statements you uploaded charged in the month, in USD, for a month they cover end to end. A card is a payment rail rather than a provider billing for its own service, so this is not counted in provider-billed AI spend - a charge to a connected provider is already in that provider's own bill. currency USD American Express

Brex

Key What it measures Type Unit Fed by
brex.card.spend What the Brex card was charged in the month, in USD. A card is a payment rail rather than a provider billing for its own service, so this is not counted in provider-billed AI spend - a charge to a connected provider is already in that provider's own bill. currency USD Brex (coming soon)

Gemini in Google Workspace

Key What it measures Type Unit Fed by
gemini_workspace.users.monthly_active People in the Google Workspace who used Gemini at least once in the month, each counted once however often: the Gemini app (gemini.google.com) and Gemini in Gmail, Docs, Sheets, Slides, Meet, Drive and the other Workspace apps together, from Google's Gemini audit log. Only uses Google marks active count; a suggestion nobody took up doesn't. Months are UTC calendar months. Read as ai_assistant.users.monthly_active when Gemini is the primary AI-assistant source. count users Gemini in Google Workspace

Glean

Key What it measures Type Unit Fed by
glean.users.monthly_active People who used Glean's Assistant (Chat) at least once in the month, each counted once however often, from Glean's Insights: distinct Assistant users for the span Roiva asks for, the UTC calendar month. People who only searched with Glean aren't counted, so it reads lower than Glean's own monthly active users, which count any activity over a rolling 28 days. Read as ai_assistant.users.monthly_active when Glean is the primary AI-assistant source. count users Glean

Looker

Key What it measures Type Unit Fed by
looker.looks.field_value The monthly value of one numeric column of one Look; the Look and the column are in the observation's dimensions. Roiva cannot infer a unit from a Look, so the type is count unless a person overrides the definition. count — Looker

Ramp

Key What it measures Type Unit Fed by
ramp.card.spend What the Ramp card was charged in the month, in USD. A card is a payment rail rather than a provider billing for its own service, so this is not counted in provider-billed AI spend - a charge to a connected provider is already in that provider's own bill. currency USD Ramp (coming soon)