Getting to AI ROI in six weeks
A week-by-week plan for answering 'what did the AI spend return?' before the board meets: what you can't prove in six weeks, what you can, and what to hand over.
When I was running a data analytics business, our CFO asked me what our AI spend was worth. My answers were less than satisfying. I focused on what the AI tools would enable for productivity and growth, but my evidence was mostly anecdotes about freed coding capacity and new product ideas we were excited about.
This is what I'd do now if I had six weeks to answer the question properly, starting from scratch. It assumes the AI tools have been live for at least a few months, so there's an "after" to measure.
What can't you produce in six weeks?
In six weeks you can't prove that AI caused the result. That takes a controlled test, with comparable groups using and not using the AI, and six weeks isn't enough to run one.
You also can't simply ask an LLM, even one connected to your data. It will produce a number, but it won't give you a baseline, costs, or an attribution rule that stand up to scrutiny, and those are what your CFO will ask about.
What can you produce in six weeks?
In six weeks you can produce a number with its method and limitations clearly stated. The aim is a defensible process your CFO can review, not certainty.
What does the six-week plan look like?
Five weeks of work, with the attribution rule written in week 1 before anything is measured. Week 6 belongs to the board pack.
| Week | What you do | What you have at the end |
|---|---|---|
| 1 | Pick one initiative, write the attribution rule, take it to your CFO | A one-sentence rule, agreed before any number exists |
| 2–3 | Total the spend and build the baseline, in parallel | The full cost of the initiative, and a baseline labeled as an estimate |
| 4 | Measure the after, on the same definition | An after comparable to the baseline |
| 5 | Apply the rule, compute, and write it up | The value the rule produces against the cost |
| 6 | The board pack goes out | Nothing new: board packs usually go out a week or more before the meeting |
Which initiative should you start with?
One with meaningful spend and a metric you can measure, and not the one you expect to look best, because your CFO will ask why you chose it. Start with a single "lighthouse" initiative rather than a business-wide or portfolio-wide answer; six weeks isn't enough to do more than one well.
What is an attribution rule?
An attribution rule is one sentence saying how much of a metric's change you will claim for the AI, and why.
Often nobody decided what would count as value before the initiative started, so decide now: the outcome the initiative is supposed to move, and whether you will claim all of the change, a fixed share, or only the activity the tool touched.
For an AI support agent, that might be: "We count savings only on tickets the AI agent closed without a human, at the baseline cost per ticket." The rule also tells you what data you'll need.
Are the savings cash or capacity?
Say which in the rule. If the work took fewer hours but payroll and contractor spend didn't change, that's freed capacity, not cash savings, and I'd call it that in the board pack.
Why write the rule before you measure anything?
Because a rule chosen after you've seen the result is hard to defend, and a CFO who has sat through a few of these can tell.
Take the rule to your CFO while it's still just a sentence and get quick feedback. If they would push back on it in week 5, you want to hear it in week 1.
What counts as the spend?
The total of what it takes to run the initiative, not just the AI line item: model API bills, AI add-ons and price increases on tools that quietly became AI tools, the cloud behind them, and the loaded hours of the people operating it.
It may come in higher than the budget. While you're in the bills, get a rough total of AI spend across the company as well, so the board can see what share of it this initiative is.
What if you don't have a baseline?
Reconstruct one from the history that exists before the initiative went live, write down how you did it, and label it an estimate.
A baseline is what the metric was before the initiative went live. For the support agent, that's cost per resolved ticket. Not having one off the shelf is common, and I wouldn't have had one either. I'd rebuild it from ticket history and conversations with the support team.
How do you measure the after?
On the same definition as the baseline: the same query, date logic and exclusions, over a representative period rather than the week that looks best. A baseline taken from all tickets and an after taken from tier 1 only will show a drop the AI had nothing to do with.
What do you hand the CFO?
A short write-up of one initiative, ready for the board pack: its full cost, the value the attribution rule produces against that cost, and how you got there.
In week 5, apply the rule, compute the value, and write it up. It covers:
- The total AI spend, and the full cost of this one initiative
- The baseline, marked as an estimate, and the after, measured the same way
- The attribution rule set in week 1, and the value it produces against the cost
- How you did it, and its limitations
- A plain statement that the other initiatives aren't measured yet, and a plan for measuring them
That gives your CFO a position to argue with, which is the point. Six weeks from scratch, certainty and proof aren't possible. A defensible process and a number to discuss and build on are.
See it against your own numbers
The free AI ROI Exposure Assessment runs Roiva's published math on your inputs — five minutes, no account, no card.
Get the next one
New field notes by email as they're published. Or subscribe to the RSS feed.