7%
of leaders report established AI ROI.
KPMG Global AI Pulse, Q2 2026
Adoption is nearly universal. Proven return is almost nonexistent. The gap between those two facts is the whole story — and it is not a measurement problem you can buy your way out of.
The fog was never the number. It was the visibility you never had.
One number opens the case.
Boards are spending on AI and cannot find the return. Everyone is asking what the return on AI is as if it were a new question. It is the oldest question in the business — and the industry is about to make the exact same mistake with AI that it made with marketing attribution.
Before the case can build to an answer, it has to take apart the numbers most leaders reach for first — the ones that feel like proof and measure nothing that matters.
Five metrics that lie, examined and struck.
Most AI measurement tracks whether people are using the tool. That is like measuring whether employees showed up to the office and calling it productivity. Here is each metric a dashboard is proudest of, struck through, with the metric worth watching in its place — the reframe is the measurement engine’s own.
You don’t need to do attribution, you just need visibility. These are very different things.”
That is the reframe that clears the fog. You were never going to attribute a dollar to a prompt. But visibility into how the work actually flows — which engagement, what the AI traceably did, whether a detection caused the save — is a different thing entirely, and it is the thing that gives you an honest shot at a real number. Look at the board: the answer was there the whole time. Three dimensions, not one.
The work behind the board
This framework was not drawn on a whiteboard. It was built across a run of Value-First Platform: AI Data Readiness episodes with Trisha Merriam — the analytics practitioner who kept asking to see the numbers, not just hear the claim.
Value-First Platform: AI Data ReadinessNovember 6, 2025
Value-First Platform: Data Readiness with AI and Trisha Merriam
Value-First Platform: Data Readiness with AI and Trisha Merriam
Watch the episodeValue-First Platform: AI Data ReadinessNovember 26, 2025
Value-First Platform: AI Data Readiness with Trisha Merriam
Value-First Platform: AI Data Readiness with Trisha Merriam
Watch the episodeValue-First Platform: AI Data ReadinessDecember 3, 2025
Value-First Platform: AI Data Readiness with Trisha Merriam
Watch the episodeThe number carries only the confidence you have earned.
Before you build anything, place yourself. These five levels are not a score — they set how much weight the number is allowed to bear. Where you honestly sit decides whether the answer is trustworthy at all, directional, or answerable with a real figure.
The level is not a score. It is the confidence the number below is allowed to carry: not yet trustworthy — you have activity, not outcome visibility. at the first two levels, directional — curated, not comprehensive. at the third, and answerable with a number that connects to real revenue and real capability. at the top two.
The confidence tiers those levels map to
- Not yet trustworthyLevels 0 to 1
- not yet trustworthy — you have activity, not outcome visibility.
- DirectionalLevel 2
- directional — curated, not comprehensive.
- AnswerableLevels 3 to 4
- answerable with a number that connects to real revenue and real capability.
This is not a meter to fill. Each tier is the confidence a number is allowed to carry at that level of visibility — nothing more. A directional number you can defend is worth more than an answerable one you cannot, and reaching a higher tier is never the point. Honest measurement is. Chasing the top tier for its own sake is the same trap as chasing the return: it makes the label the goal instead of the value underneath it.
Now assemble it with your own numbers.
Enter your AI spend, the three dimensions, and your readiness level. The measurement is arithmetic over your own entries and nothing else. If your visibility has not earned a number yet, this instrument tells you so and hands you the baseline to build instead — it will not invent one.
AI spend for the period
The denominator — measured against value created, never a value metric on its own.
Revenue Influenced
Deals AI was traceably part of — tied to the deal, not “someone used ChatGPT once.” Describe what the AI traceably did in your own words, and be honest: this is the question your board will ask, and the tool records your attestation rather than judging it. An entry with the engagement or that description left blank is shown, not counted.
Revenue Protected
Retention or expansion an AI-detected signal caused. If the team would have caught it anyway, it does not count.
Capability Built
What the team can do now that it could not six months ago. Not a dollar figure. Not hours saved.
Your readiness level
Be honest here — it sets the confidence the number is allowed to carry, and the tool checks your claim against the evidence you entered.
The level is not a score. It is the confidence the number below is allowed to carry: not yet trustworthy — you have activity, not outcome visibility. at the first two levels, directional — curated, not comprehensive. at the third, and answerable with a number that connects to real revenue and real capability. at the top two.
The metrics that do not count
Try to add hours saved, adoption, tokens, or any other metric. The tool refuses each with the metric worth watching instead.
Nothing here reaches the value read. The value picture is built only from Revenue Influenced, Revenue Protected, and Capability Built.
Both panels in full, in the page. They open over the argument as fly-ins when you summon them; this is the same content, always here, whether or not that layer is available to you.
The Case Board
7%
of leaders report established AI ROI.
KPMG Global AI Pulse, Q2 2026
The case builds to two revenue figures and a capability — never one blended number.
Where you are
Read the reckoning, then watch five metrics get taken apart.
Your case file
The denominator
AI spend for the period — the denominator, never a value metric on its own.
What the case builds to
Traceable AI-assisted revenue — tied to the deal, not self-reported.
Retention or expansion an AI-detected signal caused — not a save you would have made anyway.
What the team can do now that it could not before — ordinal, never dollarized.
Two revenue figures and a capability — never one blended number.
Excluded from your file
The five metrics that lie never enter the case — the engine measures none of them:
- Prompts per user per day
- Time saved per task
- Adoption rate / active users
- Token usage / API calls
- Emails sent / content volume
The confidence your number can carry
Where you honestly sit sets how much weight the number can bear.
The level is not a score. It is the confidence the number below is allowed to carry: not yet trustworthy — you have activity, not outcome visibility. at the first two levels, directional — curated, not comprehensive. at the third, and answerable with a number that connects to real revenue and real capability. at the top two.
- Not yet trustworthyLevels 0 to 1
- DirectionalLevel 2
- AnswerableLevels 3 to 4
Enter your own AI spend, the Three Dimensions, and your readiness level, then build the measurement. The picture that follows is arithmetic over your entries — nothing else.
Keep going
The Clearing builds the case. Two more doors open onto the same instrument, and the full framework goes deeper on every part.
Framework developed in collaboration with Trisha Merriam and Erin Wiggers on the Value-First Platform: AI Data Readiness show.