Renewal time for your AI licenses arrives, and the CFO asks the question you knew was coming: “What did we get for this?” IT brings a usage dashboard showing most seats are active, plus a few enthusiastic quotes from users. The CFO nods and says, “That tells me people use it. It doesn’t tell me it’s worth what we pay.”
That exchange is happening in a lot of organizations right now. The first wave of AI spending was approved on potential. The second wave will be approved on evidence, and most teams aren’t collecting it. If you want to measure AI ROI in a way that holds up in a budget review, you need outcome metrics, a baseline to compare them against, and an honest way to turn them into value.
The ladder of evidence
There’s a clear hierarchy in how convincing AI evidence is. Anecdotes are the weakest: they’re memorable, but every technology has fans. Adoption numbers are better, but they measure activity, not results. Business outcomes are stronger. Outcomes compared against a baseline captured before rollout, and reported regularly to leadership, are the strongest of all.
Here’s how the assessment asks the question, and the 0 to 4 ladder I score it against:
U4. How do you measure the value of AI?
- We don’t
- Anecdotes and user feedback
- Usage and adoption metrics (active users, seats used)
- Business outcome metrics for some use cases (time saved, cost, quality)
- Baselines captured before rollout; outcomes tracked and reported to leadership
Most teams I assess are at 1 or 2. The move to 3 means measuring outcomes for at least some use cases. The move to 4 depends on something you can only do at the start: capturing the baseline. That’s covered in baseline before you build, and it’s why the two questions are closely linked.
Why adoption isn’t value
Adoption and value can point in opposite directions. Usage can be high while value is low, if people mostly use the assistant for small tasks that didn’t take long anyway. Usage can be low while value is high, if a handful of people use it for a task that used to take days. A dashboard of active users can’t tell those situations apart.
Track adoption. It’s a useful early signal that something is or isn’t working. Just don’t present it as the answer to “was it worth it?”
Where AI value actually shows up
- Capacity: time freed on a task, which people can spend on other work.
- Cost avoided: hiring you didn’t need to do, outsourcing you reduced, overtime you didn’t pay.
- Quality: fewer errors, less rework, fewer compliance findings.
- Speed: shorter cycle times, which can mean faster revenue, happier customers, or quicker decisions.
- Risk reduction: harder to quantify, but real, such as fewer sensitive documents exposed because a review process now catches them.
The time-saved trap
The most common AI value claim is hours saved. It’s also the one finance teams trust least, and with good reason. Saving 30 minutes a day per person doesn’t reduce costs unless those 30 minutes go somewhere valuable, or unless the organization avoids a cost it would otherwise have had.
Be honest about it. Present time saved as capacity, and say what the capacity was used for: “the service desk handled higher volume without adding staff,” or “account managers spent the freed time on renewals.” If you can’t say where the time went, report it as capacity and don’t convert it to dollars. You’ll be more credible for it.
A simple value formula
For a time-based use case, estimate annual value like this:
Annual value = (baseline minutes per task − current minutes per task) × tasks per year × loaded cost per minute × realization factor
The realization factor is the share of freed time you believe actually turns into valuable work. It’s an assumption, so state it plainly and choose a conservative one. A CFO will accept a conservative assumption far more readily than an optimistic one. Add any direct cost avoidance and quality improvements you can quantify.
Then compare against the full cost: licenses, build time, running costs, and training. ROI = (annual value − annual cost) ÷ annual cost. The cost side matters as much as the value side, and it’s much easier to state accurately if you’ve set up budgets and tagging for AI spend.
Agree the method with finance first
The single most useful thing you can do is agree the measurement method with finance before any results come in. Sit down with someone from finance for an hour. Agree the loaded cost per hour you’ll use, the realization factor, which cost lines count, and how often you’ll report. Write it down. When the first quarterly report arrives, the conversation is about the results, not about whether your method is fair. Business owners own the metrics for their use cases, IT supplies the data, and finance validates the method. That division of labor is what makes the numbers trusted.
The asset: a one-page quarterly AI value report
Report per use case, not per tool. A tool is a cost; a use case is where value is created. One page, every quarter:
- For each use case in production: the primary metric, its baseline, its current value, the change, the estimated annual value with assumptions stated, the annual cost, and a status of healthy, watch, or at risk.
- Portfolio total: summed value and summed cost across use cases.
- What we stopped: pilots or tools you ended, and why. This builds credibility; it shows you’re managing a portfolio, not defending every investment.
- Adoption, briefly: active use of licensed tools, as a supporting signal.
- Next quarter: which use cases move from the register into pilots, and what you expect them to show.
Send it to the executive sponsor and the AI council. Over a few quarters, it becomes the basis for every AI funding conversation, including the one in how to pitch your CFO on an AI budget.
What about the stories?
Anecdotes still have a place. A specific example of how a tool helped a customer, or spared someone a miserable afternoon, makes the numbers memorable. Use stories to illustrate the metrics, never to replace them. A good pattern in a report: one number, then one sentence of story that shows what the number means.
Mistakes I see at this stage
Reporting only usage. It answers “are people using it?” when the question was “is it worth it?”
Converting all saved time into dollars. Finance will discount the entire report if one number looks inflated. Use a realization factor and state it.
No baseline. Without a “before,” every “after” is an estimate. Capture baselines for every new pilot from now on.
Cherry-picking. Reporting only the successful use cases makes the whole report less believable. Include the ones that didn’t work.
Ignoring cost. Value without cost isn’t ROI. Include every cost line, including people’s time.
Measuring once. Value drifts as processes and tools change. Quarterly reporting catches that drift early.
Where does your team actually stand?
Measuring AI value is one of 24 questions in the AI Readiness assessment, which covers six dimensions: data, security, infrastructure, skills, use cases, and governance. The free version is 10 questions and gives you a score in a few minutes.
Get your free AI Readiness Score →
Want to see what the full assessment covers first? Flip through a complete 38-page sample report.
Related guides
- Baseline Before You Build: Metrics That Make AI Pilots Fundable
- Budgets and Alerts for Usage-Based AI Spend
- The 30-Minute AI Council: Lightweight Governance That Sticks
- How to Pitch Your CFO on an AI Budget
- The 6-Dimension AI Readiness Framework, Explained
Frequently asked questions
How do you measure AI ROI?
Estimate annual value as time saved per task, times tasks per year, times loaded cost, times a conservative realization factor, then add any cost avoidance you can quantify. Compare that with the full annual cost, including licenses, build time, running costs, and training.
Are AI adoption metrics a measure of value?
No. Active users and seats used show that people are using a tool, not that the business is better off. Usage can be high for trivial tasks or low for a few high-value ones. Track adoption as a supporting signal, and report outcomes as the evidence.
Why are hours saved not the same as money saved?
Time saved only reduces cost if it's redeployed to valuable work or avoids a cost you would otherwise have had. Report it as capacity and say what the capacity was used for; if you can't, don't convert it to dollars.
What should a quarterly AI value report include?
For each use case in production: the primary metric's baseline and current value, the change, estimated annual value with stated assumptions, annual cost, and status. Add portfolio totals, what you stopped and why, a short adoption summary, and next quarter's plans.
