← Resources

Outcomes vs activity: why AI adoption is not AI return

Quick answer: AI adoption is not a good ROI metric on its own. Adoption measures activity: how many people use AI and how often. Return is measured in outcomes: lower cost, better quality, faster delivery or more revenue, expressed in dollars. High adoption can coexist with zero financial return, and the data says it often does.

Key takeaways

  • Nearly nine in ten organizations now use AI in at least one function, yet only 37% report any enterprise EBIT impact (McKinsey, State of AI 2026).
  • MIT NANDA described the pattern as high adoption with low transformation, and found 95% of organizations saw no measurable return from generative AI (The Register).
  • Activity metrics are useful as leading indicators. They become dangerous when they are reported to the board as return.

What is the difference between an activity metric and an outcome metric?

An activity metric counts what people do with AI. An outcome metric measures what changed in the business because of it.

Activity metric Why it is tempting Outcome metric to use instead
Weekly active AI users Easy to pull, always goes up Cost per unit of work in the teams using AI
Prompts or tokens per user Shows engagement Output per FTE, or cost per resolved task
Hours saved (self-reported) Sounds like money Change in contractor spend, overtime or planned hires
Percent of code written by AI Shows depth of use Cost per shipped change, defect escape rate
Number of AI use cases live Shows momentum Number of use cases with a measured payback

Why do companies report activity instead of outcomes?

Three reasons. Activity data is automatic: it comes straight from vendor dashboards. Outcome data needs a baseline that most programs never set. And activity always looks good early, which is when sponsors most need a story.

The cost is credibility. When a board hears that 80% of employees use AI but sees no change in margin, the next budget conversation gets harder for every AI program, including the ones that work.

When are activity metrics useful?

Activity metrics are useful as early warning signals. If a program has low adoption three months in, it will almost certainly not hit its outcome target. They help diagnose why a program is not working. They should sit beside outcome metrics in a report, never in place of them.

How should finance teams reframe AI reporting?

  1. Lead with dollars. Every AI program report opens with fully loaded cost and measured return, even if the return is zero so far.
  2. Move activity to a supporting row. Adoption and proficiency explain the result. They are not the result.
  3. Separate realized from potential. Time saved is capacity. It becomes return only when it changes a cost or output figure.
  4. Set a decision date. Each program gets a date by which it must show outcome evidence, or it is changed or stopped.

Is this a criticism of AI usage tools?

No. Usage and proficiency data can be strong evidence about why a program is or is not working. The point is about what gets reported as return. A finance-grade AI report starts with cost and outcomes, then uses activity data to explain them.

Frequently asked questions

Is AI adoption a good ROI metric?

No. Adoption is a leading indicator of whether a program might deliver value. ROI requires a measured change in a business outcome, converted into dollars, against the fully loaded cost.

What metrics should replace adoption in a board report?

Fully loaded program cost, the change in one outcome metric against a baseline, the dollar value of that change and payback period.

Are hours saved a valid AI benefit?

Hours saved are a valid capacity benefit. They count as financial return only when they reduce cost, avoid a planned hire or increase output that the business can sell.

How long before an AI program should show outcomes?

It varies by program, but most should show a measurable change in the primary outcome within two to three quarters of full rollout. Set the expected date before launch.

Sources

Get updates