The problem is not that someone spent $30,000 on AI. The problem is not knowing whether they created $90,000 of value or $9,000 of rework.
That is the AI ROI problem most companies are about to face.
A high AI bill is not automatically good or bad. It depends on what changed in the work.
If an engineer is using AI heavily and shipping more accepted pull requests, reducing cycle time, improving test coverage, and getting fewer review comments, that may be a great investment. Spend the money.
But if another engineer is using the same amount of AI and their pull requests are getting rejected more often, rewritten by teammates, or creating bugs downstream, that is not productivity. That is automating problems with a bonus creation cost.
Same spend. Completely different outcome.
Usage Is Not ROI
AI usage by itself tells you almost nothing.
It can tell you who is using the tool, how often they are using it, and how much they are spending. That is useful operational data. But it does not answer the business question.
The business question is: is the tool making the work better?
That distinction matters because AI can create the illusion of productivity. More prompts, more tokens, more generated documents, more code, more summaries, more activity. The dashboard looks alive. The usage chart goes up. The steering committee gets a slide.
But activity is not value.
If AI creates output that someone else has to review, reject, rewrite, debug, explain, or clean up, the company did not eliminate work. It moved the work downstream.
Measure the Work, Not Just the Tool
AI ROI needs to be connected to the work itself.
For engineering, that might mean accepted commits, pull request review rejection rates, bugs introduced, cycle time, test coverage, rework, and deployment quality.
For support, it might mean tickets resolved, escalation rate, customer satisfaction, reopened tickets, handle time, and policy compliance.
For sales, it might mean follow-up quality, pipeline movement, conversion, cycle time, manager edits, and customer response.
For analytics, it might mean fewer duplicate reports, faster decision cycles, fewer metric disputes, better data trust, and clearer ownership of business logic.
Because “used AI a lot” is not a business outcome. It is just the receipt.
Same Spend, Different Outcome
Two people can use the same AI tool with the same budget and produce completely different economics.
One person turns AI into leverage. They produce better work faster, reduce friction for teammates, shorten cycles, and improve quality. Another person creates more output, but the output requires more review, more correction, more meetings, and more cleanup.
One is value creation. The other is rework with a nicer interface.
This is why companies need to be careful with AI adoption dashboards. High usage can mean high leverage. It can also mean someone is generating five versions of mediocre work and asking the rest of the team to clean it up.
Very innovative. Very expensive.
The Better AI ROI Question
The old software question was: are people using the tool?
The better AI question is: is the tool making the work better?
That means leaders need to connect AI spend to workflow outcomes, not just user activity.
Did quality improve? Did cycle time decrease? Did rework go down? Did customers get served faster? Did decisions improve? Did the team create more value with the same resources?
Those are the questions that separate AI spend from AI ROI.
A $30,000 AI bill might be waste. It might also be the best investment you made all year.
You only know if you measure the work, not just the usage.
