"My company uses AI usage as a performance metric in quarterly reviews. We have lost the plot."
That comment got 327 likes on an AI cost video. Underneath it, an engineer at a Fortune 100 company described a $1000 per month token budget and said adoption is crucial to job prospects. Two people, same structure. Spend is the score.
Here is what happens next, and it is completely predictable. People do not start solving harder problems. They start routing work through a model that they could have done in a minute, running the premium tier on tasks that never needed it, and asking for three drafts where one would do. The behavior is rational. You told them tokens burned is the number. They burn tokens.
I watched the same failure with cloud spend. Big organizations went through a phase where migrated workloads counted as progress, so teams migrated things that should have been deleted. Nobody lied. The metric was just measuring motion, and motion is the easiest thing in the world to produce.
The awkward part is that the intent is reasonable. A company that spent two years watching AI tools go unused wants proof that people are using them. Usage is the only number sitting there ready to be counted, so it gets promoted into a target. Then the number goes up and tells you nothing about whether the work got better.
Three things are more useful and only slightly harder to collect.
Cost per completed task. Pick a repeating unit of work your team actually ships: a support resolution, a pull request, a report. Divide AI spend by units completed. If usage doubles and this number drops, adoption is working. If usage doubles and this number holds flat, you bought activity.
Rework rate. How often does AI output get thrown away or fixed by hand? This is the cost that never shows up on the invoice, and it is the first thing that gets worse when people are pushed to use a tool they do not need.
An owner per recurring AI workflow, with keep-or-kill authority. Not a reviewer. Someone who can turn it off. Workflows nobody can kill are how a $1000 budget becomes a $10,000 one without a decision ever being made.
None of this argues against adoption. Adoption is fine. Measuring it in tokens is the part that quietly changes what your team optimizes for, and by the time it shows in the bill, the habit is a year old.
I run AI cost audits, and this is usually the conversation behind the numbers. If you want to look at yours, book a time at cloudyeti.io/meet.
Book a call