Essay 02

Usage is not value.

A high adoption graph can still leave the hardest question unanswered.

I like seeing AI usage go up.

Really. I do.

It means people are trying. It means the rollout did not land with complete silence. It means someone, somewhere, thought the tool was useful enough to open again.

That matters.

But usage is still only the beginning of the measurement conversation. Not the victory lap.

Adoption tells you people touched the tool.

It does not tell you what happened to the work.

Someone can use Copilot ten times in a day and still finish the same amount of work, at the same quality, with the same stress, in the same process. Another person can use it twice and remove an hour of painful manual preparation.

The usage dashboard will notice the first person faster.

The business should care about the second person too.

Agree or disagree. I know some adoption people will push back on this. Fine. Adoption is still important. I just do not want it carrying the whole value story by itself.

Usage is evidence of attention. Value is evidence of change.

The object of measurement is the work.

When we measure AI, it is tempting to measure the tool as if the tool itself creates value.

But the value usually appears in a work unit:

• a customer meeting brief,
• a support ticket,
• a policy answer,
• a proposal section,
• a project update,
• a document review.

Those are easier to measure than a vague phrase like productivity. They have a start, an end, an owner, a volume, a quality bar, and a cost.

They also have dependencies. Data. Permissions. Policies. The annoying setup work.

If those are bad, the measurement gets noisy fast.

Three questions beat one adoption metric.

Before celebrating usage, ask three simple questions.

First: what work did AI touch?

Second: what changed in that work?

Third: did the change survive review, cost, risk, and scale?

If we cannot answer those, we are still measuring movement more than value.

A better dashboard

I do not want fewer dashboards. I want better ones.

Show usage, yes. Then show the work units that usage affected. Show median task time. Show p90 task time. Show review time. Show quality. Show escalation rate. Show cost per accepted output.

And please show confidence. If half the numbers are guesses, say that. I would trust the dashboard more, not less.

That is when adoption becomes more useful.

AI adoption is good news.

It is just not the finish line.

The real story starts when the work changes.

At least, that is where I am starting. I might adjust after a few more real examples. That would be a good sign.