AI adoption has become a metric in itself. Usage dashboards. Token consumption leaderboards. Mandates to "use AI" in performance reviews. Reports of hours saved per employee, per week, per team.

None of it answers the only question that matters: what value did it produce?

Time saved is not value if the time goes into more of the same. More output is not value if nobody needed it. A 16-page strategy document generated in ten minutes is not progress if nobody reads it, and worse if someone acts on it without understanding it. When we reward consumption, we get consumption. Teams learn to look busy with AI rather than to be better with it.

In this talk I'll look at the mistakes organisations are making as they roll out AI: maximising usage instead of outcomes, automating work that shouldn't exist, measuring activity because value is harder to measure, and confusing speed with progress. Then we'll look at what to do instead. How to start from the problem rather than the tool, how to define value before you deploy, and how to measure whether AI is changing anything that matters, accounting for what is possible today.