Anthropic tracks how often you swear at Claude Code.
Boris Cherny, who runs it, calls the dashboard the "F***s chart."
Most people saw that in the leak earlier this year, and laughed. But I saw it for the smart measurement decision it is.
Here's why:
It's industry standard to keep an eye on retention and churn. But churn is a "lagging" indicator. You only read it after the game is over, when it's too late. Sure, you can run resurrection mechanisms, but the damage is done and you've learned nothing in time to act on it.
Churn tells you who already left. Frustration tells you who's about to. It's a "leading" indicator.
And while churn is easy to measure, frustration is hard. Most people quietly get more annoyed before they leave. But if you can catch them in their frustration, then you have a strong early signal of churn. Cursing at your screen is one of the most honest, hardest-to-fake signals a user will ever hand you.
Now point that at your own team, your product, your AI rollout.
Sure, you can measure "tasks automated" and "hours saved," but you can guess the future trend if you can figure out whether people are slipping back to the old AI tool the moment nobody's watching; whether they're building workarounds instead of asking for a fix; whether they've just stopped complaining because they've stopped expecting it to get better.
Jeff Bezos: "When the data and the anecdotes disagree, the anecdotes are usually right."
Measure the thing that's still happening while you can still change it. Not the one that only tells you who already walked out.
What's the leading indicator that you're not measuring?