What I stopped measuring

We had four dashboards. One for traffic, one for errors, one the growth team owned and one nobody could remember setting up. Every Monday we looked at all four and left the meeting with the comfortable feeling of having looked at something.

In March I deleted three of them. Not as a statement — I was cleaning up a billing page and it turned out two of the four cost more per month than the server they were watching. Nobody noticed for five weeks. When someone finally did, the question was not where did the dashboard go. It was which one.

The number that survived

What we kept was the slowest page load our worst-off tenth of users saw yesterday. One number, one line, no comparison to last quarter. It is the only metric I have ever watched that changed what anyone did on a Tuesday.

The other three measured things that were true but not actionable. Traffic went up when we published, down when we did not. Errors moved when a dependency moved. The growth dashboard measured the growth team. All of it was real; none of it ever finished a sentence that started with so we should.

The test

I have started asking one question of any chart before it earns a place on a wall: if this number moved by a fifth overnight, would anyone here do something different tomorrow morning? Not would anyone care. Care is cheap. Would anyone act.

Almost nothing passes. That is the point. The charts that fail are not lies, they are just weather — worth glancing at, not worth a standing meeting. Keeping them on the wall taught the team that looking counts as working, which is the most expensive habit I know how to buy.

We are down to one number and a page of notes. The meeting is eleven minutes long now. I miss the dashboards slightly, the way you miss a desk you never used.