Almost every team we work with has an activation metric. Almost none of them can say what it predicts.
The usual definition looks like this: a user is activated when they complete steps one through five of onboarding within seven days. It is easy to compute, easy to put on a dashboard, and it moves when you change the onboarding flow. That last property is the problem.
A metric that only measures your own funnel
If activation is defined as “finished the setup wizard”, then shortening the wizard raises activation. So does pre-filling fields, so does removing a confirmation step. None of those changes made the product more valuable. They made the wizard shorter.
You can test this in an afternoon. Pull two cohorts: users who activated under your current definition, and users who did not. Then look at whether they are still around in ninety days. If the two curves converge, your activation metric is measuring onboarding friction. It is a funnel diagnostic, which is a legitimate thing to have, but it is not activation.
What a real activation metric looks like
A useful activation event has three properties.
It is a product action, not a setup action. Connecting a data source is setup. Building the first dashboard that someone else on the team opens is product.
It separates the retention curves. Plot ninety-day retention for users who did the action and users who did not. If the gap is under fifteen points, keep looking.
It happens early enough to act on. An event that only occurs in month four is a great predictor and a useless operational signal. You want something in the first week or two, where an intervention still changes the outcome.
Finding yours
Take your last six months of accounts. For every meaningful action in the product, compute the ninety-day retention of accounts that did it in their first fourteen days against accounts that did not. Sort by the gap. The top of that list is your candidate set.
Then apply judgement, because correlation will hand you some nonsense. Accounts that invited five colleagues retain better, but inviting colleagues is often a symptom of a team that had already decided to adopt. Ask whether you can plausibly cause the action, or whether it merely marks accounts that were going to stay anyway.
The uncomfortable part
Once you have a defensible activation metric, it will usually be lower than the one you were reporting. Considerably lower. That number is the one worth improving, and the drop is not a regression. It is the first time you were measuring the thing you meant to measure.
