Journal · 3 November 2025
A retention curve is a story with missing pages. Day 0 is whoever you chose to count as a beginning. Day 1 is whoever came back according to a clock that may not match the product’s rhythm. By day 7 the line has already been used in a Slack channel as evidence of character. We ask students to print the curve, put it on the table, and write in the margins before anyone types a headline.
First annotation: who is in day 0? Installs, sign-ups, or first value? A fitness app that starts the clock at install will look like a tragedy by Thursday; many people download before a class exists in their diary. Start at “class completed” and the curve is smaller and less melodramatic. Neither is wrong. Mixing them in one meeting is.
Second: what counts as a return? Any open, or the same job done again? Push notifications can manufacture day 1. So can a password reset. In Cohort Intelligence we force the room to pick one definition and keep it for the nine weeks. Changing the definition mid-programme to “make D7 kinder” is how vanity returns through a side door.
Third: mix. New OS versions, a holiday, a viral week of the wrong users, a paywall experiment — all of these bend the line. A dip after a release is not automatically a verdict on the release. We draw a small table of confounders under the chart. It looks amateur. It prevents a product manager from rewriting the roadmap overnight.
Unbounded panic usually comes from comparing your curve to a blog post about a marketplace in 2014. Apps have different heartbeats. A tax app should not look like a chat app. If your curve flattens because the job is seasonal, the honest move is to say so, not to invent a streak feature.
The house does not promise a prettier D7. We promise that you will know what the ink is made of. That is slower than a template. It is also why visiting operators in week nine ask fewer theatrical questions.