Reading retention without vanity

Blended retention is a group photograph in which the grandparents stand in front. Your oldest users still open the app. They inflate day-30. New cohorts can be falling off a cliff behind them while the slide still says “stable.” Finance did not ask for a group photograph. They asked whether last month’s joiners still pay, still ship, still log a symptom — whatever the product actually sells.

Remove people who could not have returned

If you measure day-7 retention on Tuesday, anyone who joined yesterday does not belong in the denominator. This is obvious and constantly skipped. The honest cohort is people who have had the full window. Unbounded “users in the last 30 days” charts are how teams congratulate themselves for a marketing spike that has not aged yet.

Windows are a product decision

A meditation app and a payroll app do not share a clock. Copying “D1/D7/D30” because a blog used it is how vanity sneaks back in through the side door. In class we force a sentence: “A retained user is someone who did X inside Y days, and Y exists because of Z.” Z might be a billing cycle, a delivery slot, or a clinical protocol. If Z is “Mixpanel’s default,” start again.

Resurrection is not growth

People come back because a coupon landed, a season started, or iOS nagged them. Plot a resurrection slice beside your new-user curve. If the only green line is old users returning, you do not have a retention story; you have a reminder-notification story. Those can be legitimate. They are not the same map.

The worksheet we use is ugly on purpose: one table, acquisition week down the side, window across the top, core users stripped out in a second tab. When the second tab looks worse, you are finally looking at App Analytics instead of a souvenir. The guided version lives in Retention Cartography.