19 January 2026
Reading a retention curve without panic
The first dip after day two is rarely a product emergency. Sometimes the SDK slept. Sometimes you changed what “came back” means.
Retention charts have a talent for producing urgency. A line that falls looks like users leaving, which feels like a story you can act on: rewrite onboarding, add a push, buy ads. In the studio we ask three dull questions first.
One: is this n-day, unbounded, or range retention? Mixing them between months is a rhetorical trick, even when it is accidental. Two: who is in the starting cohort — first open, first registration, first purchase? Three: did a release land that week which changed identify calls, permission prompts, or the event that marks a session?
Seams we see constantly
iOS backgrounding that starts a new session and looks like a return. Android webviews that never fire the completion event. A consent banner that delays the first event until after the user has already decided the app is empty. A “new user” definition that reset when you migrated from one analytics product to another.
None of these is exciting. All of them have sent a product team into a six-week onboarding rewrite that did not move the underlying habit, because the habit was never what the curve was showing.
A calmer drawing
We prefer retention split by first-week experience, not by a single activated flag. The flag is usually a committee. First-week experience is closer to something a user can remember: saw a live event, finished a lesson, hit a paywall, registered from a push. Those lines can fall at different speeds. That is information. A blended line is a mood.
The Atlas will not make the dip disappear. It will tell you whether the dip is a seam, a definition, or an actual corridor users do not wish to walk twice.