12 March 2026
Why Day-7 churn is a misleading KPI
Day-7 is popular because it fits a sprint review. A week is short enough to feel operational and long enough to look like a “real” window. In mobile app analytics it is also a trap: it mostly measures whether onboarding delayed disappointment, not whether anyone intended to stay.
What Day-7 actually captures
A large share of installs never wanted the product. They clicked an ad, borrowed a phone, or completed a one-time task (a government form, a concert ticket, a coupon). Those users exit between hour six and day three. If your onboarding is long, some of them are still technically “active” on day seven because they have not finished the tutorial. The KPI improves while the product has not earned a second genuine session.
We see the inverse in Thailand’s prepaid segment. A user who intends to stay may still go quiet across a payday gap. Day-7 calls them churned; day 23 often shows them back. The median silent-exit window we teach in Cohort Signals sits later than a week for a reason.
A better early clock
Pick a window from your own hazard, not from a blog. Plot survival from first meaningful action — first order, first completed workout, first guild join — not from install. Then mark where the slope steepens. If that kink is day 11, report day 11. If it is day 23, live with the awkward number.
Pair the window with session depth, not session count. Three two-second opens on day six are not loyalty. One twenty-minute session on day nine might be. Day-7 churn treats them as the same story told in opposite directions.
What to tell a board
Boards like weekly numbers. Give them two: a short onboarding completion rate, and a longer risk window named in days. Refuse to merge them into a single “health” percentage. The merge is how Day-7 keeps getting rehired.
If you want the classroom version of this argument, the vanity funnel autopsy is module one of the flagship. If you only wanted a warning label, you now have one: Day-7 is a calendar convenience, not a theory of leaving.