Journal
Why retention curves still mislead careful teams
A retention curve is often the most honest object in a product review. It is also a blender. Mix two audiences with different clocks and you get a shape that nobody lives inside.
Consider a weekend sports app and a weekday commuting utility sharing one “weekly active” line because they share a login. The curve looks stable. The sports slice is seasonal. The commute slice is tired of a permission they granted in October. App Audience Intelligence asks you to unblend before you celebrate or panic.
Smoothing is the second trap. A rolling average can hide a Tuesday crash after a push that only the most loyal slice still receives. If your push audience is not the same as your install audience, the curve is a diplomatic document.
We ask studio participants to draw the curve twice: once as shipped, once with a named exclusion (staff, library kiosks, a partner cohort). If the second drawing changes the story, the first drawing was furniture. Furniture can stay in the room. It should not get a speaking part.
Careful teams still get caught because caution feels like adding more segments until the chart is unreadable. The craft is the opposite: fewer slices, each with an exclusion list, each with a claim marked observed or inferred. Hoped lifts stay in a dated appendix.
If you want a marked attempt at this, the Claim court week in Cohort Signal Studio exists for that discomfort.