11 November 2025
How to read signup-week cohorts without overclaiming
A plain guide to comparing return rates across signup weeks, including the traps that appear when volumes are uneven.
Signup-week cohorts are useful because they freeze a group of people who arrived under similar conditions. The table then asks a simple question: of those who signed up in a given week, how many returned in later windows?
Start by locking three definitions before you compare anything: the signup event, the return event, and the windows you will inspect. If “return” means any session for one product and a paid action for another, the columns are no longer comparable across applications.
Uneven volumes
A week with forty signups will swing more wildly than a week with four hundred. When volumes are thin, prefer longer windows or merge adjacent weeks only after noting the merge. Do not hide small denominators behind a smooth percentage.
What to write in the brief
Name the pattern, the window where it appears, and one caveat that could explain it. A retention cohort reading that skips the caveat invites a product change that the table cannot support.