2 February 2026
When DAU hides a cohort collapse
Daily active users is a stock. Retention is a class photograph taken every week. Mixing them is how a product meeting congratulates itself while a single install week quietly empties.
In Retention Archaeology we ask students to hold DAU constant in a spreadsheet and vary two things: new users and the return rate of last month’s new users. It is easy to keep the daily line flat by buying installs. It is harder to notice that D7 for the January class is a third of what it was in October.
The treadmill
A treadmill looks like health because the room is never empty. People step on, people step off. App analytics vendors will happily plot the occupancy. They will not, unless you insist, plot whether the same shoes come back. Identity stitching matters here: if logout mints a new anonymous ID, your “returning” users are an accounting fiction.
Push notifications can disguise the same problem. A burst on Thursday lifts DAU and does not lift the cohort that installed in week one. We treat campaign-driven opens as a property on a session, not as proof that the product is sticky.
A practical cut
Define a cohort as people who completed a named intent in a calendar week — first_order_placed, not app_opened. Then ask what share of that class performed a second intent within your window. If you cannot name the second intent without arguing, you do not yet have retention. You have attendance.
United Kingdom apps often see a Sunday lift that is sport or shopping, not love of the feature. Annotate the calendar. A cohort that “recovers” every bank holiday is recovering from a closed shop, not from a product insight.
DAU can stay on the board. It should not be allowed to speak first.