Seven-day return is the most over-promoted number in App Analytics. It is easy to chart, easy to put in a fundraising appendix, and easy to misunderstand. A person who opens a meditation app on Sunday and not again until the following Saturday is a failure on some products and a success on others. D7 does not know the difference. You do.
In the cohort-windows week of The Release Lattice we ask students to write the names of the people their window ignores. Weekly shoppers. Parents who open an app after bedtime. Shift workers. If those people matter, D7 is a weather report from the wrong county.
Weather is still useful. You can say: in this climate, this many people came back within a week. You cannot say the product is healthy, or that a competitor is dying, or that an experiment “won” because a tile turned green. Winning requires a window that matches the job the app claims to do.
We also see D7 used to hide a release. A ship goes out on a Friday; the curve is read on Monday as if the binary had always been there. The Release Ledger exists so a cohort can be cut by version. Without that cut, you are reading last week’s weather as if it were this week’s crop.
If your board wants a single number, give them a sentence instead. “Returned in the billing month, among people who completed first session” is uglier than D7 and harder to game. Ugliness is often the point.