A retention chart is a small play. The axes are the set, the line is the plot, and the title is supposed to be the argument. Most titles we see in app analytics consoles are captions: “Retention,” as if the word were a place. It is not. It is a family of measurements that disagree with one another on purpose.
N-day retention asks whether a person did something on day N. Unbounded retention asks whether they did it on day N or any day after. The second curve is almost always prettier. Prettier is not a methodology.
Put the clock in the title
In Retention Geometry we require a title that a trustee could read without a tooltip. “Share of March authenticated users who opened the app on day 7” is ugly and sufficient. “Weekly retention” is a costume.
The product’s promise should pick the clock. A grocery app that claims to be a weekly habit can defend a seven-day view. A tax app that is honest about seasonality should not be scolded with the same clock. If your board wants a single number, give them a number and a paragraph that refuses to travel without it.
Averages that rescue cliffs
Smoothing a Tuesday-install cliff with a rolling mean is a directorial choice. Sometimes you need it to see a trend. Often you need the cliff, because it is the only honest record of a broken push certificate or a store review outage.
Theatre, in our usage, is any chart that would collapse if you labelled the window, the entry event, and the identity key. Our cohort intelligence page is the longer prejudice. The short version: grey it out until the three sentences exist.