A retention curve without a caption is a Rorschach test. One person sees product-market fit. Another sees a paywall. A third sees the week the App Store featured you and has not forgiven the chart for including it.
We teach three annotations before any colour choice. First, the birth event: what exactly placed a person in the cohort. “Signed up” is not a birth if half of those people never opened the app. Second, the return event: a session, a job completed, a paid week. Mixing them produces a line that cannot be argued with because it cannot be defined. Third, the window that flatters you. Seven-day retention on a weekly habit is a trick; call it what it is.
Churn is then a bend with a date. We ask students to write one sentence under the plot: who left, after which job, coinciding with which release or calendar fact. If the sentence cannot be written, the chart is not ready for a trustee.
A regional rail app in our winter table discovered that their “week-one cliff” was mostly people who bought a single journey and had no reason to return. The curve was honest once the birth event became “bought a season”. The number looked worse. The work became possible.
Bring a live cohort to Retention as Narrative if you want it marked in pencil. Bring pride if you want it marked twice.