A retention curve is a manner of sitting in a chair. Drawn one way, the room relaxes. Drawn another, someone asks for engineering time. App Analytics does not force the first drawing, but it rarely obstructs it either.
The flattery usually begins with the window. If “came back” means any session in fourteen days, a weekly grocery app looks immortal. If it means a completed job — a basket, a transfer, a published listing — the same population looks mortal in a week. Neither definition is immoral. Switching between them in the same quarter, without a footnote, is.
The second kindness is the mixed cohort. Employees, contractors, and the QA farm are left in because exclusion is extra SQL. On a small app this is not a rounding error; it is a subplot. In Instrumentation Studio we ask students to print the list of internal accounts they will subtract. The list is always longer than the first draft.
A third kindness is smoothing. Rolling averages have a place in operations. They have a weaker place in a board pack that is already trying not to look at week one. If you must smooth, show the raw series in the appendix. People who hate appendices still deserve the option.
A less courtly cut
Pick a job, not a heartbeat. Define the return as a second completion of that job, not as the app opening because a push notification won a glance. Cut staff. Cut the cohort that never finished onboarding, or report it separately — mixing “never landed” with “landed and left” is how App Analytics becomes fog.
Then hold the definition for a season. The Metric Council exists because someone has to stop the definition moving when the news is bad. If you change the window, change the title of the chart. Do not keep the old name as if it were a brand.
I am not against looking after morale. I am against a curve that has been dressed for dinner while the kitchen is on fire. Retention After the First Week is the room where we practise taking the dress off without theatre.