Reading week-1 retention without lying to yourself

Week-1 retention is the most argued chart in our KL studio. Marketing wants a wider window; product wants a narrower one that flatters a recent release; data wants statistical hygiene nobody has time for.

Start with three rules:

1. Pick one return event that matches habit, not any session.
2. Hold the window fixed for a quarter before comparing releases.
3. Annotate confounders — festive holidays in Malaysia, school calendars, or a viral spike that will never repeat.

We often find “improved retention” that is really a mix of bot cleanup and a dead feature removed from the denominator. A Retention Readout Session exists to force that honesty in one afternoon instead of three Slack threads.