In the last autumn desk a team showed a retention chart that had become a mascot. The curve lifted in the week they shipped a home-screen redesign. Product wanted the letter to open with congratulations. Finance, reading the same file later, asked why refunds had also lifted. Both were looking at a true picture of different people.
App analytics platforms default to a user who did anything. “Anything” includes a push that opened a splash and a paying member who completed a weekly habit. When a release increases cheap opens — a badge, a louder notification — D7 can rise while the paying contour thins. The coastline looks healthier because the tide brought driftwood.
The correction is not a more beautiful chart. It is a cut you can name in a sentence. We ask students to draw three curves on the same axes: all users, users who reached a paid or otherwise costly action in the first session window, and users acquired in the same calendar week last year. The third curve is unfashionable and usually the most honest in the United Kingdom, where December and January pretend to be product strategy.
If the paid contour falls while the all-user contour rises, the metric letter must lead with that split. Hiding it in a footnote is how a room trains itself to applaud the release. Signal Architecture spends a full evening on this; Cartograph Circle spends a day assigning the contours to terrain so nobody can “tidy” them back into one line before a board meeting.
A mild practical limit: if your identity graph still stitches on email alone, the paid contour will be noisy. Fixing that is instrumentation work, not a filter in the BI tool. We would rather you publish a noisy honest cut than a smooth fiction.