Cohort windows for subscription apps

A yearly plan measured on a thirty-day engagement clock will look like a ghost town. That does not mean the product failed. It means the map used the wrong contour interval.

Consumer playbooks love D1, D7, D30 because social products ask to be opened like a newspaper. Subscription software often asks to be opened like a solicitor: rarely, expensively, on time. If you copy the newspaper clock, your “retained” number will punish people who did the job correctly last month and will not need you until the next invoice run.

Two clocks, labelled as two clocks

Keep a billing clock and an engagement clock. The billing clock answers “did money arrive, and did they cancel.” The engagement clock answers “did they do the work the product exists for.” Mixing them produces a north-star nobody can move without cheating. In Retention Cartography we make participants write both sentences on the same page so a CFO can see the join.

Trial is not month one

A fourteen-day trial sitting inside a “first 30 days” cohort contaminates the cell. Trialists who never intended to pay look like churn. Paying customers who barely used the trial look like a miracle. Split the acquisition event: `trial_started` is not `subscription_started`. Your dictionary should already have said so; if it has not, you are doing App Analytics on a pun.

Annual plans need annual honesty

You may still watch weekly engagement for product quality. Do not brief the board that annual subscribers “churned” because they were quiet in week six. Report renewal risk with leading jobs (exports, seats invited, payroll runs) that historically precede cancel, not with a D30 copied from a meditation app. If you lack those leading jobs, that is a research problem, not a chart colour problem.

The window worksheet from week three of the flagship is the artefact people keep. If you want it marked rather than guessed, the seat is described on the Retention Cartography page.