Retention and cohorts: choosing the grain that matches the decision

Most subscription businesses eventually ask the same kinds of questions: where trials convert, which users stick around, and whether acquisition quality is drifting. The tooling varies—billing platforms, event analytics, spreadsheets—but the failure mode is often the same: the cohort is technically correct for something nobody is trying to decide.

Weekly cohorts can smooth noise when marketing wants a directional read on a channel. Daily cohorts can help product teams spot onboarding regressions after a release. The cohort “start” should line up with the business moment you care about—signup, first purchase, activation milestone—and that choice should live in a metric definition everyone cites, not only in a single analyst’s notebook.

Retention curves are easy to over-interpret. Segmenting by a few stable dimensions—channel, price tier, region—usually beats adding ten micro-slices. The goal is to find segments where curves diverge enough that prioritization changes, not to prove the dashboard can slice infinitely.

Executives rarely need more tabs; they need a short narrative: where value leaks, what to stop funding, and what experiment deserves another iteration. Good analyses end with explicit options, assumptions stated plainly, and a note on what extra data would shrink uncertainty.

When conversations turn to lifetime value and acquisition cost, ranges and sensitivities usually beat false precision. The point is to make tradeoffs legible, not to win a debate with a single hard number pulled from a black box.