Your users can leave while the account stays
A worked account-and-user example shows how team turnover, concentrated use and new participants can tell different stories about retention.
Strategy, retention, and the business of product
A metric can move because behavior changed, because the denominator changed, or because the clock did. These guides help you choose the unit of analysis, define a meaningful return, and distinguish product progress from a change in counting.
Choose whether the account, the person, or both must keep returning.
Define who can use a feature before reporting adoption.
Use a holdout to ask whether the message changed return behavior.
A worked account-and-user example shows how team turnover, concentrated use and new participants can tell different stories about retention.
A six-person example shows how valid identity joins change retention counts, plus a practical way to reconcile raw events with resolved customer histories.
Define eligible accounts, exposure and repeat use before judging a feature launch. A worked example shows why the denominator changes the decision.
Build a task ledger for background products: expected work, completion, freshness and delivery. Distinguish quiet success from automation that has stopped.
Use failed demand by location, time and job to decide whether a marketplace needs more supply, better matching or more accurate availability.
Early retention can rise while lasting use stays flat. Read the cohort definitions, observation windows, and return behavior before treating a higher curve as progress.
A reproducible hypothetical comparison separates annual-plan cash timing, monthly-normalized value and the renewal decisions a team can actually observe.
Separate customer activity, metered consumption and money due before classifying usage-based churn. A five-account example shows why zero invoices differ.
Use stable assignment, a useful return outcome and an honest comparison window to measure whether a lifecycle reminder changes product behavior.
More paid starts can conceal interrupted work and quick cancellations. Evaluate upgrade prompts across the original eligible audience and through a meaningful paid-use horizon.