Choose the event and the retained status explicitly. First payment, registration and first completed job answer different questions. A paying account that has stopped using a product also differs from an active user.
Build one table with fixed denominators
A fictional monthly service defines a cohort as customers making their first payment in a calendar month. “Retained” means still paying at the specified month-age checkpoint. The starting count stays fixed for that cohort's percentage calculation.
| First-payment cohort | Starting customers | Still paying at month 1 | At month 2 | At month 3 |
|---|---|---|---|---|
| January | 100 | 80 | 70 | 65 |
| February | 80 | 68 | 56 | Not yet observed |
| March | 120 | 102 | Not yet observed | Not yet observed |
January's month-1 retention is 80 / 100 = 80%. February's is 68 / 80 = 85%, and March's is 102 / 120 = 85%. Missing cells are unobserved ages, not zero retention.
Define exactly when each checkpoint is measured. If the business allows cancellation and later reactivation, decide how that appears and document it. Otherwise different analysts can produce different tables from the same accounts.
Compare equal ages
It would be misleading to call January worse because its month-3 count is lower than March's month-1 count. January has been observed for longer. At month 2, January and February both retain 70% of their starting groups: 70 / 100 and 56 / 80.
Even equal-age comparisons need context. Acquisition channel, customer job, offer and onboarding may have changed. A cohort with only a few starting customers can have large percentage swings from a handful of events.
Counts and revenue are different views
Suppose a separate starting group has $10,000 of recurring revenue. Cancellations and downgrades remove $1,500; expansions from that same group add $1,000. Under this stated definition, gross revenue retention is 85% and net revenue retention is 95%.
New customers are excluded from those starting-base retention measures. If new sales add another $3,000, total recurring revenue can grow even while the original group shrinks. Keep that acquisition bridge separate from retention.
These definitions should be reconciled with the metric conventions of any company report being analyzed. Do not substitute customer counts for revenue amounts or mix one-time charges into a recurring base without explanation.
Investigate the moment of departure
If departures cluster early, inspect setup and the first promised result. If they cluster at renewal, investigate value, alternatives, service quality and price. Support records, interviews and product behavior can strengthen or challenge the hypothesis.
Use the customer-promise guide to specify what success meant for that segment. When price is part of the change, connect the analysis to the price-test guide.
Choose one test and follow the cohort
Write the suspected cause, the change and the outcome you expect. Measure early completion of a useful task and later retention. A prettier denominator is not a better customer outcome.
For the cash consequence of early departures, read acquisition payback. For the test record, use the experiment guide.
Sources and limits
The wider distinction among revenue, expenses and cash is explained in the SEC's financial-statement guide. The cohort definitions, starting-base retention conventions, table and calculations are original fictional HBW examples. No universal acceptable retention rate or causal effect is claimed.