A cohort is fixed at acquisition: once a customer joins (say) the March 2026 cohort by placing their first order, they stay in that cohort, and subsequent metrics measure the same people aging, not a churning denominator. Aggregate metrics describe a calendar period across customers of mixed tenure, so period-over-period reads are noisy: the mix shifts underneath the number. Cohort metrics describe how one acquisition group behaves relative to prior groups at the same point in its lifecycle.
The methodology earns its place because it makes LTV and CAC payback defensible. Aggregate LTV is computable, but less diagnostic than a cohort-level read: it mixes maturing cohorts, older cohorts shrinking through attrition, and recent acquisitions whose lifetime hasn’t played out, so movement in the aggregate cannot be attributed cleanly. A cohort revenue curve shows one group’s cumulative spend at month 3, 6, and 12, which you can extrapolate against prior cohorts at the same age. Payback period is similarly cohort-native: it reads the contribution-margin curve of the cohort whose CAC you spent, not blended margin across all customers.
Operationally, cohort analysis surfaces acquisition-quality drift early (a March cohort retaining worse than February at month 2 is visible months before blended LTV moves), measures retention impact of product changes, and prices paid channels by the cohort LTV they produce rather than blended LTV.
The most common pitfall is comparing young and mature cohorts at calendar-aligned readouts. A 3-month-old cohort always looks worse than a 12-month-old one at the same date; the apples-to-apples view is months-since-first-order.