What is Cohort Analysis?
Cohort analysis is a method of analyzing user behavior by grouping users into cohorts that share a common characteristic — most commonly the month/week of acquisition — and tracking how each cohort’s metrics evolve over time. It separates “how is the product getting better?” from “how is the marketing mix changing?” — both can mask each other in aggregated reporting.
Cohort types
- Acquisition cohorts: Grouped by sign-up date (Jan 2025 cohort, Feb 2025 cohort, etc.). Most common.
- Behavioral cohorts: Grouped by an action (users who completed onboarding, users who used Feature X within 7 days)
- Demographic cohorts: Grouped by attributes (industry, plan tier, geography)
- Predictive cohorts: Grouped by a model prediction (high-LTV likely, churn risk high)
The retention curve
Plot cohort size over time as % of starting cohort. Three patterns matter:
- Smiley (best): Curve drops then rises — Promoter behavior + viral growth
- Flattening (good): Drops then stabilizes — sustainable retention floor
- Continuously declining (bad): No stickiness — product-market fit problem
Common cohort metrics
- % retained at month N
- Revenue per cohort over time
- Net Dollar Retention (NDR) by cohort
- CAC payback by cohort
- Feature adoption rate by cohort
Practical implications
Run cohort analysis monthly. The most actionable test: do recent cohorts retain better than older cohorts? If yes, the product is improving. If no, the early users may have been your best-fit ones (a sign of weakening ICP fit or onboarding regression). In VC due diligence, cohort retention curves are non-negotiable — be ready with at least 12 monthly cohorts.