Metrics & Analytics

Cohort Analysis

A technique that groups users who share a common characteristic (usually acquisition date) and tracks their behaviour over time to reveal retention, engagement, and revenue trends.

What is Cohort Analysis?

Cohort analysis groups users who share a common starting point (usually their signup week/month) and tracks their behaviour over time. It reveals whether product improvements are actually being experienced by real users, not just new cohorts.


Reading a cohort retention table

`

Cohort | Week 0 | Week 1 | Week 2 | Week 4 | Week 8

Jan | 100% | 45% | 32% | 22% | 18%

Feb | 100% | 48% | 35% | 25% | 20%

Mar | 100% | 52% | 40% | 31% | 26%

`

Each row is an acquisition cohort. Each column is how many users were active N weeks after signup. Improving numbers from Jan → Mar indicates the product is getting better at retaining users.


What to look for

  • Flattening curves — retention stabilising above 0% = you have a retained core
  • Improving cohorts — later cohorts retain better = product improvements are working
  • Cliff at day 1/3 — massive early drop-off = onboarding problem
  • Revenue cohorts — expansion MRR in older cohorts = strong monetisation

Frequently asked questions

What's the difference between cohort analysis and funnel analysis?

Funnel analysis measures conversion through a sequence of steps at a point in time. Cohort analysis measures what happens to a group of users over time. Use funnels to find where users drop off in a flow; use cohorts to measure whether product changes improved long-term retention.

How many users do I need for reliable cohort analysis?

At least 100–200 users per cohort for statistical reliability. With smaller cohorts, retention curves are noisy. Early-stage companies should focus on qualitative interviews alongside cohort data rather than trusting small-sample cohort curves.

Apply Cohort Analysis to your real product data

PMRead ingests customer feedback, interviews, and Slack threads — and generates PRDs grounded in real evidence.

Try PMRead free →