Cohort retention

Cohort retention is a method of measuring customer retention by grouping customers by their acquisition period (typically month) and tracking the proportion still active in each subsequent period. The output is a triangular table where rows are cohorts, columns are months-since-acquisition, and each cell is the percentage of the cohort still active.

Dr Waqas Rafique · Founder & CTO · PhD, Statistical Machine Learning
LinkedIn · About

Why aggregate retention metrics lie

A 95% monthly retention rate sounds great. But if your last cohort is retaining at 80% while older ones retain at 99%, the aggregate hides a deteriorating product. Cohorts unmask the trend.

Logo retention vs revenue retention

Logo retention counts customers (or accounts). Net revenue retention (NRR) sums revenue and divides by the cohort's revenue at month 0. NRR > 100% means the cohort is expanding net of churn — the metric SaaS investors fixate on.

How to read a cohort table

Look across rows: are recent cohorts retaining better than older ones (product is improving)? Look down columns: at month 3 retention, is it stable cohort-over-cohort or trending down? Stable bad is worse than trending bad — trending bad implies a recent change you can find and fix.

Common mistakes

1. Using calendar quarters not cohort quarters. 2. Comparing cohorts of very different sizes without showing absolute counts. 3. Ignoring seasonality (a November e-commerce cohort always looks bad in February). 4. Not separating new-business retention from upsell-driven retention.

Alternatives

Survival analysis (Kaplan-Meier curves) for cohort analysis with non-monthly time bases. Engagement-based cohorts (e.g. cohort by feature first used) for product analytics.

Frequently asked questions

What's the difference between churn and cohort retention?

Churn is an aggregate metric (the % of customers who leave per period). Cohort retention disaggregates churn by the period a customer was acquired in, which reveals whether your product is improving or degrading over time. Aggregate churn can hide a deteriorating product if your customer base is growing.

What's a good cohort retention curve?

For SaaS: M1 retention >85%, M3 >80%, M12 >70%, with the curve flattening (asymptote) by M6. For e-commerce: M1 30-40%, M3 15-25%, M12 8-15%.

How do I read a cohort heatmap?

Rows are cohorts (acquisition months). Columns are months-since-acquisition (M0, M1, M2, ...). Each cell is the % of the cohort still active. Look across rows for whether recent cohorts retain better than older ones (product improving). Look down columns for stability of retention by offset over time.