Cohort analysis groups users by a shared characteristic — typically when they first used the product — and tracks their behaviour over time. Unlike aggregate metrics, cohort analysis reveals how retention changes as the product evolves.
When diagnosing retention problems, measuring the impact of product changes on specific user groups, or understanding lifetime value by segment.
- Define the cohort attribute: month of first use, acquisition channel, user tier, etc.
- Define the metric to track: retention rate, feature adoption, revenue, session frequency
- Build a cohort grid: rows = cohort, columns = time periods after start (D7, D30, D90)
- Look for: whether retention curves flatten (healthy) or keep declining (churn problem)
- Compare cohorts: are newer cohorts retaining better than older ones?
- Investigate outlier cohorts — what's different about those users or the product?
Users acquired through playlist sharing (social cohort) show 65% D30 retention vs. 38% for users acquired through paid ads. This tells Spotify that socially acquired users are substantially more valuable — and directly changed the growth strategy and budget allocation toward viral sharing features over paid acquisition.
Please contact the author for more information on these examples at linkedin.com/in/kshitijrege
- Using cohorts with too few users — need at least 30 per cohort for reliable patterns
- Confusing correlation with causation — other factors may explain cohort differences
- Tracking too many metrics per cohort — pick one clear question per analysis
- Lean Analytics — Croll & Yoskovitz