What Is Cohort Analysis?
Cohort analysis groups users by a shared characteristic (usually install date) and tracks their behavior over time. Instead of looking at aggregate metrics that mix new and old users, cohort analysis reveals how specific groups behave as they age.
If your overall D30 retention is 12%, that number blends users who installed a year ago with those who installed last week. Cohort analysis separates them, showing whether January had 10% D30 retention while March had 15%.
Why Cohort Analysis Is Essential
Aggregate metrics can be misleading. A common scenario:
- Overall DAU is growing
- Retention rate looks stable
- Everything appears healthy
But cohort analysis reveals DAU growth comes entirely from aggressive paid acquisition while recent cohorts have worse retention. The product is actually deteriorating, masked by volume.
Types of Cohorts
Acquisition Cohorts (Time-Based)
Group users by when they started: install week, first purchase month, or subscription start date. This is the most common and default cohort type.
Behavioral Cohorts
Group by what users did: feature adoption, onboarding completion, or engagement level. Powerful for finding which actions predict long-term retention.
Channel Cohorts
Group by acquisition source. Do organic users retain better than paid? Do Meta users behave differently from TikTok users?
Revenue Cohorts
Group by spending: free vs paid, spending tier, or first purchase timing.
Reading a Cohort Table
| Cohort | Size | D1 | D7 | D14 | D30 |
|---|---|---|---|---|---|
| Jan Week 1 | 5,000 | 28% | 14% | 10% | 7% |
| Jan Week 2 | 6,200 | 30% | 16% | 11% | 8% |
| Jan Week 3 | 5,800 | 32% | 17% | 13% | 9% |
| Jan Week 4 | 7,100 | 35% | 19% | 14% | 10% |
Retention improves each week. Something changed: maybe a product improvement, channel mix shift, or seasonal effect.
Step-by-Step Method
Step 1: Define the Cohort
Decide what characteristic defines the group. Start with install week or month.
Step 2: Choose the Metric
Retention rate (most common), revenue per user, feature adoption, or session frequency.
Step 3: Set Time Intervals
Daily for the first 30 days. Weekly or monthly for longer-term analysis.
Step 4: Build the Table
Use your analytics tool or query your database. Each row is a cohort, each column a time interval.
Step 5: Analyze Patterns
- Horizontal trends: How does each cohort evolve?
- Vertical trends: Are newer cohorts performing better or worse?
- Anomalies: Any cohort that behaves very differently?
Step 6: Investigate Causes
When you spot a pattern, dig into product changes, acquisition mix shifts, or external events.
Revenue Cohort Analysis
Track cumulative revenue per user for each cohort. This reveals:
- How quickly different cohorts monetize
- Whether monetization improves with product changes
- Which acquisition channels produce highest-value users
- Whether LTV projections match reality
Common Mistakes
Cohorts that are too small. Ensure at least 500-1,000 users per cohort for meaningful data.
Ignoring external factors. A holiday promotion cohort will behave differently from an organic one.
Not acting on findings. Every analysis should end with actionable changes.
Tools for Cohort Analysis
- Amplitude - Behavioral segmentation with cohort analysis
- Mixpanel - Strong cohort reports with funnel integration
- Firebase Analytics - Free retention and revenue cohorts
- Custom SQL - Maximum flexibility for data warehouse teams