Cohort analysis

Cohort analysis groups users who share a common characteristic - typically when they signed up or first used the product - and tracks their behavior over time. Instead of looking at all users in aggregate, cohort analysis compares how different groups behave. This reveals trends, improvements, or problems that aggregate metrics would hide.

Why it matters

Aggregate metrics lie - or at least mislead. Consider a product with 10,000 monthly active users for three consecutive months. Stable, right? But what if each month you acquire 3,000 new users while losing 3,000 existing users? The aggregate looks healthy while the underlying reality is a retention crisis.

Cohort analysis matters because it reveals these hidden dynamics. By tracking specific groups over time, you see whether your product is genuinely improving - whether newer users stick around better than older ones, whether feature changes help or hurt retention, whether seasonal patterns exist.

How cohort analysis works

The basic approach:

Define cohorts. Group users by a shared characteristic. Most commonly, this is signup date (all users who joined in January form one cohort), but could be acquisition source, plan type, or any relevant dimension.

Track over time. For each cohort, measure the behavior you care about at regular intervals - Week 1, Week 2, Week 4, Week 8, etc.

Compare cohorts. Look at how different cohorts perform. Do later cohorts retain better than earlier ones? Where do cohorts diverge?

Visualize patterns. Cohort data is often presented as a retention table or chart that makes patterns visible.

The classic retention table

A cohort retention table might look like:

Each row is a cohort; each column is time since signup. This table immediately shows retention improving over time - March and April cohorts retain better than January. Something changed for the better.

Types of cohort analysis

Time-based cohorts. Group by when users joined. Most common for understanding retention trends over time.

Behavior-based cohorts. Group by actions taken. Compare users who completed onboarding vs. those who didn't; users who used a specific feature vs. those who didn't.

Acquisition cohorts. Group by how users arrived - organic search, paid ads, referral. Understand which channels produce sticky users.

Segment cohorts. Group by user characteristics - geography, company size, industry, plan tier. Understand which segments retain best.

Different cohort definitions answer different questions. Time-based cohorts show whether the product is improving. Behavior-based cohorts identify which actions correlate with retention.

What cohort analysis reveals

Product improvement. Are newer cohorts retaining better? If so, recent changes are working.

Seasonal patterns. Do cohorts acquired in certain months perform differently? Holiday signups might have different intent than typical users.

Feature impact. Do cohorts who experienced a feature launch behave differently than those who didn't?

Channel quality. Which acquisition sources produce users who stick around?

Time-to-value signals. Where do cohort curves flatten? Users who make it past that point tend to stay.

Warning signs. Deteriorating cohort performance signals problems - even before aggregate metrics show decline.

Cohort analysis in practice

Define your question first. What do you want to learn? Different questions require different cohort definitions and metrics.

Choose the right granularity. Daily cohorts provide detail but noise. Monthly cohorts are cleaner but slower to show patterns. Match granularity to decision-making needs.

Select meaningful timeframes. Track long enough to see patterns, but not so long that data becomes unwieldy.

Look for inflection points. Where do cohort curves bend? These points often indicate critical moments in the user journey.

Investigate anomalies. Cohorts that behave unusually warrant investigation. What made them different?

Beyond retention

While retention cohorts are most common, cohort analysis applies to other metrics:

Revenue cohorts. Track revenue per cohort over time. Do newer customers spend more?

Engagement cohorts. Track usage intensity per cohort. Are users becoming more engaged?

Feature adoption cohorts. Track feature usage per cohort. Are users discovering features over time?

Support cohorts. Track support tickets per cohort. Are newer users having fewer problems?

Any metric that varies over the customer lifecycle benefits from cohort analysis.

Common cohort analysis mistakes

Survivorship bias. Analyzing only active users ignores those who churned. Include all original cohort members, not just survivors.

Small cohort sizes. Small cohorts produce noisy data. Ensure cohorts are large enough for statistical significance.

Ignoring context. Cohort differences might reflect external factors - market changes, competitor moves, economic conditions - not product changes.

Over-aggregating time periods. Large time windows hide patterns. If monthly cohorts show nothing interesting, try weekly.

Stopping too soon. Some cohort patterns only emerge over months or years. Patience reveals patterns that short-term analysis misses.

Tools and implementation

Cohort analysis requires data infrastructure:

Event tracking. You need to know when users took actions and when they signed up.

Analytics platforms. Tools like Amplitude, Mixpanel, or Heap provide cohort analysis features.

Data warehouses. For complex analysis, raw event data in a data warehouse enables custom cohort queries.

Visualization. Cohort data needs good visualization to be interpretable. Heat maps, line charts, and retention tables each reveal different patterns.

Tools like Klero enhance cohort analysis by adding qualitative dimension - connecting behavioral cohorts to what users actually said about their experience. When a cohort retains unusually well, understanding why from their feedback enables replicating that success.