Cohort Analysis
Also: Cohort Retention Analysis, Customer Cohort Analysis
Cohort analysis groups users by a shared starting trait or time (such as signup month) and tracks their behavior over time to reveal retention and lifecycle patterns.
What It Is
Cohort analysis is an analytical technique that segments users or customers into groups (cohorts) based on a shared characteristic or event within a defined time window, then measures how each group behaves across subsequent periods. The most common grouping is the acquisition cohort, where users are bucketed by the period they first signed up, made a purchase, or activated. Behavioral cohorts instead group people by an action they took (for example, users who used a specific feature).
Unlike a single snapshot metric, cohort analysis follows the same group of people over time, so you can separate the effect of *when* someone joined from the effect of *how long* they have been around.
Why it matters
Aggregate metrics hide trends. A flat overall retention number can mask the fact that newer cohorts are churning faster while older ones stay loyal. Cohort analysis exposes these dynamics so teams can:
- Measure true retention and churn without survivorship bias.
- Detect whether product or marketing changes actually improved new-user behavior.
- Forecast lifetime value (LTV) and payback periods more accurately.
- Compare acquisition channels by the *quality* of users they bring, not just volume.
How it is used in practice
1. Define the cohort key: usually signup or first-purchase period (weekly or monthly).
2. Define the metric: retention, revenue, orders, active days, or feature usage.
3. Define the time axis: periods since the cohort started (month 0, month 1, and so on).
4. Build the grid: rows are cohorts, columns are elapsed periods, cells hold the metric.
5. Read the patterns: read down a column to compare cohorts at the same age, across a row to see one cohort decay.
Concrete Example
A subscription app groups users by signup month. The January cohort retains 100% in month 0, 45% in month 1, and 30% in month 3. After a redesigned onboarding flow launched in March, the March cohort retains 60% in month 1. Reading down the month 1 column shows the improvement is real and not seasonal, justifying the onboarding investment.
Tips: keep cohort sizes large enough to be statistically meaningful, and always state the metric and time unit clearly.
Frequently asked questions
What is cohort analysis?
Cohort analysis groups users or customers by a shared starting point (most often the month they signed up or made a first purchase), then tracks a metric for each group across subsequent periods. Because it follows the same people over time, it separates the effect of when someone joined from the effect of how long they have been a customer. Aggregate metrics cannot make that distinction.
What is the difference between an acquisition cohort and a behavioral cohort?
An acquisition cohort groups people by the period they arrived: signup month, first purchase, activation date. A behavioral cohort groups them by an action they took, for example everyone who used a specific feature. Acquisition cohorts answer whether new arrivals behave better over time; behavioral cohorts answer whether a given action correlates with staying.
Why does a flat overall retention rate hide a problem?
Because it averages cohorts of different ages. An overall retention number can stay stable while newer cohorts churn faster and older loyal cohorts compensate for them. Splitting by cohort and reading down a single column (all groups at the same age) reveals the degradation that the aggregate figure masks.
How do you read a cohort grid?
Rows are cohorts, columns are elapsed periods since the cohort started (month 0, month 1, and so on), and each cell holds the metric. Read across a row to see one cohort decay over its lifetime; read down a column to compare different cohorts at the same age, which is how you tell a real improvement from a seasonal effect.
What do I need to define before building a cohort analysis?
Four things: the cohort key (usually signup or first-purchase period, weekly or monthly), the metric (retention, revenue, orders, active days, feature usage), the time axis in periods since the cohort started, and cohort sizes large enough to be statistically meaningful. Always state the metric and the time unit explicitly, otherwise two grids are not comparable.