Cohort Analysis: Track How Customer Behavior Changes Over Time

Aggregate metrics lie by omission. "Monthly active users grew 15%" sounds great until you realize that new signups are masking the fact that older users are leaving at an accelerating rate. Cohort analysis separates these signals by grouping users based on when they started and tracking each group independently over time. It is the most reliable way to understand whether your product, marketing, and retention efforts are actually improving.

What Is a Cohort and Why Does It Matter

A cohort is a group of users who share a common starting event within a defined time period. The most common cohort is based on signup date: "all users who signed up in March 2025" form the March 2025 cohort. You then track what each cohort does over subsequent weeks or months.

Why this matters: aggregate metrics blend together users at different stages. A company might have 10,000 active users this month — but how many of those are users from last month who stuck around, versus brand new users who will leave next month? Only cohort analysis answers that question.

What Cohort Analysis Reveals

• True retention: Are customers from three months ago still using the product? Or are you backfilling losses with new acquisitions?
• Improvement over time: Are newer cohorts retaining better than older ones? This tells you whether product changes are working.
• Revenue expansion: Is each cohort spending more over time, or does revenue per cohort decline after the initial purchase?
• Leading indicators: If recent cohorts show worse Week 1 retention, you have an early warning of future problems before they show up in aggregate numbers.

Building Your First Retention Cohort

Step 1: Define the Cohort Event

The cohort event is what groups users together. Common choices:

• First purchase date — best for e-commerce
• Signup date — best for SaaS and apps
• First activation date — best when signup and first meaningful use are separated by onboarding
• Campaign date — useful for comparing the quality of users acquired through different marketing campaigns

Pick the event that aligns with how you think about the customer lifecycle. For most subscription businesses, signup date is the right starting point.

Step 2: Define the Retention Event

What counts as "retained" in each subsequent period? This depends on your business model:

• Subscription businesses: Still paying. This is binary and easy to measure.
• E-commerce: Made at least one purchase in the period. Monthly or quarterly windows are typical.
• Apps and platforms: Logged in or performed a core action (sent a message, created a document, ran a query). Define "core action" carefully — logging in and doing nothing should not count as retention.

Step 3: Choose the Time Granularity

Weekly cohorts are best for products with high-frequency usage (daily apps, social platforms). Monthly cohorts work for most SaaS and e-commerce businesses. Quarterly cohorts are appropriate for products with longer usage cycles (enterprise software, annual purchase patterns).

Step 4: Build the Cohort Table

A cohort table has rows for each cohort (by signup period) and columns for each subsequent time period. Each cell shows the percentage of the cohort that was retained in that period.

Example format:

Reading this table: of the 500 users who signed up in January 2025, 62% were still active in February (Month 1), 48% in March (Month 2), and so on. The improving Month 1 retention across cohorts (62% to 71%) suggests that onboarding improvements are working.

Reading Cohort Curves

Plot each cohort's retention as a line on the same chart (x-axis = time since signup, y-axis = retention percentage). This creates a set of "retention curves" that reveal patterns.

Healthy Patterns

• Curves flatten: Retention drops steeply in the first 1-3 periods, then levels off. This "flattening" means you have a core group of users who stick around long-term. The earlier the curve flattens and the higher it flattens, the better.
• Newer cohorts curve higher: Each successive cohort retains better than the one before it. This means your product and onboarding are improving.
• Curves turn upward: Some cohorts actually increase their activity over time. This is the "smile curve" and indicates strong engagement growth — rare but extremely positive.

Warning Patterns

• Curves never flatten: Retention keeps declining linearly. This means you have not found product-market fit — even long-time users eventually leave.
• Newer cohorts curve lower: Recent cohorts retain worse than older ones. Possible causes: lower-quality acquisition channels, product changes that hurt new user experience, or market saturation.
• Sudden cliff: A specific cohort shows dramatically worse retention. Investigate what changed — a broken onboarding flow, a pricing change, a product regression, or a change in acquisition source.

Revenue Cohort Analysis

Retention cohorts tell you who stays. Revenue cohorts tell you what they spend. The methodology is the same, but instead of tracking the percentage of active users, you track the total revenue (or average revenue per user) for each cohort over time.

Why Revenue Cohorts Matter

A cohort with declining user count but increasing revenue per user is actually healthy — lower-value users leave, higher-value users expand. Conversely, a cohort with stable user count but declining revenue per user signals pricing pressure or downgrade risk.

Revenue Cohort Patterns

• Revenue per cohort grows: Even as some users churn, the remaining users expand enough to offset losses. This is net negative churn at the cohort level — the gold standard for SaaS.
• Revenue per cohort declines but slowly: Churn slightly exceeds expansion. Acceptable but worth investigating how to increase expansion revenue.
• Revenue per cohort drops steeply: Significant churn with no expansion to offset it. This requires immediate attention to retention.

Behavioral Cohorts

Time-based cohorts are the most common, but you can also create cohorts based on behavior. Behavioral cohorts group users by what they did rather than when they arrived.

Examples

• Activation cohort: Users who completed onboarding in their first session vs. those who did not. Compare retention rates to quantify the impact of onboarding completion.
• Feature adoption cohort: Users who adopted Feature X vs. those who did not. If Feature X users retain at 85% versus 45% for non-adopters, you know what to push during onboarding.
• Acquisition source cohort: Users from organic search vs. paid ads vs. referral. This tells you which channels bring higher-quality users, independent of volume.
• Usage intensity cohort: Users who use the product daily vs. weekly vs. monthly. Understand the retention profile of each usage segment.

Behavioral cohorts are particularly valuable for product teams because they directly connect product actions to retention outcomes. If "users who create a dashboard in Week 1" retain at 2x the rate of those who do not, you know exactly what to optimize in the onboarding flow.

Running Cohort Analysis in clariBI

To set up cohort analysis in clariBI:

1. Connect your data source that contains user signup dates and activity records. This is typically your product database or your analytics event log. See the data source connection guide.
2. Use the conversational AI to ask questions like "Show me monthly retention cohorts for the last 6 months" or "Compare retention between users who completed onboarding and those who did not."
3. Build a cohort dashboard with the retention heatmap, cohort curves, and revenue cohort chart. The dashboard creation guide walks through the layout process.
4. Set up alerts for when a new cohort's Week 1 retention drops below a threshold. Early warning means faster intervention.

Common Mistakes in Cohort Analysis

Incomplete Data Windows

The most recent cohorts always look worse because they have not had time to mature. The April cohort's Month 1 retention is real data, but its Month 6 retention is unknown. Do not compare the incomplete trailing edge of recent cohorts to the complete data of older cohorts — it creates false alarm.

Cohorts Too Small

A cohort of 15 users is too small for reliable percentages. If 2 users churn, retention drops 13%, which looks alarming but is just noise. Set a minimum cohort size (50-100 users for weekly cohorts, 200+ for monthly) below which you aggregate into larger periods.

Ignoring Cohort Size Differences

A cohort of 1,000 users with 50% retention contributes 500 active users. A cohort of 100 users with 90% retention contributes 90. The second cohort retains better, but the first contributes more to your business. Always show cohort sizes alongside retention percentages.

Not Acting on Findings

Cohort analysis produces clear signals: onboarding is improving, enterprise customers retain better, paid acquisition brings lower-quality users. These findings require action — improving onboarding for low-retention segments, investing more in high-retention channels, intervening earlier for at-risk cohorts. Analysis without action is just expensive curiosity.

Cohort analysis is not complicated, but it requires discipline: consistent definitions, reliable data, regular review, and willingness to act on what the data shows. Once you start thinking in cohorts, you will never trust aggregate metrics the same way again. Every time someone tells you "active users grew 15%," your first question will be "which cohorts drove that growth?" That question, consistently asked, leads to much better decisions.