Products Solutions Templates Pricing Resources

Cohort Analysis: Tracking Groups of Visitors Over Time

Home / Everything About / Everything About Analytics / Cohort Analysis: Tracking Groups of Visitors Over Time

Cohort analysis groups visitors by a shared characteristic (usually arrival date) and tracks their behavior over time. Instead of looking at all visitors as one group, you track visitors who arrived in January separately from those who arrived in February. This shows whether newer visitors behave differently from older visitors, and whether product changes improve or hurt retention.

Why cohort analysis matters

Without cohorts, your retention looks flat. You see that 15% of visitors return in 30 days, but you don't know if this number is improving or declining. With cohorts, you see that January visitors had 12% 30-day retention, February had 14%, and March had 18%. Now you know retention is improving.

Cohort analysis also isolates the impact of changes you make. If you redesigned your site in February, cohort analysis shows whether the redesign improved retention or hurt it.

How to set up cohorts

Step 1: Define your cohort group. The most common is arrival date — group visitors by week or month. You can also group by traffic source (all visitors from Google Ads = one cohort), device (mobile users = one cohort), or location (visitors from the US = one cohort).

Step 2: Define your metrics. What behavior are you tracking? Returning visits, pages viewed, time on site, conversions, or engagement events. Each metric gets its own cohort table.

Step 3: Set your time window. Track behavior for 30, 60, or 90 days after arrival. Longer windows show sustained engagement, shorter windows show immediate impact.

Step 4: Build the cohort table. Each row is a cohort (January visitors, February visitors, etc.). Each column is days after arrival (day 1, day 7, day 30). Each cell is the metric (e.g., 45% of January cohort visited again on day 7).

Step 5: Review and compare trends. Look across rows to see whether newer cohorts are performing better or worse than older ones. Plot the data on a graph to visualize patterns. This shows whether your site is improving, degrading, or staying stable over time.

Common cohort patterns

The declining curve (expected): day 1 retention is 100% (everyone who arrives is part of the cohort). Day 7 is 25%. Day 30 is 10%. This is normal — most visitors leave and never return. Watch for the steepness of the decline. A steep drop means people leave fast. A gradual decline means they stay engaged longer.

The improvement pattern (good): each new cohort (newer arrival dates) has better retention than the previous cohort. January had 8% 30-day retention. February had 10%. March had 12%. This signals your site is getting better at keeping visitors.

The degradation pattern (warning): each new cohort has worse retention. January had 15% 30-day retention. February had 12%. March had 10%. This signals something changed — a redesign, a drop in traffic quality, new competition — that is hurting retention.

Using cohort analysis for product decisions

Before and after analysis: run cohorts for 4-6 weeks before a major change. Then run cohorts for 4-6 weeks after. Compare the retention curves. If the new curve is higher and to the right (more people retained, retained longer), the change was positive.

Traffic source quality: create cohorts by traffic source. Compare organic search cohorts to paid ad cohorts to social media cohorts. If one source has much higher retention, invest more there.

Seasonal patterns: in seasonal businesses, compare cohorts across the same season in different years. December 2024 cohort vs. December 2025 cohort shows whether you are improving year-over-year.

Common cohort analysis mistakes

Too small sample size: if you have only 50 visitors in a cohort, random variation creates noise. Use cohorts of at least 100-200 visitors.

Confusing cohort metrics: don't mix retention (% of cohort who returned) with frequency (total visits across the cohort). These are different — use one at a time.

Ignoring external factors: retention dropped 5% in March. Before concluding your site got worse, check: did search traffic drop (new competition)? Did you launch fewer ads? Did an influencer post about a competitor? Cohort changes sometimes reflect market shifts, not product problems.

Not segmenting by key variables: you lump all traffic sources together into one cohort. But organic search cohorts likely behave differently from paid ad cohorts. Always segment by the variables that matter: traffic source, device, geography, or user type. This reveals which segments are strong and which need work.

Waiting too long to analyze: you wait 90 days before looking at cohort data. By then, 3 months have passed and you have missed opportunities to fix problems. Analyze cohorts weekly or bi-weekly. Spot trends early so you can act faster.