Cohort analysis is an analytics technique that groups users by a shared start event — usually signup or first purchase — and then tracks how those groups behave over time. Instead of one blended average across all users, you see how the users who joined in January retained versus those who joined in March. That comparison surfaces retention and churn patterns aggregate metrics hide.
Aggregate retention numbers lie by omission. A blended "60% retention" hides the fact that older signups retain at 80% and newer ones at 30% — and that gap is the story. Cohort analysis is how every modern SaaS, subscription, and ecommerce team finds those stories.
What is cohort analysis?
Cohort analysis organizes users into groups that share a meaningful start event, then follows those groups across time intervals to see how their behavior evolves. The most common cohort criterion is signup date — everyone who joined in March 2025 forms one cohort, everyone who joined in April 2025 forms another.
The technique originated in medical research, where scientists tracked patient outcomes for treatment groups over years. Around 2010 it became standard in SaaS analytics — pushed by Y Combinator, David Skok, and the early Amplitude / Mixpanel product-analytics wave — because it was the only way to answer questions aggregate metrics could not.
Every cohort answers a variation of one question: are we getting better? If the December 2024 signup cohort retains 40% after 90 days and the June 2025 cohort retains 55%, you have proof that product changes shipped in early 2025 improved retention. Aggregate averages would blur that finding out of existence.
If you report "60% retention" across all users, you are mixing customers who signed up 3 years ago with customers who signed up last month. The old customers had time to survive; the new ones did not. A cohort view isolates each start point and shows you whether recent product decisions are actually improving retention.
Why cohort analysis matters
Four reasons every growth and product team runs cohorts:
- Retention visibility. See exactly how long each cohort stays engaged, and where the drop-off happens (week 1, month 3, day 90).
- Product-change measurement. Ship a new onboarding flow in March? Compare the March cohort to February. The delta is the impact.
- Channel quality comparison. Users acquired via organic search often retain 1.5-2x better than paid-social users. Cohort analysis makes that visible; aggregate CAC does not.
- Revenue forecasting. Once you know how each cohort spends over months 1-12, you can project future revenue with meaningful accuracy.
How cohort analysis works — the 4-step process
Whether you use GA4, Amplitude, or a raw SQL query, the workflow is the same.
- Define the cohort. Pick a grouping criterion — usually signup date or first-purchase date. Advanced teams use behavioral cohorts (users who completed key action X in week 1).
- Choose the metric to track. Retention percentage, revenue per user, feature usage, purchase frequency, or churn rate.
- Build the cohort table. Rows = cohorts (Jan, Feb, Mar). Columns = time intervals since cohort started (Day 0, Day 7, Day 30). Cells = the metric value.
- Read the table. Look at rows to see how each cohort decays, and columns to see whether newer cohorts are improving over older ones.
Types of cohorts
| Cohort type | Grouping criterion | Best used for |
|---|---|---|
| Acquisition cohort | Date of signup / first purchase | Retention, churn, LTV over time |
| Behavioral cohort | Completed a specific action (e.g. published a project in week 1) | Identifying "aha moment" behaviors |
| Channel cohort | First-touch acquisition channel (organic, paid, referral) | Comparing channel LTV |
| Geographic cohort | Country or region of signup | Localization and expansion planning |
| Plan / pricing cohort | Plan selected at signup | Understanding upgrade / downgrade patterns |
| Product-version cohort | App or feature version at signup | Measuring specific release impact |
Real cohort analysis examples
Two worked examples across SaaS and ecommerce that show how the technique earns its keep.
1. SaaS onboarding rework
A B2B SaaS team suspected their January 2025 onboarding rework improved activation. Aggregate 30-day retention across all users read 42% — a number that had barely moved in 12 months. The cohort table told a different story: January signups retained at 35% (Day 30), February at 41%, March at 48%. The onboarding shipped in early January and the impact appeared in the cohort that had time to complete the new flow. Aggregate view: no change. Cohort view: 13-point retention lift.
2. Ecommerce channel comparison
An ecommerce brand ran cohort analysis on 12 months of purchase data. Organic-search cohorts retained at 60% at 90 days; paid-social cohorts retained at 38%. Customer lifetime value differed by 2.1x — every organic customer was worth twice as much as a paid-social customer over 12 months. The team reallocated 30% of paid-social budget into SEO and content, and blended CAC dropped 22% within two quarters.
Cohort analysis vs segmentation — which to use
Both slice your user base. The difference is time.
Use cohort analysis when
- The question involves change over time
- You want to compare newer to older signups
- You need to measure the impact of a ship / launch
- You are forecasting retention or LTV curves
- Example: "Did onboarding v2 improve day-30 retention?"
Use segmentation when
- The question is about who users are, not when they joined
- You want to compare persona groups on a snapshot
- You are building marketing audiences or campaigns
- Time is not the axis of comparison
- Example: "How do enterprise users differ from SMB users today?"
7 cohort analysis best practices
- Keep cohorts big enough to be reliable. Aim for 100+ users per cohort. Below 50, a single churn event distorts the whole percentage.
- Match the cohort granularity to your volume. High-volume products use weekly cohorts. Smaller businesses stick to monthly.
- Watch the recent cohorts first. Retention degradation shows up in the newest cohorts before it shows up in aggregate metrics.
- Pair retention with revenue. A cohort can retain but downgrade. Track revenue-per-cohort alongside user-count-per-cohort.
- Overlay ship dates on the table. Mark the row where you shipped onboarding v2, pricing v3, or a new landing page. Impact is legible only with the context.
- Use behavioral cohorts to find the "aha moment." Users who take action X in week 1 usually retain 2-3x better. Identify X and route new users to it.
- Report cohorts as curves, not point-in-time numbers. One number is a snapshot; a curve is the story.
New cohorts always look better than old ones early on because they have not had time to churn. Comparing "Day 7 retention of March cohort" to "Day 180 retention of last September cohort" is meaningless. Only compare the same time interval across cohorts, and only after enough time has elapsed to make the comparison meaningful.
Common cohort analysis mistakes to avoid
- Cohorts too small. Fewer than 50 users per cohort produces noisy, unreadable retention curves.
- Wrong grouping variable. Grouping by geography when the question is about product changes gives you the wrong answer.
- No baseline. Reporting a single cohort with no prior cohort to compare it against tells you nothing about direction.
- Ignoring survivorship bias. The oldest cohort still active is a survivor group — do not assume it represents typical user behavior.
- Never tying cohorts to ship dates. Cohort tables without release annotations are pretty numbers with no story.
- Confusing correlation with cause. A cohort improvement might be caused by seasonality or a viral spike, not your product changes.
How theStacc helps with cohort analysis
Cohort analysis lives or dies on the quality of the incoming users. If your paid-social cohorts churn at 3x organic cohorts, no amount of retention work fixes the root cause — the acquisition channel is producing the wrong customer. theStacc runs the SEO and content side of that problem: we build the organic acquisition engine that reliably ships better-fit cohorts into your product, so your retention curves flatten from the top instead of being propped up by expensive paid spend.
Frequently asked questions
Cohort analysis is used to measure retention, churn, revenue per user, and repeat-purchase behavior over time. It answers questions like: are newer signups sticking around longer than older ones? Did the January product launch actually improve retention? Which acquisition channel produces customers with the highest lifetime value?
A segment is a group of users defined by an attribute (all enterprise customers, all mobile users). A cohort is a group defined by a shared time-based event (everyone who signed up in March 2025). Segments describe who users are; cohorts follow what happens to them across time.
Aim for at least 100 users per cohort for statistical reliability. Cohorts smaller than 50 users produce noisy retention curves where a single churn event can distort the percentage. High-volume products often use weekly cohorts; smaller businesses stick with monthly.
Google Analytics 4 has a built-in cohort exploration report. Product analytics tools like Amplitude, Mixpanel, and Heap have dedicated cohort views. For SaaS revenue cohorts, ChartMogul and ProfitWell surface retention and expansion curves out of the box.
Healthy SaaS products retain 40-60% of a cohort past 30 days and 20-40% past 12 months. Consumer apps often see steeper drops. The absolute number matters less than the shape: a curve that flattens after a few weeks indicates you have found genuine product-market fit for that cohort.
Related glossary terms
Sources
