Cohort analysis is the one report that tells you whether your product is getting better or quietly rotting. It groups users by a shared starting point, usually their signup week, then tracks each group separately over time. Your blended retention number can look flat and healthy while your newest signups churn twice as fast as last year's. The average hides it. The cohort shows it.
We learned this the annoying way. One of our side projects had a retention rate that barely moved month to month, so we assumed things were fine. Then we split users by signup month and the floor fell out: every new cohort was leaving faster than the one before. A shipping change three months earlier had broken activation, and the average had been smothering the signal the whole time.
This guide covers what cohort analysis actually is, the three types worth knowing, a six-step playbook to run one, and the free tools that do the job. We will also be blunt about what to cut. Most teams buy a dedicated cohort tool years before they need one.
◢What is cohort analysis?
Cohort analysis is a behavioral analytics method that splits your users into groups sharing a common trait or event, then tracks how each group behaves over time, per Wikipedia's definition. A cohort is just a group of people who started together, like everyone who signed up in March.
The magic is specificity. A blended retention rate is the average temperature across a whole country. Cohort analysis is the forecast for your street. As Lean Analytics put it, cohorts let you see patterns clearly across a customer's lifecycle instead of slicing across all customers blindly.
That shift changes the questions you can ask. Instead of "why are we losing customers," you ask "why did April signups churn faster than March." One is a shrug. The other is a lead.
◢What are the three types of cohort analysis?
There are three main types, and each points a different lens at the same users. Acquisition cohorts group by when people joined. Behavioral cohorts group by what people did. Predictive cohorts group by what people are likely to do next. Most teams only ever need the first two.
Acquisition cohorts group users by signup time, like week or month. This is the default starting point because it is dead simple to build and answers questions like "is our onboarding getting better." A January cohort retaining at 45% versus an April cohort at 32% tells you something changed, and roughly when, as Appcues lays out.
Behavioral cohorts group users by an action, not a date. Think "users who finished onboarding in their first session" versus those who did not. These are the sharpest tool for finding which behaviors predict retention. If your fast-onboarding cohort retains far better, you just found a lever for your onboarding flow.
Predictive cohorts group users by modeled future behavior, like churn risk, using signals such as session frequency and feature usage. They are genuinely useful at scale. They are also a fast way to waste a quarter if you do not yet have the volume or a data person to build the model. Cut this one until you have earned it.
◢How do you run a cohort analysis in six steps?
Run a cohort analysis by defining your question, picking a cohort type, choosing a metric, pulling the data, building the grid, and reading the pattern. The order matters: the question comes first, or you end up with a colorful chart and no decision. GeeksforGeeks and Wikipedia both stress the same starting move.
Here is the loop we use:
- Define the question. Not "how is retention," but "did the March redesign help newer users stick." A vague question yields a vague chart.
- Pick the cohort type. Testing a product change over time? Acquisition. Testing whether an action drives retention? Behavioral.
- Choose one metric. Usually retention rate at day 7, day 30, or month 3. Pick the window that matches your usage rhythm. Daily apps care about day 7; B2B tools care about month 3.
- Pull the data. You need a user ID, a cohort key (signup date or event), and the activity events you care about.
- Build the grid. Rows are cohorts, columns are time periods, cells are your metric. A pivot table does this fine.
- Read the pattern and test. Compare rows top to bottom. If newer cohorts are weaker, find the change that caused it, ship a fix, and watch the next cohort.
Step six is the whole point. A cohort chart you never act on is a vanity metric in a fancier outfit.
◢What can free tools do for cohort analysis?
Free tools cover almost everything an early-stage team needs. Google Analytics 4 ships a free Cohort Exploration report under its Explore section, with inclusion and return criteria and daily, weekly, or monthly granularity. PostHog's free tier includes 1 million events a month with built-in retention and cohort tooling. Both are real, not crippled trials.
A spreadsheet also works, and we are not joking. If you have a few hundred users, export signup dates and activity, drop them into a pivot table, and you have an acquisition cohort grid in twenty minutes. No vendor, no contract, no per-seat math.
The paid tools are real, but they are not where you start. Amplitude's free Starter plan and Mixpanel's free tier both run cohorts before you pay anything. We broke down the per-user versus per-event billing trap in our Amplitude vs Mixpanel guide. The short version: do not buy analytics before you have users to analyze.
◢What should you cut from your cohort stack?
Cut the dedicated cohort tool you bought before product-market fit. We see this constantly: a five-person team paying for a premium analytics seat to track retention for 200 users a spreadsheet would handle. That is sprawl wearing a data-team costume. Our SaaS sprawl audit exists for exactly this reflex.
Cut predictive cohorts until you have the volume to make them honest. A churn model trained on a few hundred users is astrology with a confidence interval.
And cut the habit of tracking ten metrics across twenty cohorts. One question, one metric, one decision. Everything else is a screenshot for a slide nobody reads. Tally what these tools actually cost with our stack cost calculator before the next renewal sneaks up.
◢Why cohort analysis is worth the effort
Cohort analysis connects user behavior straight to revenue, which is why it earns its keep. With churn benchmarks landing around 3.5% for B2B SaaS in 2025 per Vitally's data, small retention shifts compound into real money. A cohort chart is how you spot those shifts while you can still act on them.
It also keeps you honest about cause and effect. When you ship a change to your go-to-market motion or your support workflow, the cohort grid tells you whether newer users actually retain better, not just whether a blended average wiggled. Retention benchmarks from Pendo and SaaS Capital only mean something when you can see your own groups trending against them.
Most of all, it turns "I think it's working" into "March's cohort is up nine points." That sentence wins arguments and budgets.
◢The bottom line
Cohort analysis is the cheapest insurance you can run against a product quietly getting worse. Three takeaways: blended averages lie, so split users by when they joined or what they did; you can run a real cohort for free in GA4, PostHog, or a spreadsheet; and you should not buy a dedicated cohort tool until your free stack genuinely breaks.
Start with one question this week. Open your free analytics, split last quarter's signups by month, and look at whether newer cohorts are weaker. If they are, you just found your next sprint.
Want more no-nonsense breakdowns like this, with zero vendor kickbacks? Subscribe to our newsletter. We send the stuff we wish someone had told us before we paid for tools we did not need.
◢FAQ
What is cohort analysis in simple terms? Cohort analysis groups people who share a starting event, usually the week or month they signed up, then watches how each group behaves over time. Instead of one blended retention number, you get a row per group and a column per time period. That lets you ask sharper questions, like why January signups stuck around but April signups churned twice as fast. The whole point is to separate when a user arrived from how they behaved, because mixing the two hides the patterns that actually move revenue.
What are the three types of cohort analysis? Acquisition cohorts group users by when they joined, like signup week or month, and are the default starting point for retention tracking. Behavioral cohorts group users by what they did, like finishing onboarding in their first session, and are best for finding which actions predict retention. Predictive cohorts group users by modeled future behavior, like likelihood to churn, using usage signals to act before someone leaves. Most teams only need the first two. Predictive is overkill until you have real volume and a real data team.
Can you do cohort analysis for free? Yes, and most early teams should. Google Analytics 4 ships a free Cohort Exploration report in its Explore section, per Google's docs. PostHog's free tier includes 1 million events a month with built-in retention and cohort tools, per its pricing page. And a plain spreadsheet with a pivot table runs a perfectly good acquisition cohort if your user count is small. We have built retention charts for side projects without paying a cent. Buy a dedicated cohort tool only when the free options genuinely block you.
What is the difference between cohort analysis and funnel analysis? Funnel analysis tracks how a single group moves through a sequence of steps, like signup to activation to purchase, and shows where they drop off. Cohort analysis tracks how separate groups behave over time, usually grouped by when they joined or what they did. A funnel answers where users fall out of a process. A cohort answers whether this month's users are healthier than last month's. They pair well: use a funnel to find the leak, then use cohorts to confirm a fix actually moved newer groups.
What metrics should I track in a cohort analysis? Retention rate is the workhorse: the percentage of a cohort still active after a set period, like day 7, day 30, or month 3. Churn rate is its mirror. Beyond that, track revenue retention if you are subscription-based, since a cohort can lose users but still grow revenue through expansion. Pick one core metric tied to your actual question before you build the chart. Tracking five metrics across ten cohorts gives you a pretty grid and zero decisions.