Cohort analysis

Measures product-market fit over time

A single metric like 'overall churn' hides problems. Cohort analysis reveals whether your onboarding is improving, whether product changes increased retention, or whether customer quality changed. These insights guide product and marketing strategy.

Isolates the impact of changes

Product launches, pricing changes, onboarding redesigns—it's hard to know if they worked because many variables affect behaviour. Cohort analysis controls for timing by comparing cohorts before and after a change, giving you confidence in causation.

Predicts lifetime value

If you know a cohort's retention curve (how many survive month 1, 2, 3, etc.), you can estimate lifetime value. Cohorts that retain well generate more customer lifetime value. This prediction guides how much you can afford to spend acquiring a customer.

Group customers by when they signed up, then watch each group's behaviour over time, so you can tell which months, channels, and product changes actually produce loyal, high-value customers.

Cohort analysis means you stop looking at one big blurry average and instead split customers into groups that share a starting point, usually their sign-up month. Everyone who joined in January is one cohort, everyone from February is another, and so on. Then you follow each cohort month by month and track something that matters: how many are still around, how much they spend, whether they adopted the key feature. The trend across cohorts tells you whether the business is getting better or quietly rotting.

The reason it's so useful is that it isolates cause and effect. Say you redesigned onboarding in March. Compare March's cohort to February's: if March retains better, the redesign probably worked. Without cohorts you'd just compare 'retention this month vs last month', which mashes together customers who joined under three different versions of your product, so you can never prove what moved the needle.

The classic B2B version is retention by cohort, what share of each group is still paying six months in. Flat or rising across cohorts is healthy; a steady slide (January keeps 85%, February 80%, March 76%) is a flashing warning light to go investigate.

Concrete examples:

• Say you're running a freemium productivity app and you track retention by sign-up month in Amplitude. Your Jan, Feb and March cohorts all sit at 45% retention by month three, then May jumps to 52% and June to 55%. The thing that changed in May was a new onboarding flow, and the cohort curves prove it worked, so you keep it and keep tightening it. The average 'overall retention' number would never have shown you that, because it blends old and new customers together.

• Say you expand into a new industry vertical and want to know if those customers are as sticky as your core. You watch the new cohorts in Amplitude and pair it with session replays in Hotjar to see where the new lot get stuck. Your old cohorts held 80% at month three; the July cohort drops to 72%, August to 70%. The data tells you plainly: the new vertical is lower-fit, and you now have to decide whether to fix the onboarding for them or pull back.

• Say you want to act on a weakening cohort rather than just stare at it. You spot in your analytics that this quarter's sign-ups are adopting your key feature slower, so you build a behaviour-triggered onboarding sequence in Customer.io that nudges only the customers in that cohort who haven't hit the activation milestone. A month later the next cohort's curve has lifted, and because you changed one thing for one group, you can actually credit the email flow for the improvement.

A productivity app tracked customer retention by cohort. Customers acquired Jan-March 2024 had 45% retention at month 3. April cohort also had 45%. May cohort had 52%. June had 55%. The trend showed improving retention. What happened in May? The team redesigned onboarding. The cohort analysis proved the redesign worked; May and subsequent cohorts retained better. They kept the new onboarding and continued improving it.

A SaaS platform with variable pricing tracked revenue per cohort. January cohort generated £50K month 1, £45K month 2, £40K month 3 (some customers downgraded). By month 12, £180K cumulative revenue. CAC was £30K per customer. Payback was 2 months (month 3 approaching cumulative revenue = CAC). This cohort analysis guided their decision to raise CAC to £45K because payback was still acceptable at under 3 months.

Define cohorts by acquisition date

The most useful cohort is by acquisition month or week. Create one cohort per month for 12+ months. You'll then track how each cohort behaves over time. Monthly cohorts are standard in SaaS; weekly cohorts are too granular unless you're rapidly iterating.

Measure key metrics for each cohort

Decide what you'll measure: retention rate (% still customers), revenue per cohort, feature adoption, or churn. Track one main metric consistently so you can see trends. Retention is most common, but revenue cohorts matter more for a SaaS business (a cohort can have high count retention but low revenue if customers downgrade).

Compare cohort-over-cohort trends

Plot cohorts over time. Is January's retention 90%, February's 88%, March's 86%? That declining trend suggests a problem. Is retention improving each month? That indicates your product or onboarding changes are working. Don't look at single cohorts; look at the trend.

Investigate cohort differences

When a cohort underperforms, investigate why. Did you change pricing or product that month? Did quality of acquired customers decline? Did a competitor launch? The why matters for fixing it. A poor cohort is only a problem if you don't understand the cause.

A B2B SaaS company's Jan-June 2024 cohorts all retained at 80% month 3. July cohort dropped to 72%. August dropped to 70%. What happened? They'd changed their acquisition strategy to expand into a new industry vertical. Cohort analysis revealed the new customers weren't as sticky. Investigation showed the vertical was lower-fit; they're now considering whether to continue or refocus on higher-retention verticals.