# Cohort and retention analysis for recurring-revenue products
A product manager pulls up a triangle chart on Monday morning. Each row is a group of customers who signed up in the same month. Each column shows how many are still active 1 month later, 2 months later, and so on. The top-left corner is dense and green. The bottom-right fades to pale. In thirty seconds, she knows more about the health of the business than a quarter of revenue reports would tell her.
That triangle is a cohort retention chart, and learning to read it is the single most useful data skill in SaaS.
SaaS (Software as a Service) means customers pay on a recurring basis, usually monthly or annually, for access rather than a one-time license. In that model, the health of the business is not "how many customers do we have" but "how long do they stay and how much do they pay over time."
A single retention percentage hides the story. Say a company reports "85% retention." Is that improving or decaying? Is it the newest customers leaving, or an old batch that never fit?
A cohort answers this. A cohort is a group of customers grouped by a shared start event, almost always their signup or first-payment month. By tracking each cohort separately over time, you separate the behavior of March signups from September signups. If your onboarding got better in the summer, later cohorts will retain better, and the triangle chart makes it obvious.
Here is the structure of a typical cohort table. Rows are signup months. Columns are months since signup.
Cohort | M0 | M1 | M2 | M3 | M4
------------|------|------|------|------|-----
Jan signups | 100% | 82% | 74% | 70% | 68%
Feb signups | 100% | 80% | 73% | 69% |
Mar signups | 100% | 85% | 79% | |
Apr signups | 100% | 88% | | |Read it two ways:
The triangle shape comes from newer cohorts simply having fewer months of history.
The most common analytical mistake is comparing two retention numbers that are not measuring the same thing. There are three definitions, and they answer different questions.
N-day retention asks: was the customer active on a specific day (or in a specific window) after signup? For example, "day-7 retention" measures whether a user came back on day 7 exactly, or within a bucket around it.
This is strict and volatile. Miss the window and the user counts as churned even if they return on day 8. N-day retention is most useful for high-frequency products (a daily-use collaboration tool) where you genuinely expect near-daily engagement.
For a monthly B2B invoicing tool that a customer opens twice a month, day-7 retention is misleading. Do not use a daily metric for a monthly product.
Unbounded retention (sometimes called "range" retention) asks: was the customer active on day N *or any day after*? It counts anyone who is still around, regardless of gaps.
This is more forgiving and better matches how SaaS actually works. A finance team that logs in only at month-end is still a healthy, retained customer. Unbounded retention captures that; strict n-day would punish it.
Use unbounded retention when usage is periodic or infrequent. It answers the real business question: "are they still a customer?"
Logo counts hide money. Revenue retention tracks dollars, not headcounts, and it comes in two flavors that every SaaS professional should know.
NRRNRRNet Revenue Retention measures the percentage of recurring revenue retained and grown from existing customers over a period, including upsell and expansion, net of downgrades and churn.Voir la définition complète → above 100% is the holy grail of B2B SaaS: your existing base grows on its own. This is often cited as a benchmark for strong enterprise SaaS, though exact thresholds vary by segment and should be treated as rough guidance, not a rule.
Here is why all three matter together: a cohort can lose logos (falling unbounded retention) while gaining revenue (NRRNRRNet Revenue Retention measures the percentage of recurring revenue retained and grown from existing customers over a period, including upsell and expansion, net of downgrades and churn.Voir la définition complète → above 100%) because the customers who stay are expanding. That is common when you serve both small accounts that churn and large accounts that grow. Only looking at all three tells the full story.
For a clear primer on the revenue definitions, ServiceNow, Klipfolio, and others publish free explainers; a solid open reference is Klipfolio's metrics library.
Now to the payoff in the hook: distinguishing a leaky onboarding from a healthy long-term base.
Look at the *shape* of a single cohort's curve across the row.
The plateau level is the number to obsess over. If cohorts flatten at 70%, your long-term base is solid and the work is in reducing early leakage. If they flatten at 20% and keep sliding, you have a retention crisis dressed up as a growth story.
🎬 [VIDEO: "SaaS Retention Analysis Explained" — youtube.com — a walkthrough of building and interpreting cohort retention curves for subscription products]
1. Build the cohort triangle by signup month, using unbounded retention on logos.
2. Overlay the same triangle using net revenue retentionnet revenue retentionNet Revenue Retention measures the percentage of recurring revenue retained and grown from existing customers over a period, including upsell and expansion, net of downgrades and churn.Voir la définition complète →.
3. Compare shapes. If logo retention decays but NRRNRRNet Revenue Retention measures the percentage of recurring revenue retained and grown from existing customers over a period, including upsell and expansion, net of downgrades and churn.Voir la définition complète → holds above 100%, your small accounts churn while big accounts expand. Prioritize accordingly.
4. Read down the M1 column. If recent cohorts retain better than older ones, your onboarding changes are working.
Vérification des acquis
1. Why does a single headline retention percentage (e.g., "85% retention") often hide the true health of a SaaS business?
2. In a cohort retention triangle, what does reading DOWN a single column (e.g., the M1 column) tell you?
3. Why is a cohort defined by a shared START EVENT (like signup month) especially suited to recurring-revenue products?
4. Select ALL correct answers about how to interpret a cohort retention triangle.
Sélectionnez toutes les réponses correctes.
5. Select ALL correct answers describing situations where cohort analysis provides insight a blended retention figure would miss.
Sélectionnez toutes les réponses correctes.
Survivorship in the newest cohorts. The bottom rows of the triangle have only one or two data points. Do not draw conclusions from a cohort that is two weeks old. It has not had time to churn.
Mixing annual and monthly contracts. An annual contract cannot churn mid-year in the data even if the customer has mentally quit. They will churn at renewal, showing a cliff at month 12. Segment annual and monthly cohorts separately, or your curves will lie.
Confusing calendar time with tenure. Cohort analysisCohort analysisCohort analysis groups users by a shared starting trait or time (such as signup month) and tracks their behavior over time to reveal retention and lifecycle patterns.Voir la définition complète → measures months-since-signup, not calendar months. A macro event (a pricing change in one calendar month) hits every active cohort at once, which shows up as a diagonal pattern in the triangle, not a row or column. Watch for diagonals.
Reporting a blended NRR without cohorts. A company-wide NRRNRRNet Revenue Retention measures the percentage of recurring revenue retained and grown from existing customers over a period, including upsell and expansion, net of downgrades and churn.Voir la définition complète → can look healthy purely because one huge expanding account masks widespread small-account churn. Always check the distribution.