Cohort Retention Matrix
The triangular table that doesn't lie about retention.
Put this metric to work: explore FlowMRR's MRR retention analytics to track revenue movements, retention and forecasts from your Stripe account, or see how FlowMRR compares as a Baremetrics alternative for Stripe.
A grid where each row is a cohort — the customers whose first paid month is the same — and each column shows the share still paying N months later. FlowMRR counts people (a customer who returns under a new Stripe record stays in their original cohort), counts a returning customer again in the months they pay, and has a revenue version that follows the cohort's MRR, which can exceed 100% when customers expand. Cells a cohort has not reached yet stay empty. The most honest representation of retention: averages hide everything that matters, cohorts reveal it.
Formula
Cell = Customers of the cohort still paying in month M ÷ Customers in the cohort in its first paid monthWorked example
Jan-2025 cohort: 100% at M0 → 92% at M1 → 87% at M3 → 81% at M6 → 76% at M12. Compare to Feb-2025: same M0, but 71% at M6. Feb cohort is leaking faster — investigate what changed.
Other definitions you will see
Some tables start cohorts at sign-up or trial start rather than the first payment, which mixes trial conversion into month 1. Others use continuous survival: a customer who leaves is never counted again, even if they come back, which reads lower than FlowMRR's point-in-time count. Some count subscriptions rather than customers, or place columns on calendar months instead of months since joining.
Illustrative comparisons
M12 retention (B2B SaaS)These unsourced ranges are illustrative scenarios, not measured industry benchmarks, percentiles or recommended targets. Their qualitative labels are editorial interpretations. Compare your own cohorts using the same segment, period, currency and measurement basis.
Why it matters
A single churn number averages over years of cohorts with completely different experiences. The cohort matrix lets you see whether product changes, onboarding tweaks, or pricing moves changed retention. If your matrix shows newer cohorts retaining better than older ones, you have product-market fit improving — the single most valuable signal in early SaaS.
Common mistakes
- Reading rows without comparing them — you need at least 2 cohorts to draw any conclusion.
- Mixing continuous survival with point-in-time retention — disclose whether returning customers count as retained, and keep their original acquisition cohort.
- Truncating short cohorts (e.g. comparing M6 retention of a 5-month-old cohort and a 24-month-old one).
Calculating it from Stripe data
Build cohorts from subscriptions that actually paid: active, past_due, canceled and unpaid, with any trialing, incomplete, incomplete_expired or paused stretch left out. A customer's cohort is the month of their first paid period, not the month they signed up: someone who starts a 14-day trial on 25 January and first pays on 8 February belongs to February. Group by person, by email or company rather than by Stripe customer ID, so that someone who comes back under a new customer record stays in their original cohort.
For each month after joining, mark the customer active if any of their subscriptions was paying that month. A subscription that ended unpaid drops out after its last paid period, not when dunning finished; a past_due one still counts while Stripe retries. Decide how returning customers count: point-in-time retention counts them as retained again in the months they pay, continuous survival drops them for good at their first gap. FlowMRR keeps a returning customer in their original cohort and counts them in every month they pay.
Each cell is customers active in month M ÷ customers in the cohort at month 0. The revenue version divides the cohort's MRR in month M by its MRR at month 0 and can exceed 100% when the customers who stay expand. Leave a cell empty when the cohort has not reached that age, and treat the open month as incomplete. Split by billing interval: annual subscribers can only leave at renewal, so a mixed cohort shows a step at month 12 that monthly cohorts do not. FlowMRR shows customer and revenue matrices by billing interval and product, and a curve that averages recent cohorts and projects twelve months ahead with a fitted decay.
Reading the result
Read the matrix three ways. Down a column, compare cohorts at the same age to see whether newer customers stay longer. Along a row, find where a cohort loses most customers: early drops point to onboarding or expectations set at sale, a drop at month 12 in annual cohorts is a renewal decision. Along a diagonal, a dip that hits every cohort in the same calendar month is an event, such as a price change, an outage or a wave of failed payments, rather than a difference in cohort quality.
Weigh every conclusion by cohort size: in a cohort of 12 customers, each loss moves the cell by more than 8 points. Compare the customer and revenue matrices too. Revenue retention above customer retention means larger customers stay longer or the ones who stay expand; the reverse means you are losing your larger accounts first.
Frequently asked questions
How do you build a cohort retention table from Stripe data?
Assign each customer to the month of their first paid period, then for every later month check whether they had a subscription paying that month: active, or past_due while retried. Divide the active count by the cohort's starting size, and leave out trials, incomplete subscriptions and paused stretches.
Should cohorts start at sign-up or at first payment?
For revenue retention, at first payment. Starting at sign-up mixes trial conversion into retention: a cohort with many unconverted trials would look like it lost customers in month 1. Measure conversion separately, then follow paying customers from their first paid month.
What is the difference between logo retention and revenue retention in a cohort?
Logo retention counts customers still paying; revenue retention sums what the cohort still pays. Losing one large account barely moves logo retention but can move revenue retention a lot, and expansion can lift revenue retention above 100% while logo retention can only fall or recover from returning customers.
How much history do I need for cohort analysis?
Enough for several cohorts to reach the age you care about. Month-3 comparisons need a few months of data; month-12 comparisons, which matter for annual plans, need more than a year. Before that, compare only the ages every cohort has reached.