SaaS Cohort Analysis

A cohort is every customer who started in the same period, usually a month. Cohort analysis follows each of those groups separately as it ages, so month 3 for a cohort that signed in January and month 3 for a cohort that signed in June sit in the same column and can be compared directly.

Worked example: 120 customers signed up in August, and 89 of them were still customers three months later. Month-3 logo retention for that cohort is 89 ÷ 120 = 74.2%. If the November cohort of 162 customers has 134 left at month 3, that is 82.7% — 8.5 points better at the same age. The two cohorts are at different points in calendar time, but at the same point in customer life, which is the only comparison that means anything.

The single most common mistake is dividing by the previous month instead of the cohort start. That gives you a month-over-month survival rate, not retention, and chaining those rates hides exactly the deterioration cohort analysis exists to expose.

The same table answers two different questions depending on what you put in the cells, and healthy businesses routinely look different in each.

Run both. Revenue retention comfortably above logo retention means your larger accounts are the ones staying and expanding — good for durability, but a concentration risk worth naming. The reverse means you are keeping customers while losing the revenue they used to spend, which usually shows up as downgrades before it shows up as churn. The aggregate version of the revenue view is net revenue retention, and the aggregate version of the logo view is your churn rate. On annual contracts, where retention only moves at renewal, the renewal rate measures the contracts that actually came up for a decision in the period.

Read down a column, not across a row. A column holds every cohort at the same age, so differences between them are differences in the cohorts themselves — who you acquired, what you onboarded them into, what the product did for them. Reading across a row only tells you that customers leave over time, which you already knew.

• A steep month-1 drop that then flattens is an onboarding or expectation problem, not a product problem. The customers who make it past the first month behave well; the ones who leave never got started. Fix activation before you spend more on acquisition.
• A curve that never flattens means there is no retained core. Every cohort eventually goes to zero, and growth is a treadmill that gets more expensive with every month of scale.
• Later columns getting worse for newer cohorts is the finding that matters most and the one aggregate churn hides longest — a large, well-retained older cohort can carry the blended number for months while everything you signed this quarter falls away.
• A single anomalous row is usually a pricing change, a discount campaign, or a channel experiment that ran in that month. Annotate it rather than averaging it away.

1. 01Judging a cohort before it has aged.A cohort that is two months old has two months of data. Putting it in the same conversation as a twelve-month cohort's month-12 number is comparing different questions — which is why the last few cells of a cohort table are always empty, and should be.
2. 02Averaging cohorts of wildly different sizes. A simple mean lets a 6-customer cohort move the curve as much as a 600-customer one. Weight by cohort size — this tool does, and the weighted row is usually the one worth quoting.
3. 03Mixing self-serve and enterprise in one cohort. They retain on completely different curves, and blending them produces a shape neither segment actually has. Run the table once per motion if the two are material.
4. 04Counting trials, or counting win-backs into the original cohort. A cohort should be customers who started paying in that month. A returning customer starts a new cohort — folding them back into the old one is what produces logo retention above 100%, which cannot happen.
5. 05Reading a two-point difference as a trend. With small cohorts, a couple of percentage points is one or two customers. Look for a direction that holds across several consecutive cohorts before you act on it.

Retention is the part of a SaaS business that underwriting actually rests on. A lender advancing capital against recurring revenue is making one judgment: will that revenue still be there over the repayment term. Aggregate churn is a weak answer to that question because it is backward-looking and blended. A cohort table is a much stronger one, because it shows whether the revenue you are signing today behaves like the revenue that has already proven durable.

It also changes what the capital is worth taking for. If cohorts are holding and CAC payback lands inside the retained life you can see in the curve, spending more on acquisition compounds. If newer cohorts are deteriorating, funding acquisition simply buys more of a leak — and the honest move is to fix retention before raising anything at all. Feed the curve into lifetime value and check the result against what you pay to acquire before deciding.

For founders who would rather not sell equity to fund that growth, non-dilutive SaaS financing prices the same retention as borrowing capacity instead of as a multiple. Founderpath funds bootstrapped SaaS companies from $10K MRR against recurring revenue, repaid from revenue — no equity, no board seats. The cohort table you just built is close to the evidence that conversation runs on.