Cohort Retention Analysis for SaaS Without a Data Team

Every SaaS investor asks the same question on the second call: "what does your retention look like by cohort?"

And every SaaS founder without a data team has the same answer: a flat overall churn percentage, an apologetic shrug, and a promise to "get back to you with the proper cohort table."

That promise often takes weeks to keep. Or never gets kept. The proper cohort analysis sits on the someday-list while the founder approximates it from gut feel and the Stripe dashboard.

It does not need to.

Why cohort retention beats overall churn

If you only look at one number, monthly churn rate is misleading. It blends new customers (who churn fast), recent vintages (which might be quietly improving), and your oldest customers (who are stickier just because they are still here).

Cohort retention separates the signal. Each cohort is a group of customers who joined in the same period — usually a month. You track that specific group's retention through time. The result is a table where each row is a cohort and each column is the months that have passed since they joined.

A healthy SaaS cohort table looks like this: high attrition in the first month, a flattening curve, and a plateau where the surviving customers stick for a long time. An unhealthy table looks like a slow leak that never stops.

The pattern matters more than any single number. And you cannot see the pattern without the table.

What you need to build a cohort table

In theory, three columns:

• A customer identifier
• The date they first paid
• A monthly revenue or activity value, tracked over time

In practice, building this from your raw data takes either SQL skills or hours of Sheets work. You have to deduplicate customers, handle plan changes, decide whether downgrades count as partial churn, normalise for payment frequency. Most founders give up halfway through and just compute a rough overall churn rate.

Connect the source. Skip the build.

Anna connects directly to Stripe (the most common case for SaaS) or to your warehouse — BigQuery, Snowflake, Postgres. Read-only OAuth or read-only credentials. She reads your subscription history, identifies customers, groups them by their first paid month, and computes the retention curve.

No SQL. No CSV exports. No "let me check what's in the customer table."

The first prompt

Once Stripe (or your warehouse) is connected, paste this:

"Build a monthly cohort retention table for the last 12 months, measured by net revenue retention. Highlight any cohort where retention dropped below 90% by month 6."

Anna pulls every paid subscription in the last year, groups customers by their first-paid month, tracks each cohort's MRR forward, and produces a retention table with the standard rows-as-cohorts, columns-as-months-out shape.

The output includes the table itself, plus the things a senior analyst would surface unprompted: which cohorts are abnormal, what differentiates them, whether the differences are statistically meaningful given the cohort sizes.

A typical finding might read:

The September cohort retains beautifully. The January cohort is leaking. The recent three-month average is dragging the trend line down. Three different stories, one table.

The follow-up that actually finds the cause

A cohort table on its own is descriptive. The question your investor will ask next is "why?" — and that is where most cohort analyses stop.

Try this follow-up:

"Split the worst-performing cohorts by acquisition channel and plan. Where is the retention gap concentrated?"

Anna re-runs the cohort math sliced by channel and by plan. The honest answer is usually that the bad retention is concentrated in one specific combination — Google Ads customers on the starter plan, say — rather than evenly spread. Once you know that, the remediation is concrete: pause the spend, fix the onboarding for that plan, or both.

Net revenue retention vs logo retention

Two very different metrics. SaaS founders confuse them constantly.

• Logo retention is what fraction of customers from a cohort are still paying you anything.
• Net revenue retention is what fraction of the original cohort's MRR you still have, including expansion, contractions, and churn.

A cohort can have 70% logo retention and 110% NRR — the customers who stayed expanded enough to more than offset the ones who left. That is the SaaS dream. The opposite (90% logo, 80% NRR) is the subtle warning sign: customers are sticking around but downgrading.

When you ask Anna for cohort retention, specify which one you want. If you do not, she will compute both and ask which is more relevant to the decision you are making.

What about Mixpanel, Amplitude, or product analytics tools?

Product analytics tools have cohort views. They are great for product engagement — feature retention, day-7 active users.

They do not, in general, do revenue cohort retention. They do not see Stripe. They do not know which customer is on which plan. They cannot answer "what is the dollar retention of the customers who joined in November."

Anna can. Connect product analytics for engagement cohorts, connect Stripe for revenue cohorts, ask both questions in the same conversation.

The board prep moment

The reason this matters is the moment you need it. The investor call. The board prep. The annual planning kickoff.

A flat overall churn number gets you a follow-up question. A cohort table — with NRR by vintage, the trend annotated, the worst cohort segmented by channel — gets you a productive conversation about where to invest. And the difference is not how smart you are. It is whether the analysis is built.

Anna builds it. Two prompts and a connection.

One question to start

If you are going to ask only one cohort question this week, ask this:

"Is recent retention worse than older retention, controlling for tenure?"

It is the question your investor wants the answer to. It is the question that tells you whether your last six months of acquisition were worth it. And it is the question a flat overall churn rate cannot answer.

Connect Stripe or your warehouse. Paste the question. Try it at heyanna.studio.