# Cohort Analysis for SaaS Marketers

> How to build acquisition, revenue and behaviour cohorts, read a retention curve honestly, and judge channels months before the revenue actually lands.

Source: https://saas-marketing.net/guides/cohort-analysis-for-saas-marketers/
Topic: SaaS Metrics and Analytics
Type: guide
Published: 2026-09-11
Last updated: 2026-09-11
Publisher: SaaS Marketing (saas-marketing.net)
License: CC BY 4.0. Quote or republish with attribution and a link to https://saas-marketing.net/guides/cohort-analysis-for-saas-marketers/

## Short answer

SaaS cohort analysis groups customers by a shared starting condition, usually signup month, acquisition channel or an early behaviour, then tracks each group's retention and revenue over the same number of months. Marketers use it to compare channels on 6 and 12 month revenue retention rather than first order CAC, because a cheap channel with materially worse retention is the expensive one once the payback window closes.

## Key takeaways

- Acquisition date cohorts show whether your product is improving. Channel cohorts show which marketing spend deserves more budget.
- A retention curve that never flattens means your LTV number is a guess, and any CAC payback math built on it is unreliable.
- A channel can win on cost per lead and lose badly on 12 month net revenue retention. Both numbers belong in the same review.
- You need three fields to start: an account id, a cohort date and a monthly revenue or activity row. Everything else is refinement.
- Cohort tables mislead when recent months look strong purely because those customers have not had time to churn yet.
- Judge paid channels on month 6 cohort revenue retention before you scale spend, not on the lead volume in week two.

---

Most marketing teams find out a channel was bad about nine months too late. The spend looked efficient, the leads converted, the pipeline closed, and then the renewals did not happen. Cohort analysis is how you see that coming in month four instead of month fourteen. It is also the single analytics skill that gets skipped in almost every marketing career path, which is why so many budget decisions still rest on cost per lead.

## What a cohort actually is, and the three kinds that matter

A cohort is a group of customers who share a starting condition and are then measured over identical time windows. That last part does the work. Comparing January's customers at month six against June's customers at month one tells you nothing, which is exactly the error a blended churn number makes.

Three cohort types answer three different questions, and mixing them up is the usual reason a cohort deck lands badly in a leadership meeting.

| Cohort type | Grouped by | The question it answers | Who should care |
| --- | --- | --- | --- |
| Acquisition date | Signup or first payment month | Is the product and onboarding getting better over time? | Product, growth |
| Acquisition channel | First touch or self reported source | Which marketing spend buys durable revenue? | Marketing, finance |
| Behaviour | An action taken in the first N days | Which early actions predict survival? | Product, lifecycle |

Acquisition date cohorts are the default in every billing tool. They are useful and slightly boring. If March 2026 signups retain better at month six than March 2025 signups did at month six, something in the product or onboarding improved. That is a real finding, and it belongs to product more than to marketing.

Channel cohorts are the ones marketers should be building and almost never are. Behaviour cohorts sit in the product analytics tool and matter most for lifecycle messaging, which is where the [SaaS metrics and analytics](/saas-metrics/) work overlaps with campaign design.

Reading the bottom right corner of a cohort table as if it were data. Recent cohorts always look healthier because they have not had time to churn. Only compare cells at the same age.

## How to read a retention curve without fooling yourself

A retention curve plots the share of a cohort still active or still paying at each month of age. What you are looking for is not the starting number. It is the point where the curve stops falling.

That flattening point is the whole argument. A curve that flattens at 55 percent in month seven means you have a durable base of 55 percent, and lifetime value is a real quantity you can calculate. A curve still sloping downward at month eighteen means there is no durable base yet, and every LTV figure in your board deck is an extrapolation of a line that has not stopped moving.

**Month 4 to 8** When a healthy self serve SaaS retention curve typically begins to flatten

Two practical rules. First, always plot logo retention and revenue retention on the same chart. In B2B they diverge fast, and the gap is the expansion story that the [NRR expansion levers playbook](/playbooks/nrr-expansion-levers/) is built around. Second, never average curves across plan tiers. A self serve tier bleeding out at month three and an enterprise tier holding flat at 94 percent produce a blended curve that describes no actual customer.

**Building your first retention curve**

## Cohorting CAC by month catches rising acquisition cost early

Blended CAC hides the trend that matters. Calculate CAC per monthly acquisition cohort instead: total sales and marketing spend attributed to that month, divided by new customers acquired in that month. Then plot twelve of those in a row.

What you usually see is a slow upward drift as the cheapest demand gets exhausted, punctuated by a step change when a channel saturates. Catching the step change in the month it happens is worth more than any quarterly review. If your paid search CAC moved from $4,100 to $5,600 across four cohorts while conversion rates held steady, the auction got more expensive and your payback math needs redoing before the next budget cycle, not after.

Pair the CAC cohort with the revenue cohort and you get payback by cohort, which is the number that should gate spend increases. Our [marketing budget calculator](/calculators/marketing-budget/) handles the arithmetic if you want to stress test what a CAC drift of 20 percent does to a plan, and the [sales and marketing spend benchmarks](/research/sales-and-marketing-spend-benchmarks/) give you something to compare against by ARR band.

Marketing spend in month N rarely produces customers in month N, especially with a 60 day sales cycle. Offset your spend by your median time from first touch to close before you divide, or your CAC cohorts will look artificially cheap during growth months and expensive during flat ones.

## The cheap channel that turns out to be the expensive one

Here is the pattern that makes this whole exercise worth the effort. Two channels, same quarter, same product.

The paid social cohort is 35 percent cheaper to acquire and generates 46 points less net revenue retention twelve months in. On a first order CAC view it is the best channel in the account. On a cohort view it is the worst, and scaling it would have made the business worse while every dashboard said otherwise.

Why does this happen? Broad paid targeting reaches people with a vague problem, and vague problems do not survive a renewal conversation. BOFU search reaches people already comparing vendors, which is a stronger buying signal and a better fit filter. This is the same logic that makes comparison and alternatives pages convert well, and it is why channel quality belongs in the same conversation as channel cost.

We doubled the cheapest channel for two quarters and our net revenue retention fell six points. Nothing in the acquisition dashboard warned us.

My position: judge channels on month 6 and month 12 cohort revenue retention before you scale, and treat first order CAC as a sanity check rather than a verdict. If you only have six months of data, use month 6 and be honest that it is provisional.

## The minimum data and tooling to run this properly

You need less than people think. Three fields get you started: an account id, a cohort assignment date, and a monthly revenue row per account. Add a channel field and you have channel cohorts.

Where teams get stuck is the join. Billing data lives in Stripe or the billing system. Channel data lives in the CRM or an ad platform. Behavioural data lives in Amplitude or Mixpanel. Nobody owns the key that ties them together, which is why a [marketing tracking plan](/templates/marketing-tracking-plan/) matters more than the choice of analytics tool.

| Approach | Setup effort | What you get | What you cannot do | Rough cost |
| --- | --- | --- | --- | --- |
| Billing tool cohorts (ChartMogul, Baremetrics) | Hours | Revenue cohorts by plan and date | Channel or behaviour splits | $200 to $1,000 a month |
| Product analytics (Amplitude, Mixpanel) | Days | Behaviour cohorts, activation curves | Accurate revenue retention | $0 to enterprise pricing |
| Warehouse plus BI (Snowflake, dbt, a BI layer) | Weeks | Everything, joined properly | Nothing, if the modelling is right | $2,000 a month and up, plus analyst time |
| Spreadsheet from a CRM export | A day | A usable first channel cohort | Scale, automation, trust | Free |

Start with the spreadsheet. Genuinely. A single quarter of channel cohorts in a spreadsheet has changed more budget decisions than most warehouse projects, and it tells you whether the warehouse project is worth funding. If you are choosing between product analytics tools for the behavioural side, our [Mixpanel vs Amplitude comparison](/comparisons/mixpanel-vs-amplitude/) covers where each one wins on cohort building specifically.

For the channel attribution field itself, decide early if you are using platform reported source, a first touch CRM field, or a self reported "how did you hear about us" question. They disagree, often by a lot. The [B2B SaaS attribution tools](/tools/b2b-saas-attribution-tools/) roundup covers the tradeoffs, but the short version is that self reported data is better than most marketers expect and worse than finance wants.

## What cohort analysis costs you, and where it misleads

The honest tradeoff: this takes time you currently spend on campaigns, and it produces answers slowly. A twelve month cohort takes twelve months. There is no shortcut, and any vendor promising one is selling a model, not a measurement.

Three failure modes worth naming.

- Small cohorts produce noise. Under roughly 30 accounts in a cohort, a single churned enterprise logo swings the whole line. Group quarterly instead of monthly at low volume.
- Mix shift fakes improvement. If your enterprise share grew, blended cohorts improve without anything actually getting better. Always segment before you celebrate.
- Cohort tables invite cherry picking. Somebody will find the one good row. Agree in advance which cell you are judging on.

Pick a single gate before you look at the data. For example: no paid channel scales past $50k a month until its month 6 cohort revenue retention clears 85 percent. Deciding the rule first stops the table from being argued with afterwards.

Cohort work also has a reporting cost. Most executives have never read a triangle table and will read the wrong cell. Convert to curves for the board deck and keep the table for the working session. A standing view in your [marketing dashboard](/templates/saas-marketing-dashboard/) with two lines, month 6 and month 12 revenue retention by channel, does more than a full triangle nobody opens.

## What to do this week

Pull a CRM export with account id, first paid date, acquisition channel and monthly revenue for the last eighteen months. Build one triangle in a spreadsheet, grey out the incomplete cells, and split it by channel. Then check the month 6 column against the CAC for each channel and see whether your cheapest channel is actually your cheapest channel.

If the answer surprises you, run the same view with the [NRR and churn calculator](/calculators/nrr/) to size what a five point retention difference is worth in annual revenue. That number is usually what gets the warehouse project funded.

## Frequently asked questions

### What is cohort analysis in SaaS?

Cohort analysis groups customers who share a starting condition, most often the month they signed up, and then measures what happens to each group over identical time windows. Because every cohort is compared at the same age, you can tell whether retention is genuinely improving or whether a good looking blended number is just being propped up by recent signups.

### What is the difference between logo retention and revenue retention cohorts?

Logo retention counts how many accounts from a cohort are still paying. Revenue retention counts how much money that cohort still generates, including expansion. A cohort can lose 30 percent of its accounts and still show 110 percent revenue retention if the survivors expanded. In B2B SaaS, revenue cohorts are the more honest read on channel quality.

### How many months of data do I need before cohort analysis is useful?

Six months gives you a directional read, twelve gives you a decision. Most B2B SaaS retention curves are still falling at month three, so anything shorter mostly measures onboarding. If your contracts are annual, you need at least one full renewal cycle before a cohort tells you anything about renewal behaviour.

### Can I do cohort analysis without a data warehouse?

Yes. Stripe, ChartMogul and Baremetrics produce revenue cohort tables from billing data with no engineering work, and Amplitude or Mixpanel produce behavioural cohorts from product events. The limit is joining those to marketing source data. Once you need channel level revenue cohorts, a warehouse becomes the cheaper option.

### Why does my retention curve never flatten?

Either the product has not found a durable use case for that segment, or the cohort mixes segments with very different behaviour. Split the cohort by plan, company size and acquisition channel before concluding the product is the problem. A flat enterprise curve hiding inside a collapsing self serve curve is a common pattern.

### Should marketers own cohort reporting or should finance?

Finance should own the revenue cohort definitions so the numbers reconcile with billing. Marketing should own the channel and campaign dimensions layered on top. When marketing builds its own parallel definitions, the two teams end up arguing about numbers instead of about decisions.