Cohort Analysis: Finding Which Customers Pay Off

Your average customer is a lie. Add up everyone who ever signed up, divide by revenue, and you get a number that hides the truth: some customers stuck around for three years and referred two more, while a whole batch churned inside ninety days and burned your acquisition budget on the way out. The blended average treats both groups the same. Cohort analysis pulls them apart.

If you run B2B marketing and your reports stop at "we got 40 leads and closed 6 deals," you are flying with one instrument. You can see this month. You cannot see whether the customers you won in March are worth more or less than the ones you won in September, or which campaign quietly brings in accounts that expand and renew. That answer changes where you spend next quarter.

This guide shows how to build cohort analysis without a data team, read the patterns that matter, and turn them into budget decisions. Example numbers below are illustrative; the method is the point.

What a cohort actually is

A cohort is a group of customers who share a starting event in time. The most common grouping is acquisition month: everyone who became a customer in January is the "January cohort," February is the February cohort, and so on. You then follow each group forward and measure the same thing for all of them, month by month, from their own start.

The trick is the shared clock. Instead of asking "how much revenue did we make in June," you ask "how much revenue did the January cohort produce in their sixth month, and how does that compare to what the February cohort did in their sixth month." Now you are comparing like with like. A cohort that looks healthy at month six this year can be measured against last year's cohort at the same age.

You can build cohorts on more than signup date. Group by acquisition channel (LinkedIn Ads versus Google Ads versus referrals), by first plan purchased, by company size, or by the campaign that first touched them. Each grouping answers a different question. Signup-month cohorts tell you if your product and onboarding are improving over time. Channel cohorts tell you which acquisition source brings customers who stay.

Two cohort views you will use most

There are two readings of the same data, and people mix them up constantly.

Retention cohorts track how many customers from a group are still active over time. You start at 100% in month zero and watch the percentage fall. This shows you the shape of churn: do you lose people fast in the first two months, then stabilize, or is it a slow steady bleed? The shape tells you where to intervene.

Revenue cohorts track money instead of headcount. You follow how much each cohort spends per month or cumulatively. In B2B with expansion revenue (upsells, seat growth, add-ons), a revenue cohort can climb even as the retention cohort falls, because the accounts that stay spend more. That crossover is one of the most useful signals you can find. It means your remaining customers are getting more valuable, and your job is to win more of that type.

Read both. Retention alone makes a healthy expansion business look like it is shrinking. Revenue alone hides a leaky bucket you are papering over with a few big accounts.

Build your first cohort table in a spreadsheet

You do not need a BI platform to start. A CRM export and a spreadsheet will get you a working cohort table in an afternoon.

1. Pull the raw data. Export every customer with two fields at minimum: the date they became a customer and a monthly value (revenue, or a simple "active yes/no" flag per month). If you sell subscriptions, monthly recurring revenue is ideal. If you sell projects, use invoiced amount per month.

2. Assign each customer to a cohort. Add a column that strips the signup date down to its month, for example 2025-03. Every customer now belongs to exactly one cohort.

3. Calculate the "month number" for each data point. For a customer who signed up in March and generated revenue in July, the month number is 4 (March is month 0). This is the column that lines everyone up on the same clock.

4. Pivot it. Rows are cohorts (signup month). Columns are month numbers (0, 1, 2, 3...). Each cell holds the average value or the retained percentage for that cohort at that age. A pivot table does this in a few clicks.

5. Add a heatmap. Conditional formatting turns the grid into color. Healthy patterns and broken ones jump out before you read a single number.

Here is what a small revenue retention table looks like. Numbers are illustrative.

Revenue retention by signup cohort (% of month-0 revenue, illustrative)

Cohort	Month 0	Month 3	Month 6	Month 9	Month 12
Jan	100%	88%	81%	79%	78%
Feb	100%	90%	86%	85%	87%
Mar	100%	91%	89%	92%	98%

Read the rows left to right and the columns top to bottom. The March cohort recovers past 100% by month 12, a sign that expansion revenue from staying accounts outpaces churn. Newer cohorts holding higher at the same age means something you changed (onboarding, targeting, pricing) is working.

Reading the patterns

A cohort table is only useful if you know what good and bad look like.

The cliff. Retention drops sharply in the first one to three months, then flattens. This is an onboarding or fit problem. Customers buy, do not reach value fast enough, and leave. The fix lives in your first 30 days, not your acquisition. Tightening lead qualification upstream helps too, because a chunk of early churn is people who were never a fit. If your forms and scoring let weak-fit leads through, your cohorts will show it; this is where work on lead scoring pays back downstream.

The slow bleed. A steady few percent lost every month with no flattening. This usually means weak ongoing value or a competitor picking customers off. It is harder to fix than a cliff because there is no single moment to target.

The smile. Revenue retention dips, then curves back up above 100%. Expansion revenue is winning. Your most valuable move is to find more customers who look like the ones driving the upswing, and feed that profile back into targeting.

The improving stack. Each newer cohort sits above the older ones at the same month number. Whatever you have been changing is working, and you can attribute it to roughly when the change shipped.

One caution. Recent cohorts have fewer months of data, so the right edge of your table is always thin and noisy. Do not over-read a cohort that is only two months old. Give it time before you draw conclusions.

From cohorts to budget decisions

The reason to build any of this is to move money. Three decisions cohort analysis informs directly.

Which channels to scale. Build channel cohorts and compare 6 or 12 month retained revenue, not just cost per lead. A channel with a higher cost per lead can win on retained value if its customers stay twice as long. This connects straight to your LTV to CAC ratio: cohorts give you the real LTV input instead of a guess, broken out by where the customer came from.

What a customer is actually worth. Cumulative revenue cohorts are the honest way to calculate lifetime value, because they use observed behavior instead of a formula with an assumed lifespan. If you want the mechanics of turning cohort curves into a number, the method behind calculating LTV leans on exactly this data. Feed that into your unit economics and the payback math stops being a hopeful estimate.

Where reactivation is worth the effort. Cohorts show you which groups churned and roughly when. If a specific cohort dropped off after a price change or a feature sunset, that is a targeted list worth a customer reactivation play, not a blast to your whole dead list.

A practical rule: when two campaigns produce the same cost per lead but their cohorts diverge after month six, shift budget toward the one with the healthier curve, even if it looked equal on the surface. The surface metric lied. The cohort did not.

Common mistakes

Mixing cohort definitions mid-analysis. If you compare a signup-month cohort against a channel cohort, you are comparing two different questions and the answer is noise. Pick one grouping per table.

Judging young cohorts too early. The newest rows have the least data and the most variance. Watch them, do not bet on them.

Ignoring seasonality. A December B2B cohort may behave differently because of budget-flush buying or holiday slowdowns. Compare a cohort to the same month last year before you panic about a dip.

Stopping at retention and never looking at revenue. In any business with expansion or upsell, headcount retention undersells your real trajectory.

Building the table and never acting. The table is a tool for a decision. If it does not change where a dollar goes, you built a wall decoration.

Frequently asked questions

How much data do I need before cohort analysis is useful?

Enough customers per cohort that one or two churns do not swing the percentage wildly. As a rough floor, aim for 20 to 30 customers per cohort and at least six months of history so you can see a curve form. With fewer, group by quarter instead of month to thicken each cohort.

What is the difference between cohort analysis and customer segmentation?

Segmentation groups customers by shared traits at a point in time (industry, company size, plan). Cohort analysis groups by a shared starting event and follows the group over time. You can combine them: build cohorts within a segment, for example signup-month cohorts of only enterprise accounts.

Do I need a special tool to do this?

No. A CRM or billing export plus a spreadsheet pivot table covers the first version, and that is enough to make real decisions. Product analytics platforms and BI tools automate it and let you slice faster, which matters once you are running this monthly across several groupings. Start in a spreadsheet, upgrade when the manual work hurts.

Can cohort analysis work for long B2B sales cycles?

Yes, though the clock changes. With multi-month cycles and annual contracts, monthly cohorts can be too granular. Use quarterly cohorts and measure retention at renewal points rather than every month. The logic is identical; the time unit gets coarser.

Should I cohort by signup date or by first purchase date?

Pick the event that marks the start of the relationship you care about. For revenue questions, first paid purchase is usually right. For product and onboarding questions, first activation or signup tells you more. Whatever you choose, keep it consistent across the whole analysis.

How often should I rebuild my cohort tables?

Monthly is a sensible cadence for most B2B teams. You add a new cohort row, extend the existing rows by one month, and check whether newer cohorts are tracking better or worse than older ones at the same age. Quarterly reviews work if your sales cycle is long.

Conclusion

Cohort analysis replaces a flattering average with the truth about which customers pay off and which ones quietly cost you. A quick checklist to start:

• Export customers with signup date and a monthly value.
• Group into signup-month cohorts and line them up on a shared month-number clock.
• Build both a retention table and a revenue table, with a heatmap.
• Read the shape: cliff, slow bleed, smile, or improving stack.
• Cross-reference with channels to find which acquisition sources bring keepers.
• Act on it: move budget toward the healthier curves.

If your reporting stops at this month's leads and you suspect your acquisition mix is hiding a leaky bucket, get a focused read on it. Book a 30-minute working session with Lead The Way and we will map your cohorts against your channels, show you which sources bring customers that stay, and tell you where the next dollar earns more. Bring a customer export; we will bring the questions.