Three Ways to Cut Customer Cohorts, and When Each One Helps
Cohort analysis only tells you something useful if you cut the cohorts along a line that actually explains why some customers are worth more than others. Cut them the wrong way and you get a chart that looks sophisticated but doesn't change a single decision.
Here are three common ways to build cohorts, what each one is actually good for, and where each one falls short.
Cutting by signup month
The most common approach: group customers by the month they signed up, then track retention and expansion for each group over time. This is useful for spotting whether a specific change, a pricing update, a new onboarding flow, a product launch, affected the customers who joined right after it.
It falls short when you're trying to understand why customers behave differently, since a signup-month cohort mixes customers of every size, industry and acquisition channel together. Two cohorts might look different mainly because one month happened to bring in a batch of larger deals.
Should you cut cohorts by acquisition channel?
Grouping customers by how they found you, paid search, referral, content, outbound, often reveals bigger LTV differences than signup month does, since channel tends to correlate with intent and fit. A customer who arrived through a specific referral relationship might expand at a very different rate than one who converted from a broad paid campaign.
The limitation here is that channel data gets messier over time as attribution models change or multi-touch journeys blur the first-touch signal, so older cohorts can be less reliable than recent ones.
Cutting by initial use case or plan
Grouping by what a customer bought first, or what problem they said they were solving at signup, often explains LTV differences that channel and signup date both miss. A customer who bought specifically to solve one workflow may expand differently than one who bought a broader plan speculatively, even if they came through the same channel in the same month.
This cut takes more setup, since it depends on capturing intent data at signup that many teams don't bother collecting, but it tends to be the most actionable of the three because it points directly at what to change in onboarding or packaging.
For example, suppose two customers signed up in the same month through the same referral source. One bought a single workflow plan to solve a reporting headache, and the other bought a broad plan speculatively. Six months later the first has added seats steadily while the second has barely logged in. A signup-month or channel cohort would average them together and show a flat line. A use-case cohort separates them and shows that the narrow buyer expands while the broad buyer stalls. That points directly at a change worth testing, such as a guided onboarding path for the speculative plan, instead of a vague conclusion about the whole month.
Which cohort cut answers your actual question?
If you're asking whether a recent product or pricing change helped or hurt, cut by signup date. If you're asking which channel to invest more in, cut by acquisition source. If you're asking why some customers expand and others plateau, cut by initial use case, since that's the cut most likely to point at a fixable cause rather than just a correlation.
Running all three at once on the same dataset, rather than picking one and assuming it's sufficient, is often what reveals the real pattern. A cohort that looks flat by signup month can look completely different once you split it by use case.
Start with whichever cut you already have clean data for, and add the others incrementally rather than waiting until you can build all three perfectly before drawing any conclusions. A rough cut by use case today beats a perfectly engineered pipeline you're still building six months from now.
Mistakes that make cohort charts misleading
- Comparing cohorts that are too small to mean anything, where one unusually large or small customer swings the whole group's average.
- Measuring LTV too early in a cohort's life, before enough time has passed to see real expansion or churn behavior play out.
- Mixing self-serve and sales-assisted customers in the same cohort, when their expansion patterns are usually driven by entirely different mechanics.
- Presenting a cohort chart without the underlying customer count per cohort, which hides whether a trend is real or just noise from a small group.
A chart that omits sample size looks confident even when it shouldn't. Add the count to every cohort view by default so nobody has to ask for it separately.
What Good Looks Like
A useful cohort analysis is cut along a line that explains real differences in customer behavior, run with cohorts large enough to trust, and revisited with more than one cut before drawing a conclusion.
Building The Capability (5-Stage Skill Ladder)
How to Get Started
Frequently Asked Questions
How far back should a cohort analysis look?
Far enough that you can see real expansion or churn behavior, not just the first few months of enthusiasm. For most B2B products with multi-year value, that means looking at cohorts that are at least a year old, and treating anything younger as too early to draw firm conclusions from.
What's the minimum cohort size worth analyzing?
There's no universal number, but a cohort small enough that one customer's expansion or churn swings the group average by a large amount isn't telling you much. If a cohort has only a handful of customers, treat any pattern in it as a hypothesis to test, not a conclusion.
Should marketing or finance own cohort analysis?
Both need the output, but whoever owns the underlying data pipeline, usually a data or RevOps function, should own building and maintaining the cohorts. Marketing and finance can then cut the same clean dataset for their own questions instead of maintaining separate, inconsistent versions.
About the numbers
This guide doesn't quote a sourced benchmark. Figures in it are estimates or general guidance, so check them against your own numbers.
Related Guides
Where Conversion Rate Benchmarking Actually Leaks Marketing Budget
The specific comparisons that reveal wasted marketing spend, and why comparing your conversion rate to the wrong benchmark hides the leak instead of finding it.
How to Run a Marketing Performance Audit Before a Board Meeting
A five-step way to check marketing spend, conversion rate and creative performance before a board meeting, so you present numbers you can defend.
Building Expansion Triggers Off What Customers Actually Do
A worked example of building usage-based triggers for expansion revenue, and why they outperform a fixed check-in calendar.
Why Creative Testing Slows Down Exactly When You Need It Most
The bottlenecks that make ad creative testing slow down as a program grows, and how to keep shipping variants without cutting corners on what each test teaches.
Choosing the Right Cohorts for LTV and Expansion Analysis
A decision guide for picking cohort definitions when analyzing customer lifetime value and expansion revenue, and the mistakes that make the numbers misleading.
An SEO Audit for Terms Your Competitors Already Rank For
A practical audit for winning search positions your competitors currently hold, covering content gaps, technical blockers and link signals.