SaaS cohort analysis: how to read a retention table and turn it into decisions
SaaS cohort analysis groups customers by the month they first paid and tracks each group across its whole life, instead of one blended company-wide average that hides how customers actually behave. You read the resulting table in three directions: across a row to watch one cohort decay over time, down a column to compare cohorts at the same age, and along the diagonal to see how today’s revenue stacks up from every cohort that came before.
Most guides teach the table as one skill and stop there. But a retention table is three different charts hiding in one grid. Read the wrong axis and you misdiagnose the business.
What cohort analysis is, and why averages lie
A blended churn number averages your best and worst customers into one figure that describes nobody. Cohort analysis fixes this by freezing a group of customers on the month they joined and following only that group forward.
Say your company-wide monthly churn reads 4 percent and looks stable. That single number can hide a January cohort retaining 95 percent of its revenue while a March cohort bleeds 12 percent a month because you changed your onboarding flow. The average says “fine.” The cohorts say “you broke something in February.”
That is the whole point. Cohorts show you the truth that churn rate alone smooths over. They also let you calculate realistic, segment-specific LTV instead of a number built on assumptions that do not hold.
The retention table, and the one formula under it
Here is what a cohort retention table looks like. Each row is a sign-up cohort. Each column is months since they joined. Every cell shows revenue retained as a percent of that cohort’s starting MRR.
| Cohort | Month 0 | Month 1 | Month 3 | Month 6 | Month 12 |
|---|---|---|---|---|---|
| January | 100% | 94% | 89% | 86% | 84% |
| February | 100% | 95% | 91% | 89% | 88% |
| March | 100% | 96% | 93% | 92% | 92% |
| April | 100% | 97% | 95% | 95% | — |
The formula for any cell is simple:
Cohort revenue retention at month N = (cohort’s MRR in month N) / (cohort’s MRR in month 0)
A cohort that started at $10,000 MRR and sits at $8,400 a year later retains 84 percent. The 16 points it lost is gross revenue churn for that group.
Read across the row: is your churn front-loaded?
Reading left to right down a single row shows one cohort decaying or expanding over its lifetime. The shape of that decay tells you when you lose people.
Take the January row: 100, 94, 89, 86, 84. It drops 6 points in the first month, 5 more by month three, then only 2 points across the next nine. That is front-loaded churn, and it is the normal SaaS pattern. Retention curves usually decay fastest in the first year, with churn risk highest early and falling the longer a customer stays.
Here is the decision. If a row craters in months one through three, your problem is onboarding, not your product. People are leaving before they reach value, so fix the first-run experience and the whole curve lifts. If instead the row keeps falling steadily forever with no flattening, you have a product-value problem, which is harder and more expensive.
That is why customer onboarding is the fix that returns the most on a front-loaded curve. You are not retaining better customers. You are getting the ones you already paid for to stick.
Read down the column: did your product change work?
Reading top to bottom down a single column compares different cohorts at the same age. Rising numbers as you go down mean your onboarding and product are getting better over time.
Look at the Month 6 column: January 86 percent, February 89, March 92, April 95. Each newer cohort holds more revenue at the same point in its life. That is the clearest signal that something you shipped is working.
This column grades last quarter’s bet. Did the new onboarding flow you launched in February actually move retention, or did it just feel better? If the cohorts that started after the change retain higher at the same age, you have your answer in dollars. If the column is flat or falling, the change did nothing, and you should stop defending it.
Founders judge product changes by gut, or by a vanity metric that moved for unrelated reasons. The column read is the honest scorecard.
Read the diagonal: how concentrated is your revenue?
Summing one diagonal reconstructs total MRR for a single calendar month, built as a stack of contributions from every cohort alive that month. This is the layer-cake view of your current revenue.
Picture this month’s MRR. A slice comes from the January cohort, now aged and thinned to 84 percent of its start. A slice comes from February. A fresh thick slice comes from the newest cohort. Add those layers and you get today’s number.
The diagonal shows revenue concentration. If 60 percent of this month’s MRR sits in two old cohorts, you are far more fragile than your growth rate suggests. One enterprise logo from 2024 leaving could erase a quarter of new-business gains.
If revenue is spread evenly across many cohorts, you are durable. Concentration risk does not show up anywhere on a standard dashboard. It shows up on the diagonal.
Logo retention versus revenue retention: you need both curves
These are two different curves, and you cannot read one from the other. Logo retention counts accounts: how many customers from the cohort are still active. Revenue retention counts dollars, so it captures upgrades and downgrades that logo retention ignores.
Here is the part that breaks people’s mental model. A company can lose 1 in 5 customers (80 percent logo retention) and still post 110 percent or higher net revenue retention, because the survivors expand enough to cover the dollars walking out the door.
| Curve | What it counts | Can exceed 100%? | What it answers |
|---|---|---|---|
| Logo retention | Accounts still active | No | Are customers leaving? |
| Gross revenue retention (GRR) | Dollars kept, churn and downgrades only | No | How leaky is the bucket? |
| Net revenue retention (NRR) | Dollars kept plus expansion | Yes | Is the base growing on its own? |
So a business can be “losing customers” and “growing from its base” at the same time, and both statements are true. Watch only logo retention and you panic over a healthy expansion machine. Watch only NRR and you miss that your customer count is quietly draining. Plot both.
The three curve shapes, and what each one means
Every revenue retention curve settles into one of three shapes. Naming the shape turns a wall of numbers into a diagnosis.
- Decaying. The curve keeps falling and never stabilizes. This is the warning sign. You have no loyal core, every customer eventually leaves, LTV is low, and acquisition has to run forever just to stand still.
- Flattening. The curve drops early, then levels into a loyal core. A healthy SaaS often settles around 80 percent logo retention and roughly 90 percent gross revenue retention. The flat tail is what makes LTV worth calculating.
- Smiling. The curve dips, then climbs back above 100 percent as expansion and reactivation outpace churn. This is the top tier, the basis for net negative churn, and the shape behind products like Slack and Notion.
The shape dictates the decision. A decaying curve means stop spending on acquisition and fix retention first. A flattening curve means your model works, so pour fuel on acquisition. A smiling curve means your highest return is on expansion, so build pricing and packaging that let happy customers spend more.
What good looks like in 2025
Here are the numbers to grade your cohorts against, drawn from 2025 SaaS benchmark reports.
| Metric | Healthy benchmark (2025) | Top tier |
|---|---|---|
| Net revenue retention | approx 101 to 106% median | above 130% |
| Gross revenue retention | approx 90% | above 95% |
| Annual logo retention ($3 to 8M ARR) | approx 80% top quartile | higher |
| LTV:CAC ratio | 3:1 starting benchmark | 3.6:1 median (2024) |
NRR splits hard by size. Companies below $10M ARR run around 98 percent. Companies above $100M ARR sit near 115 percent. GRR moves the same way, from roughly 85 percent at $1 to 10M ARR up to 94 percent at scale.
Bootstrapped SaaS in the $3 to 20M ARR range ran median NRR near 104 percent in SaaS Capital’s 2025 data. Always compare inside your own ARR band. A 98 percent NRR is on-target for early-stage and a problem for a scaled business.
LTV by cohort: use the empirical number, not the formula
The formula most people use for LTV is wrong for real companies. The shortcut, (ARPA times gross margin) / churn rate, assumes churn and ARPA stay constant forever. Early-stage companies break both assumptions every quarter.
The honest method is empirical cohort LTV: sum all the gross-profit-adjusted revenue a cohort produced across every month it has been alive, then divide by the number of customers the cohort started with.
Work it. A January cohort started with 100 customers. Across 12 months it produced $144,000 in gross-profit-adjusted revenue, including the dollars from customers who later churned. Empirical LTV so far is $144,000 / 100 = $1,440 per customer, and it keeps growing as the surviving cohort ages.
The detail that matters: you divide by the customers the cohort started with, not the survivors. You already paid acquisition cost up front for every customer, and that cost does not disappear when they leave. Counting only survivors inflates LTV and wrecks your LTV:CAC ratio. A cohort LTV calculator does this division correctly. A back-of-envelope formula usually does not.
Turning cohorts into a revenue forecast
Cohort curves forecast revenue from your existing base before a single new sale. You take the flattening tail of your average retention curve, project it forward, then multiply the surviving revenue in each future period by ARPA.
Early-stage models often assume decay settling around 2 to 5 percent monthly churn after year two. So if your cohorts flatten at 3 percent monthly churn, you can project each live cohort’s surviving revenue month by month, sum the layers, and get next year’s revenue from today’s customers. Add expected new cohorts on top and you have a full forecast.
This is also where a defensible LTV comes from. Discount those future revenue streams back and you get a number you can put next to CAC. The LTV:CAC ratio explained only means something when the LTV underneath it came from real cohort behavior, not a formula. Pair the forecast with an MRR forecaster and you can see the base and new business separately.
Stop drawing these by hand
Cohort tables are painful to build in a spreadsheet. You are pulling raw billing events, bucketing by sign-up month, and recomputing every cell when a customer upgrades. The whole thing is stale the moment you finish.
Mowt generates segmented revenue-retention cohorts automatically from your Stripe data, so the table redraws itself the second a customer upgrades, downgrades, or churns. You get the three reads, the logo and revenue curves side by side, and LTV computed empirically per cohort, without a single VLOOKUP. See how cohort analysis and segmentation work on live Stripe data, or read the longer SaaS cohort analysis guide.
Pull your last 12 cohorts and name the shape. If it decays, fix retention before you spend another dollar acquiring. If it smiles, your next dollar belongs in expansion.
FAQ
What is cohort analysis in SaaS?
Cohort analysis groups customers by a shared start point, usually the month they first paid, and tracks each group’s retention, revenue, or expansion over time. Instead of one blended company-wide churn number, you see how each monthly cohort behaves as it ages, which exposes whether retention is actually improving and lets you calculate realistic, segment-specific LTV.
How do you read a cohort retention table?
Read it in three directions. Across a row shows one cohort decaying or expanding over its lifetime. Down a column compares different cohorts at the same age, so a rising column means onboarding and product are improving. Along the diagonal you reconstruct a single month’s total revenue as a stack of contributions from every cohort alive that month.
What is a revenue retention curve?
A revenue retention curve plots a cohort’s recurring revenue at each month of its life as a percent of its starting revenue. Unlike a logo-retention curve, which counts only accounts, the revenue curve captures expansion and contraction, so it can climb above 100 percent when upgrades from survivors outweigh the dollars lost to churn.
What does a good cohort retention curve look like?
There are three shapes. A decaying curve keeps falling and never stabilizes, which is a warning sign. A flattening curve drops early then levels into a loyal core, often around 80 percent logo retention and roughly 90 percent gross revenue retention, which is healthy. A smiling curve dips then rises back above 100 percent as expansion outpaces churn, which is the top tier.
How is cohort analysis used to forecast revenue?
You take the flattening tail of your average retention curve, project it forward (early-stage models often assume decay settling around 2 to 5 percent monthly churn after year two), then multiply the surviving revenue in each future period by ARPA. Summing those layers across all live cohorts forecasts revenue from your existing base before any new sales, and discounting the streams gives a defensible LTV.
About the Author
Matt Smith
Serial entrepreneur and former big 4 consultant turned SaaS operator. Built and scaled analytics and data warehouses platforms at multiple enterprise Stripe companies before founding Mowt. Passionate about making complex metrics accessible to every founder.