AS | Ankit Sarawagi|Founder, CFOmatrix·July 2026·9 min read | Metrics deep-dive |
- A cohort is customers grouped by start period. Tracking that fixed group over time separates the behaviour of old customers from the noise of new sign-ups.
- Logo retention counts customers; revenue retention counts money. Logo retention can only fall; net revenue retention can rise above 100 percent when expansion beats churn.
- A good curve flattens into a plateau. A curve that keeps decaying month after month means a leaking bucket.
- Around 12 percent annualised churn is acceptable. A failed POC is not a leak; it is a sales cost, part of customer acquisition cost.
- In AI SaaS, read cohorts on margin. Compute varies per customer, so a cohort can grow in revenue while margin quietly erodes.
| >100% Net revenue retention when expansion beats churn | ~12% Annualised churn ceiling for healthy SaaS | ~30% POC to customer conversion (a sales cost, not churn) |
What a Cohort Is, and Why It Beats Blended Churn
A cohort is a group of customers who started in the same period: everyone who signed up in January 2025, say. The trick is that the group is fixed. Nobody joins it after January, so as you track it forward you are watching the same people age, not a moving crowd blurred by new sign-ups. That is why cohort analysis tells the truth where a single blended churn number lies.
Imagine a company adding customers fast. Its overall churn looks fine because a flood of fresh sign-ups masks the fact that last year’s customers are quietly leaving. Split those customers into monthly cohorts and the leak is obvious: each cohort thins out over time. Blended metrics hide age; cohorts expose it. This is the foundation for every retention number in net revenue retention and churn and for the whole picture in our SaaS finance pillar guide.
| Cohort | M0 | M3 | M6 | M9 | M12 |
| Jan 2025 | 100% | 98% | 101% | 106% | 112% |
| Feb 2025 | 100% | 97% | 99% | 104% | 108% |
| Mar 2025 | 100% | 95% | 96% | 100% | 103% |
| Apr 2025 | 100% | 90% | 85% | 82% | 79% |
Read the grid two ways. Down a column compares cohorts at the same age (is a newer cohort retaining better or worse than an older one?). Along a row follows one cohort as it matures. The Jan and Feb cohorts here climb above 100 percent by month twelve, expansion is outrunning churn. The April cohort decays, a warning that something about who you acquired, or how you onboarded them, changed.
Logo Retention vs Revenue Retention
The single most important distinction in retention is what you are counting. Logo retention counts customers. Revenue retention counts money. They can point in opposite directions, and confusing them is how founders misread their own business.
So one cohort is simultaneously 88 percent on logos and 112 percent on net revenue. Neither is wrong. They tell different truths: you are losing some (usually smaller) customers, while the ones who stay grow. Investors read both, because a healthy net figure can mask a logo problem at the bottom of the base, and a weak logo number in a land-and-expand business can still be fine if the accounts that matter are expanding. Three flavours are worth naming:
- Logo retention: percentage of customers retained. Simple headcount.
- Gross revenue retention (GRR): revenue kept from the cohort, counting churn and downgrades but not upsell. Capped at 100 percent.
- Net revenue retention (NRR): revenue from the cohort including expansion. Can exceed 100 percent, and usually the headline number.
GRR and NRR always travel together. GRR tells you how leaky the bucket is; NRR tells you whether expansion is filling it faster than it leaks. Report both, or you can flatter a leaky product with a strong NRR driven by a handful of big accounts.
Retention Curves: Flattening (Good) vs Decaying (Bad)
Plot a cohort’s retention against time and you get a retention curve. Its shape is the single clearest read on product-market fit. There are two shapes that matter.
The teal curve is the one you want. It dips in the first few months as poor-fit customers self-select out, then flattens into a plateau. That plateau is the signal that you have found a durable core of customers who keep using the product. The height of the plateau is roughly your long-run retention.
The red curve is the warning. It never flattens; it keeps decaying toward zero. There is no stable core, so every month of growth depends on refilling a bucket that never stops leaking. On a revenue basis the very best businesses do something even stronger: the curve bends upward over time, because expansion within the survivors outweighs all the revenue lost to churn. That is what a net-revenue-retention-above-100-percent cohort looks like when you plot it.
Do not let failed proofs of concept pollute your churn number. A leak is a live, paying customer who leaves. A POC that does not convert is a sales cost, effectively part of customer acquisition cost, not a retention loss. POC to customer conversion typically runs around 30 percent, or 40 to 45 percent if you are selective. Keep those two buckets separate or your churn will look far worse than it is.
How Expansion Lifts Net Retention Above 100 Percent
Expansion is upsell and cross-sell inside your existing base, and it is the mechanism that lets a cohort grow in revenue even as it loses a few logos. When expansion outweighs churn and downgrades, net revenue retention climbs above 100 percent, and the business compounds without adding a single new customer.
“Net revenue retention is the single most important metric, for the founder and the investor. And you raise it on two fronts at once: growing consumption within the account through customer success, and land-and-expand across new departments through sales.”
Ankit Sarawagi, from working across SaaS and AI SaaS startupsThis is exactly why the logo curve and the revenue curve diverge. Logo retention can only fall as customers leave, but every rupee of expansion on a surviving account adds back on the revenue side. Read the mechanics end to end in net revenue retention and churn, and note the cash angle: expansion within a locked-in annual contract is the most efficient revenue you will ever book, which ties straight into lifetime value and LTV:CAC.
The AI SaaS Twist: Read Cohorts on Margin, Not Just Revenue
In traditional SaaS, revenue retention is nearly the whole story, because the cost to serve a customer is close to zero and roughly the same for everyone. AI SaaS breaks that assumption. Compute cost varies by customer and by usage, so two customers paying the same can carry wildly different margins. A cohort that looks healthy on revenue retention can be quietly eroding on margin if the survivors happen to be your heaviest compute users.
In the AI SaaS companies I have worked with, revenue retention alone can mislead you. You have to layer gross margin onto the cohort, per customer and per cohort, because compute is variable. Set up metering and per-customer tags from day one; it is not complicated, but it is very hard to reconstruct later. A cohort that expands revenue at falling margin is a repricing signal, not a win.
So the AI SaaS cohort grid gains a second layer: alongside revenue retention per cohort, you track gross margin per cohort. When both rise together, the cohort is genuinely healthy. When revenue climbs but margin falls, you have a pricing problem to fix at renewal, exactly the kind of per-customer decision that per-customer gross margin exists to surface. It is the same discipline that runs through the whole SaaS finance pillar: in AI SaaS, blended numbers hide the truth, and the customer-by-customer view is where the real signal lives.
Frequently Asked Questions
What is cohort analysis in SaaS?
Cohort analysis groups customers by the period they started (say, everyone who signed up in January 2025) and tracks that fixed group over time. Because the group never changes, you can see honestly how much of it survives and how much revenue it generates in month one, month six, month twelve and beyond. It separates the behaviour of old customers from the noise of new sign-ups, which blended metrics hide, and it is the cleanest way to see whether your product actually retains people.
What is the difference between logo retention and revenue retention?
Logo retention counts customers: if 100 customers start and 88 are still active a year later, logo retention is 88 percent. Revenue retention counts money from that same group, including upsell and downgrades: if those survivors now pay more than the original 100 did, revenue retention can exceed 100 percent even though some logos left. Logo retention can only fall toward zero; net revenue retention can rise above 100 percent because expansion offsets churn. Investors read both, because a company can lose small customers while growing the ones that stay.
What does a good SaaS retention curve look like?
A healthy retention curve drops in the first few months as poor-fit customers leave, then flattens into a stable plateau, which means you have found a core of customers who keep the product. A bad curve keeps decaying month after month with no plateau, which means you are constantly refilling a leaking bucket and growth is fragile. On a revenue basis the best curves bend upward over time because expansion within surviving accounts outweighs the revenue lost to churn.
How does expansion affect retention?
Expansion is upsell and cross-sell inside your existing base, and it is what lets net revenue retention climb above 100 percent. You raise it on two fronts: growing consumption within each account through customer success, and selling into new departments and use cases through a land-and-expand motion. A cohort can lose a few logos yet still grow in revenue if the survivors expand faster than the leavers subtract. This is why net revenue retention, not logo retention, is usually the single most important SaaS number.
What churn rate is acceptable for SaaS?
Around 12 percent annualised churn is generally acceptable for SaaS; materially above that signals a retention problem. One important clarification: a failed proof of concept is not churn. A leak is a live, paying customer who leaves. A POC that does not convert is a sales cost, effectively part of customer acquisition cost, not a retention loss. POC to customer conversion typically runs around 30 percent, or 40 to 45 percent if you are selective about which POCs you take on.
How do AI SaaS cohorts differ from traditional SaaS cohorts?
In AI SaaS you cannot read a cohort on revenue alone, because compute cost varies by customer and by usage. A cohort can look healthy on revenue retention while margin quietly erodes, if the surviving customers are the heaviest compute users. So you track per-customer and per-cohort gross margin, not just per-cohort revenue, using metering and per-customer tags set up from day one. A cohort that expands revenue but at falling margin is a repricing signal, not a success.
This is general educational information for founders, current to mid-2026, drawing on the author’s experience across SaaS and AI SaaS startups, and is not legal, tax or investment advice. Cohort tables and curves shown are illustrative, not real client data. Benchmarks are indicative and vary by stage, motion and business model.
AS | Founder, CFOmatrix | Finance Strategy & Equity Compliance CFOmatrix is a knowledge platform focused on how finance actually works inside growing companies. This guide draws on hands-on experience across SaaS and AI SaaS startups, from cohorts and retention to per-customer margin, pricing and the finance function. |