Reading a Revenue Retention Curve, Shape by Shape

A revenue retention curve plots how much money a single cohort of customers still spends with you in each month after they signed up, expressed as a percentage of what that cohort spent in month zero. Read left to right, it takes one of three shapes: it decays toward zero, it flattens onto a plateau, or it bends back upward into a smile. The shape tells you whether the cohort will ever repay what you paid to acquire it, and it does so far more honestly than any blended, all-customers-at-once retention number your dashboard shows by default.

I've built a lot of these curves, usually the night before a board meeting, usually because someone asked "so is the new cohort actually better or does it just look better." The curve answers that. But only if you read it correctly, and there's one reading that trips up almost everyone: a beautiful upward smile can sit on top of a business that's quietly bleeding customers. More on that below, because it's the part that'll get you in trouble.

What the curve actually measures

Start with the raw material. You pick a cohort by acquisition date, say everyone who first paid you in January. You record their combined revenue that month and call it 100%. Then, month by month, you record what that same group of accounts spends, and you divide by that January baseline. No new customers get added to this line ever. That's the whole point. A cohort is a closed room, and you're watching what happens to the people already inside it.

Because you never add fresh signups, the curve can only move for three reasons: accounts leave (revenue drops), accounts shrink or grow their spend (revenue drifts either way), or accounts upgrade and buy more seats (revenue climbs). The interplay of those three forces is what bends the line into one of the classic shapes.

Aggregated retention hides all of this. Runway makes the point cleanly in its 2025 write-up on cohort analysis for finance: aggregate metrics give you the headline, while cohort analysis tells you the real story, because tracking groups that joined at the same time uncovers patterns the blended number smooths over. When your overall retention looks stable, it's often because a decaying old cohort and an improving new one are cancelling each other out on the same chart. Split them and the truth falls out.

The three shapes, and what each one is telling you

Here's the table I keep pinned. If you take one screenshot from this article, take this one.

Curve shape	What it looks like	What's happening inside the cohort	CAC verdict
Decaying	Falls month over month, no floor, heads toward zero	Churn outruns expansion; nobody is sticking	You almost certainly never recover CAC
Flattening	Drops early, then holds a stable plateau (say 70%)	Early churn burns off, a loyal core remains	You recover CAC if the plateau clears payback math
Smiling	Dips, bottoms out, then climbs back above 100%	Surviving accounts expand faster than others leave	Best case for LTV, but read the fine print

The decaying curve is the one you want to catch early. It means each month erodes the last with no stabilizing floor, and total lifetime revenue from that cohort converges to a finite, often disappointing, number. If the area under that curve doesn't exceed your blended CAC, the cohort loses money over its entire life. No amount of new-logo growth fixes a product that leaks like this; you're pouring acquisition spend into a bucket with a hole in the bottom.

The flattening curve is the workhorse of healthy SaaS. Revenue falls for the first few months as the tourists and mismatched buyers cancel, and then it settles onto a plateau held up by customers who genuinely need the product. That plateau height is the single most important number on the chart. A cohort that flattens at 70% and stays there for years is a good business. A cohort that flattens at 25% is a subscription with a leak that merely slowed down.

The smiling curve is the trophy. It dips, finds a bottom, then climbs back past its starting point because the customers who stayed keep buying more, faster than the leavers subtract. When people talk about "negative net churn" or net revenue retention above 100%, this is the picture. And it's real, and it's wonderful, and it is also where the traps live.

Why a smile can lie to you

Here's the thing the smile doesn't tell you on its own: whether it's built on breadth or on a handful of whales.

A revenue retention curve is measured in dollars, not in customers. That means a cohort can lose a third of its logos and still smile, as long as the accounts that remain expand enough to cover the loss. Optifai's 2025 dataset of 939 B2B SaaS companies found a median gross revenue retention of 91% against a median net revenue retention of 103%, a roughly 12-point gap driven by seat expansion and tier upgrades. Their read on it is blunt: a big gap between the two "reveals potential structural issues where businesses are losing roughly one-third of customers annually but masking this with expansion from surviving customers."

So the dollar curve smiles while the logo count sinks. If you only ever look at the revenue shape, you'll conclude the cohort is thriving right up until the expansion engine stalls, at which point the shrinking base has nothing left to hide behind and the smile collapses into a decay. I watched this happen to a company whose net retention chart was the prettiest thing in the deck for four straight quarters. Their logo retention had been falling the entire time. When two large accounts renewed flat instead of expanding, the whole curve rolled over inside a single quarter, and suddenly the "growth" was gone.

The fix is boring and non-negotiable: plot two curves, not one. Put the revenue retention curve next to a logo retention curve for the same cohort. If revenue smiles and logos hold, you have a durable business. If revenue smiles and logos decay, you have a countdown. SaaS Capital's 2025 benchmarking makes the same argument from the investor side, noting that strong net retention can temporarily mask deep underlying churn, which is exactly why they dig into gross retention first to see the floor of the business.

For a fuller taxonomy of curve shapes and the behavioral reasons a product's retention bends the way it does, our friends at the product analytics handbook cover the shape decoding in more depth.

Reading the curve for CAC recovery

The reason a CFO cares about this shape at all is repayment. Every cohort arrives with a bill attached — the sales and marketing you spent to acquire it. The retention curve tells you whether, and when, that cohort's cumulative revenue clears the bill.

Mechanically, you sum the monthly revenue under the curve, subtract cost of goods, and watch the running total climb until it crosses accumulated CAC. The month it crosses is that cohort's payback point. A flattening curve with a high plateau crosses quickly. A decaying curve may never cross, which is the mathematical way of saying the cohort is unprofitable no matter how long you wait.

The benchmarks are worth holding in your head as a sanity check. Optifai's 2025 figures put median SaaS CAC payback at 16 months, an improvement from 18 months in 2024, with top-quartile companies recovering in six months or less. That same data shows payback stretches hard with deal size: companies selling contracts above $100,000 had a median payback of 24 months, versus roughly nine months for those under $5,000. So a two-year payback isn't automatically alarming for an enterprise-priced product, and a nine-month payback isn't automatically great for a self-serve one. Context first, then judge the curve.

One more caution on benchmarks generally, because it's my job to be the killjoy here: most published retention and payback numbers are survivorship-biased. The companies that report into benchmark surveys are, by definition, still around to answer the survey. The cohorts that decayed hardest belong to companies that already died and aren't in the sample. Treat any median NRR you read — including the 106% Benchmarkit reported for 2025 — as the retention of the survivors, not of everyone who ever tried the model. Your own decaying cohorts are the ones nobody publishes.

If you want the full machinery for turning these curves into a defensible payback-and-LTV model, I laid it out in the unit-economics dashboard template. And for teams doing this on mobile, where you're forecasting a cohort's value from day-zero signals long before the curve fills in, the predictive-LTV approach for SKAN is the companion read.

Building the curve without losing a week to SQL

The classic way to produce a cohort revenue curve is a self-join in SQL: bucket accounts by their first-payment month, sum revenue per account per calendar month, index each month against the cohort's baseline, pivot into the triangular cohort grid. It works. It's also finicky, and every analyst I know has shipped a version with an off-by-one on the month index at least once, usually the version that made it into a board deck.

Spreadsheets do it too, if your account count is modest and you enjoy dragging formulas across a hundred columns. Purpose-built tools take different routes. Some product analytics platforms have a dedicated cohort or retention report where you pick the metric and the shape draws itself. Others, the newer AI-native ones like Kixo, let you describe the cohort curve you want in plain language and generate the chart from a prompt rather than hand-building the query, which is handy when a CFO asks a follow-up mid-meeting and you'd rather not disappear into a SQL editor for twenty minutes.

None of these removes the judgment. Whatever tool draws the line, you still have to decide whether to index on recognized revenue or bookings, whether to strip involuntary churn, and — the big one — whether to trust a smile before you've checked the logo curve underneath it. The tool renders the shape. Reading it correctly is still on you.

A short field guide, in order

When a fresh cohort curve lands on my screen, I read it in this sequence, and I'd suggest you do the same:

1. Find the shape. Decay, flatten, or smile — name it before anything else.
2. If it flattens, read the plateau height. That number is the durable core.
3. If it smiles, immediately pull the logo retention curve for the same cohort. Breadth or whales?
4. Sum under the curve against this cohort's CAC. Does it cross payback, and when?
5. Compare to the previous cohort's curve at the same age. Better, worse, or noise?

That fifth step is the one people skip and the one that actually pays for the analysis. A single curve is a snapshot; the trend across successive cohorts is the diagnosis. If each new cohort flattens a little higher than the last, your product is getting stickier and every future dollar of acquisition is worth more. If each new cohort decays a little faster, you have a problem that new-logo growth is currently papering over, and the paper is thin.

FAQ

What's a good plateau for a flattening revenue curve? There's no universal number because it swings with price and segment, but the median gross revenue retention landmarks help: Benchmarkit put 2025 median GRR at 90% annually and Optifai at 91%. If your annual revenue plateau sits well below the mid-80s, early churn is eating more than the survivor base can carry, regardless of how the net number looks.

Is a smiling curve always better than a flattening one? In pure dollar terms, yes, a curve that climbs past 100% returns more lifetime revenue than one that plateaus below it. But a smile built on expansion from a shrinking set of accounts is more fragile than a broad, boring plateau. Durability beats a pretty shape when the expansion engine hiccups.

Why does my blended retention look fine when individual cohorts look bad? Because blending mixes an improving new cohort with a decaying old one and averages away both signals. That's the entire reason to cohort in the first place. Split by acquisition month and the divergence appears.

How many customers do I need before the curve is trustworthy? Enough that no single account moves the line meaningfully — I get nervous below about 30 accounts per cohort. Under that, you're reading one or two customers' renewal decisions dressed up as a trend, and you should caveat the chart accordingly.

The revenue retention curve is one of the few charts I trust to survive contact with a skeptical CFO, precisely because it's hard to game once you're plotting logos alongside dollars. Draw both. Name the shape. Then decide whether the cohort earns its keep.