%20scilla.studio.svg)
Product-market fit is what you have when the people who use your product would be genuinely upset to lose it, and they keep coming back without being chased. The most cited proxy is the Sean Ellis survey: ask users "how would you feel if you could no longer use this product?" and if 40% or more say "very disappointed," you supposedly have fit. It's a useful smoke signal, but it's a survey, not the thing itself. The real evidence of product-market fit lives in your retention curve: do your cohorts stop churning and settle onto a flat plateau, or do they bleed to zero? That plateau, rather than a 40% threshold, is the strongest evidence you can get.
The 40% test became famous because it gives a tired founder a single number to chase, and a single number is comforting. But fit is a behaviour, not an opinion. People can tell a survey they'd be "very disappointed" and then quietly stop logging in. The retention curve doesn't lie the way a survey does.
Product-market fit = a defensible match between what your product does and a market that demonstrably wants it, shown by users who retain (come back on their own) and, ideally, expand and refer. The Sean Ellis 40% test is one input. The harder evidence is:
None of these is a single magic number. They're a shape and a set of relationships. That's the part most "is it fit yet?" guides skip.
Sean Ellis ran growth at Dropbox, LogMeIn and Eventbrite, and noticed that the companies that took off shared a survey pattern: roughly 40% of users said they'd be "very disappointed" without the product. Below that line, growth efforts mostly fizzled; above it, they compounded. So the 40% benchmark was an observation, not a law of physics.
That origin matters because it tells you exactly where the test is weak:
So run the survey: it's cheap and the open-text answers are gold for finding your most-loved use case. Just don't mistake the proxy for the territory.
This is the single clearest signal, and it's the one the survey can't give you. Plot the percentage of a cohort still active over time. Three shapes are possible:
The number that matters isn't the height of any single point. What matters is whether the curve goes flat. The way our benchmark tool frames it: look for curve flattening, not just absolute numbers. A 25% plateau that holds is healthier than a 45% that's still sliding.
What "flat" looks like by month is heavily context-dependent, which is the whole point, but here are sourced reference bands to read your own curve against.
| Retention point | B2B SaaS (avg) | Consumer apps (avg) |
|---|---|---|
| Day 1 | 50% to 70% | 20% to 30% |
| Day 7 | 40% to 60% | 8% to 15% |
| Day 14 | 35% to 55% | 4% to 8% |
| Day 90 | 25% to 35% | 1% to 4% |
Sources: Pendo Product Benchmarks, Amplitude, Mixpanel (B2B); Adjust, AppsFlyer, UXCam (consumer).
Look at the gap. A consumer app with 3% Day-90 retention can have excellent product-market fit, while a B2B tool at the same 3% is dying. Different growth mechanics, different physics, which is why comparing your B2B numbers to a consumer benchmark (or vice versa) will lie to you. Read the curve against the right reference class: a B2B number judged against a consumer band will flatter or terrify you for no reason.
The other tell, hiding in the table: by Day 14 the drop should be flattening. In the way our benchmark tool reads it, a steep decline that hasn't levelled by then signals a weak core loop. The slope tells you more than any single point. (For the curve-by-curve breakdown, see retention-rate-benchmarks.md.)
A flattening curve for one cohort is good. The stronger signal is comparing cohorts over time. Group users by the month they joined and plot each cohort's retention separately.
This is why a single aggregate retention figure is dangerous: it averages a loyal early cohort with a churning recent one and hides the divergence. Fit is a question you have to keep asking of each cohort, not answer once.
Here's the interesting part: product-market fit rarely shows up in one metric. It shows up in how your metrics relate. A few of the interactions that matter most:
So which signal is "the" signal? None of them alone. Fit is the state where retention has plateaued, cohorts are holding, and the unit economics and any loop all point the same way. When they fight each other (great survey, leaking curve; great ratio, decaying cohorts), you don't have fit yet, you have a number that's flattering you.
You'll have noticed I keep giving ranges and then immediately undercutting them. That's deliberate, and it's the core thesis of how we think about this.
A benchmark is a reference class, not a goal line. "Good" Day-90 retention is 25% to 35% for B2B and 1% to 4% for consumer, but your "good" depends on your business model (freemium vs. self-serve vs. enterprise sales move the bar a lot), your contract length, your switching costs, and what job your product is hired to do. A frequency-of-use product should have a higher, flatter curve than a once-a-quarter tool, and both can have genuine fit.
The failure mode is treating a benchmark as a target: you optimise to hit 40% on the survey, or to drag Day-7 up to the middle of the band, and you start gaming the proxy instead of building the thing. The benchmark's job is to tell you which questions to ask (is my curve flattening at all? are my newer cohorts holding? do my economics rest on real retention or forecast retention?), not to hand you a finish line.
In my honest, experience-earned opinion: the teams who win don't ask "did we hit 40%?" They ask "has the curve gone flat, and is it staying flat as we scale?" That's a harder question, and a truer one.
A practical sequence, in order of how much each step tells you:
Do these in this order and the survey becomes what it should be: a way to understand your fans, not a gate you're trying to clear.
If the diagnosis comes back ugly, the levers are roughly in this order too: fix the curve before you fix the funnel. Sharpen the ICP so you stop acquiring users who were never going to stay, deepen the core loop that drives repeat use, and only then pour fuel on acquisition or virality, since a leaky bucket just churns faster the more you fill it. The retention and unit-economics breakdowns below go deeper on each lever.
The hard part is not knowing that retention should flatten. The hard part is reading your own curve against the right reference class without fooling yourself. That's exactly what our free benchmark tool does: paste in your retention by month and your unit economics, and it charts your curve against the B2B and consumer bands, flags whether you're plateauing or still sliding, and shows how your retention, K-factor and LTV:CAC relate, as context, not as targets. It takes a couple of minutes.
Check your product-market fit signals → benchmark.scilla.studio
For the wider picture, see our full B2B SaaS growth benchmarks for 2026.
Is the Sean Ellis 40% test still valid? It's a useful proxy, not proof. The 40%-very-disappointed threshold came from observing a handful of high-growth products, so it's directional rather than a universal pass/fail line. Run it for the open-text insight into who loves your product and why, but verify fit with revealed behaviour (a flattening retention curve), since surveys only sample surviving users and capture stated, not actual, behaviour.
What's the clearest signal of product-market fit? A retention curve that flattens onto a stable plateau instead of decaying to zero. That flat tail is a base of users who found durable value and keep returning on their own. The height of the plateau depends on your market (a 25% B2B plateau and a 3% consumer plateau can both indicate fit), but in every case it has to flatten.
What's a good retention rate for product-market fit? For B2B SaaS, roughly 50% to 70% Day-1, 40% to 60% Day-7 and 25% to 35% Day-90 (Pendo, Amplitude). For consumer apps, 20% to 30% Day-1, 8% to 15% Day-7 and 1% to 4% Day-90 (Adjust, AppsFlyer). These are reference bands, not targets, and the slope flattening matters more than any single point.
Can you have product-market fit with low retention? Yes, if you're consumer and your curve still flattens. A consumer app with a 3% Day-90 plateau can have real fit, while a B2B tool at the same level is dying. Always read your numbers against the right reference class (B2B to B2B, consumer to consumer) because they follow different growth mechanics.
How is product-market fit different from traction? Traction is growth you can sometimes buy with spend or hype. Product-market fit is durable: it shows up as users who retain, cohorts that hold as you scale, and unit economics that rest on real retention rather than forecast retention. You can have traction without fit (a leaky bucket filling fast); fit is what makes the traction compound instead of leak.
