You can see product-market fit coming before revenue. The early signs are pull rather than push: users return without reminders, a meaningful share would be very disappointed to lose the product, retention curves flatten instead of falling to zero, strangers refer others unprompted, and people try to pay before you ask.
Revenue confirms fit, but it arrives late and it can be bought with discounts and founder-led selling. The signals below arrive earlier, cost little to measure, and work with a few dozen users. None is proof on its own. Three or more pointing the same way is worth taking seriously.
What product-market fit means, and what it feels like
The definition most founders work from is Marc Andreessen's, from his 2007 essay The only thing that matters: "Product/market fit means being in a good market with a product that can satisfy that market."
The same essay describes the difference in practical terms. Without fit, customers are not quite getting value, word of mouth is not spreading, usage is not growing fast and sales cycles drag. With it, "the customers are buying the product just as fast as you can make it." That second state is unmistakable once you are in it. The work is noticing the signals that come before it.
Signal 1: pull, not push
The simplest test is who is doing the work. If every active user is active because you messaged them, onboarded them on a call or reminded them this week, that is push. If users return on their own, file feature requests, complain when something breaks and ask when the next thing ships, that is pull.
How to measure it: for two weeks, stop all reminder emails and personal nudges to a subset of users, and compare their activity with the rest. If usage collapses without your effort, the product is not yet pulling. Also count inbound messages that were not replies to you: bug reports, requests and "can my colleague use this?" are all pull.
Signal 2: a retention curve that flattens
Plot what share of each weekly or monthly cohort is still active over time. Almost every curve drops at first. What matters is whether it then levels off at some share of users who keep coming back, or keeps sliding towards zero. A flat tail, even at a modest level, means a group of people have made the product part of their routine.
For a sense of scale, Lenny Rachitsky's benchmarks for good retention, gathered from operators and investors, put good six-month user retention at around 40% for consumer SaaS and around 60% for SMB and mid-market SaaS. Treat those as a rough reference, not a target: your category, price and usage frequency move them a lot.
How to measure it with few users: cohorts of ten are noisy, so look at shape rather than exact percentages, and define "active" as the action that delivers value (a report generated, an invoice sent), not a login.
Signal 3: the 40% "very disappointed" test
Sean Ellis's survey asks users one question: how would you feel if you could no longer use this product? The options are very disappointed, somewhat disappointed and not disappointed. In Rahul Vohra's account of how Superhuman built an engine to find product-market fit, Ellis found that "the magic number was 40%": companies where at least 40% of users answered very disappointed tended to have fit.
Vohra's write-up adds three practical details. First, the question is a leading indicator, unlike revenue and growth, which lag. Second, results become directionally correct at around 40 respondents, so you can run it early. Third, the score is something to work on rather than just read: Superhuman's started at 22% and reached 58% after three quarters of deliberate work.
How to run it: survey only people who have used the product meaningfully in the last two weeks, ask the question before anything else, and add the follow-ups Superhuman used: what type of person would most benefit, what is the main benefit, and how could it be improved. The very disappointed group's answers describe who your product is really for.
Signal 4: unprompted referrals
A user recommending you when asked is politeness. A user recommending you when nobody asked is evidence. Watch for signups you cannot attribute to anything you did, new users who name an existing user, and team accounts that grow by themselves.
How to measure it: add a single optional "How did you hear about us?" field at signup and read the free-text answers weekly. A small but growing share naming a person rather than a channel is one of the most reliable early signals, because nobody refers a product that embarrasses them.
Signal 5: people try to pay
Before revenue, the strongest signal is an attempt to give you money: a pre-order, a deposit, a paid pilot, a signed letter of intent with a price in it, or a user asking how to upgrade before a paid plan exists. Each costs the buyer something, which is what separates it from interest.
How to measure it: put a real price in front of engaged users, even if billing is manual. Count how many say yes and how many ask about price unprompted. The customer interviews guide explains why a stated intention to buy is weak evidence and an actual commitment is strong.
Signal 6: growth that does not depend on you
Growth rate is a lagging signal, but it is still worth tracking from the start because it tells you whether the other signals are compounding. Paul Graham's essay Startup = Growth gives the benchmark Y Combinator uses: "A good growth rate during YC is 5-7% a week." For startups that are not yet charging, the same essay says the next best thing to measure is active users.
The useful question is not the rate itself but its source. Growth that comes from referrals and returning users is fit. Growth that tracks how many hours you spent on outreach that week is effort.
False positives that look like traction
| Looks like | Usually means | What to count instead |
|---|---|---|
| Friends and family using it daily | They are being kind, and they know you | Usage from people with no personal connection |
| A big launch-day spike | Curiosity from a single traffic source | How many from that cohort are active four weeks later |
| Thousands of free-tier signups | The price was zero, so the decision cost nothing | Signups who completed the core action twice |
| "I'd definitely pay for this" | A polite prediction about future behaviour | A pre-order, deposit or signed letter of intent |
| Usage that follows your reminder emails | Push, not pull | Usage in a cohort you stopped nudging |
| One large customer who loves it | Fit with one buyer, possibly custom work | Whether the next three similar buyers say the same |
The common thread: count behaviour from strangers who had an easy way to leave, and discount anything you caused.
When the signals are weak
Weak signals are information, not failure. Usually they point to one of three causes: the problem is not painful enough for this group, the product does not yet solve it well, or you are serving several groups and only one of them cares. The last is the most common and the cheapest to fix. Segment your survey and retention data by user type, and you will often find one group with strong numbers hidden inside a weak average.
If no group shows pull, go back a step and test the problem itself before building more. Is my startup idea good? covers the ten questions to revisit, and how to evaluate a startup idea covers scoring what you find without more code.