How do you know if you have product-market fit?
Product-market fit is less about hitting a number and more about a qualitative shift in how your product moves—from something you push to something that pulls. The clearest early signal is retention and word-of-mouth behavior you didn't manufacture. Before you can read those signals honestly, you have to be close enough to your users to see them.
The push-to-pull transition is the real signal
Patrick Collison's description of Stripe's early momentum—captured in Paul Graham's 'Do Things That Don't Scale'—is the best plain-language definition of product-market fit most founders will encounter. The product stops feeling like a boulder you're shoving uphill and starts accumulating its own forward force. Users come back without prompting. They tell other people without incentive. Support requests shift from 'how do I do X?' to 'can you add Y?'—meaning people are already using the core and want more, not trying to figure out what you built.
The challenge is that this transition is felt before it's measured. Founders often look for a single dashboard metric to confirm what they already sense—or deny what they suspect isn't there yet. The honest test is simpler and harder: if you stopped all outbound tomorrow, would usage decay immediately or would it hold? If the answer is 'it would collapse,' you don't have fit yet. If there's a cohort of users who'd genuinely be upset if the product disappeared, you're getting closer.
Retention curves are the most reliable quantitative proxy. If your weekly or monthly active user curve flattens and holds after the initial drop-off, that flat line is fit. If it slopes continuously toward zero regardless of how many new users you add, you're refilling a leaky bucket—and no growth tactic fixes a product that isn't working for the people who've already tried it.
You can't measure fit from a distance—you have to be inside the usage
Paul Graham makes a point in 'Do Things That Don't Scale' that founders often underestimate: the quality of feedback you get from direct, close observation of early users is categorically better than anything you'll get later through surveys, analytics dashboards, or focus groups. At scale, you're inferring behavior from aggregate data. In the early stage, you can sit next to someone using your product and watch exactly where they hesitate, where they succeed, and what they do immediately after.
This isn't just a nice-to-have. It's how you distinguish 'they said they liked it' from 'they actually used it again.' Users are polite. They'll tell you the product is great in a Zoom call and then never open it again. Watching someone use your product in their real context—at their desk, in their workflow, under actual time pressure—reveals friction that no survey will surface. The user who 'loves the product' but takes three minutes to complete a task that should take twenty seconds is telling you something important that they'd never think to say out loud.
Fit lives in repeated, unprompted use. If you're engineering every session—sending reminders, personally following up, doing manual work behind the scenes to make the product appear to function—you haven't found fit, you've found politeness. The signal you want is the user who comes back before you reach out.
Narrow depth beats broad mediocrity as a proving ground
One of the most counterintuitive things about finding product-market fit is that the path to it usually runs through a smaller market than founders want to admit. Paul Graham uses Facebook's early Harvard-only strategy as a case study in deliberate constraint: by limiting the initial market to a tiny, specific group, the product could be dense with relevance for those users in a way that a general-audience product almost never is in its early form. Those users felt like the product was built specifically for them—because it was.
For most startups, this means resisting the temptation to generalize the product prematurely. Fit happens in a specific context, for a specific type of user, solving a specific version of a problem. If you're getting strong signals from one segment and weak signals from others, the answer isn't to average them out by building for everyone—it's to go deeper on the segment where the pull is already happening. A product that a small group finds genuinely indispensable is much closer to fit than a product that a large group finds marginally useful.
This also reframes how you think about early user selection. Picking early users who are genuinely representative of the problem you're solving—not just the most accessible people or the friendliest ones—is what makes early feedback valid. If your first hundred users are friends doing you a favor, their retention and usage data tells you nothing useful about whether the market wants what you built.
What fit doesn't look like—and the traps founders fall into
Garry Tan has pointed out that early-stage founders often mistake looking like a successful company for being one. This shows up in the product-market fit conversation as over-reliance on vanity metrics: download counts, sign-up rates, or favorable press coverage that creates the appearance of traction without the substance of it. None of those things measure whether users are getting durable value from the product. They measure whether you're good at distribution and PR, which is a different skill from building something people actually need.
Another common trap is confusing early enthusiasm for fit. Early adopters—especially in tech—will try almost anything new and give it generous initial reviews. The test is what happens in month two and month three. Did usage deepen? Did they start using features they didn't ask for because they discovered value on their own? Did they tell someone else? The drop-off from first week to sixth week is where most unfit products reveal themselves, and founders who aren't tracking that cohort curve carefully can spend months mistaking early excitement for a sustainable signal.
Finally, founders sometimes confuse sales effort with product pull. If every new customer requires a long sales process, significant hand-holding, and ongoing manual support to stay active, that's a sign the product isn't yet doing the work on its own. Fit doesn't mean zero sales effort—but it does mean the product should be doing most of the convincing once a user is inside it.
“It tipped from being this boulder we had to push to being a train car that in fact had its own momentum.”
— Patrick Collison, via Paul Graham, source
The one thing to do
Stop outbound for two weeks and watch: if a real cohort comes back and uses your product without prompting, you're close—if usage collapses, go back to watching users in their actual environment until you find the version that pulls.
Frequently asked questions
Is there a retention number that definitively signals product-market fit?
No single number is universal—it depends on your product category and usage frequency. What matters is whether your retention curve flattens into a stable cohort rather than declining toward zero. A flat curve with even a modest percentage of retained users is a stronger signal than high initial adoption that decays completely.
Can you have product-market fit in a small niche?
Yes, and that's often where fit is found first. Strong fit in a narrow segment—where users find the product genuinely indispensable—is the right starting point. The question is whether the niche is a foothold into a larger market or a permanent ceiling.
What's the fastest way to test whether you have fit right now?
Stop all outbound activity for two weeks and watch what happens to usage. If a core group of users keeps coming back and engaging without prompting, you have something worth building on. If usage flatlines immediately, the product isn't yet self-sustaining.
How do you distinguish genuine fit from users being polite?
Watch behavior, not words. Users who are being polite will tell you the product is great and then quietly stop using it. Users who have real fit will return unprompted, use the product in contexts you didn't anticipate, and tell others without being asked.
Sources
- Do Things that Don't Scale — Paul Graham
- gstack: skillify/SKILL.md — Garry Tan