How Many Responses Do You Need for a Reliable PMF Survey?

Almost every founder who's about to run their first Sean Ellis survey hesitates at the same spot: "I only have a few dozen active users — is that even enough to mean anything?" It's a fair worry. A PMF score you can't trust is worse than no score, because you'll make real decisions on it.

So let's settle it with actual numbers — both the practical thresholds you can act on and the statistics underneath them — and then the part most guides skip: why the size of your sample matters less than who's in it.

The short answer

For the Sean Ellis test specifically, here are the thresholds that have held up across hundreds of startups:

"Valid responses" means responses from engaged users, excluding the "N/A — I no longer use it" answers that don't count toward your score. Forty real responses beat two hundred half-remembered ones, every time.

Why the number is smaller than you fear

People assume PMF surveys need the thousand-plus samples you see in political polling. They don't — and the reason is reassuring once you see it.

A PMF score is just a proportion: the share of users who say "very disappointed." The uncertainty around a proportion shrinks with the square root of your sample size, which means it drops fast at the start and then flattens. Going from 25 to 100 responses buys you an enormous amount of confidence. Going from 1,000 to 2,000 barely moves the needle. You hit "good enough to act" much sooner than intuition suggests.

The actual margin of error

If you want the real statistics: for a proportion, the 95% confidence interval is roughly ±1.96 × √( p(1−p) / n ), where p is your score and n is your number of valid responses. Near the 40% threshold — the decision that matters most — that works out to:

Read that table honestly and two things jump out. First, at 40 responses your score is genuinely fuzzy — which is exactly why we call it "directional." A 40% reading on 40 people is encouraging, not proof. Second, you don't need to chase ±1 point. For a founder deciding whether to keep iterating or start scaling, knowing you're "comfortably above 40" or "clearly below it" is what matters, and ~100 responses gets you there.

Quality beats quantity — by a lot

Here's the part the statistics can't capture: a clean sample of 50 engaged users is worth more than a noisy sample of 500. Sample-size formulas assume every response is a fair draw from the population you care about. The moment you survey the wrong people, the math is measuring the wrong thing — precisely.

Two ways founders quietly poison the well:

• Surveying everyone. Blast your whole email list and you fill the sample with people who barely used the product. They answer "not disappointed" out of indifference and drag your score down for no real reason.
• Surveying only superfans. Ask just your hand-picked beta users and you get a flattering number that vanishes when normal users show up.

Both produce a precise, confident, wrong answer. The fix isn't more responses — it's the right responses. Survey engaged users only (used the core product at least twice recently), and 40–100 of them will tell you more than a thousand random signups ever could. Here's exactly how to pick and reach them.

When you genuinely need more responses

The thresholds above are the general case. Push for a larger sample when:

1. You're right on the 40% line. At 38–42% with a ±15 margin, you truly can't tell which side you're on. Get to 100+ before you make a scaling decision on it.
2. You're about to make an expensive bet. If a fundraise or a big hire hangs on the number, the extra responses are cheap insurance.
3. You want to segment. Cutting the score by plan, persona, or acquisition channel splits your sample into smaller buckets — each segment needs its own ~40 to be readable.

You don't need them all at once

The trap is treating sample size as a wall you must clear before you learn anything. You don't. Run the survey now with the engaged users you have, take the directional read, and keep it running — responses accumulate, and your score sharpens from "rumor" to "reliable" over a few weeks without any extra effort.

That's also the difference between a one-off survey and a tracked PMF score: the tracked version is always gathering its next response, so your sample is forever growing and your trend line gets more trustworthy with every passing week. A rough 33% you measured today beats a perfect score you're still waiting to collect.