Segment-level PMF: why your average score hides your ICP

Say your last PMF survey came back at 34%. Below the 40% line, but not catastrophic, so what do you do? Most founders read that single number as a verdict on the whole product. It usually isn't. A blended score is a weighted average of everyone who answered, and averages are exactly the statistic that hides the thing you most need to see: one segment that already has product-market fit, sitting inside a number that looks like it doesn't.

How a real ICP hides inside a mediocre average

Take a concrete version of the 34% example. Say 100 people answered your survey: 20 of them fit your intended ICP tightly, and 80 don't, they came from a broad launch, a random newsletter mention, or a feature that happens to attract the wrong crowd. If the 20 ICP-fit users score 70% very disappointed, and the 80 poor-fit users score 25%, the blended math is (20 × 0.70 + 80 × 0.25) / 100, which lands at 34%.

That 34% is mathematically correct and strategically misleading. It reads as "not there yet, keep iterating broadly." The segment cut reads as "you have strong product-market fit with a narrow group, and 80% of your respondents are the wrong audience." Those two readings lead to opposite roadmaps: one says build more features for everyone, the other says stop marketing to everyone and go all-in on the 20%.

The segment cuts worth running

You don't need a different survey to do this, you need to tag respondents with attributes before you look at the "very disappointed" split. The most useful cuts, roughly in order of how often they surface something:

• Plan tier or price point. Paid users almost always score differently than free-tier or trial users. If your free score is dragging the average down, that's not a PMF problem, it's a targeting or upgrade-path problem.
• Company size or persona (B2B). A 25-person startup and a 2,000-person enterprise often want completely different things from the same product. See B2B vs B2C PMF for why the sample math differs here too.
• Primary use case. If your product does two or three distinct jobs, users doing "job A" and users doing "job B" can score very differently. This is where Jobs to Be Done and the survey overlap directly.
• Acquisition channel. Users who found you through a specific channel (a targeted launch, a partner, organic search for a specific problem) are frequently better-fit than users from a broad channel like a general newsletter mention or a paid campaign with loose targeting.
• Tenure. New users and users who have been active for 3+ months can score very differently, since the value of some products compounds with data or habit.

Don't cut a segment you can't trust

The same sample-size discipline that applies to your overall score applies per segment, and it's easy to forget once you start slicing. Roughly 40 valid responses give a directional read; 100 or more is what you want before you call a number investor-grade. If your ICP segment only has 12 respondents, a 70% score there is a genuinely interesting hypothesis, not a claim to build a strategy on yet. Keep surveying, and specifically make sure your next batch of responses includes more of that segment, don't just wait for volume to accumulate randomly.

This is also why segment cuts work best as a lens on the very-disappointed / somewhat-disappointed / not-disappointed split you already have, rather than a wholly separate analysis. You're asking the same three-way question, just per group instead of blended across everyone.

What to do once you find the strong segment

Finding a segment scoring well above 40% isn't the end of the analysis, it's the start of a much sharper one:

1. Read their open-ended answers specifically. "What type of person would most benefit" and "what's the main benefit you get," filtered to just this segment, usually hands you positioning language you couldn't have written yourself.
2. Aim acquisition at people who match their profile. If the strong segment is 20-50 person B2B SaaS companies doing a specific job, stop spending equally on everyone and go find more of exactly that.
3. Consider narrowing your beachhead. A segment already at 70% is a strong candidate for the single market you nail before you expand to the rest.

The founders who get stuck here usually keep optimizing for the blended average, chasing incremental gains for the whole mixed population instead of doubling down on the group that already said yes. The math rewards the second approach almost every time: it's far easier to get from 70% to 80% in a segment that already loves you than to drag a 34% blended average up to 40% by trying to please everyone at once.