Measuring product-market fit

Andreessen's definition is a lagging description — by the time bankers are interested, you already have fit. You also need leading signals: retention curves, and a survey that asks how upset people would be to lose the product.

Feelings lie. Demand that outruns you is real, but you still need numbers so you do not declare victory on a good week.

The big idea

Retention is what people do. A survey is what they say. Lock-in can look like retention without love. Courtesy can look like love without retention. If the two disagree, segment before you celebrate.

Walking through it

Watch cohort curves first: they should flatten above a usable floor after the initial drop. Rising organic or referral share, deeper usage among actives, and a shorter sales cycle at similar price help confirm. Later, net revenue retention matters. Caveats: the floor depends on category, founder-led sales inflate win rate, and vanity DAU without retention misleads.

The Sean Ellis test

Sean Ellis found a leading survey proxy after working growth at Dropbox, LogMeIn, and Eventbrite:

How would you feel if you could no longer use this product?

• Very disappointed
• Somewhat disappointed
• Not disappointed

Across many startups, companies where ≥40% of qualifying respondents said "very disappointed" tended to grow; those below struggled. Slack early on reportedly scored around 51% in one oft-cited sample.

Survey hygiene:

• Poll users who experienced the core product recently (Ellis often suggested at least twice in the last two weeks — adapt to your cadence)
• Do not poll one-time tire-kickers
• Around ~40 respondents can be directionally useful; more is better
• Segment: your best-fit customers may love you while the wrong segment dilutes the score

Superhuman's engine

Rahul Vohra turned Ellis's question into a continuous loop:

1. Measure the "% very disappointed"
2. Identify the high-expectation customer (who loves you and why)
3. Double down on what fans love
4. Fix the blockers that hold somewhat disappointed users back
5. Resurvey; treat the score as a product KPI

Superhuman's public story: the score started around 22%, rose after segmenting to the right customer, and climbed to roughly 58% over later quarters. The method is disciplined listening with a numeric north star.

Common mistakes

• Polling tire-kickers. One-time visitors dilute the score toward "not disappointed."
• Skipping segments. A blended 35% can hide a 48% beachhead.
• Trusting the survey alone. Retention without love can be lock-in. Love without retention can be courtesy.
• Treating 40% as a trophy. It is a threshold to keep retesting after pricing and ICP changes.

Check yourself

Your Ellis score is 35% blended and 48% in one ICP. What do you do?

Narrow to that ICP, fix the blockers those fans named, and stop optimizing for the segment at 12%.

Why is retention the tie-breaker when the survey looks great?

Surveys are stated preference. Retention is whether people actually come back.

Key takeaways

• Retention cohorts are behavioral truth. Ellis's "very disappointed" question is a leading proxy.
• The usual bar is ≥40% very disappointed among people who used the core product.
• Superhuman's loop: measure, find fans, double down, fix fence-sitter blockers, resurvey.
• Always segment. A blended miss can hide a beachhead.

Further learning

How Superhuman Built an Engine to Find Product/Market Fit (First Round Review)Superhuman founder and CEO Rahul Vohra walks us through the framework his startup used to make product/market fit more actionable, detailing the survey and four-step process that were key to measuring and optimizing it.First Round How Superhuman built an engine to find product-market fitFinding product-market fit is an obsession for every startup. Superhuman CEO Rahul Vohra built an engine to find it — here's how...Superhuman Blog

Next: false positives.