Superhuman's PMF Engine: How Rahul Vohra Made Product-Market Fit a Number

The problem the framework solves

“Product-market fit” is the most-cited concept in startup advice and the worst-defined. Most founders know they want it. Few can tell you whether they have it. Almost none can tell you whether they are getting closer to it month over month.

Marc Andreessen’s original 2007 framing was felt, not measured: “you can always feel product-market fit when it’s happening.” That is a great quote and a terrible operating tool. You cannot run a quarterly planning meeting on a feeling.

Rahul Vohra’s contribution, published in First Round Review in 2018, was to convert PMF into a number you can measure, segment, and engineer. Built on Sean Ellis’s “very disappointed” survey question (originally proposed in 2009), the Superhuman PMF Engine is now the canonical operational measure of product-market fit in modern SaaS.

The core question

Send a one-question survey to your active users:

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

1. Very disappointed
2. Somewhat disappointed
3. Not disappointed (it isn’t really useful)

The percentage who answer “very disappointed” is your PMF score. Sean Ellis’s original heuristic: above 40% suggests product-market fit; below suggests you don’t yet have it.

That is a heuristic, not a law. Slack hit ~50%. Superhuman hit ~58% after running the engine. Some perfectly good companies plateau at 35%. Use the number as direction, not destiny.

What “active user” means

The single most common mistake is surveying the wrong people. Survey signups, trialists, or one-time users and your score will be artificially low. Survey only your evangelists and it will be artificially high.

Vohra’s working definition: active = completed the core product action in the last 14-28 days. For an email client, that means actively reading and sending email. For a project tool, actively creating or updating items. Pick the action that defines real usage of YOUR product, then survey only the people who did it recently.

The four segments

After running the survey, you have three groups:

The third group is the trap. Founders see those answers and think “we need to win them over.” Don’t. Vohra is explicit: those users are not your market. Trying to keep them costs you fan-experience focus and lowers the overall score.

Profile your fans

Look at everyone who said “very disappointed” and find the common pattern. For Superhuman it was: tech founders, executives, and managers who live in their inbox. Not “everyone who uses email.” A specific shape of person whose workflow makes Superhuman’s speed obsessions actually pay off.

Ask:

• What industry are they in?
• What role?
• What team size?
• What use case do they share?
• What did they switch from?
• How did they find you?

The answers narrow your ICP from a vague guess to a concrete profile. Now you know who to acquire more of, and who to ignore.

Build the roadmap: half/half

Two follow-up questions in the same survey:

1. To everyone: “What is the main benefit you receive from this product?”
2. To the somewhat-disappointed group: “What would you need to upgrade your answer to very disappointed?”

The first answer tells you what to double down on. (For Superhuman: speed, keyboard shortcuts, AI summaries.) The second tells you what to fix. (For Superhuman, at the time: mobile, calendar, attachments.)

Vohra’s split: roughly 50% of engineering effort on deepening the loved features, 50% on closing the on-the-fence blockers. Re-run the survey every quarter. The PMF score should rise. If it does not, the roadmap is wrong; iterate.

Why “very disappointed” beats NPS

NPS asks would-you-recommend. People can recommend a product they barely use. They cannot honestly say they would be very disappointed to lose a product they do not depend on. The PMF question demands the user has integrated the product into a workflow they would mourn losing. That is a structurally stronger signal than recommendation.

This is also why PMF score is harder to game. You can manufacture an NPS spike with a great support interaction. You cannot manufacture “very disappointed” without actually being load-bearing in the user’s day.

Common mistakes

1. Surveying the wrong audience. Active users only. Not signups, not trialists, not churned users.

2. Treating 40% as binary. It is a directional heuristic. Movement matters more than the absolute number.

3. Optimizing for the not-disappointed group. They are not your customer. Building for them dilutes the experience for your fans.

4. Running the survey once. Quarterly cadence or you cannot tell whether you are getting closer or further from PMF.

5. Confusing PMF score with NPS. Different questions, different signals. Use both, but don’t substitute.

ShipFit and the Superhuman PMF Engine

ShipFit treats the Superhuman PMF Engine as the canonical post-launch PMF measure. Stage 7 (Will They Pay?) defines pre-launch demand proof; once you have 40+ active users, the Engine becomes the loop you run to confirm whether early demand is hardening into PMF or leaking out the back. The ICP definition you locked at Stage 2 is what you segment the survey responses against — refining it for the “very disappointed” cohort is how you sharpen positioning over time.

You don’t get PMF by accident. You get it by measuring, segmenting, and pointing roadmap effort at the things that move the score.

Further reading

• Rahul Vohra, “How Superhuman Built an Engine to Find Product/Market Fit”, First Round Review, 2018. The original article. ~30 min read, fully worked example.
• Sean Ellis, “The startup pyramid”, 2009. Where the “very disappointed” question first appeared.
• Product-Market Fit glossary entry. Short definition + Andreessen’s 2007 framing.
• Lean Startup validation framework. Complementary discipline for validated learning.
• Buyer Persona Canvas. What to do with the high-expectations customer profile once you have it.