How Superhuman Went From 22% to 58% PMF Score — And How to Copy the Engine

Here's the part of the Superhuman story nobody puts on a pitch deck: it started as a failure-in-progress.

By 2017, Rahul Vohra's team had spent years building what might be the most lovingly crafted email client ever made. It was fast. It was gorgeous. Power users wrote poetry about it. And when Vohra ran the standard product-market fit survey on his users, the number came back: 22%.

Twenty-two percent. Just over half of what Sean Ellis says you need to call it fit. By the usual founder mythology — the one where PMF is a lightning-bolt moment you either feel or you don't — this was a clear "no." Pack it up.

Vohra did something far more interesting. He decided that if a single number could measure fit, then fit was a variable. And variables can be moved. So he stopped treating product-market fit as a feeling and started treating it as an engineering problem with a target metric. That decision is the whole story.

The trap that catches almost every founder

When your PMF score comes back low, the instinct is to listen to everyone. All that feedback! Every complaint feels like a clue. So you build a little for the people who want integrations, a little for the people who want a cheaper plan, a little for the ones who want it to work differently — and you slowly sand your product down into something that's fine for everybody and essential to nobody.

That's the 22% death spiral. And it's seductive precisely because it feels productive. You're shipping! You're responsive! You're also diluting the exact thing that made a slice of users love you.

Superhuman's engine is, at its core, a disciplined refusal to listen to the wrong users. Here's how each part works.

Step 1: Segment by disappointment, not by demographics

Vohra started with the same Sean Ellis question every founder should run: "How would you feel if you could no longer use Superhuman?" Very disappointed / Somewhat disappointed / Not disappointed.

But the magic wasn't asking the question. It was what he did with the answers. He split his users into three buckets by their response — and then treated each bucket completely differently:

That last row is the one founders can't bring themselves to do. Ignoring users feels wrong. But the "not disappointed" group is a trap — chasing their feedback is how you build a worse product for the people who actually matter.

Step 2: Let your fans define your ICP

Next, Vohra zoomed in on the "very disappointed" users and asked a different question: who are these people, exactly?

When the team profiled them, a sharp picture emerged. The fans weren't "everyone who does email." They were a specific type — founders, executives, managers, business development people. People who live in their inbox, for whom email is the job, and for whom speed is worth paying for. The folks who were "not disappointed"? Often people who barely used email or treated it as a chore to minimize. No app was ever going to make them passionate.

This is the move most teams skip. Your ideal customer profile isn't a guess you write on a whiteboard at the offsite. It's sitting in your survey data. The "very disappointed" segment is, almost by definition, the group that gets the most value from what you've built. Profile them — role, company stage, use case, what they were using before — and you've found the people you should be building for and selling to.

Step 3: Find out what your fans would miss

Then a follow-up question, asked only of the fans: "What is the main benefit you get from Superhuman?"

The answers clustered around speed. Not the design, not the cleverness — the raw, blistering speed of getting through email. That was the core value, straight from the mouths of the people who'd be devastated to lose it. Which meant the roadmap had a north star: protect and amplify speed above all else.

This is how you stop guessing what to build. You don't brainstorm features in a vacuum. You ask the people who already love you what they love, and you make more of that.

Step 4: Find out what holds the "somewhat" group back

Now the growth lever. Vohra went to the "somewhat disappointed" users — the on-the-fence crowd — and asked what was stopping them from depending on the product. Their answers were a feature list: mobile app, certain integrations, specific workflows.

But — and this is critical — he didn't just build all of it. He filtered that wishlist through the ICP and the core value. A request only made the roadmap if serving it would deepen the product for the kind of user who could become "very disappointed." Requests that would have pulled the product toward the wrong audience got dropped, even when they were popular.

Step 5: The 50/50 roadmap

That quote is the whole engine compressed into one sentence, and it's worth internalizing as a literal split of your effort:

• ~50% — double down. Take what the "very disappointed" love (for Superhuman, speed) and make it even more pronounced. Widen your moat with your fans.
• ~50% — convert. Take the specific blockers from the "somewhat disappointed" (that also fit your ICP) and remove them, one by one, to pull people across the line into fandom.

Half the roadmap defends and deepens the love you've already earned. Half the roadmap manufactures new love. Nothing goes to the people who were never going to care. That balance is what makes the score climb instead of just wobble.

Step 6: Re-measure, and watch the number move

Then the loop closes. Superhuman didn't run the survey once and frame the result. They ran it continuously, watching the percentage of "very disappointed" users climb as the roadmap did its work. Survey, segment, build, re-survey. Again. And again.

The score went from 22% to 58% — past Slack's ~51%, deep into must-have territory. And here's the sequence that matters most: the growth investment came after the score moved, not before. Vohra didn't pour fuel on a 22% fire. He engineered the fire to 58% first, then poured.

The catch: Superhuman built a whole tool to do this

Read the original First Round story and you'll notice something. To run this engine, Superhuman's team built custom internal software — to collect responses, segment users, profile the very-disappointed cohort, and track the score over time. They had the engineers and the runway to do that.

Most founders don't. And so the engine, as inspiring as it is, usually dies at the implementation step. You read the article, you nod, and then you face the reality of stitching it together: a Typeform for the survey, a CSV export, a Google Sheet to calculate the score, a pivot table to segment, manual tagging to find your ICP, and a slide deck to show anyone. By the time you've built that pipeline once, you never want to run it again — which kills the one thing that makes the engine work: re-measuring.

How to run the engine without the engineering team

This is exactly the gap PMFtracker was built to close. The Superhuman engine, productized:

• Survey — the Sean Ellis question, pre-loaded, sent to your engaged users.
• Score — calculated automatically, tracked as a trend line, not a one-time snapshot.
• Segment — users auto-split into very / somewhat / not disappointed.
• ICP — your very-disappointed cohort profiled and the themes in their open-ends surfaced for you.
• Report — an investor-ready view of the trend, the ICP, and what users would miss.

In other words: the thing Superhuman needed a team of engineers to build, ready in five minutes. The point isn't to admire Vohra's discipline from a distance — it's to run the same loop on your own product, this month, and start walking your own 22% toward 40% and beyond.

What to take from this

Strip away the email-client specifics and Superhuman leaves every founder three transferable lessons:

1. A low score is a starting line, not a verdict. 22% became 58%. Your number can move too.
2. Your ICP and your roadmap are already in your survey data — in who's very disappointed and what the somewhat-disappointed are missing.
3. The loop only works if you keep running it. Measure once and you have a snapshot. Measure continuously and you have an engine.

Superhuman didn't get lucky. They got systematic. The good news is that the system is fully copyable — and you don't need their engineering team to copy it.