← Writing

Product-Market Fit: How to Know You Have It

Published
Reading
17 min
Contents · 14 sections
  1. What it feels like, and what it doesn’t
  2. Measure 1: The “very disappointed” survey
  3. Measure 2: Segment the survey (the Superhuman method)
  4. Measure 3: Retention curves that flatten
  5. Measure 4: Organic growth and word of mouth
  6. Measure 5: Selling gets easier
  7. False positives: signals that look like fit but aren’t
  8. Fit for small and bootstrapped businesses
  9. A worked example: Bookwell
  10. Reading it
  11. Before fit and after fit
  12. Before fit
  13. After fit
  14. Key ideas

On June 25, 2007, Marc Andreessen, the co-founder of Netscape, published the fourth part of his guide to startups. He gave it a blunt title: The only thing that matters.

The essay asks which of three things matters most to a startup: the team, the product, or the market. His answer is the market. A great market, he argues, pulls a product out of a startup, even a messy one. A dead market kills a startup, even a brilliant one.

Then he gives the idea a name, crediting the investor Andy Rachleff for the thinking behind it. Product-market fit means being in a good market with a product that can satisfy that market. And his rule for founders is short: getting there is the only thing that matters.

The definition is easy. The hard part is the question every founder asks at some point: “Do we have it yet?” This post covers what fit feels like, five ways to measure it, the signals that fool people, what fit means for a small business, and a worked survey example on 60 fictional users.


What it feels like, and what it doesn’t

Andreessen says you can always feel product-market fit when it’s happening. You can also feel it when it isn’t. His description of both is worth paraphrasing, because it’s more useful than most formal definitions.

When it isn’t happening: customers don’t quite get value from the product. Word of mouth doesn’t spread. Usage grows slowly. Press reviews are lukewarm. Sales take too long and many deals never close.

When it is happening: customers buy as fast as you can deliver. Usage grows as fast as you can add servers. Money from customers piles up in the bank. You hire support and sales people as fast as you can. Reporters call because they’ve heard about you. In his words, even “investment bankers are staking out your house.”

Nothing in either list is about how good the code is or how hard the team works. Fit is something the market does to you, not something you do. You can’t push your way into it. You can only build until the market starts pulling.

The “not happening” list is the default. If you’re unsure whether you have fit, that’s usually a sign you don’t. Andreessen splits a startup’s life into two parts, before fit and after fit, and his advice for the “before” part is to do whatever it takes: change the product, change the people, change the market.

But “you’ll feel it” is not a measurement. Feelings lie, especially to founders who want to believe. So let’s look at things you can count.


Measure 1: The “very disappointed” survey

The most famous test comes from Sean Ellis, who ran early marketing at Dropbox and later coined the term “growth hacking.” In a 2009 interview with Venture Hacks, he described a simple survey he used to judge whether a company was ready to grow.

The key question is:

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

Users pick one answer:

  • Very disappointed
  • Somewhat disappointed
  • Not disappointed

You count the share who say very disappointed. Ellis’s rule of thumb: if 40% or more say “very disappointed,” you probably have product-market fit. Below 40%, you probably don’t yet.

Where does 40% come from? It’s not a law of nature. Ellis compared survey results across many startups (Rahul Vohra later wrote that it was nearly a hundred) and noticed a pattern. Companies that struggled to grow almost always scored below 40%. Companies with strong traction almost always scored above it. So treat 40% as a benchmark from experience, not a precise threshold. 38% isn’t doom, and 42% isn’t safety.

Why ask about disappointment instead of satisfaction? Because satisfaction is cheap. People say they like lots of things they’d happily stop using. “Very disappointed” asks something harder: would losing this hurt? That’s much closer to whether people will keep coming back and paying.

A few practical rules make the survey trustworthy:

  • Ask the right people. Survey users who have actually experienced the core product recently. Vohra’s version used people who had used it at least twice in the last two weeks.
  • Get enough answers. Vohra says results become directionally useful at around 40 responses. Below that, one or two answers swing the percentage a lot.
  • Ask each person once. Re-surveying your happiest users inflates the score.
  • Add the follow-up questions. The single number is less useful than the answers to “who is this for?”, “what’s the main benefit?” and “what should we improve?” More on those next.

Measure 2: Segment the survey (the Superhuman method)

The survey got a second life in 2018, when Rahul Vohra, the founder and CEO of the email app Superhuman, wrote How Superhuman Built an Engine to Find Product/Market Fit for First Round Review.

His story starts with a problem. Superhuman had been building for a long time, and he couldn’t say with any confidence whether the product had fit. So in the summer of 2017 he ran the Ellis survey. The result was 22%. Well short of 40%.

What makes the essay worth reading is what he did next. He didn’t treat 22% as one number. He treated it as an average hiding several different groups of users. His survey had four questions:

  1. How would you feel if you could no longer use Superhuman?
  2. What type of people do you think would most benefit from Superhuman?
  3. What is the main benefit you receive from Superhuman?
  4. How can we improve Superhuman for you?

Then he worked through the answers in four steps.

Step 1: Segment. He grouped respondents by their answer to question 1 and gave each one a persona, a short description of their role. The “very disappointed” group was full of founders, managers, executives and business development people. So he focused on those personas and set the others aside for now. Looking only at those segments, the score rose from 22% to 33%. Nothing about the product had changed. He had simply stopped averaging in people the product wasn’t built for.

Step 2: Describe the high-expectation customer. Using an idea from the marketer Julie Supan, he wrote a detailed profile of the most demanding, most discerning user in the group that loved the product. Superhuman even gave her a name, “Nicole”: a busy professional who lives in email and wants to get through it fast.

Step 3: Double down on what users love. Question 3 told him what the “very disappointed” users valued most. The top answer was speed. So half of the roadmap went to more of that: more speed, more keyboard shortcuts, more automation, more design polish.

Step 4: Fix what holds the others back. He looked at the “somewhat disappointed” users and split them by their main benefit. If speed was their main benefit, they were almost-fans: people who wanted the same thing as the core group but hit something that stopped them. Their answers to question 4 became the other half of the roadmap: a mobile app, integrations, better attachments, calendaring, better search and more. The “somewhat disappointed” users who didn’t care about speed, and the “not disappointed” users, were politely ignored. They wanted a different product.

Then he measured again. And again. Superhuman tracked the score weekly, monthly and quarterly, surveying new users as they arrived. According to the essay, within three quarters the score had reached 58%.

The lesson I take from this is that the average score is a blur. Somewhere inside it there’s usually a group that loves the product, a group that almost does, and a group that never will. Fit starts with the first group.


Measure 3: Retention curves that flatten

Surveys measure what people say. Retention measures what they do.

Retention means the share of users (or customers) who are still active some time after they start. If 100 people sign up in March and 30 of them are still using the product in June, three-month retention for the March group is 30%. That March group is called a cohort: a set of users who started in the same period.

Every product loses people at first. The question is what happens after that.

  • No fit: the curve keeps sliding toward zero. Month after month, the cohort shrinks until almost nobody is left.
  • Fit: the curve drops, then flattens. Some share of the cohort stays, month after month.

A flat curve at 20% can be a real business. A curve that’s still falling at 50% is a warning, because it’s on its way to zero. The level matters less than whether it levels off.

How to build and read the cohort table is the whole of Cohort Retention: The Chart That Tells the Truth. The short version: if your cohort curves never flatten, no survey score should convince you that you have fit.


Measure 4: Organic growth and word of mouth

Andreessen’s “not happening” list includes word of mouth that doesn’t spread. That’s a good test on its own.

When a product fits, people tell other people about it without being asked. You start seeing signups from sources you didn’t create: “a friend told me,” “my accountant uses it,” “someone in a Facebook group recommended you.” Organic growth is growth that comes without paid ads or your own outreach.

How to watch it:

  • Ask every new customer how they heard about you. One question at signup, free-text, not a dropdown. Count the answers each month.
  • Track the share of new customers from referrals and direct search (people typing your name into Google). If that share grows over time, people are talking.
  • Watch for unprompted requests. Customers asking if they can add a colleague, a second location, a sister company.

Organic growth takes months to show up, but it’s hard to fake.


Measure 5: Selling gets easier

The last signal is about friction. Andreessen’s description of no fit includes long sales cycles and deals that never close. With fit, the opposite starts happening.

You can see it in your own numbers:

  • The sales cycle shortens. Fewer meetings from first contact to paid.
  • Win rate rises. A higher share of demos or trials turns into customers.
  • Fewer objections, and the same ones. Instead of every prospect raising a different concern, you hear the same one or two, which you can fix.
  • Price pushback drops. People argue less about price when the product clearly saves them something.

If you’ve been doing founder-led sales, as in Your First 10 Customers, you’ll feel this before you can measure it. The pitch gets shorter.

None of these five measures is enough on its own. Fit is when several of them agree. A strong survey score, cohort curves that flatten, a growing share of word-of-mouth customers and an easier sales process together tell a story you can trust.


False positives: signals that look like fit but aren’t

Founders want to believe, so it’s worth naming the signals that most often fool them.

Looks like fit Why it can mislead What to check instead
Early adopter excitement Some people love trying anything new. They sign up, praise it, then move on to the next thing. Do they still use it three months later? Would mainstream users in the same segment?
Growth from paid ads Ads can buy signups for a product nobody keeps. Turn the ads off and growth stops. Retention of paid cohorts, and the share of organic signups.
Friends and family They’re kind. They’ll use it to support you, not because they need it. Remove them from the survey and the retention numbers.
One big customer One large contract can look like a business. It may only be one buyer’s taste, and losing them ends everything. How many customers like them do you have? Could you sell the same thing to ten more?
Press and launch spikes A popular article or a launch day brings a crowd that leaves just as fast. Retention of the spike cohort, a month later.

The common thread: every false positive is about getting people in, not keeping them. Fit shows up in what happens after people arrive.

Investor interest is a trap too. Raising money doesn’t mean the market wants your product. It means investors think it might.


Fit for small and bootstrapped businesses

Andreessen wrote with venture-backed startups in mind. If you’re building a small software business or a niche tool, the picture still applies, at a different scale. My view is that fit for a small business has the same core, people who’d be very disappointed to lose you and keep paying, but it can live in a much narrower segment.

A few things change:

  • The segment can be tiny and still be enough. A tool that 60% of independent dental clinics in one country would hate to lose may be a great business, even if nobody else cares. You don’t need the whole market. You need one group that pulls.
  • Profit matters as much as growth. For a bootstrapped company, fit is only useful if each customer earns more than they cost. That’s the subject of Unit Economics in Plain English.
  • Smaller samples. You may never have 40 active users to survey. Lean more on retention and conversations, and treat the survey score as a rough guide.

A common mistake is trying to fit everyone. A booking tool for “all service businesses” is fighting on many fronts at once. A booking tool for salons with three to eight chairs can win one front. Which brings us to the worked example.


A worked example: Bookwell

Bookwell is a fictional company I made up for this post. It’s a small appointment-booking app for local service businesses. After a year it has a few hundred users, and the founders can’t tell if they have fit. So they run the Ellis survey on 60 active users (each used Bookwell at least twice in the last two weeks), and they also record which kind of business each one runs.

Here are the raw answers:

Segment Very disappointed Somewhat Not Total
Hair and nail salons 13 6 3 22
Physiotherapy clinics 3 6 5 14
Private tutors 2 5 8 15
Personal trainers 3 4 2 9
All users 21 21 18 60

The founders compute the percentages in Python:

data = {
    "Hair and nail salons": (13, 6, 3),
    "Physio clinics":       (3, 6, 5),
    "Private tutors":       (2, 5, 8),
    "Personal trainers":    (3, 4, 2),
}

totals = [0, 0, 0]
for segment, (very, somewhat, not_d) in data.items():
    n = very + somewhat + not_d
    totals = [a + b for a, b in zip(totals, (very, somewhat, not_d))]
    print(f"{segment}: {n} answers, {100 * very / n:.1f}% very disappointed")

n_all = sum(totals)
print(f"All users: {n_all} answers, {100 * totals[0] / n_all:.1f}% very disappointed")

Output:

Hair and nail salons: 22 answers, 59.1% very disappointed
Physio clinics: 14 answers, 21.4% very disappointed
Private tutors: 15 answers, 13.3% very disappointed
Personal trainers: 9 answers, 33.3% very disappointed
All users: 60 answers, 35.0% very disappointed
Share "very disappointed", by segment (fictional Bookwell survey) 40% benchmark All users (60) 35.0% Salons (22) 59.1% Physio clinics (14) 21.4% Tutors (15) 13.3% Trainers (9) 33.3% Bar length = percent of answers in that segment; value shown on the right. Only salons clear 40%.
The overall 35% hides one segment well above the line and three below it.

Reading it

The blended score says “not yet.” 35% is under 40%. Stopping here, the founders might start adding features for everyone.

The segments say something different. Salons score 59.1%, far above the benchmark. Everyone else is below it. Take salons out and the other 38 users score just 21.1% (8 of 38). The product isn’t mediocre for everyone. It’s loved by one group and lukewarm for the rest. The average was mixing them.

Go one level deeper. The founders also asked salons how many chairs they have. Of the 14 salons with three or more chairs, 10 said “very disappointed” (71.4%). Of the 8 solo salons, 3 did (37.5%). So the strongest pull is from small teams, where juggling several staff calendars is the real pain. That gives them their high-expectation customer: a salon owner with three to eight stylists who loses money every time two clients get booked into the same chair.

Be honest about sample size. 22 salon answers is below the 40 that Vohra suggests, and 14 and 8 are smaller still. A couple of answers either way would move these percentages by several points. So this is a strong hint, not proof. The next step is to survey more salon users, and to check that salon cohorts actually retain better than the others. If both agree, the hint becomes a decision.

What they do next, following the Superhuman playbook:

  1. Focus marketing and sales on salons with three or more chairs, for now.
  2. Ask the “very disappointed” salon users what they value most (question 3) and build more of it.
  3. Read the “somewhat disappointed” salon answers (6 people) to find what holds them back, and fix those blockers.
  4. Stop building features that only tutors or trainers asked for. Not forever. Just until the core segment is solid.
  5. Re-run the survey on new salon users every month.

This is the loop: measure, segment, double down, fix blockers, re-measure. It turns a vague “do we have fit?” into a short list of things to do on Monday.


Before fit and after fit

Andreessen’s split of a startup’s life into “before” and “after” is the most practical part of his essay, because the right behaviour is almost opposite on each side.

Before fit

  • Keep costs low. You’re still searching. Every extra hire or big tool shortens the time you have to find the answer. Default Alive or Default Dead? is the runway check that keeps you honest.
  • Change things fast. The product, the segment, the pricing, even the market. Andreessen’s advice is to do whatever is needed.
  • Don’t scale marketing. Pouring money into ads before people stay is filling a leaky bucket. You pay to acquire people who leave.
  • Narrow, don’t widen. When in doubt, serve fewer people better.

After fit

  • Now spend on growth. The bucket holds water, so filling it makes sense. This is where marketing, sales hires and paid channels start to pay off.
  • Make things reliable. More users find more bugs. Support and stability matter more every month.
  • Watch the edges. Fit is not permanent. Markets shift, competitors arrive, and new segments behave differently from your first one. Keep running the survey and keep watching cohorts.
  • Expand carefully. When you move into a new segment, you’re partly back to “before fit” in that segment. Measure it separately.

The biggest error is acting like you’re after fit when you’re still before it. Hiring a sales team, signing long contracts for tools and running big ad campaigns all make sense only once people stay. Scaling before fit doesn’t create fit. It just makes the lack of it more expensive.


Key ideas

  • Product-market fit means being in a good market with a product that satisfies it. Marc Andreessen’s 2007 essay calls reaching it the only thing that matters.
  • The Sean Ellis survey asks how users would feel if they could no longer use the product. 40% or more “very disappointed” is a widely used benchmark, drawn from experience, not a law.
  • Segment the answers. As Superhuman showed, an average score can hide a group that loves you. Find them, build for them, fix what holds back the almost-fans, and re-measure.
  • Back the survey with behaviour. Cohort retention curves that flatten, growing word of mouth and easier sales are harder to fake than any opinion.
  • Beware false positives: early-adopter buzz, paid growth, friends and family, one big customer, press spikes and free users all bring people in without proving they’ll stay.
  • Small businesses can have real fit in a narrow segment. One group that pulls hard, with unit economics that work, is enough.
  • Before fit, stay cheap and change fast. After fit, scale. Doing the second before the first only makes failure more expensive.

Sources: Marc Andreessen, Part 4: The only thing that matters (June 25, 2007); Venture Hacks, Sean Ellis interview (December 14, 2009); Rahul Vohra, How Superhuman Built an Engine to Find Product/Market Fit (First Round Review, November 2018).

Related: Cohort Retention: The Chart That Tells the Truth, Validate an Idea in One Week, and Unit Economics in Plain English.