Skip to main content

Strictly necessary · always on

Remembers this choice in your browser. No tracking.

A Startup Founder's Guide to Achieving Product-Market Fit
Product-Market Fit

The Path to Product-Market Fit: What Actually Counts as Evidence

Marijana Vukovic··Updated ·11 min read

Product-market fit gets measured, not felt. Validate demand with three to five design partners, ship an MVP that tests your riskiest assumption, then measure with the Sean Ellis 40% test and your retention curves. Plus the four levels of PMF, how long it really takes, and how AI-era traction fakes it.

Product-market fit is the point where demand starts pulling harder than you can push. Getting there is not mystical. You validate demand with three to five design partners before you build, ship an MVP that tests your riskiest assumption, measure with the Sean Ellis 40% test alongside your retention curves, and work up the four levels of PMF that First Round Capital mapped from B2B case studies. Two numbers set the stakes. Poor product-market fit is the second most-cited cause of startup death, named in 43% of the failed VC-backed startups with an identifiable cause in CB Insights' March 2026 analysis of 431 shutdowns, behind only running out of capital at 70%, which is usually the symptom rather than the disease. And strong fit means at least 40% of your users would be "very disappointed" if you took the product away.

We have spent more than a decade taking products from an idea to a deployed system that earns money, lately in industries where iteration collides with regulation. Reaching PMF is a discipline, not luck. The four phases we have always run are intact here: validate, build, measure, recognize. What we have added is the part founders actually ask us about. Named frameworks. Real thresholds. And the specific ways 2026 traction lies to you.

What is product-market fit, and how do you know you have it?

Product-market fit exists when your solution meets a validated market problem so well that your main challenge flips from finding users to keeping up with them. Lenny Rachitsky calls the tell a shift "from push to pull": leads start arriving on their own. Todd Jackson of First Round puts it less politely. At strong PMF, "the fish are jumping into the boat."

Nobody gets a single moment of confirmation. The signals show up separately and then converge:

  • Cohort retention curves flatten instead of decaying toward zero. Of everything on this list, that is the signal we trust most.
  • Organic growth accelerates with no paid push behind it.
  • Sales cycles shorten and inbound appears, because prospects start the conversation.
  • Acquisition cost falls while willingness to pay holds.
  • Users complain. Bug reports and feature requests mean people now depend on you.
  • Customers explain your value back to you, pay without discount gymnastics, and ask for more.

Notice what is missing from that list. Press coverage, a funding round, a marquee pilot logo. Those measure how hard you are pushing.

How do first-time founders validate demand before launch?

You validate demand before launch by talking to the people who have the problem and then asking them to commit to something. Interviews first, about the current state: how they work today, what they have already tried, what it costs them. Then sell before you build. A letter of intent, a pre-order, a signed design-partner agreement. Smoke tests help only if they measure willingness to act rather than politeness. Jackson's definition of the first level of PMF is the concrete target: "find three to five customers with a problem worth solving, engage with them, deliver a solution, and validate that solution."

Most startups that die of poor fit built an elegant solution to a problem nobody was funding. The work in this phase is buying evidence before you spend engineering money on it:

  • Define the ideal customer profile precisely. Who feels this most acutely, in what role, at what kind of company?
  • Validate the pain, not the pitch. Listen more than you talk, and be honest with yourself about whether this is a hair-on-fire problem or a mild annoyance.
  • Check that the segment is large enough to sustain a business and that the pain is severe enough to carry a price.
  • Ask for commitment rather than compliments. An LOI or a paid pilot is validation. "I would definitely use that" is a sentence people say to be nice.

When validation comes back negative, pushing harder is the expensive answer. Jackson's prescription is to reassess one or more of the four P's: persona, problem, promise, product, usually in that order, because the cheapest pivots sit earliest in the chain. Demand validation is a product strategy exercise, not a survey exercise.

What should your MVP be designed to prove?

An MVP earns its name by testing your riskiest assumption on real users faster than anything else you could build, not by being a small version of your roadmap. Every scoping decision should follow from that. Identify the minimum feature set that delivers the value proposition you validated, then cut the rest without sentimentality.

Minimal still has to work. A buggy or confusing core workflow generates noise rather than feedback, and you will spend a month arguing about whether users rejected the idea or the implementation. Pick technologies and an architecture that make change cheap rather than ones that anticipate scale you do not have yet. And instrument everything before launch, including the survey plumbing for the measurement below. You cannot iterate on what you cannot see.

How do you measure product-market fit?

Measure PMF with one attitudinal test and one behavioral curve. The attitudinal test is Sean Ellis's question, "How would you feel if you could no longer use this product?", where the benchmark is at least 40% answering "very disappointed." The behavioral curve is cohort retention flattening at a healthy level. Either one alone can be gamed. Together they are hard to fake.

The best-documented application is Superhuman's PMF engine. Rahul Vohra's team started at a 22% PMF score and raised it to 58% in under a year by running the survey continuously, segmenting for the high-expectation customers who answered "very disappointed," working out what that group loved and what was holding the "somewhat disappointed" back, then splitting the roadmap between doubling down and clearing objections and re-tracking the score weekly.

Two caveats we would want a founder to take seriously. The Ellis test is attitudinal and suffers sample bias, because only engaged users answer surveys. And a strong score sitting on top of a decaying retention curve means you found passionate users, not a market. Treat NPS the same way: a supporting, trailing indicator, never the headline metric. A healthy score alongside flat usage tells you people like you, not that they need you.

What turns measurement into progress is the iterate-measure-learn loop from our original playbook, which we have not changed. Engage your target users deliberately. Gather the qualitative why from interviews and session recordings alongside the quantitative what from activation, engagement, and cohort retention. Form hypotheses specific enough to be wrong. Ship the change, then measure that change rather than the aggregate.

What are the stages of product-market fit?

PMF is not binary. It comes in levels, and knowing your level tells you what to work on next. The staging model we use with founders is First Round's four levels of product-market fit, developed by Todd Jackson from B2B case studies:

LevelWhat it looks likeTypical timeline
1 — Nascent3–5 paying customers who are genuinely delighted; you deliver the outcome almost by hand12–18 months
2 — DevelopingA repeatable acquisition channel emerges; unit economics start working (watch new ARR against acquisition cost)the level where most founders stall
3 — StrongInbound demand and referrals compound: "the fish are jumping into the boat"varies
4 — ExtremeYou are the category default; the organization scales past 100 people2–6 years from founding

Almost everyone we meet is stuck at Level 2 on the demand side. The product works for the customers you hand-carried, and the market is not pulling. The prescription is the same four P's from the validation phase. Revisit persona, problem, promise, and product in that order, instead of adding features and hoping one of them changes the slope.

Why do AI startups get false-positive product-market fit?

The 2026 danger is not missing PMF. It is believing a false positive. Enterprise experimentation budgets buy pilots that look exactly like revenue on a chart, and then the budget line moves and the revenue goes with it. Bessemer's PMF playbook for AI founders (July 2025) opens with the operating principle: "PMF isn't binary; it's a spectrum." Bessemer partner Adam Fisher names the mechanism: "The main challenge with PMF in the AI era is that while fast-growing wedge products have become more attainable than ever, they're not nearly as sticky as SaaS products."

The benchmarks that used to settle the argument no longer do. Bessemer notes that 60–70% retention at 90 days "may still just be a sign of early experimentation" for AI products. So the metrics have to get more forensic. Look at retention by use case rather than blended retention, and name which specific jobs keep pulling people back. Watch the second-bite rate, which was Perplexity's earliest PMF metric: the share of users who ran a second query, 80% at the time, with an explicit push toward 100%. And separate durable ARR, meaning revenue from workflows now embedded in operations, from experimental pilot spend. Honestly. In your own board deck.

Miqdad Jaffer, product lead at OpenAI, adds the part that makes this exhausting: "In the AI world, PMF is a moving target. Your users' definition of 'intelligent enough' changes every month". He calls treating PMF like a checkbox the biggest mistake he sees AI founders make. Fit you measured last quarter is a hypothesis this quarter.

How long does product-market fit take?

Longer than the highlight reels suggest. Lenny Rachitsky's research across B2B startups puts the average at roughly two years before founders start feeling PMF, and Jackson's case studies put extreme PMF at two to six years from founding. Fisher goes further: "I don't like to declare victory on PMF until a very late stage." Plan the runway and the cadence around those numbers. A founder who expects fit in six months makes desperate decisions in month seven, and we have watched several of them do it.

What changes when your market is regulated?

Every framework above quietly assumes you can ship whenever you learn something. In healthcare, automotive, and the rest of the regulated world, iteration runs inside design controls and change management, IEC 62304 for medical device software and ISO 26262 for automotive, and you cannot A/B-ship your way to fit when every change needs verification evidence behind it. The path does not change. The cost of each loop does, which changes how you run it.

Front-load validation harder than you would elsewhere: design-partner agreements and simulated-use evidence do the work smoke tests do in consumer software, and you want those commitments before the regulatory clock starts. Draw the MVP boundary by risk classification rather than by core value alone, because the scope of your first shippable product is partly a regulatory decision and getting it wrong costs quarters. Then automate evidence generation, or accept that documentation becomes the throttle on your learning speed. That last one is the mistake we get called in to unwind most often.

This is the lane we work in. On Canary Medical's smart-knee implant platform, the measurement loop had to run inside medical-device constraints, and the V&V documentation and automation we delivered became part of the evidence package that contributed to the client's own De Novo authorization. Finding fit under constraint is a repeatable discipline, and it is the core of how we structure engagements across healthcare and automotive.

What to take from this

  1. Buy evidence before you build: problem interviews plus real commitments from three to five target customers.
  2. Scope the MVP around your riskiest assumption, and instrument it before launch.
  3. Measure with the pair, the Sean Ellis 40% test and a flattening retention curve, and segment your high-expectation customers the way Superhuman did.
  4. Find your level. If you are stuck at Level 2, the answer is persona, problem, promise, product, in that order.
  5. Audit AI-era traction skeptically. Retention by use case, second-bite rate, durable ARR versus pilot spend.
  6. Budget roughly two years to feel PMF in B2B, and do not declare victory early.
  7. In regulated markets, build evidence generation into the product loop before compliance becomes your speed limit.

Working toward product-market fit where regulation raises the cost of every iteration? That first validated, evidence-backed deployment is what our First Build exists to de-risk. Scope a First Build: about a month, your riskiest assumption tested on real users, and a verdict on whether the demand signal you already have will survive a second cohort.

Tagged:Product-Market FitStartupsProduct StrategyMVPValidation
Your Turn

Facing a problem like this one?

Tell us what the business needs to win and who owns it on your side. If it is a fit, a First Build starts within weeks. If it is not, we will tell you why.