Product-Market Fit: How to Measure It Honestly (Ellis, Vohra, Rachleff)

Most product-market fit content credits Marc Andreessen. That’s a mistake of historiography. Andy Rachleff coined the concept at Benchmark, building on Don Valentine’s Sequoia thesis that “a great team without a great market will fail; a great market with a merely adequate team will often win.” Andreessen gave us the phrase most people quote. Rachleff gave us the idea.

The distinction matters because Rachleff’s framing is structural — market first, product second — where Andreessen’s definition (“being in a good market with a product that can satisfy that market”) is easier to confuse with “our customers say they like us.” A lot of founders and boards confuse the two, and the consequences of that confusion cost real money. I’ve seen it up close in private-equity due diligence: a CEO claims PMF, the retention curve says otherwise, and the value creation plan is now built on sand.

In 2026, this matters more than it did five years ago. AI has collapsed the cost of building a working prototype of almost anything. You can now ship ten wrong products in the time it used to take to ship one. That’s either a superpower or an expensive way to burn capital, depending entirely on whether you can measure PMF honestly and stop building the ones that don’t have it.

Product-market fit is the state in which a specific product serves a specific market well enough that users pull the product from the team rather than the team pushing it into the market. It is measured through a combination of the Sean Ellis 40% test (≥40% of qualified users would be “very disappointed” without the product), cohort retention curves that flatten into a “smile”, and organic growth that is disproportionate to marketing spend. PMF is per-product, not per-company, and it decays over time as markets and channels shift.

Sign up (free) to allocate dedicated teams to hunt product-market fit

What Product-Market Fit Actually Is

Rachleff’s original framing was structural. You pick a market with real need, then build a product that serves it better than the alternatives. The market does the heavy lifting — the right market pulls a merely adequate product to success, and the wrong market buries a technically superb one.

Andreessen made the concept famous in his 2007 “The Only Thing That Matters” essay. His definition is worth re-reading carefully: PMF is “being in a good market with a product that can satisfy that market.” Not “a product that customers like.” Not “a product that has users.” The structure — market — is doing at least half the work in that sentence.

Marty Cagan’s four product risks — value, usability, feasibility, viability — map cleanly onto what PMF actually tests. PMF is principally a value risk test at scale: will enough of the right people find this valuable enough to change their behaviour? Usability is necessary but not sufficient. Feasibility is presumed by the time you’re measuring. Viability — can you sustain the economics — is the part most start-ups discover they don’t have even when user love is real.

The most useful modern extension is Brian Balfour’s Four Fits framework — Market-Product, Product-Channel, Channel-Model, Model-Market. Balfour’s point is that PMF is necessary but not sufficient; a product that has it can still fail if the channel doesn’t fit the model or the model doesn’t fit the market. This is the framing PE diligence actually uses. A business with PMF but no channel that scales economically is a hobby, not an investment.

How to Measure Product-Market Fit

There are three measurement tools that belong in every PM’s toolkit. Use at least two of them; don’t trust any one in isolation.

1. The Sean Ellis 40% Test

Sean Ellis, first marketer at Dropbox, LogMeIn, Eventbrite and several other unicorns, noticed a pattern: the companies that broke out all had ≥40% of qualified users answer “very disappointed” to the question “How would you feel if you could no longer use [product]?” — with “somewhat disappointed” and “not disappointed” as the other options.

The survey is only useful if you qualify respondents properly. Ellis’s rules:

• Respondents must have used the product at least twice
• Within the past two weeks
• And experienced the core value (not just signed up)

If you survey everyone who ever visited your homepage you’ll get noise. If you survey recent active users, you get a usable signal. A score ≥40% suggests PMF is plausible. Below 25% it almost certainly isn’t there. Between 25% and 40% you have the option of running Rahul Vohra’s engine to pull the score up.

2. The Rahul Vohra / Superhuman PMF Engine

Vohra’s 2018 First Round Review piece is the most quoted case study on moving a PMF score deliberately. Superhuman took their Ellis score from 22% to 58% in three quarters by:

1. Segmenting the “very disappointed” cohort to identify the High-Expectation Customer (HXC) — who specifically loves the product and why.
2. Analysing the “somewhat disappointed” cohort — what friction is stopping them from being very disappointed?
3. Running a 50/50 roadmap — half the capacity on doubling down for the HXC (the things that made them fall in love), half on removing friction for the “somewhat disappointed” (the reasons they’re not there yet).
4. Re-surveying weekly so the score becomes a live metric, not a one-off snapshot.

The engine works beautifully for narrow, founder-identifiable HXC SaaS like Superhuman. It works less well for platforms, marketplaces, or dev tools with diffuse buyer personas. When the engine fits, it is the sharpest PMF-hunting tool available. When it doesn’t, reach for retention curves instead.

3. Retention Curves — The “Smile”

The most honest PMF metric is the cohort retention curve. Plot the percentage of each monthly sign-up cohort still active 1, 2, 3… 12 months later. A product with PMF produces one of two shapes:

• A curve that flattens (users retained at a stable non-zero rate after an initial drop-off)
• A curve that turns up into a smile (users returning after initial churn as they come back for new reasons)

A curve that decays monotonically toward zero is the shape of a product without PMF. A founder who points at the first three months of a cohort and calls it PMF is describing slope, not state.

Pair the retention curve with NRR (Net Revenue Retention) for B2B products. NRR <100% over multiple cohorts is the PE signal that there’s a leak at the bottom of the bucket that no amount of top-of-funnel will fix.

The 2026 Reframe: AI Made PMF Measurement More Important, Not Less

Here is the asymmetry that defines early-stage product strategy now:

The cost of building a product has collapsed to near zero. The cost of proving enough people want it has not changed at all.

This changes the entire early-stage economics conversation. It used to take six months and half a million pounds to build a credible prototype. The cost alone forced you to do some diligence on the problem first. Now you can have something in front of users in a week. The old forcing function is gone.

The paradox: because building is so cheap, a team can churn out ten plausible products in parallel and still miss PMF on all of them — because none of them had the distribution, the Channel-Model fit, or the real customer pull in the first place. Ten no-PMF products is worse than one, not better. You have ten maintenance burdens, ten sales cycles, ten support queues, and no breakout.

What the AI era demands is more disciplined PMF measurement, applied earlier. Don’t wait until you’ve built a full product to measure PMF. Start measuring problem-solution fit (see the upcoming article on problem-solution fit ) before writing a line of code. Use riskiest assumption tests to kill ideas cheaply, before the cheap build cost tempts you into accidentally shipping ten things. Use the Mom Test to extract honest signal from customer conversations. The frameworks that were nice-to-have in 2019 are survival gear in 2026.

This also sharpens the connection with the product lifecycle : AI has effectively shortened the Introduction stage, because prototypes are cheaper to produce. What it hasn’t shortened is the time it takes to cross the chasm . If anything, the chasm is wider, because your pragmatist prospect’s inbox is now full of AI-generated pitches from competitors who also built their MVP in a weekend. The bottleneck is trust, distribution, and references — none of which AI accelerates.

Products, Not Companies — PMF Is Per-Product

Every PMF article I’ve ever read implicitly assumes a one-product start-up. Real companies don’t look like that. Microsoft has decisive PMF for Excel and none for a dozen products Satya has quietly killed. Stripe has PMF for Payments and is still reaching for it in new product lines. Notion has PMF for the core editor and is still hunting for it in Notion AI.

The practical consequence: you measure PMF per product, not per company. A mature company has a portfolio, and each product in that portfolio is at its own point in the lifecycle. Some have PMF and are earning it out (cash cows). Some are in the early-majority grind trying to extend it. Some are new bets that don’t have it yet and are legitimately in discovery. Conflating “the company is doing well” with “every product has PMF” hides the fact that one product is carrying several others, which is both normal and dangerous when unexamined.

See the Run / Grow / Transform lens for a capacity-allocation frame that makes this visible, and the BCG Growth-Share Matrix for a portfolio-diagnostic that pairs well with per-product PMF measurement. The question every board should be asking is: of our products, which ones have PMF, which ones don’t, and are we allocating dedicated capacity to the ones that don’t in a way that could credibly get them there?

The Coupling Anti-Pattern in Acquisitions and Multi-Market Products

A specific failure mode worth naming. When a company with PMF in one market acquires or launches into an adjacent market where PMF has not been achieved, the instinct is almost always to consolidate — one platform, one roadmap, one set of squad objectives. That instinct is usually wrong.

Coupling the innovation lifecycles of a PMF product and a pre-PMF product is how both of them get worse. The PMF product carries the weight of features it hasn’t validated in its own market. The pre-PMF product effectively stops hunting PMF because its roadmap is now subordinated to the larger product’s priorities. A year later the PMF gap is wider than it was at acquisition, and the consolidation rationale that looked so clean on a slide has produced two half-optimised products instead of two separately-optimised ones.

The better pattern: leave the products completely separate. Let each iterate on its own timeline with its own goals. Migrate markets, customers, or countries onto the PMF platform only after those markets have achieved their own PMF signal on whichever platform they’re currently on. Consolidate deliberately, segment-by-segment, not as a day-one integration decision. The coupling is the cost; treat it as a cost, not a given.

The same principle applies inside a single company launching a new product line. If your core product has PMF and you’re launching an adjacent new product, don’t share the roadmap. Don’t share the prioritisation queue. Don’t share the squad. The PMF-hunting work is fundamentally different from PMF-defending work, and mixing them corrupts both.

Try RoadmapOne to map PMF per product across your portfolio — free sign-up

The Side-of-Desk Anti-Pattern

This is the single most common failure mode I see in early-stage product work. A CEO or founder has an idea. Nobody writes a proper business case. Nobody commits to a measurable target. The idea gets allocated to a team who already has a full day job. Six months later the “new product” is half-built, nobody owns it, and the CEO is frustrated it hasn’t taken off.

The fix is embarrassingly simple and almost nobody does it:

1. Write a proper business case . Not a one-pager. A real document with revenue assumptions, customer assumptions, timing, and investment — that someone personally signs up to. A tight 3-page PID that every ExCo member signs is the Trayport-tested minimum viable version.
2. If it clears the hurdle, allocate a minimum viable team. Two engineers and one product person. Dedicated. Full-time. Nothing else.
3. Measure the team on outcomes, not output. Ellis score, cohort retention, reference customers. Not feature velocity. See outcome-based roadmaps and the deeper treatment in outcome vs output vs input .
4. Protect the team from interrupt work. Use WIP limits and avoid priority whiplash .

Most companies can afford two engineers and a product person. What they cannot afford is the theatre of claiming to pursue a new product with no dedicated capacity behind it. That’s how bright ideas die — not because they were bad, but because nobody was paid to prove them.

PMF Is Not Binary and It Decays

Almost every ranking article treats PMF as a milestone you reach. That framing is wrong. PMF is a continuous state that erodes. Markets shift. Channels saturate. New entrants arrive. The product that had decisive PMF in 2022 may have middling PMF in 2026 because the market has moved underneath it.

A useful mental model — borrowed from Coursera’s PMF article — is the four levels: Nascent, Developing, Strong, Extreme. Nascent is an early signal. Developing is enough to justify investment. Strong is Superhuman at 58%. Extreme is the kind of pull that produces unpaid word of mouth in quantities the marketing team can’t explain. Your product moves between these states continuously; the survey run quarterly tells you which direction you’re moving.

The related pitfall is what Balfour warns against in Four Fits: Channel-Model drift. Your Ellis score might be stable but your CAC has doubled because the channels that fed you last year are saturated. You have PMF on the product axis and no PMF on the channel axis, and your unit economics are quietly collapsing. This is the pattern the PE due-diligence team spots in three days and the founder has been blind to for six months.

The PE / NED Diagnostic for Fake PMF

When I look at an early-stage portfolio company claiming PMF, here’s what I actually check — in order.

1. Retention curve, by cohort, 12+ months. Is there a smile, a flatten, or a slide toward zero? Founder-reported “90% retention” is meaningless if it’s an average of the last 30 days.
2. NRR (B2B) or revenue cohort curve (B2C/consumer). Is the dollar-weighted curve holding up? A product with good logo retention and 80% NRR has a leaky-bucket problem.
3. Qualified Ellis survey, run independently. Not the one the marketing team ran that produced a 53% score from an unqualified list.
4. Organic vs paid split. What percentage of new logos came from outbound vs inbound? A product with PMF leaks at least some organic signal; a product with no organic is entirely pushed, not pulled.
5. Concentration. Is 60% of ARR in three customers? That’s not PMF in a market. That’s three bespoke engagements that happen to use the same codebase.
6. Pipeline composition. Is the pipeline dominated by prospect-specific feature requests? That’s the bad-salesperson pattern — asking for more features because the existing product can’t be sold as-is. A good salesperson sells what you have; a bad salesperson always asks for more features.

Usually by signal 2 or 3 the question is answered. Fake PMF looks like genuine PMF in the founder’s deck and looks like nothing at all in the cohort spreadsheet.

Cagan’s Four Risks as the Pre-PMF Filter

Before you can measure PMF, you have to have a product to measure. Cagan’s four risks — explained in full in the SVPG product risks article — are the right pre-PMF filter:

• Value risk: will customers choose to use this? (This is what PMF tests empirically, but it starts as a pre-build hypothesis.)
• Usability risk: can they actually use it? (Fast-moving research during prototype.)
• Feasibility risk: can our engineers build it? (In the AI era this is usually yes; in pre-AI it was the binding constraint.)
• Viability risk: does the business case work? (Channel-Model-Market fit, unit economics.)

Most early-stage bets die at value or viability. Feasibility is rarely the constraint any more. Usability fails produce churn but not strategic wipe-out. Running the four-risk diagnostic before spending engineering capacity is much cheaper than running the full PMF engine after. See the product discovery cluster — particularly discovery mistakes — for the discovery techniques that test these risks cheaply before you commit a team.

The dual-track agile operating model is the structural enabler: discovery work happens continuously alongside delivery so value and viability risks are being tested before delivery capacity is committed.

How RoadmapOne Helps

RoadmapOne exists to make the allocation decision visible. A grid view of squads × sprints shows you whether your supposed new-product squad is actually dedicated, or whether those two engineers are also tagged against three other squads keeping the Run product alive. The objectives-and-key-results model (see objectives to key results and OKRs for product teams ) lets you set a PMF-hunting team’s OKR as cohort retention, Ellis score, or reference customer count rather than feature velocity. And tagging objectives by Run / Grow / Transform means the analytics can tell the board exactly how much capacity is going into the new bet vs the existing product — usually much less than anyone thinks.

Frequently Asked Questions

What is a good product-market fit score?

On the Sean Ellis 40% test, ≥40% of qualified users (used the product at least twice in the past two weeks and experienced the core value) answering “very disappointed” is the traditional threshold for plausible PMF. Below 25% you almost certainly don’t have it. Between 25% and 40% you’re developing — the Rahul Vohra engine is the fastest path from there to 40%+. Don’t trust unqualified surveys; they inflate the number dramatically.

How long does it take to find product-market fit?

The honest median for B2B SaaS is 18 to 36 months from first paying customer to decisive PMF — and most products never get there. Consumer can be faster when virality is real, slower when it isn’t. The useful mental model is that PMF isn’t a waiting game; it’s an active hunt. You run a diagnostic (Ellis survey + retention curves), identify the largest blocker (Channel-Model fit? value-prop fit? HXC segment clarity?), fix that one thing, re-measure. Teams that treat PMF as a milestone they’ll “arrive at” wait a lot longer than teams that treat it as a weekly measurement.

Can a company have product-market fit without product-channel fit?

Not sustainably. Brian Balfour’s Four Fits framework says PMF is necessary but not sufficient — you also need Product-Channel, Channel-Model, and Model-Market fit. A classic failure mode: a B2B product with 58% Ellis score but a CAC that only pays back via outbound sales, which doesn’t scale economically. The product is loved; the business is a hobby. You need PMF and a channel that fits the product and a revenue model that fits that channel’s economics.

Is product-market fit the same as customer validation?

No. Customer validation tests whether the problem is real and the proposed solution is directionally right — it’s a pre-build, pre-revenue question. PMF is a post-launch, post-usage state in which real users (ideally paying) behave in ways that prove the product is genuinely serving the market. Customer validation is upstream; PMF is the milestone that confirms the validation was correct. You can have validated customers and still fail to reach PMF because the Channel-Model fit is wrong.

How is product-market fit different for AI-era products?

The shape of PMF is unchanged. What’s changed is how easily a competitor can copy your functionality. When build cost was high, functionality itself could be a moat for a year or two. In 2026, functional parity is assumed within weeks of a successful product shipping. That means PMF increasingly lives in the parts of the product AI can’t replicate: network effects, proprietary data, switching costs, brand, trusted reference customers, and deep workflow integration. If your PMF thesis is “we have feature X”, you probably don’t have PMF — you have a head start on a commodity.

Can you lose product-market fit?

Yes — routinely. PMF decays. Markets shift (new buyer personas, new use cases); channels saturate (Facebook ads CPA triples); models stop fitting (a freemium product hits enterprise demand and the self-serve model doesn’t convert it). The product with decisive PMF in 2022 may have middling PMF in 2026 without any change to the product itself, because the environment around it has moved. The fix is to re-measure regularly — quarterly Ellis, monthly cohort retention — and to treat PMF as a state to defend, not a milestone passed.

Sign up (free) to measure PMF per product, not per company

Conclusion

Product-market fit is the only milestone that matters for an early-stage product, and it is the most over-claimed status in the start-up world. Andy Rachleff’s original framing — a great market doing most of the work — is the one worth remembering. Sean Ellis’s 40% test and Rahul Vohra’s engine are the measurement tools worth using. Cohort retention curves are the signal PE diligence trusts because they can’t be spun.

AI has changed the cost structure around PMF hunting but not the discipline required. If anything, the collapse in build cost makes PMF measurement more important: you can now produce ten plausible products in the time it used to take to ship one, and ten no-PMF products is materially worse than one. The teams that win the next decade will treat PMF as a continuous state they measure weekly, not a milestone they’ll reach eventually — and they’ll allocate a dedicated minimum viable team to hunt it, rather than hoping an existing team does it side-of-desk between delivering the Run roadmap.

If a new product is worth pursuing, it’s worth a proper business case, a dedicated team, and a PMF hunt run with the same discipline the rest of the business runs its quarterly results. Anything less is theatre, and theatre is expensive.

Baxter image prompt (photorealistic, 4:3): Baxter the wirehaired dachshund as a Victorian-era pharmacist in a three-piece tweed waistcoat, standing behind a brass-trimmed apothecary counter, peering through a pince-nez at a single glass beaker held up to the light. The beaker contains a faintly glowing amber liquid labelled “PMF ≥ 40%”. Behind him, wooden shelves of unlabelled bottles in different colours — most empty, a few half-full. Warm gaslight, shallow depth of field, slightly sceptical expression. The card in front of the counter reads “We don’t guess. We measure.”