5 steps to find product-market fit in an unstable world

Finding product-market fit takes five steps: determine your target customer and their needs, specify a value proposition, scope a minimum viable product, test it with real users, and measure the result with signals you can trust.

Product-market fit itself is the state where what the market needs and what your product delivers line up closely enough that customers start pulling the product from you.

The difficulty in 2026 is that the alignment doesn't hold still. Markets move, capabilities that used to be your differentiator ship as a checkbox in someone else's product, and a fit you earned last year can stop being one without any single dramatic event. These five steps work best as a loop.

What product-market fit really is

Product-market fit is the state where the market's needs match what your product delivers, closely enough that demand starts to run ahead of your ability to serve it.

The term was coined by Andy Rachleff, co-founder of Benchmark Capital and founder of Wealthfront, building on ideas from Sequoia Capital founder Don Valentine. Marc Andreessen popularized it in his 2007 essay "The only thing that matters," and his description of the two states still holds up:

"You can always feel when product/market fit isn't happening. The customers aren't quite getting value out of the product, word of mouth isn't spreading, usage isn't growing that fast, the sales cycle takes too long, and lots of deals never close. And you can always feel product/market fit when it's happening. The customers are buying the product just as fast as you can make it, or usage is growing just as fast as you can add more servers."

What that leaves out is that fit comes in degrees. Todd Jackson's PMF Method at First Round breaks it into four levels, and most companies stall at the second one.

Why product-market fit keeps expiring

Product-market fit expires because both halves of the alignment move independently, and the market half moves without warning you.

Chegg is the clearest case. Its homework-help business depended on students finding answer pages through Google. Once free chatbots answered the same questions and Google began generating answers directly in search results, the subscriptions went with the traffic. Chegg has lost roughly 99% of its market value, cut 22% of its staff in May 2025 and another 45% that October, and sued Google over AI answers.

Stack Overflow adds a wrinkle worth sitting with. It posted 1,304 questions in July 2026, against a peak of roughly 207,000 a month in March 2014. But the slide began around 2014, well before ChatGPT existed, driven by aggressive question-closing that made new contributors feel unwelcome. AI finished a decline the community's own moderation culture had started. The same pattern runs at feature scale: dozens of companies building PDF processing tools were effectively eliminated when ChatGPT added native PDF support. If your core function fits inside somebody else's roadmap, your fit expires on their release schedule.

What re-fitting looks like at scale

Established companies aren't exempt, and Salesforce is the most instructive case I know of. Agentforce launched at Dreamforce 2024 at $2 per conversation, moving off per-seat licensing because an agent doing work doesn't map onto a seat. The market pushed back, since a "conversation" was hard to budget. Salesforce introduced Flex Credits in May 2025 at roughly $0.10 per action, added per-user options later that year, and in June 2026 announced pay-per-resolution: $2 when the agent resolves an issue on its own, nothing when it fails. Four pricing models in under two years, each one re-fitting the same product to a market still deciding what it would pay for, as AI+SaaS products increasingly perform work rather than merely support it.

Running out of money appeared in 70% of those same 431 postmortems, which CB Insights treats as the symptom rather than the cause. Companies die of building something the market stopped wanting; the empty bank account is where it becomes visible.

How product-market fit turns into revenue

Fit becomes revenue through a chain with four links, and every link has to hold. The product solves a problem the customer would otherwise pay a person or another tool to solve. That creates a behavior: the customer comes back without being prompted or re-sold. Repeated behavior shows up as retention, the one place fit can't be faked. Retention compounds into revenue, because customers you keep cost nothing to reacquire, which pulls down your payback period and frees the budget you were spending to replace churn.

The chain breaks most often at the second link, and AI products break it in a measurable way. RevenueCat's 2026 report found AI-powered apps churn 36% faster while generating 41% more revenue per user in year one. ChartMogul's analysis of 3,500 software companies, 200 of them AI-native, found median gross revenue retention of 40% among AI-native products against 82% median net revenue retention for B2B SaaS.

Price sorts the outcomes almost perfectly. In ChartMogul's data, AI products under $50 a month held 23% gross revenue retention; between $50 and $249, 45%; above $250 a month, 70% GRR and 85% NRR, which is ordinary B2B SaaS territory. Cheap AI products largely measure curiosity. Expensive ones have to be embedded in someone's work to survive a renewal, and the ones that survive it have fit.

There's a practical warning in that for anyone reading their own dashboard. Signup and revenue curves can look like fit for two or three quarters while the underlying behavior never repeats once. Retention tells you the difference, and it takes time you have to plan for.

5 steps to determine product-market fit

1. Determine your target customer and their needs

Steve Jobs said "people don't know what they want until you show it to them," and the line gets used as permission to skip customer research. What it describes is the opposite kind of work: knowing a customer well enough to anticipate a need they can't yet put into words. That's a harder research problem than asking people what they want.

Apple's own Vision Pro is the counterexample. It launched at $3,499 into a market that hadn't asked for it, sold somewhere around half a million units, and after the M5 refresh in October 2025 failed to move demand, Apple reportedly halted development, canceled the lighter Vision Air, and moved the team toward smart glasses. Reading a market wrong is expensive even when you're Apple.

So confirm a market exists, then find out what it needs. Talk to people in your target segment until you stop hearing new problems, usually somewhere between ten and twenty conversations per segment; the saturation point matters more than the number. Structured UX interviews surface the reasoning behind what people tell you, which is where the useful material tends to be. You can also build an AI-assisted prototype to put something concrete in front of users, validate the idea, and only then commit to professional software development in order to scale it.

Questions to answer at this stage:

• Who is your customer and what are they trying to do?

• What do they need?

• How can you help them?

• What problem are you solving?

• How exactly does your product solve this problem?

2. Specify the value proposition

You found a market, which means you almost certainly have competitors. Your value proposition is the promise you make about why someone should choose you, and it has to attach to a real problem rather than a feature list.

Being genuinely different is worth more than being incrementally better. Research from MIT Sloan's Luca Gius, covering 67 venture competitions, found that judge disagreement about a startup predicted its success. Distinctive propositions split expert opinion, while consensus-friendly ideas underperformed. If everyone in the room nods along at your pitch, that's a signal worth examining rather than celebrating.

Serving multiple audiences usually means writing a separate value proposition for each, because a promise aimed at two segments tends to persuade neither. And don't discard an idea because something similar exists. Facebook wasn't the first social network, and Slack arrived years after the first team chat tools. Both won on execution.

Questions to answer at this stage:

• Who is your competitor?

• What are the key features of the product?

• What distinguishes you from your competitors?

• What do you offer that no one else does?

3. Specify a minimum viable product (MVP) feature set

An MVP is a minimal form of your product that's tested on the market. It gives you something real to learn from while keeping the cost of being wrong low.

Choose the features your product can't exist without: the ones carrying the core promise from step two. Anything that would be satisfying to build, or that a competitor happens to have, belongs in a later release. Deciding what to leave out is the harder half of the exercise, which is why our MVP development work starts with a workshop rather than a backlog.

AI tooling has changed the economics here. Teams can assemble a working MVP considerably faster than they could two years ago, which lowers the cost of testing an assumption and makes it easier to build the wrong thing quickly. The discipline of steps one and two carries more weight than it used to. Resist polishing at this stage; there will be time once you know which parts people actually use.

Questions to answer at this stage:

• Which features are the most essential for the existence of your product?

• Are these functions directly related to solving the original goal?

• Does the implementation of these features require a lot of effort?

4. Test

Testing is where your assumptions meet people who don't share them. Once the MVP is in front of real users, the job is to find out where your model of their behavior is wrong.

Run A/B tests, collect both qualitative and quantitative data, and watch what people do alongside what they say, because the two frequently disagree. UX research methods like usability testing and structured interviews exist to catch that gap.

The people you ask matter as much as the questions. Friends and family answer generously, which is what makes their feedback misleading. Look for people who have the problem you're solving and a budget to fix it, and pay particular attention to the ones who tried your product and stopped.

If users keep coming back, you have something to build on. If they report gaps, return to step two or three and adjust. Most teams cycle through this loop more than once, and that's the process working as intended.

Questions to answer at this stage:

• What is the main goal for using this product?

• How was the first experience with the product?

• Were user expectations met, unmet, or exceeded?

• What services or features are missing, if any?

• Why did users choose to use your product over other options?

5. Measure

Measuring product-market fit means finding evidence that the behavior repeats, and no single number gives you that. The best-known signal is the 40% rule, developed by Sean Ellis, who led growth at Dropbox, LogMeIn, and Eventbrite. Ask existing customers how they'd feel if your product disappeared tomorrow. After benchmarking close to a hundred startups, Ellis found that companies with strong traction almost always had 40% or more respond "very disappointed," while the ones struggling to grow came in below. A Google Form is enough to run it.

Treat 40% as a reading rather than a verdict. It's a heuristic from a small sample of mid-2000s SaaS companies, and it only reaches people still engaged enough to answer, which under-samples the users who already left. Pair it with signals drawn from behavior instead of opinion.

Total addressable market, calculated by multiplying average revenue per user by the total potential customers, sizes the prize rather than your claim on it. NPS and LTV have their place too.

Our guide to finding product-market fit once you've released your MVP goes deeper, including cohort analysis and how to segment survey responses so the average doesn't bury the signal.

Question to answer at this stage: What is the one metric that matters most?

What has to be true for this to work

These five steps assume four conditions. When one is missing, the process still produces confident answers, and the answers are wrong.

You need access to real buyers. Interviewing users when a procurement committee makes the decision produces a fit with the wrong party, so in B2B you have to reach both, which is slower and unavoidable. You also need a measurement window long enough for retention to mean something, because cohort curves take months to form. If your board expects a verdict in six weeks, you'll report acquisition numbers and call them fit, which is the trap the AI retention data exposes.

You need pricing that filters. Free and near-free tiers fill your dashboard with people who were curious, and ChartMogul's price gradient shows how badly that distorts the reading. Charging enough that someone has to make a decision is itself a validation instrument.

You need a build cadence fast enough to act on what you learn. A discovery phase that ends in a document nobody can implement for two quarters has produced accurate information about a market that no longer exists.

For AI products there's a fifth condition: buyers increasingly want evidence rather than a demo. McKinsey's review of 150 software vendors found only 30% publish quantifiable ROI from real deployments, while enterprise customers report AI tools raising software costs by 60 to 80% without matching reductions in headcount. If your value proposition rests on savings you can't demonstrate, expect renewal conversations to get uncomfortable.

Finding and maintaining product-market fit in an unstable world

The product, the market, and the people you're serving all drift, usually at different speeds, and the gap between them opens well before it shows up in revenue.

The teams that hold fit longest build the checking into their operating rhythm. A standing habit of talking to customers, a retention dashboard somebody actually owns, and a willingness to revisit the value proposition while the numbers still look fine. That last one is hardest, because nothing in a healthy quarter tells you to go looking.

Chegg, Stack Overflow, and Salesforce are three versions of the same lesson at different scales. Two ran out of time to re-fit; the third has re-fit four times and is presumably not finished. Fast-moving markets reward teams who treat that re-fitting as ordinary work. If you're looking for a product development partner to do it alongside you, Monterail is here to help.