What PMF looks like in the data | The PMF Playbooks

Product-market fit is easy to recognise in hindsight and hard to see up close. Nick Frosst, co-founder of Cohere, says he does not love the term for exactly that reason: like fitness in evolutionary theory, it is only ever visible after the fact. He is right that nobody can forecast it. But founders on The Product Market Fit Show describe the moment they knew in remarkably similar ways, and the evidence shows up in roughly the same order. This playbook is about that evidence and the decision each piece of it should trigger. Nick Frosst, Cohere

A working definition

Marc Andreessen's original description still holds: being in a good market with a product that can satisfy that market. In the data, that shows up as three things happening at once:

Customers keep using the product without being pushed. Retention curves flatten.
Customers pull. They ask for more, refer others, expand, and get upset when something breaks.
The unit economics can work. What a customer pays over their life is clearly more than what it costs to find and serve them, even if you are not yet efficient.

You can have any one of these without the others. Fit is all three, in one segment, for a sustained period.

Bob Tinker adds a fourth thing that fit is not. After MobileIron closed its first customers he hired salespeople, burn spiked, and growth flatlined. He had product-market fit and no go-to-market fit: no repeatable way to find and win the next customer without a founder in the room. The two arrive separately, and the second has its own data signature, which is new-customer velocity that survives the founder stepping out of deals. Product-market fit vs go-to-market fit

Finding product market fit is surfing. You have to find a big wave first, some sort of big wave of change that your surfboard is riding on. Then your product is your surfboard on that wave.
Bob Tinker, MobileIron

The moment founders describe

Ask a founder how they knew and you get a story, not a number. The stories rhyme.

Strangers sign up. About 50 to 60 companies used Gusto before launch. When it launched, for new companies in California only, roughly 100 companies the founders had never spoken to signed up self-serve in the first month. Josh Reeves says that was the moment. How Gusto got to $1B in revenue
Customers demand, rather than evaluate. Anurag Goel says Render's moment came right after its launch at TechCrunch Disrupt in 2019: a flood of users and, more telling, a flood of feature requests from people who were already invested rather than kicking tyres. Render reached $10M ARR in about four years with no marketing spend. Render, Anurag Goel
Customers ask you to solve the next problem. Accrual, AI for accounting firms, dates its fit to the month enterprise firms started requesting features in adjacent areas nobody had shown them. How Accrual found PMF
Someone calls you when it matters. Peregrine's founders worked inside a police department as unpaid analysts for 18 months. The signal was a homicide detective phoning them for help on a cold case. Peregrine is now valued at $6.8B. How Peregrine built a $6.8B company
The calendar collapses. Matt Watson says true fit feels like being overwhelmed: onboarding, support and product all go crazy and you are holding onto the rocket. If you feel in control with time to spare, you probably do not have the pull. Matt Watson
Referrals arrive unasked. Adam Robinson, who bootstrapped Retention.com to $25M ARR and $14M in annual profit, thinks this is the only signal that counts.

The only thing you should be doing until you are getting referrals that you are not asking for, many every week, is talking to prospects or customers or writing code.
Adam Robinson, Retention.com, Bootstrapped SaaS and product-market fit

And the contrast, in the same founder's words. Before fit, his previous company sat at $3M ARR, cash-flow positive and growing slowly, and he could not change that curve. Ben Alarie at Blue J describes eight years of what he calls partial fit: real customers, real revenue, no market pull. He put the product in maintenance mode and gave the team six months to rebuild on large language models. Net promoter score went from 20 to 84, the company went from $2M to $25M ARR in two years, and it was adding about ten new firms a day. Ben Alarie, Blue J

The retention curve, and how to read it

Retention by cohort is the closest thing there is to a single PMF test. Build the table described in the previous playbook: rows are the month a cohort activated, columns are months since activation, cells are the share of that cohort still active. Then plot each row.

There are only three shapes.

Falling to zero. The curve keeps dropping every month with no sign of levelling. Customers try the product and drift away. This is the pre-fit shape. No amount of acquisition fixes it; new customers just leave through the same door. Tome's 20 million users had this shape.

Flattening. The curve drops in the first one to three months, then levels at a plateau and stays there. The plateau is the share of customers for whom the product became a habit. This is the fit shape. Zach Lloyd at Warp refused to spend on marketing until he saw it. Zach Lloyd, Warp

Smiling. The curve drops, flattens, then rises, because surviving customers expand or bring back colleagues. This is the strong-fit shape, and it is what net revenue retention over 100 percent looks like at the account level. SolidRoad's 186 percent net revenue retention is a smiling curve.

What counts as a good plateau depends on the product and the buyer, but for B2B software the rough ranges founders and investors use:

Product type	Month-6 logo retention that suggests fit
Small-business tools, monthly billing	55 to 70%
Mid-market software, annual billing	75 to 85%
Enterprise, multi-year	85 to 95%
Usage-based infrastructure	Judge on revenue retention: 100%+

Two cautions. First, you need at least three cohorts of at least ten customers each before the shape means anything. Second, measure on usage, not billing. An annual contract hides churn for up to a year.

Two companies, same revenue, different futures

Two founders each had $30,000 in monthly recurring revenue and 45 customers. The first had cohorts that fell to about 20 percent by month six and were still falling. The second had cohorts that dropped to 65 percent by month three and stayed there. The first company's revenue was almost entirely new sales replacing churn; the founder was running to stand still and did not know it. The second company had fit in its segment and could grow by adding customers to a stable base. Revenue said they were the same. Retention said they were different companies.

Pull: the qualitative signals

Retention tells you fit exists. Pull tells you how strong it is and where. These are the signals founders report noticing, usually before the numbers confirmed them.

Inbound you did not cause. Prospects arrive who heard about you from a customer. Track the source of every lead; "referral" rising as a share is the signal. AJ Loiacono spent nothing on marketing for two years at Capital Rx, now Judi Health, and raised $400M at a $3.25B valuation.
Expansion without selling. Customers add seats, sites or volume and you find out from the invoice. Roy Moussa at GetVocal watched one deployment grow to 80 agents across a single enterprise.
Complaints about downtime. Nobody complains about a product they do not depend on. Keith Peiris says the difference between Tome and Lightfield was customers who were hopping mad when something broke.
Users tolerating rough edges. Robinson's secondary signal: customers put up with a poor interface and still give you a high score.
Feature requests that are specific and operational. Render's flood of requests. Accrual's adjacent problems.
Buyers who know the price before you say it. They have already discussed you with a peer.
Sales cycles shortening in the core segment while the price holds or rises. Tenex's first big financial-services customer bypassed its own procurement to close in under 60 days. Terra Security sells against budget lines that already exist, so cycles are short. The distinction that matters: compression the customer initiates is a signal; compression you buy with discounts is not.
Word-of-mouth you can trace. Ask every new customer how they heard of you, and write it down.

The survey version of pull is the Sean Ellis question: how would you feel if you could no longer use the product? Treat 40 percent "very disappointed" as the bar, and segment the answers to find who the product is really for. Watson prefers it to net promoter score. Use it as a check between cohorts, not as the answer.

Speed after the switch

One pattern from the show is worth its own section. When fit arrives, it tends to arrive fast, and the speed is itself a signal.

Axonius: $1M ARR in under twelve months, $10M in 18, $100M in four and a half years, with a net promoter score in the 70s. The first paying customer signed after a single slide showing 20 bugs fixed in a day or two each, in a market where incumbents measured response time in quarters. How Axonius hit $100M ARR
Blockthrough: pre-revenue for two and a half years, then more than $100,000 a month within eight months of the fourth version shipping.
Within: Andrew Antos sold a $10M business to go all in on a side tool. It did about $350,000 in ARR in its first five or six weeks, over $1M the next quarter, and about $20M in 18 months. Andrew Antos, Within
Thatch: 80 times growth after the pivot. Wispr Flow: 20 percent paid conversion against a 3 to 4 percent benchmark. Product-market fit after a pivot

The show's own summary is that real post-pivot fit shows up in the numbers quickly. If you have been at "almost" for a year, that is information.

The exception is deep enterprise, where the pre-fit period can be years of nothing. Arvind Jain left Rubrik to start Glean, raised $15M with no revenue and went about two years with zero, because enterprise search is hard to build. Kyle Hanslovan bootstrapped Huntress for three years, was rejected by around 60 investors and slept in his car for the better part of a year. Both companies are now worth billions. The signal, when it came, was late and steep. Glean and Huntress at Web Summit

The stages, and what to do at each

Fit is not a switch. It usually arrives in a sequence, and the right action changes at each step.

Stage 0: nothing sticks

Retention falls to zero. Users are polite. No inbound. The action is not to grow; it is to narrow. Pick the segment where usage lasted longest, talk to every customer in it, and rebuild onboarding and the core workflow for them alone. Founder-led sales only; no hires, no paid acquisition. Robinson's rule applies: talk to customers or write code, nothing else.

Stage 1: fit in a pocket

One segment flattens at a low plateau while others fall. Pull appears only from that segment. The action is to cut everything else. Say no to customers outside the segment, even paying ones. SolidRoad fired its sales-team customers to follow usage into support teams. Jay Madheswaran fired every customer of a $3M ARR business to follow the pull into plaintiff law. Make the product excellent for the pocket and measure retention there alone. Raise the price for the pocket; it will hold.

Stage 2: fit in the segment

Cohorts in the core segment flatten at a healthy plateau. Referrals arrive. Expansion appears. This is when the motion can become repeatable, and it is the moment to write Tinker's one-page playbook and make the first sales hire. Spend on acquisition that targets exactly this segment. Resist the urge to widen; the plateau is a fact about this segment, not about everyone. Guy Podjarny's warning from Snyk belongs here: the company burned $4M to reach $100,000 ARR because developers loved the product and security leaders held the budget. Once it sold to both at once, it added $4.5M ARR in twelve months. User love without buyer fit produces usage data and no revenue data. Guy Podjarny, Snyk

Stage 3: fit with a business

Net revenue retention is above 100 percent, sales efficiency is improving, and you can predict next quarter's revenue from this quarter's pipeline. Now you can add adjacent segments one at a time, each with its own cohort table and its own test. This is go-to-market fit in Tinker's terms, and it is where MobileIron went from 10 to 20 new customers a quarter to 400 to 500.

The most common mistake at every stage is acting one stage too early: hiring at stage 0, spending on ads at stage 1, expanding segments at stage 2. The cohort table is the discipline.

Signals it is time to change course

Retention that keeps falling is the clearest, but there are earlier warnings:

Activation is high and month-one retention is low. The product delivers once, then has no reason to be reopened. Consider whether you are a workflow tool or a one-time service, and price accordingly.
Retention is fine and expansion is zero. The product solves a bounded problem. That can be a good business, but not a venture-scale one without a second product.
Pull only comes from users, never from buyers. The Snyk problem. Find who can pay, or find a product for the users' bosses.
Every customer needs a different product. Custom work in every deal means you have consulting fit, not product fit.
Sales cycles are lengthening despite a stable price. The urgency that drove early deals has faded. Look for a new trigger or a new segment where it is fresh.
The market keeps asking for something else. Seven of Thatch's eight prospects did. That is the pivot signal in its purest form.

Changing course does not usually mean a full pivot. Most of the time it means a narrower segment, a different buyer in the same company, or a different pricing metric. Full pivots are rarer and expensive, and even then the biggest wins on the show, among them Mercury, Ada, Wrike, Thatch and Amplitude, found fit only after one. Do the smaller moves first, and run each one against the cohort table for two or three months before judging.

Signals to keep going, even when it feels slow

Founders quit on fit more often than the data would justify. Keep going if:

A cohort flattened. Any cohort, at any plateau. Something works for someone.
Referrals exist, even a handful, and you can trace them.
Customers are angry when things break. They depend on you.
The plateau rises cohort over cohort. The product is getting better for the people it is for.
Revenue per customer grows while logo count is flat. Depth before breadth is normal.

Slow, flattening growth in a real segment beats fast growth in a leaky one every time. The first compounds; the second does not.

The one-page PMF review

Do this once a month, in writing, and keep the pages.

Cohort table and curves for the core segment. Shape: falling, flattening or smiling?
Plateau level for the last three cohorts, and the trend.
Net revenue retention for customers older than six months.
Share of new leads that came from referrals or unprompted inbound.
Sean Ellis "very disappointed" share, if you run the survey, by segment.
Sales cycle and close rate in the core segment, and the trend. Who initiated any shortening?
The list of customers lost this month with the reason in their words.
The stage you believe you are at, and the one action for next month that matches it.

Do this week

6 actions

01Build the cohort table for your core segment and name the shape honestly.
02Add "how did you hear about us" to every new deal and start tracking referral share.
03Count the referrals you received this month that you did not ask for.
04Run the Sean Ellis question with your active users and segment the answers.
05Write down which stage you are at and check whether your current plans belong to that stage or a later one.
06Schedule the one-page PMF review for the first Friday of every month.

Product-market fit is not a feeling, and it is not a revenue number. It is a shape in a table, a set of behaviours from customers you did not ask for, and a business that can afford itself. Look for all three, in one segment, and let them tell you what to do next.