"Do we have product-market fit?" sits behind big early calls: whether to hire a sales team, raise on a growth story, or start spending on acquisition. One famous answer is a feeling. In a 2007 essay, Marc Andreessen wrote: "You can always feel when product/market fit isn’t happening."
That works when the signal is loud, and less when a few customers love the product while the rest are politely lukewarm. This guide swaps the feeling for three tests you can run this month: the 40% "very disappointed" survey, a cohort retention curve, and a check for demand that pulls. The rest covers leading and lagging signals, false positives (AI products included), what to do on each side of the line, and a checklist.
What product-market fit means
The definition comes from Andreessen’s post "The only thing that matters," dated June 25, 2007:
"Product/market fit means being in a good market with a product that can satisfy that market."
He credits Andy Rachleff, formerly of Benchmark Capital, "who crystallized this formulation for me," and names a rule after him, Rachleff’s Corollary: "The only thing that matters is getting to product/market fit."
Andreessen’s picture of life without it will sound familiar: customers aren’t getting value, word of mouth isn’t spreading, usage isn’t growing fast, the sales cycle drags, and deals never close. With it, customers buy "just as fast as you can make it" and you hire sales and support staff as fast as you can.
Both descriptions are outcomes. The useful work happens earlier, while the evidence is still mixed.
Three tests that tell you whether you have it
Run all three. The survey measures intensity, retention measures durability, and pull measures whether buyers want it badly enough to act.
The 40% survey, run properly
Sean Ellis’s test asks existing users one question: how would you feel if you could no longer use the product? Superhuman’s answer options were "very disappointed," "somewhat disappointed," and "not disappointed."
The benchmark is Ellis’s own, from his post "The Startup Pyramid": "In my experience, achieving product/market fit requires at least 40% of users saying they would be ‘very disappointed’ without your product." He is frank about it: "Admittedly this threshold is a bit arbitrary, but I defined it after comparing results across nearly 100 startups."
Treat 40% as an informed line, not a law. Two details decide whether your score means anything:
- Who you ask. Rahul Vohra says Superhuman followed Ellis’s recommendation to survey people who had used the product at least twice in the last two weeks and had recently experienced its core. Surveying sign-ups who never activated measures your onboarding, not your fit.
- How many. Vohra found results become directionally correct at around 40 respondents.
For a reference point, Vohra relays that Hiten Shah put Ellis’s question to 731 Slack users in 2015 and 51% said "very disappointed."
The score shows how many users would miss you. It can’t tell you why deals close or stall; for that, pair it with buyer conversations, such as a 10-interview win/loss process.
Worked example: how Superhuman went from 22% to 58%
Vohra’s First Round Review account (November 2018) shows the survey run as an operating metric. Superhuman asked four questions:
- How would you feel if you could no longer use Superhuman?
- What type of people do you think would most benefit from Superhuman?
- What is the main benefit you receive from Superhuman?
- How can we improve Superhuman for you?
In summer 2017, 22% said "very disappointed," well under the line. The team studied who the very disappointed users were and defined their high-expectation customer with Julie Supan’s framework. Among that segment the score was 33%.
Superhuman politely disregarded "not disappointed" users. It split "somewhat disappointed" users by whether the main benefit, speed, resonated, and paid attention only to those who valued speed. The roadmap rule: "spend half your time doubling down on what users already love and the other half on addressing what’s holding others back."
Then they made the score a standing metric. Superhuman surveyed new users continuously, never surveyed anyone twice, and tracked the number weekly, monthly, and quarterly. Within three quarters it reached 58%.
The lesson for a founder stuck at 22% isn’t "ship more features." Find the segment that already scores high, build for them and for fence-sitters who value the same benefit, and stop chasing people who will never love the product.
A retention curve that flattens
The survey records what people say. Retention records what they do. Brian Balfour’s 2013 essay on product-market fit describes the method: for each signup cohort, plot the percentage still active over time. "IF it flattens off at some point, you have probably found product market fit for some market or audience." He notes that B2B SaaS is the slowest to read this way.
For benchmarks, Lenny Rachitsky and Casey Winters gathered estimates from about 20 growth practitioners in June 2020: six-month user retention of around 60% is good and 80% great for SMB and mid-market SaaS, and 70% and 90% for enterprise SaaS. These are venture-scale bars, and the same post says they are "not required for product-market fit," because "a flat retention curve that drives a scalable acquisition strategy is enough to keep your business alive."
Retention has one real weakness: it lags. Mark Roberge of Stage 2 Capital calls long-term retention the best statistical representation of fit, but recommends defining a leading indicator in the form "P% of customers achieve E event(s) within T time." His examples include Slack (70% send 2,000+ team messages within 30 days) and Dropbox (85% upload one file in one folder on one device within an hour).
A counterweight: in his guide to finding B2B product-market fit, Lenny reported that none of the 20+ B2B founders he interviewed used retention as their PMF signal. Early on you may not have enough cohorts, which is where the third test helps.
Demand that pulls
Andreessen’s essay already has the idea: in a great market, "the market pulls product out of the startup." Rob Snyder’s PULL framework turns it into something you can check in a sales conversation. Snyder has advised 50+ startups from zero to $1M ARR and discussed the framework on Lenny’s Podcast on October 25, 2026.
Snyder credits the underlying idea to author Gorick Ng and defines PULL as "the state of having demand." The episode page gives the final letter names; the explanations below paraphrase his May 2025 post, which used slightly different names for the two L’s:
- Project: the buyer has a task queued on their to-do list.
- Unavoidable: it has to be handled now.
- List of options: they look into ways to get it done.
- Limitations: those options fall short for a critical reason.
When all four are present, buyers pull. When you are explaining why a problem matters to someone with no project, you are pushing. In "Retention is the hard part", Snyder ties the leading indicator of retention to that project and observes that a typical first 10 customers splits into about 2 "hell yes," 5 to 6 "not unhappy," and 2 to 3 who are unhappy. Treat that as one practitioner’s observation. It is still a useful check if you expect every early customer to be delighted.
Leading vs lagging signals
The "type" column is our editorial reading of the sources.
| Signal | Type | What good looks like | Watch out for |
|---|---|---|---|
| Ellis survey score | Leading | 40% or more "very disappointed" among recently active users | Inactive sign-ups; under about 40 responses |
| Activation event (P% do E within T) | Leading | An early action most new customers reach | An event never checked against retention |
| Pull in sales calls | Leading | Buyers name an unavoidable project and weak options | Polite interest with no project behind it |
| Cohort retention curve | Lagging | It flattens; for AI products, judge from month 3 | Annual contracts hiding disengagement |
| Six-month user retention | Lagging | About 60% (SMB and mid-market SaaS) or 70% (enterprise) is "good" in the 2020 estimates | Treating venture-scale bars as a PMF requirement |
| Market behavior | Lagging | Customers buy as fast as you can deliver | Growth bought with spend while churn stays high |
Leading signals tell you where to look; lagging signals tell you whether you were right.
False positives that look like PMF
AI tourists. In "Retention Is All You Need" (September 10, 2025), a16z’s Santiago Rodriguez and Alex Immerman warn about "AI tourists who sign up and churn after a couple months," which make early curves less predictive than in SaaS or consumer internet. Across hundreds of AI companies, "most often, the curve begins to flatten around M3," so they rebase retention to month 3 instead of month 0 and use M12 divided by M3 as an early predictor. If the curve keeps falling after M3 or never flattens, "initial customer delight isn’t translating into durable utility." The scope is AI companies with self-serve, month-to-month billing.
Growth with churn. Snyder’s same post opens bluntly: "Product-market fit is measured by customer retention, not customer acquisition. If you’re growing 100% MoM but have massive churn, you do NOT have product-market fit."
Fit with one segment only. Coda’s Shishir Mehrotra, quoted by Lenny, says "you always have product-market fit with one group and not with the next." Superhuman’s 22% overall versus 33% in one segment cuts both ways: segmenting can reveal fit, and a blended number can hide it. Know which group your score describes before you scale into the next one.
Annual contracts. The same a16z piece notes that with monthly billing the retention curve is a direct PMF signal, but with annual contracts high retention can give "a false sense of security." Track engagement as well.
What to do before PMF, and what changes after
Before
Andreessen’s instruction: "focus obsessively on getting to product/market fit." Ellis adds guardrails: keep burn low, keep founders in direct contact with users, and "avoid bringing in VPs of Marketing and Sales to try to solve the problem." Keep marketing cheap and aimed at learning, as in a pre-PMF content playbook.
After
Ellis’s pyramid lays out the order once you clear 40%:
- Promise. Highlight the benefits your "must have" users describe. The answers to Superhuman’s third question are raw material for a positioning statement.
- Economics. Get the business model working; a SaaS metrics calculator shows what your churn rate means for LTV, LTV:CAC, and CAC payback.
- Optimize. Build repeatable customer acquisition.
- Scale.
Roberge frames it as product-market fit (customers succeed consistently), then go-to-market fit (consistently and scalably), then growth and a moat. a16z agrees on timing: once the retention curve stabilizes, more aggressive go-to-market investment is justified. Choosing the motion is its own decision; see product-led, sales-led, or hybrid growth.
PMF also isn’t binary. First Round’s Levels of PMF, built for sales-led B2B, describes four levels across satisfaction, demand, and efficiency: Nascent (3 to 5 customers, $0 to $500K ARR), Developing (5 to 25 customers, $500K to $5M), Strong (25 to 100, $5M to $25M), and Extreme (100+, $25M+).
A product-market fit checklist
- Survey users active at least twice in the last two weeks, and wait for about 40 responses.
- Segment the "very disappointed" group, note the main benefit they cite, and build for "somewhat disappointed" users who value it.
- Plot retention by signup cohort and check whether it flattens (from month 3 for AI products).
- Define one leading indicator: P% of customers do E within T.
- In sales calls, listen for an unavoidable project and options that fall short.
- Hold off on scaling sales and marketing hires until the survey and the curve agree.
FAQ
What is product-market fit?
Marc Andreessen defined it in 2007 as "being in a good market with a product that can satisfy that market." In practice it shows up as users who would be very disappointed to lose the product, cohorts whose retention flattens, and buyers with a problem they must solve now.
How do you measure product-market fit?
Run the Sean Ellis survey on recently active users and look for 40% or more "very disappointed." Plot cohort retention and check whether it flattens. Listen in sales calls for an unavoidable project and options that fall short. The survey and pull lead; retention confirms.
What is the 40% rule?
It is Sean Ellis’s benchmark: at least 40% of users saying they would be "very disappointed" without your product. He calls the threshold "a bit arbitrary" but set it after comparing nearly 100 startups. Survey active users, aim for about 40 responses, and segment before concluding anything.
How long does it take to find product-market fit?
In 24 B2B startups Lenny Rachitsky hand-picked and charted in 2023, the median time from idea to first feeling PMF was roughly two years, typically 9 to 18 months from a working product. Figma, Airtable, and Slack took 4+ years. His rule of thumb: start worrying past two years, and worry seriously past three.
What comes after product-market fit?
In Sean Ellis’s pyramid: sharpen the promise, fix the economics, then build repeatable acquisition before you scale. Mark Roberge calls the next milestone go-to-market fit, customer success that is consistent and scalable. Keep surveying, because fit with one segment doesn’t guarantee fit with the next.
