PMF Is Not a Funnel. It Is a Network.

In 2023 I was on a Zoom with the YC-backed founding team of an AI-powered personal productivity app in Silicon Valley. They were pre product-market fit. They were already planning for scale before fit. Customer acquisition cost in one corner of the deck. Lifetime value in another. A funnel with three A/B tests waiting on Friday's readout. The CEO asked what I would prioritise. I opened with usability tests and task-completion studies because that is what my retainer was for. Here is what I should have said first: in a category where value rises when more people use the product, you do not find product-market fit through a funnel. You find it through a network. Their deck measured drops through a pipe. PMF in their market would show up as density: a small group of users who kept returning because each new person made the product harder to leave for everyone already inside. They were polishing a canopy while the soil stayed empty.

Funnels Are Billable. Density Proves Fit.

Product-market fit is the moment retention proves the product earns a place in someone's life. For a solo tool, that proof can come from one user and a workflow that sticks. For a networked product, the proof is collective: enough people in the same graph that leaving costs something real. Founders at seed and Series A still lead with funnels because funnels are legible. Cost per click has a number. Onboarding tests fit a sprint. Investors ask for CAC and LTV before anyone asks whether the category tips on participation. That stack is billable. It is also the wrong first question when your product gets more valuable as the room fills.

NFX, the venture firm that has catalogued network effects for two decades, argues that among the four remaining defensibilities in software (brand, embedding, scale, and network effects), network effects are by far the strongest. nfx I have watched pre-PMF teams spend six figures on step three of a funnel while a thinner competitor found fit because fifty teams in one vertical already shared habits inside someone else's product. I have invoiced for that mistake. The error at early stage is not skipping user research. It is treating PMF as a conversion problem when the category demands a connection problem first.

Stage One Kills Most Startups

Andrew Chen spent years inside the machine before he wrote the manual. As head of rider growth at Uber, he oversaw an increase from 15 million to 100 million active users, city by city, atomic network by atomic network. a16z He later joined Andreessen Horowitz and published The Cold Start Problem (Harper Business, 2021). Chen's framework is not a post-PMF growth playbook. It is a pre-PMF survival guide for anything that gets better when strangers or colleagues join.

Chen's Cold Start Theory splits network growth into five stages: the Cold Start Problem, the Tipping Point, Escape Velocity, Hitting the Ceiling, and the Moat. Stage one is where most startups die before anyone writes "we have PMF" on a board slide. Chen calls the early force anti-network effects: new users churn because not enough other users are present yet, and "small networks want to self-destruct." chen Slack is useless until colleagues join. Uber is useless until drivers and riders overlap in the same place at the same time. Polite sign-ups and pretty onboarding do not disprove that. Sparse retention does.

The escape hatch is the atomic network: the smallest stable group where one more user makes the product harder to leave for everyone already inside. Chen opens the concept with Bank of America's 1958 credit-card launch in Fresno, not with a SaaS dashboard. Slack's atomic unit can be three people in one team. Uber's earliest wedge was narrower than a city: "5pm at the Caltrain station at 5th and King St." Stewart Butterfield, Slack's co-founder, told Reid Hoffman on Masters of Scale that early Slack had "very, very little network effect between companies, but a huge binary, 100% or just nothing, inside of companies." mastersofscale That is not marketing trivia. It is how you manufacture PMF signals in a networked category before you have a city, a country, or a category. At the Tipping Point, Chen documents a pattern called "come for the tool, stay for the network": Dropbox users arrived for file sync, then stayed for shared folders with colleagues; Instagram began as a stripped-down photo tool cut from a cluttered app called Burbn, then became a social graph. chen PMF showed up in the network layer, not the feature list.

You Cannot A/B Test a Full Room

The physics predates pitch decks. A telephone alone is furniture. A telephone when someone else might answer is infrastructure. Katz and Shapiro formalised the mechanism in the American Economic Review in June 1985: when utility rises with the number of other people using a compatible product, consumers must form expectations about how large each network will become before they commit. jstor At early stage that expectation is your problem. A founder interviewing ten users in isolation learns whether one person likes the UI. A founder seeding an atomic network learns whether ten people in the same workflow stay once an eleventh joins.

Metcalfe's Law, the shorthand that network value scales with the square of users, is contested. Briscoe, Odlyzko, and Tilly argued in IEEE Spectrum in July 2006 that n² dangerously overstated value during the dot-com boom and proposed n log(n) instead. ieee The exponent is debatable. The direction is not: value grows faster than linear as connections multiply. Pre-PMF teams misread polite trial users as fit because they never held the graph constant long enough to see anti-network effects churn the room back to zero.

The Fuller Room Won

Hold both truths at full strength. Truth one: merit still matters at seed. Clear UX, honest pricing, and fast support win the first conversation. Truth two: in networked categories, PMF is a density reading, not a feature score. WhatsApp did not win on feature depth. It won because your contacts were already there. That is personal utility network effects: value rises with the people you already know.

When expectations shift, the scoreboard changes. Parker, Van Alstyne, and Choudary wrote in Harvard Business Review in April 2016 that demand-side economies of scale, their term for network effects, are the driving force behind platform strategy. hbr By January 2015 Apple's App Store offered 1.4 million apps. By 2015 the iPhone alone generated roughly 92% of global profits in mobile phones while most legacy handset makers earned little or none. The lesson for a pre-PMF founder is not "build a platform moat overnight." It is harsher: if your wedge depends on other people showing up, a solo user's delight is necessary and insufficient. Retention in an empty room is not PMF. It is a false positive.

VC Bets on Cold Start

Venture firms do not fund funnels. They fund businesses where each new user raises the value of the last once density arrives. Andreessen Horowitz published Chen's book because network effects are the asset class. NFX exists to name, measure, and fund them. nfx At seed, the bet is rarely "will this product work for one person?" It is "will this team reach an atomic network before runway runs out?"

The attention economy runs the same physics at scale. Meta alone generated $160.6 billion in advertising revenue in 2025 by auctioning predicted attention slots, not by selling you products directly. pwc Creators who farm content are feeding the hard side of a network someone else owns. I have written about how engagement dashboards treat users as inventory to optimise, not people to serve. gaganmalik-contempt The early-stage mirror is simpler: if your PMF hypothesis requires a crowd, buying traffic into an empty room is not validation. It is renting strangers who leave when the subsidy ends.

I Sold Him Polish. He Needed a Wedge.

Marcus, the CEO on that call, is a composite built from three founder engagements I advised between 2022 and 2024. The incentives are not. His team had shipped an AI productivity app with genuinely thoughtful design. They were pre-PMF and cold: no integration density, no vertical wedge, no plan for the first fifty power users in one founder community. I sold what I knew how to sell. Usability studies. Onboarding audits. Task-completion tests. All polish levers. All legible on a monthly report. What Marcus needed was Chen's atomic network: the smallest group where one more person makes the product harder to leave for everyone already inside. a16z We never mapped it. After nine months the co-founder left. Runway gone. Users were polite and sparse. I kept invoicing until the account closed. Marcus did not fail because the product was ugly. He failed because the team never reached the density where retention could prove PMF.

Your First Fifty Are the Lab

You do not need a billion-user platform to run this at seed. You need a graph small enough to hold in your head. Ten finance teams in London sharing templates in a private channel. Three colleagues on Slack who cannot leave without losing the thread. A newsletter where replies and forwards create recognition among readers. Each is an atomic network. PMF shows up when one more member raises the cost of exit for everyone already there.

The mistake at early stage is copying growth playbooks from companies that already cleared cold start. Your job before PMF is not "scale acquisition." It is name the wedge, seed the room, and read retention inside that room only. Chen's book sits next to venture decks for a reason: investors are betting on whether you can reach stage two before cash runs out. If you are pre-PMF in a networked category, your roadmap question is not "what is step three of the funnel?" It is "who has to be in the room together for this product to stop feeling empty?"

The Sapling Does Not Grow Alone

You have felt this without the vocabulary. A colleague leaves a Slack channel and the room goes quiet. A group chat dies when two people stop replying. The product did not break. The density did. Walk through an old-growth forest and you see individual trees. You do not see the mycorrhizal network beneath the soil, the fungal web that moves water and nutrients between roots. Foresters sometimes call it the wood wide web. A sapling plugged into that web inherits resources it cannot reach alone. Pull it out and it wilts, even with the same sunlight as its neighbours. At seed stage the canopy is your demo. The mycelium is the atomic network you are trying to stand up before you declare PMF.

That analogy does not claim every company is a forest. A solo SaaS tool can find PMF with one user and a workflow. The mistake is planting a networked sapling and calling polish a strategy while the soil stays sterile.

The Strongest Objection

The honest objection deserves a full paragraph. Many products find PMF without network effects. A solicitor's practice, a hairdresser, a direct-to-consumer brand, a single-player SaaS tool: these prove fit on quality, reputation, and distribution. Metcalfe's n² was dangerous hype. Briscoe, Odlyzko, and Tilly showed in IEEE Spectrum in July 2006 how that arithmetic fed dot-com overbuild. ieee Forcing viral loops onto products that never get more valuable when strangers join is how pre-PMF teams waste quarters chasing mechanics that will never close. If your category is genuinely single-player, network vocabulary is category error. Not everything is Uber. Not everything should be.

That objection is correct about fit and incomplete about stakes. Parker, Van Alstyne, and Choudary argued in Harvard Business Review in April 2016 that when a platform business built on network effects enters a market that never gets better when strangers join, the platform nearly always wins. hbr The iPhone did not merely outsell Nokia on specs. By 2015 it captured the profit pool because developers, users, and accessory makers expected the largest graph around the device. Legacy handset makers owned factories. Apple owned the connection graph. The debate at seed is not network effects versus no network effects. It is whether your PMF hypothesis requires density you are not building on purpose. Marcus's category was networked enough that integrations and shared habits mattered. He tuned the deck for conversion. The incumbent reached the room where retention could compound.

In a networked category, product-market fit is not a funnel outcome. It is a density outcome. If you cannot name your atomic network and who pays the cold-start tax, every pound on polish buys polite trials in an empty room. Marcus is preparing for product interviews at companies that already found PMF in the wedge he never mapped, and his co-founder is back in Palo Alto billing funnels for founders who still trust them.