Finn Brooks: Hey — how's your week been, genuinely, because mine has been chaotic and I need this conversation to be a reset.
Clara Bennett: Mm, same kind of week — but honestly the prep for today helped, because this is a topic that just gets quieter and more interesting the longer you sit with it.
Finn Brooks: That's the right word for it — quieter. Because network effects sound like a hype phrase and then you hit a number like seventy percent of all tech value created since 1994 traces back to them, and it stops being hype really fast.
Clara Bennett: Seventy percent. And that's the number we should probably just let sit for a second.
Finn Brooks: No seriously — sit with it. That's not a contributing factor, that's basically the whole story of how tech generates value. And today we're trying to figure out what network effects actually are, where they came from, and — this is the part I care about most — why most people who think they have them are wrong.
Clara Bennett: Right, and the definition is more precise than people treat it. A product becomes more valuable for all existing users as additional users join — each new participant adds utility to the whole network. That compounding curve is the mechanism. It's not just 'more users equals good.'
Finn Brooks: Okay and the oldest version of that I found — this one genuinely surprised me — was Theodore Vail in AT&T's 1908 annual report. He wrote that a telephone without a connection is 'one of the most useless things in the world.' A hundred and fifteen years ago, someone already named the exact mechanism.
Clara Bennett: And useless-to-useful is a phase transition, not just an improvement — that's what makes the framing so precise.
Finn Brooks: Wait, say more on that — phase transition versus improvement, that's a real distinction?
Clara Bennett: In practice, yes. An improvement means the thing you had gets better. A phase transition means the category of the thing changes entirely — one telephone user means the telephone is worthless, two telephone users means it has a function. That's not incremental.
Finn Brooks: Which connects directly to the PayPal story — they hit that phase transition instantly because eBay's buyer-seller network was already past critical mass. Fewer than ten thousand users to five million in a single year. They didn't have to earn the phase transition, they inherited it.
Clara Bennett: And that's the central tension for the whole episode — network effects are the most powerful competitive moat in modern business, and also the most misapplied concept. Most platforms that claim them don't structurally have them. Working out that difference is what we're here for.
Finn Brooks: But here's what trips me up — if the mechanism is that simple, value grows as people join, then isn't that just... scale? Like, bigger company, more users, more good? Where's the actual line?
Clara Bennett: That's the conflation that breaks strategy. Scale economies make your unit costs cheaper — you manufacture more, each unit costs less. Network effects make the product itself more valuable to every existing user when someone new joins. Those are completely different mechanisms.
Finn Brooks: Wait — to every existing user?
Clara Bennett: Every single one. Now, imagine a group chat where you're the only member. It's worthless — not bad, not limited, worthless. Your friend joins. Now it has a function. Ten more join, and it becomes the coordination layer for your whole social life. The app didn't change. WhatsApp didn't ship a feature update. The people arrived, and that's what created the value.
Finn Brooks: Okay that is actually the cleanest version of this I've heard.
Clara Bennett: And it splits into two flavors, which matter a lot. Direct network effects — same side helping same side. WhatsApp, Facebook, the telephone network Vail was describing. Each new user on your side directly makes it better for you. Then there are indirect network effects, which are — actually, the Visa example is the one that snaps it into focus.
Finn Brooks: Walk me through it, because I always half-understand Visa and then lose the thread.
Clara Bennett: Two completely different groups — cardholders and merchants. More cardholders join, which means merchants have a reason to accept Visa, which makes Visa more useful to cardholders, which pulls in more merchants. Neither side is directly helping its own side. Growth on one side feeds the other. That's the cross-side loop — indirect network effects.
Finn Brooks: So it's not one network, it's two networks pulling each other forward. And if either one stalls—
Clara Bennett: The whole thing collapses, yes. That's also why the cold-start problem is twice as hard for platforms like Visa — you have to seed both sides simultaneously, or neither takes off.
Finn Brooks: Okay but wait — switching costs, brand loyalty, that stuff also locks people in. So how is a network effect actually distinct from, I don't know, people just really liking the product?
Clara Bennett: In practice, the test is this: if every other user left, would the product still be valuable? With brand loyalty — say, a well-designed piece of software — yes, it might still be worth using alone. With a real network effect, the answer is no. Facebook with one user isn't a worse Facebook, it's a non-functional Facebook. That's the distinction. Network defensibility isn't just that people stay — it's that leaving means abandoning the entire network, not just the product.
Finn Brooks: But that's — wait, that's the part that breaks my brain, actually. Because if leaving means abandoning the whole network, then how does any network ever get started? Like, day one, you are the only person there. It's worthless by definition.
Clara Bennett: That's the cold-start problem. And it's not a small operational hiccup — it's a genuine paradox. You need users to generate value, but you need to offer value to attract users. The chicken-and-egg runs all the way down.
Finn Brooks: PayPal — and I want to slow this down because I think people just say 'oh PayPal did eBay' and move on — but like, what actually happened was they didn't have a cold-start problem because eBay already solved it for them. The buyers and sellers were already there, already transacting, already desperate for a payment layer that wasn't a personal check.
Clara Bennett: Five million users in one year.
Finn Brooks: Five million. In one year! They didn't engineer a network — they plugged into one that was already past critical mass. That's not a cold start, that's a hot start borrowed from someone else's work.
Clara Bennett: And critical mass is the threshold — that's the specific thing. Before it, the platform is fragile, every user who leaves matters enormously. After it, the feedback loop becomes self-sustaining. PayPal crossed someone else's critical mass line. Most platforms have to find their own.
Finn Brooks: Which is — I mean, how often does that work? Like, is there usually a host network sitting there waiting?
Clara Bennett: Rarely. And the 1894 case is — this is the one that makes it feel real to me. Bell's original telephone patent expired, and within a short period, roughly 6,000 independent phone companies appeared across the country. Six thousand. And they were nearly worthless, individually, because you could only call people on your own network.
Finn Brooks: Wait — 6,000 separate phone networks? That aren't connected to each other?
Clara Bennett: Each one marooned. Which is the cold-start problem at industrial scale — every single one of those 6,000 companies had to independently solve it, and most couldn't. Vail's move at AT&T was to see that the fragmentation itself was the problem, and start acquiring them. Not because the companies were valuable — because the connections between them were.
Finn Brooks: He was buying network mass. Not assets, not customers — just the mass. That is actually kind of a wild reframe of what an acquisition even is.
Clara Bennett: And it's worth holding that question — whether cold-start is ever truly solved, or whether most platforms are just subsidizing early users until critical mass hits and then calling it strategy in hindsight. That tension between genuine network design and expensive imitation of one... that's where the winner-take-all debate gets a lot messier than it looks on paper.
Finn Brooks: But hang on — that tension you're describing, genuine network design versus subsidized imitation, that's actually where the winner-take-all thing falls apart for me. Because if network effects are so bulletproof, why do I have four messaging apps on my phone right now?
Clara Bennett: That is exactly the wrinkle. And Evans and Schmalensee — they published through CPI in 2017 — made the argument that economists and practitioners invoke network effects to predict monopoly outcomes without empirical grounding. Meaning: the theory says tip to one winner, the market says WhatsApp, Signal, iMessage, and Discord coexist.
Finn Brooks: Wait, an actual academic challenge to the monopoly assumption? That's — I did not know that had a name.
Clara Bennett: It doesn't just have a name, it has a really concrete lived example, which is — tell me about Thursday night.
Finn Brooks: Oh man, okay so — this literally happened. Trying to plan drinks with a group, and one person's on Discord, three are on WhatsApp, two prefer Signal because they're privacy people, everyone's technically on iMessage but nobody checks it — and the whole coordination just collapses across four platforms because nobody wants to be the one to say 'can we all pick one.' That's not a moat, that's just friction that everyone accepted.
Clara Bennett: And that phenomenon has a name — multi-homing. Users participating in more than one competing platform simultaneously. Which should be impossible if winner-take-all actually held.
Finn Brooks: So the network effect is real — those apps genuinely have them — but they're not producing a monopoly. Why not?
Clara Bennett: Because the lock-in mechanism isn't the network effect itself. It's what leaving actually costs. And this is — I mean, this is where I think the framing has to shift entirely. With Visa, you leave and you literally cannot transact at millions of merchants. The exit is structural. With Facebook, you leave and you abandon your social graph — every contact, every shared history. With Slack, you just open a second tab.
Finn Brooks: Wait — so the social graph is the actual lock, not the product.
Clara Bennett: Right. And WhatsApp is the clearest case of this. Even users who are genuinely dissatisfied — they stay because leaving means convincing their entire contact network to move with them. That's not inertia, that's a structurally painful exit. The switching cost isn't 'I have to relearn a new app.' It's 'I have to negotiate a collective migration with everyone I know.'
Finn Brooks: That's actually dark. Like, the product can get worse and it doesn't matter — you're trapped by other people's choices, not your own.
Clara Bennett: Which is exactly why NFX mapped sixteen distinct network effect types — because not all of them produce that kind of structural exit cost. Some network effects are strong enough to dominate, some just create friction that people tolerate. The depth of lock-in varies enormously by structure, and the winner-take-all outcome only follows from the ones where departure means abandoning something irreplaceable.
Finn Brooks: So the framing isn't wrong, it just — it only applies when exit is structurally painful. Not just inconvenient.
Clara Bennett: That's it. The formal term for the underlying mechanism is network externality — your decision to join confers value on other users without them compensating you for it. That spillover is real in all sixteen types. But the monopoly outcome? That only follows when the spillover runs in reverse on exit — when leaving takes value away from everyone else in your network, making their exit more likely too. That's the cascade. And that's what most platforms claiming network effects don't actually have.
Finn Brooks: And that's the part I keep getting stuck on — the same externality that makes the telephone worth having in 1908, that Vail was describing, that exact same spillover is what traps people on platforms they genuinely resent. It's not two different mechanisms. It's one mechanism doing both things at once.
Clara Bennett: That's the regulatory knot. When you're trying to decide whether a platform is creating real value or just exploiting lock-in, those two things look identical from the outside. Users stay. You can't tell whether they stay because the network genuinely serves them better, or because leaving costs too much.
Finn Brooks: Right — but the part that doesn't fit is, we built seventy percent of everything on this mechanism without actually settling that question.
Clara Bennett: From 1994 onward, yes. Once marginal connection cost dropped to essentially zero, the network externality stopped being a telecom curiosity and became the primary engine. And we layered an entire economy on top of it before anyone worked out where the value creation ends and the trap begins.
Finn Brooks: I genuinely don't know how you resolve that. Like — I don't have a clean answer. Whether it's the next generation of platforms, or regulators looking at Facebook right now, that distinction between real network value and pure lock-in is still... I mean, it's still the question. Nobody's cracked it.
Clara Bennett: Mm. It's genuinely open. And I think that's the honest place to land.
Finn Brooks: Vail named it in 1908. It's been a hundred and fifteen years. We're still figuring it out.
Clara Bennett: Good conversation today. Genuinely.
