Network effects
A Socratic walk-through of network effects — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why does a product become more valuable simply because more people already use it?
Most goods are worth what they do for you. A kettle boils water whether one person owns it or a million do. So consider the odd case: a messaging app that nobody else uses is worth nothing at all, and the app itself has not changed. Where is that value coming from, if not from the thing?
Reasoning it through
REASONING #Start with the simplest version. The value of a telephone is the set of people you can call, so each new subscriber adds something to every existing subscriber's holding. Value rises with adoption not because the product improved but because the reachable set grew. That is a direct network effect: users benefit from users of the same kind.
How fast does it rise? Metcalfe's old claim was roughly the square of the number of users, on the reasoning that a network of n people has about n squared possible pairs. It is a useful intuition and almost certainly an overstatement — Odlyzko and Tilly argued that value grows more like n times the logarithm of n, because you do not value all connections equally; most of your calls go to a handful of people. The shape matters less than the direction: the curve bends upward, at least for a while.
Now the second kind, which is where most modern platforms live. A games console is not more fun because other people own consoles. It is more valuable because developers write games for the console that has an audience, and players buy the console that has games. Two distinct groups, each valuable to the other. That is an indirect, or cross-side, effect, and it is the analytical heart of two-sided markets — Rochet and Tirole's work is the standard reference. It explains something a one-sided view cannot: why a platform will deliberately lose money on one side, subsidising developers or riders, to build the side the other side wants.
Given a loop like that, what should competition look like? Not smooth. If a small early lead makes the product genuinely better for the next joiner, the next joiner is more likely to choose it, which enlarges the lead. Positive feedback compounds small differences, so the market can tip: the share does not settle at a comfortable split but runs toward one winner. And expectations become part of the mechanism. If everyone believes a platform will win, joining it is the correct individual choice, and the belief makes itself true. Two platforms of identical quality can end with wildly different fates depending on which one people expected to prevail.
Once tipped, why does a better rival not simply displace it? Because switching is not an individual act. Leaving costs you the network unless the network leaves with you, and no one moves first. That is lock-in, and it is a coordination problem rather than a quality problem.
The analogy
THE ANALOGY #A language works this way. Nobody adopts a language because its grammar is elegant; they adopt it because of who they can talk to. That single fact explains the tipping toward a lingua franca, the failure of superior constructed languages, the difficulty of spelling reform, and why bilingualism is the practical hedge.
No one owns a language or sets its price, whereas a platform has an operator who can actively engineer the loop — paying one side to join, restricting who may connect, or bundling the network with something else — and a language tips over generations while a platform can tip within a year.
Clarifying the model
THE MODEL #Three qualifications, all of which the enthusiastic version of the story tends to skip.
First, the curve does not rise forever. Additional users can impose costs on existing ones: congestion on a road, noise and spam in a forum, moderation load, or simply the fact that a crowded marketplace is harder to be found in. These are negative network effects, and they can flatten or reverse the curve well before the whole world has joined.
Second, users need not pick one. Multihoming — carrying two ride-hailing apps, listing on two marketplaces, accepting several card networks — weakens tipping considerably, because it lets a challenger get in front of users without asking them to abandon anyone. Where multihoming is cheap, "winner takes all" is usually wrong.
Third, and most often confused: many advantages described as network effects are really economies of scale. Falling unit cost with volume is a supply-side property. It is a real advantage, but it does not make the product better for the user as others join, it does not produce self-fulfilling expectations, and it can be competed away by a rival who reaches efficient scale. The diagnostic question is direct: if a new user joins, does an existing user become better off? If the honest answer is only "the firm's costs fall", it is scale, not a network effect.
Related is locality. Many effects are local rather than global — what matters is whether your friends or your city's drivers are present, not the worldwide total. Locally-structured networks fragment into regional winners rather than one global one, which is why several platforms with strong effects nonetheless coexist in different countries.
A picture of it
THE PICTURE #How to readBoth lines start near zero, because a network with nobody on it is worth nothing regardless of how good the product is. The upper line is the textbook case: value accelerates as the reachable set grows, then flattens once the people you actually wanted are already there. The lower line is the same product with congestion — moderation load, noise, competition for attention — which turns the curve down past a point, so the best size for a user is not the largest size. The units are illustrative; only the two shapes are the claim.
What became clearer
WHAT CLEARED #The value sits in the connections, not the product, which is why adoption feeds itself and why markets with strong effects tip rather than settle. But the same reasoning bounds the story: the loop runs only while extra users help rather than crowd, only while switching means leaving people behind, and only where the benefit is genuinely demand-side. Ask of any claimed moat whether a new user makes an existing user better off — most of the time, that question is the whole analysis.
Where to go next
ONWARD #- How a platform decides which side to subsidise, and what it charges the other side.
- Why interoperability and number portability are the standard remedies for lock-in.
Key terms
TERMS #| Term | What it means |
|---|---|
| Direct network effect | value to a user rises with the number of users of the same kind, as with a telephone network. |
| Indirect (cross-side) network effect | value to one group rises with the size of a different group on the same platform, as with console owners and game developers. |
| Tipping | the tendency of a market with positive feedback to run toward a single dominant option rather than a stable split. |
| Multihoming | using several competing platforms at once, which weakens both tipping and lock-in. |
| Economies of scale | falling unit cost as output rises; a supply-side advantage frequently mislabelled as a network effect. |
Every term the collection defines is gathered in the glossary.