Strategy.

The framework

1. 01

   Run the test — value to users, not popularity with them.

   Ask it in one sentence: when your next customer joins, does the product improve for the customers you already have? A messaging app clearly passes — every contact who joins makes yours more useful. A well-loved brand clearly fails — a million other buyers do nothing for your experience of it. Keep the distinctions clean: virality is users bringing users (a distribution engine), economies of scale are costs falling with volume (a supply-side edge), and a network effect is users enriching users. Businesses can have one, two or none, and each is defended differently — misdiagnosing which you have means defending the wrong thing.

2. 02

   Name the type — they are not equally strong.

   Direct network effects are the strongest: every user can connect to every other, so value compounds fastest — telephones, messaging, payment networks. Two-sided effects run through a marketplace: more buyers attract more sellers attract more buyers — powerful, but leakier, because sellers happily list on rival platforms too. Data network effects — usage improves the product for everyone through data — are real but usually the weakest and the most overclaimed, because the value of additional data plateaus long before the pitch deck admits it. Name yours precisely, because its type sets both how fast it compounds and where it leaks.

3. 03

   Solve the cold start with an atomic network.

   The curse of a network business is that the moat works in reverse at the beginning: below critical mass, each user finds an empty room, and the same maths that rewards density punishes its absence. Andrew Chen's The Cold Start Problem (2021) names the escape: the atomic network — the smallest group inside which the product is already self-sustaining — and the discipline of winning it completely before widening. Facebook launched to one campus, marketplaces launch in one city or one category. Constraining your launch to a niche is not thinking small; it is the only way a network gets born.

4. 04

   Defend density, not totals — and watch the multi-homing leak.

   Aggregate user counts flatter and mislead: a ride-share network is not one network but a city-by-city collection of them, and it can be dominant in one city while dying in the next. Value lives in local density — how connected each user is to the users who matter to them — so deepen engagement inside each cluster before adding new ones. Then watch the two leaks that drain even dense networks: multi-homing, where users run your rival in the next tab because switching costs them nothing, and stagnant clusters, where the network exists but nothing new happens inside it. A network that is easy to leave is a network you are renting.