Cold Start Problem

Metcalfe and Reed explain why network effects are powerful once you have them. Neither explains how to get them — and this is where most network products quietly die.

The Cold Start Problem is the observation that you can't start with a big network. You have to start with what Chen calls an atomic network: the smallest group of users that's genuinely useful to itself without needing anyone else to join.

For Uber, the atomic network wasn't 'a city'. It was closer to '10 drivers and 100 riders in three square miles around the financial district on a Wednesday night'. For Slack, it wasn't 'a company'. It was 'one 8-person team that would use it every day and stop using email for internal chat'.

Your job is to find the smallest version of the network that's useful inside itself, launch there, get it stable, and only then worry about the next atomic network next door. Teams that skip this and try to 'launch broadly' end up with a network that's evenly thin everywhere and valuable nowhere.

A 'social investing' feature launches on a brokerage app: users can follow other users, see their trades, comment on positions. The rollout is to all 4 million users simultaneously, and marketing runs a campaign about the social experience.

Three months post-launch: 2.1% of users have followed anyone, 0.4% post publicly, and engagement with the feature is below the threshold to justify its maintenance cost. The atomic network was never found.

Contrast with the successful pattern: launch to a single curated community of 200 users who already talk to each other outside the app (a trading Discord server, a university investing club), wait until that community is self-sustainingly posting 5+ times per day, then allow controlled invite-based expansion to adjacent communities.

The first version spreads evenly and dies. The second compounds — because at every stage the network has enough density to be fun without reinforcement from outside.