Cold-start problem

The cold start decides which products get the chance to have network effects at all. A network effect cannot begin until the product works for a small group, and that group is the only asset the founders control at launch. Uber’s 2019 filing described starting with supply to build marketplace liquidity and said incentives on both sides could reduce margins before sufficient scale.

Density matters more than headcount. Andrew Chen argues that growing city by city, campus by campus, or team by team can create relevant connections, raise engagement, and make referrals more likely. A national launch with a thousand scattered users in a thousand places produces none of this.

The methods that work are usually not scalable, and Chen says as much: launch in the simplest form, target a tiny network, and do whatever it takes even if it is unscalable or unprofitable. Reddit’s founders made up users. Uber’s local managers used an internal tool called Starcraft to text drivers to go to a train station where riders were gathering. These look like hacks, and they are deliberate bets on the earliest density.

Finally, the cold start shapes cost. Each subsidy or hand-built activity is paid for before revenue exists. A company that cannot say what happens when the subsidy stops is buying activity, not building a network.

Solving the cold start by handing out the product can bring the wrong participants. The Fresno mailing was meant as a controlled experiment, but within months the bank heard that a competitor was planning a similar program and expanded fast. The 99% Invisible episode says fraud and theft were rampant, that the bank had mailed cards to basically every one of its customers in Fresno and then across California, and that about one in four cardholders were not paying their debts. The problem was not a lack of users. Bank of America had to set up a collections department and an anti-fraud unit, revoke cards from people who did not pay, and wait until 1961 for a profit. Federal law later made it illegal to mail cards to people who had not asked for them.

Reach does not substitute for density. Google said in 2018 that its consumer social network had not achieved broad adoption, and that nine in ten user sessions lasted fewer than five seconds. Google announced the shutdown of the consumer service. It did not say why sessions were short; the figure is consistent with visitors finding too little relevant activity, but does not prove that explanation.

Subsidies buy density that may not last. Uber’s filing said it had used driver incentives and consumer discounts, had incurred significant losses since inception, and faced rivals that could use incentives to weaken its liquidity advantage. The question after a cold start is whether users stay when an incentive ends. A network built on a payment is a rented one.