Cold Start Problem

Products that work fine alone never run into this. A budgeting app does its arithmetic whether it has 100 users or 100,000, and one person gets the same thing out of it either way. The early job is simply to let people know it exists.

The trouble starts with products that need other people to be worth anything. Open a secondhand marketplace and week one has twenty listings on it. Buyers arrive, find nothing to buy, and leave. Sellers see no buyers and don't bother listing. Both sides are waiting for the other to show up first, and neither does.

The state got a name as these products multiplied through the 2010s. Until then, weak early numbers were read as not enough marketing or a weak product, and it became clear there was a third case where neither explanation fit. No amount of advertising helps when the thing people arrive at is empty — they arrive and leave at the same rate.

It usually surfaces as a conversation like this. An investor asks why churn is so high among signups, the founder says there aren't many listings yet, and the next question is why listings aren't growing. The answer is that there aren't many buyers. Whether there's a plan to break that loop is the real question being asked.

The mechanism is the network effect run backwards. A network effect means each additional user makes the product more valuable to every other user, which also means that when there are few users, each one gets correspondingly little. Same property, opposite end.

The loop turns in four beats.

1. Few users, so the product delivers little value. You installed a messaging app and the contact list is empty.
2. Low value, so people who arrive leave within days. They installed it; nothing gives them a reason to open it.
3. They leave, so the user count doesn't grow.
4. It doesn't grow, so the value stays where it was. Back to step 1.

The loop has a line running across it. Past that line, value grows on its own and users start bringing in users. Short of it, everything reverts the moment you stop paying for traffic. It's like lighting damp wood: hold a flame to it long enough and it catches by itself, pull away early and it goes out.

There are broadly three ways across. The first is shrinking the arena. Don't launch nationwide — run inside one neighborhood, one campus, one company first. A thousand users scattered across a country produce nothing; the same thousand inside one university are already plenty.

The second is filling one side by hand. Staff post the early listings themselves, or suppliers get paid to show up before there's demand. It doesn't scale and it isn't supposed to — it only has to hold until the other side arrives on its own. The third is shipping something useful to a single user first — make the photo editing good enough on its own, gather users, and open the sharing features later.

Whether you've crossed the line shows up in the viral coefficient, the average number of new users each existing user brings in: 100 users bringing 30 makes it 0.3. When that number starts climbing and unpaid word of mouth mixes into it, the loop has begun turning the other way.

The classic case is the marketplace, a product that connects suppliers with buyers. Because it sits on a two-sided market — one that only functions when both sides are present — the loop runs in two layers at once. For food delivery: no restaurants means no orders, and no orders means no restaurants sign up. Which is why early delivery companies pick one neighborhood, sign twenty restaurants directly, and advertise only there.

Social products and communities hit the same wall. No posts means nothing to read, and no readers means nobody writes. That's why the founding team so often writes the early posts and answers the early comments themselves.

Collaboration tools take a slightly different shape. They arrive company by company, so what matters isn't total users but how many people inside one team are in. Five users out of a 500-person company will eventually leave too, because the people they work with aren't there. The same five inside one 20-person team have everything they need, which is why adoption tends to spread team by team rather than evenly across a company.

Recommendation features use the phrase as well. A brand-new account has no history, so the system has no basis for choosing what to show. The usual answer is to ask about interests at signup, or show what's popular overall, and move to personalized results once behavior accumulates.

"The numbers aren't a marketing problem right now, they're a cold start problem."

More advertising won't help while what people land on is still empty; the spend converts into arrivals that leave again.

"Let's get past the cold start one neighborhood at a time. Nationwide comes after."

A proposal to narrow the arena and build density inside it first.

"We solved the supply-side cold start by signing the first 200 venues by hand."

One side was filled manually to break the loop.

At Pagedeck, a feature for publishing documents publicly and sharing them outside the company launched last year. The hope was a loop: public documents accumulate, search traffic finds them, visitors create accounts, and those accounts publish more documents.

Month one produced 40 public documents and about 12 search visitors a day. With nothing to read, most of them left. Mia narrowed the scope to one document type — product specs — and Grace asked ten companies she knew to seed 60 good ones. Three months later search traffic was around 300 a day, and some of those visitors had started publishing documents themselves.

The most common assumption is that more users solve it. It looks like a problem of small numbers, so that reads as the fix. But 10,000 users scattered across a country produce less value than 500 inside one neighborhood. Density within one pocket matters more than the total. Spend the advertising budget broadly and users arrive and drain at the same rate, leaving only the invoice.

The other is expecting it to resolve with patience. It feels like things should accumulate slowly over time, but while the loop is turning, arrivals keep leaving, so nothing accumulates. Somebody has to push the product up to the line, and it's worth deciding in advance where that push comes from.