Many small rooms

A lot of things that are now used by nearly everyone started out belonging to a small group. It is tempting to read that as a strategy: win a niche completely, then spread outward from it. For products that are worth something only when other people use them, half of that is right. A niche is where such a product first becomes complete, and that is why its first users stay. But the niche is not what carries the product to everyone else. One company's data, the best we have on the question, suggest that people join something when it reaches them from several separate parts of their life, and that a tight group of friends, however enthusiastic, counts as roughly one voice.

A messaging app with one user is useless, and so is one with a million users none of whom you know. What matters to any one person is how many of their own people are there. Launching to everyone at once spreads the first users thinly. Each newcomer finds few of their contacts and leaves, so the network never fills in anywhere.

Facebook is the standard example of doing the opposite. It was created at Harvard in February 2004 and reached other US colleges in stages over the next two and a half years, opening to the general public only in September 2006.¹ The staggering was enforced by requiring a verified college email address.¹ At the first hundred colleges, by September 2005, there were on average 86 Facebook users for every 100 undergraduates.¹ Inside each campus the product was close to complete: nearly everyone a student might want to reach was there.

The economists who reconstructed that rollout, Luca Braghieri, Ro'ee Levy and Alexey Makarin, list three likely reasons for it, drawing on earlier accounts: limited server capacity, a wish to keep a flavour of exclusivity, and a desire to keep "the fraction of users who likely knew each other offline artificially high".¹ The rollout was also not random. More selective colleges got access earlier.¹ If those reasons are right, the famous niche launch was partly capacity, partly status and only partly a deliberate bet on density, which makes it hard to copy by intention alone.

What actually pulled people in

The expected next step is that a dense niche spreads through its edges: friends at other schools, siblings, colleagues, each arriving with some of their people already there. Facebook's own data complicate that.

In 2012 Johan Ugander, Lars Backstrom, Cameron Marlow and Jon Kleinberg studied 54 million invitation emails sent by Facebook users in the summer of 2010.² Two of the authors worked at Facebook, and the data are the company's.² For each person invited, they looked at the people already on Facebook who had that person's email address, and asked what predicted whether the invitation was accepted. The answer was not how many such contacts there were. It was how many separate groups they formed: contacts who were not connected to one another.² Once the number of separate groups was held constant, having more contacts was, in the authors' words, "largely a negative indicator" of joining.²

The finding was sharper still. Two contacts who were not friends on Facebook but had been tagged in the same photo converted the invitee at about the same rate as two contacts who were friends.² People who had been at the same event behaved, for this purpose, like one source.

The same pattern held for staying. Among about ten million people who joined in 2010, those with no tight group of friends on the site a week after joining were much less likely, three months later, to be using it on at least six days a week.² Those with several separate groups did better again, by what the authors call "a sizable additional increase".² Their summary of the lesson for anyone trying to spread a behaviour is that "it may be less important that they receive many endorsements than that they receive the message from multiple directions."²

The floor and the route

So the niche changes job. A dense group is the floor: without at least one place where most of your people are, a networked product does not hold you. It is not the route. Ten close friends recommending something reach an outsider as a single recommendation, repeated. What makes someone outside join is meeting the same product in their family, then at work, then among old school friends, as separate and unrelated signs. Mass, on this reading, is many small rooms that happen to contain the same people.

The college rollout may have worked less because each campus was dense than because there were many dense campuses at once. A student whose high-school friends had scattered to a dozen colleges would hear about Facebook from several of them, each group unconnected to the others. Nobody has measured that for 2004-06. The invitation data come from 2010, six years after the college launch.²

The strangest consequence concerns the devoted niche. A group that adopts something completely and talks mostly to itself produces exactly the kind of endorsement that, in this data, moved outsiders least: many voices from one place. The more a product belongs to one community, the more its recommendations sound like one community's taste. That would explain products that are adored by a group and never leave it, though how often that happens, and how often a niche did scale, is not known.

The evidence is one company's, at one moment, through one channel; the authors note that email invitations were only a small part of Facebook's growth.² Products that are useful to a single person, a calculator or a note-taking app, never needed a network at all, and for them "start with a niche" is a question of focus, not mechanics. And every story about a product that started small is told by a survivor.