Facebook simulates itself up a better, more gradual product launch

When you’re launching a new social media product, like an image-sharing app or niche network, common wisdom is to make it available to everyone as soon as it’s ready. But simulations carried out by Facebook — and let’s be honest, a few actual launches — suggest that may be a good way to kneecap your product from the start.

It’s far from a simple problem to simulate, but in the spirit of the “spherical cow in a vacuum” it’s easy enough to make a plausible model in which to test some basic hypotheses. In this case the researchers crafted a network of nodes into which a virtual “product” could be seeded, and if certain conditions were met it would either spread to other nodes or “churn” permanently, meaning this node deleted the app in disgust.

If you’re familiar with Conway’s Game of Life it’s broadly similar but not so elegant.

In the researchers’ simulation, the spread of the product is based more or less on a handful of assumptions:

• User satisfaction is largely governed by whether their friends are on the app
• Users start using the app at a low rate and use it either more or less based on their satisfaction
• If a user is unsatisfied, they leave permanently

Based on these (and a whole lot of complex math) the researchers tried various scenarios in which different numbers and groups nodes were given access to the product at once.

It wouldn’t be unreasonable to guess that under these basic conditions, giving it to as many people as possible (not everyone, since that’s not realistic) would be the right move. But the model showed that this isn’t the case, and in fact creating a few concentrated clusters of nodes had the best results.

If you think about it, it becomes clear why: When you make it available to a large number of people, the next thing that happens is a large die-off of nodes that didn’t have enough friends at the start or whose friends weren’t active enough. This die-off limits the reach of other nearby nodes, which then die off as well, and although it doesn’t start an extinction-level event for the virtual app, it does permanently limit its reach due to the number of people who have churned.

On the other hand, if you seed a few clusters that are self-sufficient and keep usage high, then introduce it to others adjacent at a regular rate, you see steady growth, low churn, and a higher usage cap since far fewer people will have bounced off the product at the beginning.

You can see how this would work in real life: get the app to a few small, active communities (socially active photographers, celebrities, or influencers and their networks) and then create adjacent nodes through invitations sent out by existing users.

Turns out, lots of apps already do this! But now it’s supported by science.

Will this affect the next big Facebook product rollout? Probably not. Chances are the people in charge have a few other factors that figure into these decisions. But research like this, simulating crowds and group decision-making, will surely only increase in accuracy and usage.

The study, by Facebook’s Shankar Iyer and Lada A. Adamic, will be presented at the International Conference on Complex Networks and their Applications.