Behind the Build: How did we get from spies to marketing?

About 15 years ago, I was doing research on gamers and I came into possession of data from the game EverQuest II, thanks to Raph Koster and Scott Hartsmann.

This was back in the day when “big data” wasn’t yet a thing, and when game data was used for debugging, not analytics. For a team of researchers, it was a very exciting time. They had data on what people did, and that just like never happened. And they also had data on who was connected to whom, which also never happened. That lets you build a network. 

We started building predictive models out of these data. Things like “Hey, a player who gets to level 12 and plays a troll is more likely to _____”. You could predict all kinds of things. Personality, voting habits, age, gender, and, interestingly, whether the player could speak Russian or several other languages.

That piqued the interest of our spy agencies, who weren’t so much interested in finding spies in games (they would be later), but in learning how to build these kinds of models. So they funded our work, and for several years we were able to play around with troves of data, and connect them to surveys. That work led to a bunch of dissertations, careers, and papers. One of those was Nishith Pathak’s brainstorm of how we might use those network graphs. His insight was that when a player did something, it might predict what the other players they were connected to would do. It did, and the rest is, well, if not history, it’s what we’re still working on today.

Knowing which people cause others to act is the holy grail of marketing, and it started with a bunch of nerds playing video games, then working with spies. Truth is stranger than fiction.