Todd Yellin is pretty sure his wife would have never watched Jessica Jones if he hadn’t tricked her. She’s not a fan of shows based on comic books, and even though he’s a Netflix exec, nepotism wouldn’t have been enough to make her tune in. Yet he knew she’d love it. So Yellin did what Netflix often does to woo users: He played to her interests by mentioning the strong female lead and rave reviews.
It worked. They were three episodes in before she noticed the Marvel logo that opens the show, but by that point she was hooked.
As Netflix’s VP of product, figuring out what people like is Yellin’s job—and that often means spending a lot of time with data scientists and machine learning engineers. But his efforts to parse the data coming from the The Defenders, which brings the superheroes Jessica Jones, Luke Cage, Iron Fist, and Daredevil together to defend New York, started at home. “The unique opportunity The Defenders gives us is because of the team-up thing,” Yellin says. “My wife wouldn’t watch Iron Fist or Daredevil. But because she likes Jessica Jones should we suggest Defenders to her?”
More importantly, should Netflix suggest a show like The Defenders to you?
The Defenders provides Netflix with a unique case study. Instead of merely allowing it find out of someone who likes, say, House of Cards also will like Daredevil (yes, BTW), it tells them which of the people who landed on Daredevil because of House of Cards will make the jump to The Defenders. Yellin’s wife enjoyed Jessica Jones because she likes female protagonists. Will that apply to a show where her beloved character is part of an ensemble? Will future crossover shows draw viewers? And, if so, which viewers should Netflix target? Those are just a few of the questions Yellin hopes to answer with data from The Defenders.
Wildly different programs lead people to The Defenders’ standalone shows. The top lead-in show for Luke Cage? Narcos. But for Iron Fist, it’s a Dave Chappelle special. Someone who watches Jones probably will watch Cage, but beyond that the groups of people—Netflix calls them “taste communities”—gravitating toward those shows enjoy very different programming. “Take someone like my 75-year-old mother-in-law, who enjoys Grace and Frankie. Are we going to necessarily suggest Iron Fist to her? Absolutely not. But there is going to be a subset of people it appeals to, so we have to get smart about which people,” Yellin says. “So when something like The Defenders comes out, we’ll look at what percent of people we showed it to actually clicked play, what percent of people had to go and look for it when we should’ve presented it on their homepage at the top.”
That Grace and Frankie example isn’t a joke. Internal data shows that lots of people who like the Jane Fonda-Lily Tomlin vehicle also dig Iron Fist. The taste community for Luke Cage, on the other hand, draws from Stranger Things and Easy. If you like Jessica Jones, you probably watch Master of None. And people who watch Friends tend to enjoy both Jones and Daredevil. (Maybe they really love establishing shots of New York City.) Netflix has some 2,000 taste communities worldwide, and every time the company’s algorithms get a data set they get smarter about recommending stuff to those communities. Defenders lets Netflix learn more about how they overlap.
That knowledge will eventually make its way to your queue. Every Netflix user belongs to three or four taste communities. It’s easy to say that this influences what appears in your recommendations, but it’s not quite that simple. Membership in those communities does more than dictate the top 10 comedies appearing in a row of your queue, it determines whether comedies appear there at all. “It determines what rows we show you and the order of the titles in those rows,” Yellin says. “You might get drama as your second row; I might get it as my 15th. Someone else might not even see a drama row.” Each time you open Netflix it exposes you to 40 or 50 titles. Netflix considers it a win if you choose one of them.
The Defenders premiered last week, so it will be awhile before Netflix gains an exhaustive understanding of viewers. But people like Yellin are watching the data closely. A thorough understanding of it will make Netflix more adept at recommending programs, yes, but it will also provide Netflix with a better idea of the shows it ought to make in the first place. Netflix has never brought stars from four shows together in an ensemble cast before, and the experiment could help determine whether it should do it again. Only a small fraction of viewers watch all four programs that inspired The Defenders, but many people watch at least a couple of them. Yellin hopes to learn “what happens when you put the mélange together” and what that might mean for future team-ups.
“Even if you look at historical things like the Avengers series in movies that Marvel does—there you can do big surveys, but you never know exact behavior. But we will,” Yellin says. “What we’re going to be learning as a company from both the algorithm side and the content side is what happens when you pair up shows that aren’t exactly the same thing.”
Get ready for a much more refined taste community in Hell’s Kitchen.