Some businesses would love to get basic information like your name, age and gender to better understand how to market to you in the future. Netflix could care less.
Netflix used to place an emphasis on collecting these types of biographical details about its users, but it eventually decided the data wasn't particularly useful. "It really doesn't matter if you are a 60-year-old woman or a 20-year-old man because a 20-year-old man can watch Say Yes To The Dress and a 60-year-old woman could watch Hellboy," Todd Yellin, VP of product innovation at Netflix, told Mashable in an interview this week.
In recent years, the video streaming and delivery service has instead focused on tracking the kind of data that users wouldn't even think to provide themselves. It tracks what you like and just as importantly, it increasingly tracks what you don't like.
The goal of any video service is to offer the best selection of content, but the heart of Netflix's business is making sure users can easily find the right video to watch at the right time. The former might interfere with the latter if not for the centerpiece of the Netflix experience: recommendations.
Between 75% to 80% of the videos that Netflix users end up watching on the service come directly from the company's recommendations about what to watch next. To put that another way, just one fifth of the content viewed on the site is from users visiting Netflix and choosing to go through the steps of typing out the name of something they want to see. The better the suggestions Netflix can make, the more videos users will stream, and the more customers will want to continue paying for the service.
To get to the point where it could make better suggestions, Netflix had to move away from relying mostly on what Yellin calls "explicit" data like biographical information or asking users to say whether they prefer comedies to movies with sad endings. (The exception is the introduction earlier this year of user profiles, which are intended to tackle the problem of multiple members of one household using the same account and messing up the recommendations.)
"You want to take people at face value. When someone tells you they are always watching foreign films and documentaries, you want to show them that," Yellin said, noting that's how traditional retailers function. "The truth of it is that some people are posing. Some people are really showing you their aspirational self because some people just want to watch Christmas Vacation with Chevy Chase for the 15th time, and that's what they really want out of their night because it's been a long day."
With that in mind, Netflix started downplaying features like ratings and predicted ratings, and focused more on "implicit" data showing what users had watched and were likely to watch next. The most obvious thing Netflix tracks, at least obvious to anyone who glances at their homepage, are the videos users have watched previously. But Yellin and his team, which now includes "dozens" of data scientists and "hundreds" of engineers, also track things like the "velocity" of how fast a user makes it through a video and whether or not they stalled out five or ten minutes into it. They track whether the user is more likely to view an edgier sitcom, like It's Always Sunny In Philadelphia, late at night or watch comedies on a particular day of the week to better dole out recommendations.
More recently, Netflix has started to track how users scroll down the page and where they click to see which suggestions they ignore. "It's one thing to know what people play. It's another to know what they didn't play," Yellin says. "If we know what you saw in front of you, we can know how many times you saw that title." All the other data may suggest that the user should want to watch Skyfall, but if they repeatedly ignore it, Netflix will eventually stop suggesting it.
In short, Netflix is trying to know its users better than they know themselves.
That would have been virtually impossible when Netflix first launched back in 1999 as users only rented DVDs rather than streamed videos. But now, Yellin says, "That implicit data is becoming more and more powerful because we have more and more of it in the streaming world."
Sometimes Yellin and his team imagine what they could do if they processed exactly the data needed to make the perfect recommendation to each user at each moment.
"The perfect utopia that we joke about here... is why show thousands of titles? Why not just show one tremendous gorgeous image of one title because we've read your mind and know what it's going to be?" Yellin says. He adds that one could take this even a step further and just have a video cued up to start playing as soon as the user visits Netflix. "Could it happen that we have such confidence in what you want to watch that we start autoplaying something? We're working toward that."
Whether Netflix can or should ever reach that point is anyone's guess, but in the short term the goal is to keep improving Netflix's picks so that the number of videos viewed from recommendations begins to rise above 80%. "We're glad that 20% people search for [videos] on their homepage, but we'll keep bumping it down," Yellin says.
"Do we want it to be 100%?" he asked rhetorically when pressed on whether that's the end goal. "We want to make it super easy. It's hugely important for us and the consumer. They don't come to Netflix with a machete in their hand looking to chop through a ton of content. It's good for them because it makes it easy to find something great to watch and it's great for us because we want to win the moment of truth."
Image: Netflix