Netflix is always testing and scheming. Each time you click play, pause, or – heaven forfend – stop watching TV altogether, it gathering data on your preferences. Spread across more than 300 million user profiles, this is a colossal amount of information. And it all feeds back into what you see when you next look for something to watch.
Todd Yellin, the vice president of product who has been at the company for ten years, explains that data informs everything Netflix does. "If you click play nowadays in the streaming world, it tells volumes more information that is a lot less superficial than getting someone's gender and age," he says. Netflix doesn't include age or gender in its recommendation system as it doesn't believe they're useful.
Instead, here are some of the ways Netflix and its algorithms customises what you watch.
Ready for some bingeing? These are the best shows on Netflix, the best films on Netflix and WIRED's guide to what is new on Netflix.
A/B tests
Netflix runs 250 A/B tests each year, Yellin says. These tests present users with two slightly different experiences to see how they respond, varying from changes to the way the Netflix player looks or the mechanisms by which people find shows.
These are partly why one person's experience of Netflix can be completely different to another. Tests give one set of users one experience and another a slightly different one: this can vary from how the Netflix player looks, design, and how people find shows.
Around 100,000 people are randomly selected for each test, with another 100,000 used as a control group. If one version gets more people watching a show, Yellin says, it may be incorporated across the whole service.
Landing cards
Netflix creates multiple different landing cards – the images that are shown as people scroll through shows – for each of its titles. The idea is to find the most popular options. One day Stranger Things' landing card may be an image of lead character Eleven, the next time you come to look at it Dustin, Lucas, Mike and Will may be featured.
The landing cards people click on the most are adopted more widely. The next evolution is likely to be find success of autoplaying trailers. "We're going to likely test on that," Yellin says.
Read more: Inside Netflix's $8 billion plan for world entertainment domination
Recommended shows
On average a person views 40 to 50 titles before they pick what they're going to watch. Perhaps the biggest personalisation in Netflix is the rows of shows a user is presented with. These are largely based on watching history
"There are tens of thousands of rows we can show you," Yellin says. "I can show you new releases row I could show you an action adventure row, I can show you a witty action adventure row or dark action adventure row or a silly action adventure or romantic comedy or romantic drama". Some of the rows can seem a little bizarre: corporate corruption, raunchy comedies, ripped from the headlines, European TV dramas based on books are just some examples.
No user will be shown exactly the same combination rows, but Netflix will occasionally throw in new shows and types of shows it thinks a person may be interested in. If you watch a lot
If you start watching a series but don't get to the end of it, Netflix's algorithm will occasionally resurface the unfinished show in a bid to tempt you back onboard. "We're not going to bury it in the UI," Yellin says. "We're going to do more occasionally, and it will get less frequent, show that to you and give you a little nudge."
Timing
Netflix customises its recommendations based on when you're watching. Yellin doesn't reveal how much of a difference it makes if you open the app at 23:00 at night compared to 17:00, but says recommendations are weighted differently depending on the time.
"We experiment with a lot of signals," Yellin says. Netflix may show you shorter programmes, or ones you're halfway through, when you login late at night and may not be looking to watch an entire show from scratch.
This article was originally published by WIRED UK
