How Netflix knows what you want to watch before you do
Jordan Canning is one of the most influential taste-makers in Canada, even if most of the 1.5 million people who rely on her judgment and advice every month have never heard of her.
Ms. Canning spends hours upon hours watching Netflix content that is about to be made available to the online video service's subscribers, meticulously classifying movies and television shows and filling in the spreadsheets that ultimately determine what programming is offered to each Netflix user when they log into the service.
She is one of four Canadian taggers – content experts for hire who are the brains behind the intricate algorithms that predict what viewers in this country are likely to enjoy watching based on their previous choices.
In a digital world dominated by databases, she's using her very human analytical skills to get viewers to devour more content. And when they watch, they leave a digital record of their habits that viewers of regular television do not.
"This is pretty much the best gig ever," said Ms. Canning, who makes films when she isn't watching them for the California-based company. "Everyone always wants to know how they can get a job like this when they hear about what I do."
And while her decisions are at the core of most Canadians' Netflix experiences – the company estimates that 70 per cent of the choices viewers make about what to watch are based on what is suggested to them – the service is now using the data to better understand what viewers want and to decide what content it will produce on its own.
The stakes are high: If Netflix can use its algorithms to predict what will be popular with viewers, it will be able to steal a bigger share of the market from established cable and satellite companies, which have dominated the market for decades. It believes it has the secret formula that has large television networks (and cable and satellite companies) looking over their shoulders and wondering whether the company has an edge in the battle for subscribers. Its second original series, House of Cards, a drama about a politician played by actor Kevin Spacey, recently debuted to critical acclaim, and the Toronto-filmed Hemlock Grove will be released next month.
"When we first heard about House of Cards, we immediately started getting into the data," said Ted Sarandos, Netflix's chief content officer, explaining how the company decided whether to buy exclusive rights to the show.
Netflix studied how many people watched movies featuring Mr. Spacey, whether they tended to follow him from one to another and what sort of roles drew more viewers for him. The spreadsheets showed viewers likely to enjoy the actor in serious roles were the same people who were likely to watch a political drama. So the company cut the $100-million cheque to produce the first season.
Data have been at the heart of the Netflix experience since it was founded in 1997 as a mail-order DVD service. In its early days, it would send staffers to visit subscribers at home to take notes on how it was being used. It developed its "Cinematch" algorithm in 2000. It found that if it recommended movies based on what users were renting, they were more likely to return their discs and keep using the service.
"They have 14 years of customer data now," said Gina Keating, author of Netflixed: The Epic Battle for America's Eyeballs. "Very little emotion goes into the decisions they make."
That's certainly true for Ms. Canning as she fills in the spreadsheet for Madagascar 3: Europe's Most Wanted. The spreadsheet mostly allows her to choose from pre-set categories – is there nudity, are any characters gay, is there a happy ending, is there constant violence, is it hilarious or merely funny – but it's essential she be as specific as possible.
"Some are things easy to know off the bat – I know there won't be any sex or gay and lesbian content in Madagascar 3," she said. "But I know I can select 'family friendly' and 'warm-spirited.'"
Without the tags, every viewer would have the same welcome screen greet them when they log in. But each subscriber gets a unique slate of movies and shows based on their viewing history, and the system generates categories specific to each account. Some are so detailed, such as "visually striking father-son films" and "emotional period pieces featuring a strong female lead," they sound like a joke.
But there's nothing funny about the level of competition for viewers. Consumers across North America are increasingly looking to reduce their television bills by using services such as Netflix, but are often dissuaded by a lack of programming that caters to their tastes.
With help from the company's taggers and a little bit of complicated math, that could soon change as Netflix looks to produce more mainstream content. But that doesn't mean the company will target every niche its computers tell it may lead to increased viewership – bad news for anyone hoping for a comic-retro-time-travel-romance-with-an-animal-lead option on date night.
"This interest in reaching niche audiences probably won't end in micro-targeting every tiny sub-genre," Mr. Sarandos said. "Netflix is not ready to go that far."
Chris Berube is a freelance writer