How Data Analytics Is Shaping What You Watch

By Knowledge@Wharton | October 31, 2015

When Netflix decided to go into the entertainment-producing business by commissioning the streaming series House of Cards, there was a lot of data to be crunched before the first download ever took place, said Dave Hastings, Netflix’s director of product analytics, data science and engineering.

“You do not make a $100 million investment these days without an awful lot of analytics,” Hastings said at the recent Wharton Customer Analytics Initiative Conference in Philadelphia. While some may claim that Netflix over-analyzed House of Cards, Hastings is adamant that it was the way to go. “We have had a whole catalogue of things we did not make. We wanted early on to be careful, and it has worked out.”

Just four years after the March 2011 release of that first season of House of Cards, Hastings said Netflix has as many as 60 series seasons in the works or soon to arrive, from season two of comic book adaptation Marvel’s Daredevil to another House of Cards season to a new comedy talk show with Chelsea Handler. All of them will undergo thorough analytics before release.

“We live and breathe the customer,” said Hastings. “We want to be the world’s leader in Internet TV. When you have aspirations of being the largest, that means shaping demand. What is Internet TV and how does it differ from linear TV — the questions around that are what we use analytics for.”

In terms of corporate maneuvering, the speed at which Netflix has altered course in its business practices could give anyone along for the ride whiplash. Started in 1997 after CEO and co-founder Reed Hastings got irked about a $40 video store late-fee for the movie Apollo 13, it rapidly became the No. 1 DVD-by-mail rental service, shipping out its billionth DVD by 2007.

However, recognizing that streaming video would eventually displace physical movie rentals, its management team slowly de-emphasized the DVD mailing business and expanded into streaming. By 2010, that business had become its calling card. The move into producing entertainment, said Hastings, was the natural next step, and it was heavily planned, taking into account a lot of big data.

“House of Cards is an outstanding show, but it was a huge bet from the company to get into that business,” Hastings noted. “We were a technology company, and it took some highly creative education to move into entertainment production. There was a lot of analysis.

“But it was not what urban legend had it,” he added. “People said the whole show concept was made analytically — that we needed Kevin Spacey as the lead in a political drama, and so forth. I mean, in theory you could do those things with analytics, but there is a far gap between that and what we did. On the other hand, we hope now to be the biggest original studio in a few years.”

Netflix had to establish a strong analytics infrastructure early on, Hastings noted. The company, for instance, had to determine how many DVDs of which movies it would need on hand to meet demand. And that focus only became more important as the company moved to streaming.

“The foundation of the streaming business was analytics,” he said. “Our catalogue was extremely limited when we started streaming. We had to be right in what we were trying to deliver.”

Getting to Know Your Customer

Hastings said that currently the company is using analytics to see how to expand services in a variety of ways. For instance, it uses data on different regions to see what to promote in the American South or in urban areas or in Australia – Netflix’s newest streaming market. The company also needs to know what sorts of customers are using the service on televisions versus those who are using hand-held devices.

“You integrate all this together, and you really get to know who your customer is,” he said.

Bill Franks, chief analytics officer for Teradata, which works with Netflix on big data, was on the dais with Hastings, and noted that it’s not enough for a company to have powerful analytics tools; applying them correctly to an end product is the key.

“Analytics cannot replace execution,” said Franks. “If you can’t deliver a good experience, analytics are irrelevant. Netflix made sure those videos were delivered and they were not stopping every 10 minutes.

“Sometimes you go overboard on theories, but Netflix was pragmatic and looking for the best solution with the data they got,” he said.

Hastings said Netflix can use as much data as it can get on customers, distribution and the like, and it’s always looking to improve its algorithms. He noted that Netflix’s streaming services are so vast that the company is the largest user of downstream North American web traffic, accounting for as much as a third of traffic.

“It is a huge amount of data to wade through,” he said. “You have an issue then of scale with the analytics you do.”

Though Netflix has $5.5 billion in sales, it has only 1,600 employees. Hastings said the company hired 187 people last quarter, which seemed a lot to those there, but was, he realized, a small number for a company with its reach. Because of that, Teradata’s Franks said, Netflix has to be careful in what its staff does with analytics to ensure that it does not overreach.

“We are getting to the point, with all the data we can have, that there will be tradeoffs,” said Franks. “The focus has to be on achieving a gain, not just analytics for its own sake.”

That said, according to Hastings, there is little in the new entertainment realm that will go online without a lot of advance study.

“I can say that no changes in Netflix’s products are not tested and validated, and we do not just test to test,” he said. “If we do not believe it will improve, it will not be tested. We have 300 major tests of products and dozens of variations within. When you go into Netflix, hopefully we are improving your experience. How do we increase streaming? How do we increase customer retention? We mine lots of data for those sorts of things.”

‘Morphing and Twisting’

Sometimes, though, Hastings admits, the results of analytics get overruled at Netflix. For instance, he said, data would indicate that the company should make it hard to cancel its service – that making it difficult to quit would increase retention rates. But, he said, the founders of the company early on knew they wanted to make it easy for a customer to cancel: It would, they hoped, give the company a good reputation, and they realized that ultimately, making their customers frustrated would not be a good thing.

“The data said we should not have done that,” he noted, but he admits it worked, enhancing Netflix’s reputation in the long run.

Teradata’s Franks said that if it wanted to, Netflix could almost be a stand-alone analytics firm.

“It is getting blurry out there as to what companies are in,” said Franks. “AT&T is not just a phone provider. It is providing TV, and then has its own data. Nike is now manufacturing high-tech electronics and housing data in its data center, not just doing knitted sportswear. Companies can commercialize the use of their data outside of their core business. It is a fascinating time, and a lot of companies will be morphing and twisting.”

Hastings said, however, that despite all the change the company has gone through in its 18 years, Netflix is not looking to be constantly shifting businesses.

“We are really careful about doing things outside our core,” he said. “If you are distracted, you waver from being the best. There are [ways to use our data that] could theoretically make money, but it would be a distraction from using those analytics to be the best we can be in our core businesses.”

According to Franks and Hastings, there are sometimes things analytics cannot measure effectively, at least at not given the current state of the art.

“Mood is an interesting thing to try to go after. When you sit down in front of your iPad, we do not know what your mood is. There is only so much information we have,” said Hastings. “Maybe we can tell certain things by the time of the day, but I don’t know what I am going to do tomorrow and I don’t know what I will want to watch tonight, so getting that mood right is not part of analytics right now.”

Franks agreed, but said that outliers in analytic data have to be expected.

“Even if I go to such-and-such restaurant every day and order such-and-such sandwich all the time, today I might decide to go with the other submarine sandwich,” said Franks. “Even perfect data will not always lead you to a perfect conclusion. But, like Netflix, if you go with the aggregate, you will be successful.”