Content Recommendation Tech Becomes Essential Tool for NSPs

Pete Docherty, founder & CTO, ThinkAnalytics

Pete Docherty, founder & CTO, ThinkAnalytics

July 24, 2012 – If, as Netflix researchers claim, 75 percent of what people watch on the over-the-top subscription service is from “some sort of recommendation,” is it any wonder that a stampede has begun among pay TV service providers to capitalize on the power of recommendation engines in their navigation systems?
 
As on-demand viewing accounts for an ever greater share of hours spent consuming TV programs and movies on devices of all types, discovery techniques driven by advanced algorithms and specialized applications of metadata are due to become as much a factor in traditional TV viewing as Netflix is experiencing in the OTT domain. But unlike Netflix, NSPs in general have to deal with a market base saturated with old set-tops running old EPGs.

A handful of content discovery vendors vying for NSPs’ business are reporting accelerating success across North America, Europe and Asia. One of the leading suppliers, ThinkAnalytics, says it has added three unnamed North American MSOs to a customer base that includes BSkyB, Virgin Media, ITV, Telenet, UnityMedia and 15 other providers outside North America. Meanwhile, other leading players such as DigitalSmith, Jinni and Red Bee Media are capturing business from many other SPs.

As the experience at Netflix makes clear, there are a lot of issues to consider in choosing suppliers. Reporting on recent advances in their personalized recommendation system Netflix research engineers Xavier Amatriain and Justin Basilico say they have gone beyond using recommendation algorithms that precisely rank movies and TV programs for users based on how content matches personal preferences in order to account for other factors that influence how popular the content will be with any given viewer.

“We also need to take into account factors such as context, title popularity, interest, evidence, novelty, diversity and freshness,” they say in a recently posted two-part blog series. “Supporting all the different contexts in which we want to make recommendations requires a range of algorithms that are tuned to the needs of those contexts.”

These efforts are responsible for the recently achieved 75 percent ratio of recommendation-driven viewing, they say. “We reached this point by continuously optimizing the member experience and have measured significant gains in member satisfaction whenever we improved the personalization for our members,” they note. That satisfaction and the attendant subscriber retention gains correlate “with maximizing consumption of video content. We therefore optimize our algorithms to give the highest scores to titles that a member is most likely to play and enjoy.”

In other words, successful recommendation technology has become essential to success of the Netflix business model. Judging by metrics generated from the network service provider side, the same may soon be true of NSPs’ TV service business models.

According to ThinkAnalytics founder and CTO Pete Docherty, data from its customers verify that intelligent recommendations reduce churn and add significantly to perceived value.
“We see recommendations are really driving consumer behavior,” Docherty says.

Recommendation technology is also enhancing revenues by driving more VOD purchases and creating upsell opportunities on linear tiers. “Our data show that subscribers receiving recommendations purchase 50 to 60 percent more on-demand content than average purchase rates,” he notes. As for upsell, NSPs are using the technology to expose content not available in a given subscriber’s package and, when free trials are offered, to give the trial users greater exposure to what they might get if they take the higher tier.

But transforming user experience with the pay TV service, now saturated with hundreds of channels and thousands of VOD options, is the biggest benefit to be gained from content recommendation support, Docherty notes. “We can’t expect consumers to recognize value in programming that they don’t know exists,” he says.

“We’re finding from real feedback that users like having recommendations,” he adds. “Out of millions and millions of recommendations delivered, they’ve been given the opportunity to say thumbs up or thumbs down or to rate the recommendations on a one- to five-star basis. Over 90 percent of the feedback is positive.”

To get to that level of satisfaction requires innovation. “We are delivering discovery and recommendations that go way beyond techniques that simply deliver ‘people like you like this,’” he says. “True personalization, search and social recommendations are what we really deliver via intelligent navigation.”

The ThinkAnalytics platform is now delivering recommendations from live TV, EPGs, VOD, social media and OTT content to about 70 million subscribers worldwide, he notes. They’re accessing personalized recommendations on set-top boxes, the Web and second screens, including tablets, smartphones and smart TVs.

ThinkAnalytics has gone to great lengths to create a personalized recommendation platform that is sensitive to myriad factors influencing user preferences, including the devices they access content on, Docherty says. “You have different use cases on different devices where a user accessing content on an iPhone may be looking for a different type of content than they’d be accessing on an iPad, for example,” he notes. “So we think it’s important to have multiple algorithms that allow you to support those different use cases.”

Of course, he adds, one of those devices is the set-top box, where the use case not only may be different but is not as personal, given the shared nature of the viewing experience. But even in this case ThinkAnalytics gets to some level of personalization. “Rather than just household-labeled recommendations, we separate viewing patterns according to what happens at different times of the day and different days of the week so that we can more closely tune the recommendations to whoever is watching at a given time,” he says.

Multi-device usage of content from multiple sources requires not only sophisticated algorithms to generate personally optimized viewing options but also a much more extensive metadata base to draw from than is intrinsic to traditional feeds. “We’re using the structured metadata with keywords and tags that MSOs have developed, but then we’re adding more key words and tags partly based on analysis of unstructured information that we analyze,” Docherty says. “And we also we have our own pre-tagged library of content, including a half million movie titles that we’ve tagged with recognition-quality metadata.”

ThinkAnalytics works through middleware suppliers such as NDS and NAGRA or directly with NSPs to set up internally or cloud-hosted content discovery mechanisms. Once they’re installed, ease of use is vital to maximizing the benefits.

“People need to be able to change the parameters over time,” Docherty says. “They have to be able to react to implement not just new promotional offers or upselling messages but also to make adjustments as their goals change.”

For example, he explains, the initial emphasis with how the system works on the EPG might be on what Docherty calls the “consumer-friendly” approach where content recommendations are all about the content users have access to within their existing service package. Over time, marketing people may want to change the balance to where some measure of what’s being recommended is outside the package so as to encourage people to move to a higher tier. “Our interface into the recognition engine allows you to tailor the system to what you want to accomplish without getting IT involved,” he says.