March 15, 2013 – As wariness gives way to aggressive embrace of recommendations and search in pay TV navigation, service providers are confronted with a surprisingly broad array of issues that must be factored into choosing a vendor solution that will serve their needs over time.
Some issues may be top of mind: Which platform is best suited to delivering a reliable discovery experience that really helps subscribers find something appealing to watch from the pay TV lineup on any screen wherever they happen to be? Or, if the screen in use is the living room TV, how can something as personal as recommending content suited to a particular individual’s tastes be made useful in the shared viewing experience on the big screen?
Other issues may not be so top of mind. For example, on the linear programming side, what is the right balance between using discovery to make subscribers more aware of things they don’t normally know about within their subscription package versus trying to upsell them by recommending content that’s not in their package? Or how can the operator employ discovery to maximum advantage in a universal navigation strategy that encompasses OTT as well as pay TV content?
And then there are some issues which might not be on operators’ minds at all at this point but which could well become important as market trends unfold, such as, how can the information generated through tracking each customer’s viewing preferences be used to facilitate advanced advertising? Or how important is voice-driven search to the navigational experience?
Nothing better illustrates the pace of uptake in advanced discovery among MVPDs (multichannel video programming distributors) and virtual MVPDs (over-the-top distributors) than the growth curve recorded by ThinkAnalytics over the past several months. Since the summer, when the company reported its Recommendation Engine product suite was used by 25 pay TV operators reaching 70 million subscribers, the count has climbed to over 40 customers with a subscriber count topping 100 million, says founder and CTO Peter Docherty.
“2012 marked the coming of age for the media recommendations engine and content discovery market with most operators commencing field trials or starting full deployment rollouts,” Docherty says. This includes a break in the ice in the U.S. where Cox Communications has become the first multichannel operator to offer individualized recommendations for video on TV sets.
The distinction, individualized versus household-wide types of recommendations, is significant, given operators’ search for ways to distinguish their subscribers’ experience in the pay TV realm from what they get with OTT content. While Netflix and other OTT providers have long delivered recommendations, they are based on whatever the viewing patterns are on a given device, which means a family sharing a TV may get a set of recommendations that range from SpongeBob SquarePants to Sleepover to The King’s Speech.
“We are making significant investments in our video experience based on customer and enhanced usability insights,” says Len Barlik, executive vice president of product management and development at Cox. “Customers want their viewing experiences to reflect their personal interests. Our new video products and services take their experience to the next level.”
The ThinkAnalytics Intelligent Navigation solution, which was recently introduced as an enhancement to the Cox Trio Guide running on the MSO’s Advantage TV package, delivers personalized recommendations for up to eight individual profiles per household, encompassing viewing histories tied to live TV, VOD and DVR. Officials report the recommendation system will later be extended into the multiscreen domain on the user interface provided for accessing Cox service on smartphones, tablets and other IP connected devices.
The many strategies ThinkAnalytics customers are pursuing with their implementations of its recommendations and search technologies demonstrate how far discovery has come in providing operators means to improve subscriber retention and drive revenues. They also reflect how unsettled the market is on best approaches to using such tools.
“Since our first customer (BSkyB in the U.K.) launched our Recommendation Engine in the 2005-2006 timeframe we’ve learned a lot about what our customers want,” says ThinkAnalytics chairman Eddie Young. “We learned that at all points you have to be flexible in your approach to recommendations because not everybody wants the same thing. Fortunately, we designed the platform from the outset to have the flexibility to adjust to these needs, which made it easy to update. It’s a much more mature system today than when we started.”
Today one wouldn’t think that the navigation systems on Sky and Virgin Media in the U.K. are supplied by the same company, but they are, Docherty notes. “You can have two companies in the same country where one, for example, might focus on VOD while the other focuses on over-the-top,” he says. “And their use of the technology changes over time. What Sky wanted in 2005 isn’t the same mix as what they want today.”
The emphasis on the role of recommendations varies as well, sometimes as a function of whether the platform is employed in a legacy programming guide environment or on a next-generation user interface. Getting recommendations into next-gen UIs is now fundamental to everyone’s strategies, Docherty says. For example, Liberty Global, using ThinkAnalytics’ technology with its rollout of the next-generation Horizon service in The Netherlands, Switzerland, Spain and elsewhere, is “completely driven by recommendations,” he says. “They’re trying to get you not to use the grid.”
But the majority of ThinkAnalytics’ deployments are on legacy systems, as in the case of Sky, where 80 percent of the base is the older generation set-tops. Despite the limitations imposed by these interfaces there’s a growing realization that the technology can make a huge difference on the basic goal of minimizing churn. “Operators are seeing how they have to improve how people find content so they don’t think there’s nothing of value for them to watch,” Docherty says.
This makes recommendations on live programming a vital part of the strategy. Starting from the beginning with Sky, ThinkAnalytics has always brought recommendations into linear TV as well as VOD. As Young notes, this is a bigger challenge owing to all the things that have to be taken into consideration with linear viewing beyond the viewer’s personal history, such as the type of program the viewer seems interested in at a given time or which channels are available with any given subscriber’s package.
“If you’re looking for sports on your linear channels, we know which types of sports, which teams you’re most interested in,” Young says. “Recommendations are very important on linear but they must be relevant.”
Indeed, adds Docherty, while the upside to bringing recommendations into the TV space is the fact that consumers are well prepared for the concept owing to their experiences with Netflix, Amazon and other Web outlets, “the downside is they’re used to getting bad recommendations, so we have to be really good at overcoming those negative expectations. Customers have to trust the recommendation engine, to know it understands them.”
There are many nuances that come into play with how recommendations are used to enhance subscribers’ perception of service value. “You have to spread the recommendations across the channels they’ve paid for beyond whichever ones they’re used to tuning into,” Docherty observes. “If they’re subscribers to premium packages, you have to make sure you’re not drawing all their recommendations from the basic package.”
Similar considerations come into play with options presented through the ThinkAnalytics Search Engine. Docherty notes that some customers already have their own search mechanisms in play and so only use ThinkAnalytics for the recommendations. But often they don’t have the personalization capabilities that ThinkAnalytics includes, in which case “they pass the results from their search algorithms over to us for personalization,” he says. Other customers are relying on ThinkAnalytics for all the search functionalities.
A massive amount of information processing goes into identifying content for recommendations, especially when it comes to keeping pace with live programming where millions of instances of what’s being broadcast across all the channels worldwide must be continually tracked and categorized. ThinkAnalytics ingests into its database all the basic metadata associated with each program and also performs natural language analysis on restructured metadata, which is to say, the idiosyncratic modes of describing content that differ from one producer to the next.
This analysis is also applied to any program synopses that are available from producers and networks. In addition, the company has its own reservoir of pre-existing metadata tags that can be brought into play for identifying any given piece of content. All of this is processed algorithmically into vectors that allow the Recommendations Engine to instantly associate relevant content with specific users. “We’re doing this for every program in the linear as well as VOD space,” Docherty says.
Of course, once linear programs receive this treatment, they are also set up for recommendation processing when they’re moved into storage for free VOD access. Depending on how operators want to orchestrate their recommendations, a natural way to generate greater usage of FVOD and therefore greater appreciation of what the pay TV service has to offer is to factor in FVOD recommendations with the recommendations offered when viewers are in linear viewing mode.
Commercial Recommendation Strategies
This strategy is very much in keeping with the churn reduction agenda, where everything that’s recommended is available within the subscriber’s chosen service tier. But there’s also a place for VOD recommendations with linear viewing that gets into the commercial side of the strategy, which is to say, those instances where a movie or TV title that’s available for paid viewing is brought into play with recommendations tied to the linear viewing experience.
This is a step beyond the most basic area of commercial promotions, which is the recommendation of titles when viewers are in pure VOD search mode. When commercially biased recommendations are extended to the linear realm, the options can include programs from premium or other channels not part of the subscriber’s package as well as VOD titles.
This commercial side of recommendations and search is now in wide use among ThinkAnalytics customers, but it must be used with care, Young says. “It’s important that recommendations don’t become advertising,” he warns.
ThinkAnalytics has built its rules engine to provide maximum flexibility for marketing departments to use Intelligent Navigation to increase ARPU in ways acceptable to viewers, he notes. Most customers want to leverage the ARPU capabilities but all have different approaches to balancing the degrees of emphasis between churn reduction and increasing revenue, he adds.
For example, the operator can set various proportions of recommendations of VOD content when viewers are watching linear. “I can tell the API I only want linear or I can just let the recommendation engine call up whatever is relevant to that user whether it’s linear or VOD,” Young says. “Or I can tell the business rules to set it at 40/60 or 50/50.”
At the outset of the engagement ThinkAnalytics works with customers to set the rules, typically with heavy if not exclusive emphasis on churn reduction. After that, the operator can assign access to the rules engine across various departments for internal management of how everything is done, or it can utilize the ThinkAnalytics cloud service to manage those changes as instructed by the marketing and other departments.
“If marketers want to go in and manage the rules themselves there’s a browser-based business rule management system they can use,” Docherty says. “We train them how to use it, and they take it from there. But other customers want us to set the parameters, so we do it as a managed service.”
Most often, though, operators choose to set the balance between consumer-friendly and commercial recommendations and let the system “keep learning and reconciling,” Young says. As for the minority of operators who take a more proactive stance, “they’re not changing the rules every day,” he says. “But week to week and month to month they are proactive.”
While it’s hard to measure the impact of recommendation technology on churn reduction other than to observe patterns over a long period of time following introduction of such capabilities, the impact on commercial applications can be quantified with great accuracy. “We know everything that’s been recommended and what has been requested,” Docherty says. “We can drill into individual data and slice and dice it hundreds of ways, telling you what’s been purchased, what’s most watched, most recorded, most searched, etc.”
Variations in Operator Approaches
Among the proactive users of the ARPU-generating side of the Intelligent Navigation platform, FreeView of the U.K. is a typical example of how this capability gets brought into the recommendation process, Young notes. “At the outset they didn’t want to scare customers off so they focused on very consumer friendly recommendations within their service packages,” he says. “Six months later they were confident enough in the trust they’d built with subscribers that they were able to turn on the commercial bias to begin upselling people.”
What and how things get promoted is up to marketers, who are given a wide range of subtle and not-so-subtle ways to recommend content that requires a purchasing decision by the user. For example, the marketing department can influence the order of recommendations so that, in a given genre, a particular title might be placed at the top. And the granularity of how this is done can be taken to another level so that, in the case of users whose profiles suggest a title that otherwise would be pushed to the top is not one they would like, it is buried further down.
The way Cox is using the Intelligent Navigation platform with its Trio guide offers another example of the nuances that can be employed in the mix of commercial and non-commercial recommendations. “Trio can aggregate content recommendations around what’s on now, what’s on later today or what’s coming up in the future,” Docherty says. “Maybe with what’s on now there’s no commercial bias in the recommendations, but with what’s coming up in the next few days there’s content you don’t already pay for, like a movie from a premium channel or a pay-per-view boxing match.”
Cox is also employing an option available on the platform that lets users to grade recommendations by liking, disliking or suspending them. “Some customers are using this to help refine the recommendations process, some aren’t,” Docherty says.
With all these approaches to using the ThinkAnalytics platform now in play the company and its customers are beginning to look at new ways to do things. One of these is to promote universal search and recommendation, which is to say the ability to generate results for subscribers on the pay TV service from OTT as well as walled-garden content.
This introduces another order of complexity where operators may want their customers to appreciate that the pay TV navigation system is the only place they can go to discover content of interest across all outlets but not to the point of de-emphasizing their own offerings. This means, for instance, that if a recommended movie is available from various OTT sites but is also stored in the operator’s VOD system or on one of the pay TV channels, the latter options will be more prominently displayed.
“On the one hand you gain from the trust you build with consumers who find that not everything that’s recommended is coming from the operator,” Docherty says. “But you don’t want to be recommending Netflix all the time.”
Universal navigation is gaining momentum as a point of serious discussion within the operating community, he adds. “Most are talking about it, and some are actually doing it,” he says.
Also now coming into play is the use of data gathered through the ThinkAnalytics platform for advertising purposes. Given the scope of information that can be drawn and configured for various purposes having to do with generating recommendations to individuals, there’s an opportunity to process information into useful, anonymous profiles for advanced advertising purposes, Docherty notes.
“In 2009 we started looking at the possibilities with CableLabs in conjunction with the use of SCTE 130 and the Subscriber Information Service API that is part of that whole environment,” he says. “We’re generating incredibly valuable data that can be fed into ad management systems to determine which ads to send.”
But, he adds, “Before things move in this direction the whole advertising industry needs to move beyond the CPM model to a better pricing system for advanced advertising. Some operators are looking at experimenting with this to gauge the value of the data that can be fed into these other systems and how that affects the whole business model.”
Also on the ThinkAnalytics agenda is the introduction of voice recognition technology into the user interaction with the search and recognition processes. Here the driver is interest from CE manufacturers who are adding voice recognition to the bells and whistles of some smart TV models.
“We’re looking at this and talking to manufacturers,” Docherty says. “We see it’s likely to happen and so have it ready for use in search.” Interest in speech recognition for use on media gateways and set-tops has been minimal from the pay TV service provider side, he adds. “There may be some more forward looking operators who will put it in their RFIs,” he says, “but the CE people are the main players now.”