Your browser is not supported. Please update it.

26 June 2014

Swisscom puts recommendation center stage with ThinkAnalytics

The tussle between ThinkAnalytics and Jinni for the Swisscom account has been watched with interest as a head to head between two of the leading standalone contenders in the increasingly hot TV recommendations arena.

The ability to recommend not just VoD but also linear content on the fly has become an important differentiator and revenue generator for pay TV operators, recognized as essential for boosting both stickiness and ARPU. This has led to consolidation within the sector, with two of the main adversaries having been acquired by middleware vendors plugging recommendation holes in their portfolio. Firstly Tivo acquired Digitalsmiths for $135 million in January, followed by Rovi buying Veveo for $69 million in February. These moves highlighted how a recommendation engine had become a must have for pay TV software players and left the UK’s ThinkAnalytics and Israel’s Jinni as the major dedicated vendors in the field.

At Swisscom Jinni drew first blood in July 2012 when the operator announced it was deploying its semantic discovery engine to help subscribers tap the increasingly large volumes of on demand content available. At the time Swisscom indicated that Jinni’s engine would be employed both for VoD and linear programming, but it turned out that it has been confined largely to the latter.

Although Jinni has been deployed for live content elsewhere and insists that it is perfectly capable of this, it looks like it did not convince Swisscom that its approach focused on personalized taste based channels would work so well for scheduled linear programming, where the choice of content at a given point in time is much smaller. Swisscom indicated then that the shift away from linear to on demand viewing was itself driving innovation in content discovery and called for more advanced recommendations capable of taking users beyond their normal zone of viewing and suggesting items that would broaden their horizons. This in turn would lead to increased ARPU as well as customer satisfaction, so Swisscom hoped.

But since then the value of effective recommendation applied to scheduled channels, especially in combination with all the VoD content, has become apparent. At first Swisscom had assumed Jinni would take this role, but over time decided to consider other contenders. In the end ThinkAnalytics was chosen because of its greater proven pedigree for live recommendations. Jinni might have hoped it still had a role for VoD, but ThinkAnalytics’ chairman Eddie Young insisted that Swisscom has abandoned Jinni entirely, which makes sense given the desire for an integrated linear/VoD recommendations engine.

Neither Jinni nor Swisscom responded to our offer to comment, so it looks like Young’s version of events is correct. Swisscom’s head of TV and portal Development Volker Dietzel has stated, ‘we turned to ThinkAnalytics because we wanted a partner with proven experience in live, linear TV deployments, not just VOD.’ That sounds fairly clear cut.

Swisscom has also confirmed that the ThinkAnalytics recommendation and search engine is being incorporated in the EPG of its new TV 2.0 Android IPTV set tops, as well as legacy boxes. It is being used across the full range of services, including live, linear TV channels, the seven day catch up service called Replay, and the VoD platform.

It looks like ThinkAnalytics was preferred for its implementation and performance rather than underlying technology or approach. There is some continuity, with the ThinkAnalytics deployment including specific personal and recommendation categories, such as ‘content for you,’ ‘most watched/rented,’ and ‘highest rated.’ This dovetails quite well with Jinni’s personal channels determined by different facets of the user’s tastes. Jinni’s idea was to present users with several channels aligned to more subtle aspects of mood beyond broad genres like horror, comedy, costume drama, or ball sports, to perhaps ‘sentimental romance’, or ‘silent movies without Charlie Chaplin,’ identified through analysis of past user choices.

This does not seem to have worked for Swisscom, largely because the approach requires quite a lot of data to become fine-tuned enough to be effective, especially again for live/linear programming. The same applies probably to Jinni’s Explore feature, which is supposed to enable discovery through use of natural language metaphors, such as “mind bending, stylized and truth seeking’, or “feel good but never quite works out.’ The problem is that the actual discovery probably never quite lives up to the promise and presumably has failed to give Swisscom the uptick in views it had hoped for.

Jinni does at least have some other good feathers to its cap, such as AT&T U-verse, Belgacom, Prisa in Spain, Singapore’s SingTel, and Canada’s Telus. Jinni also provides the discovery engine for Xbox consoles and has been integrated with Vudu, Wallmart’s electronic sell-through service.

Meanwhile ThinkAnalytics has Fox, Cox Communications, Liberty Global, BSkyB, Viaplay, and Portugal’s Zon among its 50 plus pay TV customers.

On the face of it both companies have adopted similar technology for engines, both having dismissed the collaborative filtering approach made famous by Amazon as too crude for video content. Collaborative filtering is based on the assumption that people are clustered around shared tastes, so that if two individuals both like apples and one of them also likes oranges, then the other will most likely enjoy oranges as well. This assumption is only partially correct and merely yields probabilities of association that works well enough for lists of products displayed on a web page which users can browse, such that only one or two need to catch the attention and prompt a purchase. In the case of video recommendation, users are more likely to be turned off by being given predictable lists of suggested titles to watch.

The approach taken by Jinni and ThinkAnalytics probes deeper into the nature of content the user has decided to watch rather than making assumptions related to the behavior of others which superficially appear to have similar tastes. This led to the mood based approach espoused by Jinni, but that needs a critical mass of data to get off the ground, as well as lack of proven evidence that it really does provoke increased viewing and VoD purchase.

ThinkAnalytics has implemented recommendation slightly differently and at Swisscom has set up a range of recommendation techniques, including personalized ‘content for you’, ‘most watched/rented’, highest rated, category-based, related content and intelligent search.

To an extent Swisscom has gone back to basics here and perhaps Jinni has also, even in its case incorporating some of the previously condemned collaborative filtering. Jinni has not made this concession directly but through its choice of partners, at any rate with video analytics company Ooyala. The two teamed up in January 2014 to incorporate big data analysis directly into Jinni’s recommendation engine in order to enable operators to hit the ground running faster with effective content suggestions, rather than having to wait months for the system to learn about each subscriber’s tastes. Ooyala’s analytics does include use of collaborative filtering.

The two companies say their objective is to enable semantic discovery to be tailored to individual users at a much more granular level than before, to enable personalized channels and custom programming guides, as well as mood-based browsing and search. This is exactly what Jinni had already been promising, but had failed to deliver, for Swisscom at least. With the joint platform currently being tested at selected platforms and not due for commercial launch until later this year, it comes too late to rescue Jinni at Swisscom, but may keep its other top tier customers happy. The role of collaborative filtering here will be to help crank the recommendation engine up initially.

ThinkAnalytics though says it has already sorted out this aspect of getting up and running quickly. The key point is to seed the system on day one, but according to ThinkAnalytics this should be done using only data sourced within the geographical region and within the particular pay TV provider in order to ensure relevance. ThinkAnalytics is rather vague about how this works, because it is part of their secret sauce, but does emphasize that the seeding is relied on just to get the engine started, with the fuel then coming from actual actions performed by the user.

ThinkAnalytics is also vague over the wider question of how effective its recommendation engine is, talking merely of increasing ARPU by upwards of 30%. At least it can point to pay TV operators voting for it with their wallets.