AIBroadcastersContent RecommendationFeaturesMonetisationHow UKTV Used AI in its New VOD Apps Suite

Kauser Kanji9 months ago4 min

Last month, Britain’s biggest multichannel broadcaster, UKTV, announced that it had added elements of machine learning and AI – specifically in terms of personalisation, content recommendation and building up taste profiles – to its iOS, Android and Amazon FireOS VOD apps.

AI is a hot topic in the industry right now and I wanted to learn more, so I spoke to Oliver Davies, UKTV’s Director of Digital Product, about these product enhancements. Here are four take-outs from our conversation:

#1. Machine Learning in Content Recommendation

UKTV’s apps now benefit from the use of a personalisation engine which surfaces content to viewers. Conceptualised in-house and developed by external vendors, the new personalisation engine builds on the work done by UKTV’s editorial teams to classify content by category, sub-category and programme brand. The company also created its own taxonomy and these information layers help to power and improve recommendations.

Davies told me that,

The personalisation engine has a notion of “people like me” so it looks at programmes but also genres, channels, collections and all of the other constructs within the system and matches my taste profile with content. It doesn’t just look at the category breakdown because in real life, people like all sorts of things. It will find like-minded individuals and see what they’re watching and then make decisions based on that.

Using these content-based and collaborative filtering techniques – as well as pattern matching – Davies found that,

…in the mode of trusting the algorithm, [the personalisation engine] was making links between programmes that we may not have thought about. A great example is if someone likes historical documentaries, the algorithm was making links to historical-based dramas too like “The White Princess” based on a similar period of history.”

#2. The Long View of the Long Tail

Davies also said that UKTV had recently switched metadata providers and now had a “richer, more robust set of metadata around programmes on an episode, series and brand level which we can add into the analysis of personalisation”. In practice, this means that the personalisation engine could potentially recommend related content based on the location of a TV show (London, Colchester, the Highlands etc.) as well as its cast and contributors.

As a simple example, if you’ve watched Judge Romesh, the personalisation engine could tell you that Romesh is also featured in Season 3 of something else. Taking that one step further, you could also look at discretely personalising imagery relating to the show on the OTT platform… Sentiment analysis, location, all of this is potentially available in the programme metadata which influence signposting and programme imagery.

And continuing the theme,

We wanted to bring the worlds of linear and VOD as close together as possible. Most people still treat them separately… My take on this is if you click on “Watch now” for a simulcast stream, you’re interested in watching that particular programme, rather than that channel. And yet, on a simulcast stream, you often don’t see information on what you’re watching and how there might be 50 episodes available on VOD. And vice versa: if you’re watching something on-demand, you don’t often see a notice telling you when the show is next premiering on TV. It’s bonkers! So, we deliberately set out to break down these barriers because ultimately, they’re all opportunities to watch content. Ultimately, we thought, wouldn’t it be great to take everything we know about you and what you like and not only recommend VOD but also stuff to watch on TV.”

#3. Combining Datasets to Retain and Attract Viewers

Viewers have to be logged in to watch content on UKTV’s apps (UKTV now has over 2 million registrants) but taste profiles take time to build up. By combining viewing history with third-party ‘lifestyle’ data,

…we can obtain extra information about the individual very quickly after registering. Right now, that’s only used to determine ad segmentation, but by analysing things like hobbies, household make-up and magazine subscriptions we can build a taste profile almost straight away to try to eliminate the ‘cold start’ syndrome. So, if we know that someone in that household subscribes to a motoring magazine, we could surface motoring shows like Top Gear by the time they’ve finished watching their first show. That’s wringing as much as you can out of the data.”

Paying more attention to consumption habits: we’re not currently analysing types of content preferred on different devices, times of day, days of the week. We have a hypothesis that people prefer watching short-form content on mobile devices, but we haven’t confirmed that. And if we can tune recommendations to take these ‘environmental’ factors into account, it makes things richer.

Customer acquisition is hard but retention is an equal challenge that relies on personalisation, relevancy and effective communications. One of the things we want to do, as part of the next iteration of the personalisation platform, is to treat off-platform comms, like CRM, in the same way as on-product personalisation. These things should come from the same data, have the same view of the user, use the same taste profile. And whereas with product, when you’re boiling this down to an individual moment in time, when you take that same information you can use it to create on-the-fly segments which can then be used to create comms – emails, notifications etc.”

We want to avoid silos of decision-making and analysis. Using AI, we want to have a much more sophisticated multi-layered, multi-faceted view. And if we pull all this data together, we can have so much more insight into usage and behaviours.”

#4. A Special ROI Case wasn’t needed to Start Using AI

I asked Davies directly about return on investment and he replied that introducing AI into the personalisation engine was a natural evolution of UKTV’s OTT products.

Spec’ing the next generation of the product, its core features, that’s just something we do as part of our roadmap – we don’t need an ROI case for it. It’s only when you think about additional features like simulcast or subscriptions… that’s when you need an ROI. We saw this project as a huge enabler for viewers and the response to it [the upgraded personalisation engine] has been incredibly successful.

Shorts

  • UKTV served over 140m VOD views last year across all platforms, a YOY rise of 36%
  • The broadcaster’s new generation product is available on web, iOS, Android, Fire OS, Samsung TV, Roku and NowTV, and coming soon to Fire TV, YouView, Freeview and Freesat.
  • Work on the AI-enhanced personalisation engine started in October 2017. Development commenced in February 2018 and the first launch was in October 2018.

Kauser Kanji

I've been working in online video since 2005 and have held senior roles at NBC Universal, Trinity Mirror and Virgin Media. I've also worked on VOD projects for the BBC, Netflix, Sony Pictures, DR in Denmark and a host of broadcasters and service-providers all over the world.

Reporting from the Future of Television

VOD Professional contains news, reviews, interviews and analyses from the future of TV. Online video is at a pivotal moment in its evolution and our focus is on writing about monetisation, product development and the people – from service-providers and vendors alike – who help bring brilliant content and services to our screens.

Get our Weekly Industry News Digest

Consulting & VUIX Library

We have 15 years of experience in working on strategic consulting projects for Netflix, Sony Pictures, Virgin Media and a host of commercial and public service broadcasters in Europe, North America, APAC and MENA. Drop us a line if you need some help or just want to have a geek-out about the industry! And if you’re a product developer, check out our VUIX Library resource which tracks over 500 VOD products, across platforms, from all over the world.

VOD Professional, 2019 © All Rights Reserved