Data: the new frontier for viewer engagement
Between steaming services, AVOD, FAST channels and of course traditional linear broadcasters, there is a tsunami of TV content out there at the moment. With such a vast amount of choice available for audiences – not to mention competition from other screen usages such as video games and social media – it’s getting more and more difficult for content providers to ensure viewer engagement. Fortunately, they have found a trusted ally in technology: Artificial Intelligence and analytics are powerful tools to allow them to package, promote and recommend content in ways that engage and retain viewers more efficiently.
Two leading professionals from the entertainment content industry talked about this during a panel at MIPCOM 2023 in Cannes, explaining their strategies, tools and plans when it comes to leveraging technology in the form of data and AI to boost viewership. These professionals were Nathalie Gabathuler-Scully, EVP Revenue Distribution and Data Operations at well-known music video platform Vevo, and Kamran Lotfi, VP Product Management at Gracenote, a company that provides music, video and sports metadata and automated content recognition to entertainment services.
AI in media for audience retention
First of all, Gabathuler-Scully and Lotfi were asked what role AI and data play in their services, and how they are using it. “The largest task we have to tackle is correlating all of it,” began Nathalie Gabathuler-Scully. “We have 200+ FAST channels that we distribute through 35 different distribution partners, we have our own apps, and the data we get from all those partners is different, so our operations team is constantly taking the apples and oranges and turning them into pineapples for us to even start utilizing that data and then leveraging it.”
Once synthesized it, the data is then used to build algorithms that support Vevo’s programming team regarding what type of music to programme at certain times of the day, for instance. “We also use it to leverage on the ad sales side, connecting the advertisers to the content, and when we work with our distribution partners to figure out what the right channels are for their platforms where we’re talking about FAST channels for example.”
Gracenote faces a similar technology challenge of processing vast amounts of information in order to render it usable. “We curate over 67,000 different channels, so you can imagine the amount of data involved,” said Lotfi. “ID normalization is a key part of what we do because the same show can be broadcast over streaming platforms, linear channels, FAST channels, AVOD… At the end of the day, most consumers just want to be able to watch that show they’re interested in without having to jump through five different catalogs. That’s where we help platforms, by providing one ID and telling them ‘this is five different places you can watch this particular piece of content’.”
It turns out, the two companies work together, one feeding information to the other to feed their digital tools and help them better capture audiences thanks to AI in media. “We provide our metadata to Gracenote and they work to tie that into their internal warehouse,” explains Gabathuler-Scully. “That allows us to have better searchability. When someone is looking for Tailor Swift, that metadata connection allows our content to surface up, whether it’s a video, her catalog or a channel that she’s on.”
Data in itself is a crude raw material that must be processed in order to be exploited by AI tools. That’s what Gracenote does with all the viewing information it collects. “We enrich it in certain ways. For example, imagery is a big thing: a number of services require a crazy number of aspect ratios, formats, titled imagery, descriptions in different lengths, etc.: we facilitate that to make sure they get what they need. This is really critical for discovery, because if the image isn’t right on the screen, ultimately the consumers when they’re seeing a wall of posters, they’ll just keep scrolling. So the metadata is a key part of making sure you have a good discovery process.”
“There are a few ways we look at data as a content owner,” continued Gabathuler-Scully. “There’s the metadata that we use in a partnership with Gracenote for example, which helps us become searchable on a platform, but then there’s also what we collect from the distribution partners on things like time spent. There are all sorts of layers that we use to help us understand what programming we should put together to keep people engaged. What’s the right promotion to provide if we see a pattern where people stay on the country channel but then they like to move to the pop channel: what are we doing to make sure they stay engaged across all our platforms? We have to be creative in the tools we can use to pull a picture or ID graph together so that we can make smart decisions on how we’re surfacing content.”
Nuanced recommendations in media thanks to AI algorithms
The key challenge here is discovery. Enabling searchability is a “solved problem”, as Lotfi put it, but getting audiences to discover content they have never been exposed to, that’s where the real challenge resides, and where technologies like AI in media and analytics tools can make a real difference. “Say it’s Friday night, you feel like watching something relaxing and inspiring… Good luck, that’s a real challenge. The algorithms are getting better and better, but at the end of the day, the fuel that we provide to them, which is the data, really has to improve.”
He goes on to give an example of two viewers who both like drama series, but nevertheless have different tastes. Recommendation algorithms need to be sufficiently refined to tell the difference between what those viewers might enjoy watching. “You might like Sex Education and I like The Bridge, they’re very different shows, but to a recommendation engine, if the only information it has is ‘it’s a drama’, it won’t know the difference. So the question is how do you get much more nuance, how do you get much deeper into content recommendation? We have a product called Video Descriptor that goes much deeper. It has 16,000 keywords in a hierarchy of 16 different categories, that’s the fuel for the recommendation engine, so it knows about the subtle nuances.”
From relevant recommendations to personalisation, there is just one step. Nevertheless, leveraging the data at hand to create relevant content personalisation through AI tools is the holy grail for content providers – hard to attain but they’re putting a lot of energy into it. “The level of personalisation depends on where you’re watching our content,” said Gabathuler-Scully. “Our on-demand app has a bit more personalisation based on your watching patterns. On the FAST channels, we have niche channels that are not about capturing everybody but having a variety of channels that can reach the right person. That’s an initial layer of personalisation, which is more in tune than just one Vevo Music channel with everything in it, where you might loose the audience because they don’t know what to watch. But then within those channels, we’ll add personalisation, because we know that folks like to watch fun party songs on a Friday or Saturday night, but they prefer something more chill on a Monday morning for example. That type of personalisation will add as layers in our programming, and that’s something that our programming team does using data science models, in addition to human curation.”
So you have to take the consumer through that journey. Presenting something they’ve never seen is a real challenge – Hiba
The difficulties of leading viewers to discover new content is something these platform managers tackle with on a daily basis. Kamran Lofti addressed this problem by talking about a stat Gracenote has uncovered where users tend to spend 10.5 minutes finding something to watch, and then one in five users simply give up. “Why is it so challenging? I think that unlike music streaming where you can just flip on a song, and ten seconds in realise you don’t really want to listen to it so you flip to something else, with video content, it’s more of a ‘consider purchase’, you have to think about it for a minute. Users need awareness, consideration, intent and then finally purchase. How many of us have watched 15 minutes of a show or a movie and then decided it was no good, and then your evening is kind of messed up, because you only had an hour, so what do you watch now? You don’t need to have many of those experiences before you realise ‘I really want to know whether this is going to be a good use of my time’. So you have to take the consumer through that journey. Presenting something they’ve never seen is a real challenge.”
So what are the main ingredients to making personalisation successful, so that viewers not only find something to watch, but consistently come back for more because they were satisfied with what their service suggested to them thanks to AI in media. One key element here is merchandising your content – and guess what? Artificial intelligence can do a lot to help TV services do that effectively. “For personalisation to be successful, one of the aspects is visually merchandising it,” continued Lotfi. “What we mean by this is there are multiple aspects to any programme. Take Lupin for instance, there’s the gentleman burglar aspect where you’re wondering what he’s going to come up with this time, there’s the whole aspect of his family, with a lot of flashbacks to his childhood… So depending on what aspect intrigues you, you could personalise it, visually merchandise the show accordingly, showcase one aspect rather than another.”
The right data for monetisation
The root of the problem? Well, it’s back to square one on that one: the plethora of available content. “If you’re looking at a wall of posters, the aspects that resonate to you are the ones you’re going to gravitate to,” added Lotfi on the topic of visual merchandising. “We tested this, and we saw an 11% lift in terms of hours watched and a 7% lift in terms of titles watched for consumers. So it definitely works, and we’re seeing more platforms adopt this.” Gabthuler-Scully: “It’s true. We have something similar with thumbnails that we use, and if you change the visualisation of these thumbnails, you can see spikes in engagement. We’ve had our data science team build a model that tells us exactly what would be a more engaged view of a given video or artist, and we’ve seen considerable spikes in engagement. I also think variety in the thumbnails is important, not showing the same images all the time. Constantly renewing them draws new interest.” And what makes it possible to constantly renew the images illustrating your content offer? Artificial intelligence-based tools, of course.
But all this usage of AI in media would be moot debating if we lost track of the ultimate objective, which is monetisation. Going through all this work of merchandising, recommending and personalising content is only aimed at making sure operators can generate cash from their catalog. “Talking about those 10 1/2 minutes users spend finding content,” said Lotfi, “if you’re on FAST or AVOD, there’s no ads to show during that time, so that hurts monetisation. Everything you can do to get the consumer right into content that they want to engage with will ultimately drive monetisation. So we’re looking at data to provide to the algorithms to help in that.”
