Two decades ago, a then-nascent DVD rental service adopted an emerging technology: a relatively basic algorithm that made simple recommendations to users based on what their fellow members liked. The service was Netflix, its software was Cinematch and it represented a turning point in the history of entertainment as we know it.
As a precursor to the recommendation systems that power streaming today, Cinematch's algorithm was based on collaborative filtering. It took millions of its members' film ratings and used them to predict how much other members would enjoy the same movies. It had its limitations. The recommendations were only as accurate as the ratings from members. It struggled to recommend films to users who were yet to rate titles. It failed to factor in that multiple people may be using the same profile. Yet with Cinematch, Netflix saw the benefits that personalising people's experiences could have on engagement. Alongside the likes of Amazon, YouTube et al, it realised, early on, that in order to increase brand loyalty and conversion rates in an increasingly digital world, there was a need to offer highly tailored content experiences.
These early algorithms were primitive by modern AI and machine-learning standards. Yet they were the start of things to come. The concept upon which they were based would later become so ubiquitous that today it underpins the majority of interactions we have online – interactions that are increasingly driven by our need for ever more customised entertainment, and which are set to get even more personalised over the coming years.
According to ABI Research estimates, 2022 will see a boom in AI and machine learning across media and entertainment.
"Due to competitive pressures from direct-to-consumer services, it has become a necessity for incumbents (i.e. pay-TV operators, broadcasters) to reduce costs, limit churn and extract as much value from existing customers [as possible]," explains Stuart Carlaw, chief research officer at ABI Research. To achieve this, Carlaw believes AI and machine-learning's role will increase to the point where people will see highly personalised promotions and adverts move from their phones and laptops onto their TVs. But what does this mean in reality?
Today, many of us – up to 70 per cent according to a recent McKinsey report – expect personalisation across all of our digital services. At the same time, despite the diversification of content across streaming platforms, the TV is still the heart of the home. In the UK, 63 per cent of us watch video content on a television set more than any other device, and more than half of Europeans (55 per cent) bought a connected TV in the past two years.
The major pay-TV and cable providers in the UK – Sky, Virgin, TalkTalk, BT Sport – have all found ways to personalise their content-management screens, typically by using the types of recommendation engines made popular by Netflix et al. A lot of data mining and algorithm training goes on behind the scenes to serve these recommendations, but even then they can fall short. This is largely because they're trying to cater to an entire family or multiple people, each with potentially conflicting viewing tastes, but also because the operators have their own interests to serve, in the form of promotions, especially of their own content.
The incumbents have also started to make moves to bridge the content gap. You can now access streaming content alongside regular TV content on your connected TV or set-top boxes. The recommendations, for the most part, remain relatively siloed and you still have to have separate subscriptions with these providers. But shows, films, series and catch-up TV from the likes of Netflix, Amazon Prime, Disney+, ITV, YouTube and more are now effectively presented, and searchable as a single library.
Yet what Carlaw and others like him predict is that this personalisation is set to go wider and deeper.
Given the monetary incentives involved, the biggest drivers of deeper TV personalisation are likely to be advertisers. And the technology is already in play.
AdSmart from Sky allows businesses to pay for targeted, so-called addressable ads that can be personalised and swapped in and out based on who is watching and where they're watching. It goes far beyond broadcasters sharing demographic data with advertisers, or local TV channels promoting businesses from a certain region. Instead, AdSmart takes billions of data points from thousands of sources to place viewers into small and highly defined target groups. Businesses are then able to serve these tightly specific audiences with the most effective and useful ads for the viewers' needs, thus increasing their chance of conversion.
In order to achieve such personalisation, Sky effectively turns set-top boxes into local ad servers. When watching an AdSmart-enabled channel, stored ads are automatically played during commercial breaks. A vegan cat owner, as one niche example, could be shown an advert for meat-free pet food on the day they usually go shopping while another viewer, who recently searched for travel tips, could be served an advert for holidays during the same ad break.
If there are no AdSmart ads available, a regular, non-personalised ad is shown instead. This is both a key benefit and a potential danger of personalised TV – people are unlikely to know there's a difference; that what they're seeing isn't what's being seen everywhere else.
Image credit: E&T
Sky has used AdSmart across its services since 2014 and outsources it to other providers, including Channel 4 and Virgin. The benefits this technology brings is clear: purchase intent has been shown to be as much as 20 per cent higher with addressable ads than generic linear TV advertising, yet data also suggests it benefits viewers too. People have been found less likely to switch channels when they're shown targeted ads, by as much as 48 per cent.
"Contextual advertising that's complementary to viewing experiences is better for everyone," according to a spokesperson for Comcast Technology Solutions. "Content that connects consumers with like-minded advertising is much more prone to buying behaviour. It creates a mutualism of benefit: campaign ROI improves, which elevates the value of advertising on a particular platform or destination, with the end result being an experience that makes viewers happier and more prone to keep watching."
In a market where consumers and businesses are increasingly opting for ad-free subscriptions, shunning third-party cookies, or turning to adblockers – 81 per cent of participants in a recent survey said they actively block adverts across the web – this could be significant for advertising and the businesses that rely on it.
If broadcasters can do this for advertising, what's to stop them using AI to serve more personalised content? Not just in the form of deeper content recommendations, but in the form of unique content made specifically for you as an individual.
Dr Jian Li, head of machine learning at Sky, recently presented a case for using machine learning to recommend content based on people's moods. "Humans have moods. Recommendation machines 'talk' in genres," Li explains. "It is difficult for machines to translate moods to genres, so opportunities to promote the most relevant content get missed." As part of this concept, Li's team built a model to learn the correlation between moods and keywords. From here, the keywords were mapped to titles across its catalogue. Using such a model, people could, in theory, tell the content provider they were feeling sad but wanted to watch uplifting films, or were stressed and wanted to watch relaxing, funny shows.
Sky's patent-pending technology is still largely conceptual, but it offers insight into where personalised TV recommendations and entertainment could be heading – as does the idea of using AI to plan, as well as produce, unique content.
Until now, entertainment has been largely driven by so-called hit culture – the lowest-common-denominator media that attract the largest fanbase. With content providers offering seemingly unlimited choice, audiences today are much more diverse. This, coupled with what each platform's AI knows about its users, allows content creators – or more accurately their algorithms – to choose what to commission next, or to create personalised content for smaller, yet more engaged groups.
"Historically, if a filmmaker did something different they would have had to have gained a lot of attention to make it a success," says Jude Sheeran, VP EMEA at real-time data experts DataStax. "As such, new genres were hard to get off the ground. Today, such projects don't need as much attention. Producers can very quickly determine, using a small sample, if something has potential to be popular, and this has the benefit of both creating more personalised content as well as helping creators find new areas of interest and unlock creativity."
Taking things a step further, future content creation could go beyond personalised to become individualised. Building on the fact that AI assistants can already identify individuals by their voice, and assistants like Alexa are increasingly embedded into connected TVs, the next generation of sets could feature a range of sensors that allow them to instantly recognise who is watching. The content and recommendations would then be highly specific and unique to that individual.
"There is a notional future world where programming is adapted to the individual tastes of each viewer, through the creation and allocation of consumer-specific synthetic faces," says Dr Alex Connock, head of department of the Creative Business MA at the National Film and Television School and author of 'Media Management and Artificial Intelligence: Understanding Media Business Models in the Digital Age'. He continues that this could include solutions where viewers are shown content that features digital versions of themselves, or avatars that are most likely to generate a specific positive response.
With the push for greater personalisation, though, comes an even greater pull on personal data. After all, recommendations are only as good as the algorithms on which they're based, and these are only as good as the data they're trained on.
Sky's AdSmart technology pulls information from its own and third-party sources in order to know what technology, vehicle or property we own to our recent purchase history and even our Experian credit score. Yet in June this year, a report found that while the majority of consumers (63 per cent) are willing to hand over information in return for personalised experiences, it's on the condition that brands use "their own first-party data and not data purchased or rented from third parties."
A pilot scheme, launched by Channel 4 and ITV in 2017, introduced the idea of personalised ads to a small group of people. While the concept was generally welcomed, when the personalisation became too specific, in terms of using individuals' names or other highly personal data in the adverts, the group described it as creepy and invasive.
When it comes to TV then, how can broadcasters offer deep personalisation without alienating customers or disrespecting privacy? Apple is a lead proponent of 'differential privacy', the statistical science of learning as much as possible about a group while learning little about the individuals within it. It's not as insightful as if they had access to the individual data, but it's not a million miles away from how current recommendations work today, in terms of looking for patterns and paths of behaviour. It also combines useful data with high accuracy without sacrificing privacy.
There are also ongoing concerns of personalisation versus polarisation. There is an argument that the deeper the personalisation of content, particularly news, the greater the risk of the so-called filter bubble. It is posited that personalised TV could essentially lead to a rise in advertisers exploiting the personalised, trusted connection to broadcast political or other polarising views. The increased regulation of broadcasting would probably prevent this, but it's another hurdle that providers in this space need to consider.
Yet for all the concerns and dystopian views, personalised TV offers a lot of promise, as Prashant Natarajan, vice president of strategy and products at AI cloud company H2O.ai and author of 'Demystifying AI for the Enterprise' concludes.
"Fundamentally it comes down to this – data is the best surrogate we have today for people's emotions and behaviours in an increasingly digital world, and AI is simply a representation of this data. A way of putting this data and what it represents to use. There will be challenges, but I believe the conversations we're having around AI and personalisation will ultimately create a new digital ecosystem, one which is far more fulfilling for us as humans."
Letters to the editor: volume 17, issue 11
In the December 2022 issue of E&T, readers discuss how we should assess the success of technological progress, the economics of hydropower, and more.