Winning as a modern content business: what would you recommend?

< Blog home

Bibblio co-founder Mads Holmen was at the The AI Summit in New York last week with the Department of International Trade, explaining how recommendations powered by AI have become a driving force in digital content. Read on to find out what it's worth to Netflix, and how we ended up in a war for attention.

"Hello guys, I’m here to talk about content recommendation and AI. It’s something we all interact with every day, but you might not be aware of this. So, let me give you a quick introduction. One of our clients is AFAR.com - a travel site. They’re a nice example of how sites can use content recommendation to engage users. Someone is reading an article, they get to the bottom of the page and, rather than just a blank space, there's some more content suggested to them:

AFAR.com's recommendations
Content recommendations at the bottom of articles on afar.com

"Bibblio’s role here is powering the AI that's figuring out what should be suggested to each user. Now that I’ve given you a frame of reference, I actually want to start by rewinding a little. The story of recommendations is far bigger than Bibblio and our clients: it’s really the story of the web as a whole. About ten years ago, if you spoke to any of the big websites they would say that 80 or 90% of their traffic came directly to their site. These days, that same 80 to 90% is coming through Google or social media. Most of the content we now consume is ‘recommended’ - I’ll come back to this later - by one of those sources.

The article page is the new front page

"This pattern of consumption has disrupted the way sites work, because most people rarely go to the ‘front page’ any more: they land straight on a piece of content. So, we jokingly say the content page is actually the new front page, because that’s where the majority of users land. That’s changed the way sites need to operate, unless they want to end up at the mercy of the big platforms as just search results and social media links. They need to really invest in their article pages, because they only get that one page to try to engage people - to turn one page view into two and create loyal subscribers in the future.

"More broadly, what this actually highlights is that there’s a real war going on for online attention, and the search and social giants are winning. This is Herbert Simon. In 1971, he looked into his crystal ball and said these words: “Information consumes the attention of its recipients.” He was right, and it follows that when the amount of information increases exponentially, the attention span of recipients must curve the other way.

The war for attention

"Lo and behold, when we look across the web now, users spend over 10 billion hours every week on Instagram, Netflix, YouTube and Facebook, and increasingly less time in other places. In many ways, these companies have taken control over the rest of the content ecosystem. In the publishing and media world they’re often called frenemies, and that’s being pretty friendly.

"I call this situation the 100-billion-dollar question, because that’s about the amount of money Google and Facebook have taken out of the publishing world in the last ten years. And it goes further: Reed Hastings, the CEO of Netflix, recently said that their biggest competitor is now sleep. At the end of the day, anything that consumes someone’s attention is increasingly competing with everything else. And at some point, if you’re scrolling enough on Facebook and Instagram, that competition becomes sleep, or time with friends or your better half, and everybody suffers. There are only 24 hours in a day, and time is the one truly finite commodity that humans have.

10 billion hours a week on social media

"At the moment – and I’ve spent 10 years in the online advertising industry – we, people, are being sold as a commodity. Our time and attention is aggregated and sold. To capture more and more of it these companies are using content recommendation systems.

"Think about Facebook’s newsfeed. It’s a recommendation engine. In fact, over 85% of time that people spend on these platforms is a result of algorithmic recommendations. It’s a never-ending personalized feed, a never-ending personalized reflection of what you’ve liked in the past. And we’re being fed more and more.

"Netflix actually quantified what content recommendation means to them annually: one billion dollars - and that was last year, the figure’s almost certainly higher now. That’s about 12% of their revenue attributed to content recommendation. So recommendation is good for business. But what is a good recommendation?

The ingredients of a good recommendation

"This is something I can talk about for hours, but I’ll give you a very brief intro. There are broadly three things that matter to a recommendation. Firstly, understanding the content – specifically, its relevance to other content. So say I’m reading an article on North Korea, perhaps I’m interested in reading more about this topic. At Bibblio we use a whole bunch of semantic analysis, topic modelling and clustering algorithms to figure out which content is most relevant at that moment to the piece that you’re reading about North Korea.

"Secondly, there’s understanding generalized user behavior. The best example is probably the way Amazon recommends products – others users who liked that item you’re looking at also like this. It’s actually not a true personalization algorithm, but a general behavioral one.

"Thirdly and finally, you can’t do without the truly personalized models. Personalization, understanding users at the individual level, is really hard. Algorithms are still not very good at understanding humans. But a great example here is Facebook Newsfeed. It looks at about 120 signals to figure out what it should show every individual.

"Where AI really features in content recommendation is in the data network effects. The more behavioral and user data we get, the better we can train the models. This raises some very interesting ethical questions. What should recommendation do? Should it give people more and more of the stuff they like, or should it also sometimes challenge them? Perhaps we need to show them stuff even when we don’t know if they are going to ‘like’ it, because if you keep just showing people more of what they like, they might never know what else is out there.

Our recommendation AI delivers improved results for both users and publishers over time

"One of the most fascinating things about recommendations is that whenever we choose to show people something, we also choose not to show them a whole bunch of other stuff. We’re making very real choices here that are beginning to shape people’s worlds. I won’t dive deep into fake news right now. One thing I do want to say though is that things like the Facebook Newsfeed algorithm have done a lot to encourage publishers to create sugary content that gets clicked on and shared but which might not be very good for people.

"We talk a lot about an obesity epidemic in the physical world: we have also an obesity epidemic in the digital world! We get fed a lot of sugary content that’s designed to drive clicks and shares but doesn’t help users stay ‘healthy’ or follow any ethical guidelines. How sustainable is this in the long run?

"This talk has hopefully given you a bit of insight into why recommendations are important. Get in touch if you’re interested in finding out how recommendations can lift your site’s performance, drive better engagement or want to know more about the clients we’re working with. If content is your main product, you should speak to us! Thank you."

Recommended for you