“Automated content recommendations emerged victorious”: MEDIAHUIS doubles down on AI

Editors note: We are republishing an article by WNIP that looks into the experience of Mediahuis in using AI-engine Froomle. This piece was originally published on What’s New in Publishing.

Mediahuis is one of Europe’s leading multi-platform publishers with national newspaper brands covering Belgium, Germany, the Netherlands, Luxembourg, and Ireland including renowned brands Het Nieuwsblad (Belgium) and De Telegraaf (Netherlands) as well as a number of regional titles. 

In 2017 Mediahuis partnered with Froomle, a leading AI-driven self-learning recommendation engine, to increase audience engagement via hyper-personalized online experiences. Froomle was originally created in partnership with researchers from the University of Antwerp, Netflix, and Apple, and the platform is being continually tested by academics and data scientists. 

The specific challenge? To increase reader engagement and loyalty through a variety of channels, including email, push notifications, websites, and mobile applications. After showing consistent improvements with use cases across its Belgian brands, Mediahuis brought in Froomle to work with De Telegraaf – the largest Dutch newspaper in the Netherlands – to test whether hand-picked article selections on their homepage or an automated personalized solution would result in higher engagement.

As part of the test, De Telegraaf chose to implement the Froomle ‘Recommended for You’ module in two places on its homepage: almost directly below the top header and then further down the homepage.

After a few weeks of testing, the results showed a marked increase in reader engagement as well as a broadening of home page article diversity. Specific results included a 25% higher CTR on automated personalized positions vs handpicked recommendations (20% on mobile, 30% on desktop) as well as 4.5% more clicks overall on the home page with the personalized experience. Froomle recommendations also contained 100% more diversity in terms of content topics compared to hand-picked recommendations.

Paul Janssen, Editor-in-Chief of De Telegraaf, told WNIP, “On the one hand, you shouldn’t personalize news 100%, but you also shouldn’t be afraid to use automated personalization strategically because the impact on the overall relevance of news for users is undisputed.”

“Getting started with something is complex, getting started with something as crucial (and difficult) as personalization is even more complex. So, the fact that we, together, were able to set things in motion is a great plus for us. We now do have to take next steps of course, that’s crucial for getting sustainable positive results. We are indeed by far not there yet”, said Riske Betten, Product Director Mediahuis Netherland

Increasing session length

After becoming a major shareholder in Medienhaus Aachen GmbH at the beginning of 2022, a key goal of Mediahuis was to increase the time spent by readers on the Aachener Zeitung website, a leading German news website.

The ‘Continue Reading’ solution it settled upon, once again in partnership with Froomle, used AI algorithms to optimize recommendations for the specific audience of the newspaper. Since implementation, the Continue Reading AI module generated an additional 30k pageviews each month with 5% longer sessions on average than users who did not receive ‘Continue Reading’ recommendations. 

Regional news recommendation

Having worked with Flemish Newspaper Het Nieuwsblad since the first days of the collaboration, Froomle has been constantly updating and testing new algorithms on its homepage. Most recently, they were challenged with offering relevant regional news high up on the homepage. The solution involved running an A/B test on a specific position: a baseline that had non-personalized recommendations, consisting of the most popular regional articles across regions vs the Froomle personalized recommendations that were based on user profile and reading behavior. 

The results of this test showed CTR doubled with the Froomle recommendations, from 1.8% to 3.6%. More information about the research done by the R&D team at Froomle on this type of recommendation can be found here.

Conclusion

Many publishers still use hand-picked content recommendations to good effect, increasing user engagement and dwell times. However, as artificial intelligence continues to evolve, the evidence now points clearly to the fact that AI-driven content recommendations far outstrip their human mentors – the evidence from a variety of European publishers is conclusive. On top of better results, AI also reduces the workload for newsrooms in repetitive tasks while still giving them control they need. And this AI is only going to get better.

Source of the cover photo: https://unsplash.com/

---