Aller Media achieves personalisation and monetisation with its own system, Xavier

By Kjetil Laumann

Aller Media AS

Oslo, Norway

Connect      

A traditional tabloid news site has different needs than a revenue-driven commercial department. But at the end of the day, they both want the same things: 

  1. To present the right message to the right user at the right time.
  2. To not waste valuable space on the wrong content. 

To meet these needs, Aller Media AS created its own personalisation system: Xavier. 

Fast, solid, adaptive, and flexible 

Around 50 milliseconds after a user has entered one of our pages, Xavier has run its models and is presenting pixel-perfect recommendations. Depending on the user’s behaviour and how the models are tuned, Xavier by then will have removed articles that a user has already read or been exposed to, and will have selected the best articles to present based on popularity, collaborative filtering, and other models. In the case of premium content, the models are tuned to drive subscriptions. 

That speed allows us to personalise any part of our pages in real time. Behind that speed lies a solid, versatile, and flexible system, where having core control over vital components like tracking, user data, content, models, and algorithms is a prerequisite for getting out of the starting block. 

Collaboration, mutual respect, and finding the balance 

While a solid system is crucial to getting anything done, Xavier also has to support editorial and commercial needs. The commercial side is primarily about earning money, while the editorial side has more varied goals: getting users to come back and read as much as possible, delivering the right editorial profile, and supporting journalistic integrity and ethical decisions. 

The commercial side can be handled entirely by Xavier, while the less tangible editorial side requires a balance of automation and manual control. One step we took to provide that editorial control was the introduction of “safe pools” of articles, which are articles that the editors have pre-approved for Xavier to recommend.

These safe pools can be quite broad, such as all articles published in the last two days, or quite restrictive, like a hand-picked list of videos. Single articles can easily be removed from or added to the safe pools when needed. Larger safe pools mean better recommendations, so this is a constant balancing act. 

Large parts of Aller’s pages are handled automatically by Xavier, but editors do have the power to override some of the recommendations. To ensure editors use that power for the right reasons, we constantly supply them with analysis to show them the consequences of their actions. As a rule of thumb, manual overrides should only be used for editorial reasons. While overrides are used to drive traffic and provide readers with a variety of our content, letting Xavier operate automatically is almost always the best option.

Knowing static content has a negative effect on our numbers and that users rarely read the same article twice, we made a solution to provide two different “types” of override: Strong Mode and Weak Mode. In Strong Mode, the editor takes full control over the position and the same content is shown to all users. In Weak Mode, the editor decides what content to display in a given position, but that choice is only presented until the user has either seen the content three times or clicked on the content.

After that, the position will automatically be optimised for that user. Weak Mode ensures that all users see the selected content, but not more than necessary. 

This graph shows that the chance of a user clicking on an article is reduced each time he sees it, and the numbering shows in which order the articles should be recommended for the users. It also indicates that sometimes it can be more valuable to show the same article twice than to show a new one. It is not possible for an editor to act on this manually. But Xavier can do just that.
This graph shows that the chance of a user clicking on an article is reduced each time he sees it, and the numbering shows in which order the articles should be recommended for the users. It also indicates that sometimes it can be more valuable to show the same article twice than to show a new one. It is not possible for an editor to act on this manually. But Xavier can do just that.

On the horizon and beyond 

Xavier is widely used across Aller Media. On Dagbladet, Aller’s biggest digital publication, the majority of the front page is controlled by Xavier. That includes all ad positions, all premium content, the vast majority of videos, and all content marketing, as well as free site articles further down on the page. This has allowed us to apply different models to solve different problems, while using the same underlying technology. 

As a result, we are able to use Xavier to optimise the types of content we show to our users. For example, all content marketing goes through Xavier, and when Xavier decides that there is no content marketing worth showing to a user, it automatically replaces the position with a display ad. At times when display ads are less valuable, we show premium content or videos instead. 

Screenshot of personalised content from Xavier on the frontpage of Dagbladet.no.
Screenshot of personalised content from Xavier on the frontpage of Dagbladet.no.

Our goal is to make Xavier the center of a fully automated, yield-optimised solution that gives the user the most valuable content at any given time — whether that is a video, premium content, Aller’s most important articles, a valuable advertisement, or something completely different.

Image courtesy of Anna Shvets from Pexels.

About Kjetil Laumann

By continuing to browse or by clicking ‘I ACCEPT,’ you agree to the storing of cookies on your device to enhance your site experience. To learn more about how we use cookies, please see our privacy policy.
x

I ACCEPT