Overview of this campaign

Exploratory Goals

• Test a new approach to content personalization using collaborative filtering instead of a topic-based ranking system.
• Experiment with ranking models to identify personalization thresholds that drive engagement and sales.
• Mitigate the algorithm’s bias toward male users aged 55+.

Traffic & Business Goals

• Increase Click-Through Rate (CTR) and Clicks Per User (CPU) across all sections of the automated front page.
• Boost sales performance by improving CTR and CPU on sales teasers.
• Broaden content exposure and enhance engagement, particularly among users aged 20–30.

Process Goals

• Optimize cross-functional collaboration between Aftenposten and Schibsted to tackle complex personalization challenges.

Results for this campaign

Overall Performance

• CTR increase subs (automated front page): +23.5%
• CPU increase (automated front page): +65.2%
• CTR increase logged in non subs (sales teasers in automated section): +32.2%

Impact on Key User Groups

Men/women 20-30 years.

CTR increase subs (automated front page) +29.2%
CTR increase logged in non subs +34.3%

Content Exposure Growth

Articles displayed per user session: +120% (from 130 to 278 per day).

Process Learnings

Close collaboration between data scientists, editors, and engineers was crucial in balancing algorithmic optimization with editorial integrity. Iterative testing refined ranking models to enhance content diversity while maintaining engagement. Addressing bias underscored the value of cross-team expertise in creating a more inclusive recommendation system.

Lessons Learned (and a few bloopers!)

By year-end, 40% of the ranking signal on the front page was powered by collaborative filtering. But we didn’t stop there—we pushed it to 100%. And what happened?

A fully personalized front page didn’t work. Without a mix of ranking signals, the page lost its news pulse, becoming less dynamic and leading users into an over-personalized rabbit hole. Engagement dropped—CTR and CPU plummeted.

The key takeaway? Personalization works best when balanced with editorial judgment and diverse ranking signals.

Contact

To contact a company representative about this campaign, click here for the INMA Member Directory