Schibsted AI Boosts Subscription Sales – Case Study

Key Takeaways from the ‍Text:

This text details how Schibsted implemented a new real-time content proposal system to substantially boost subscription sales. Here’s a breakdown of the key points:

Data Efficiency: The team‍ successfully ⁤built a highly effective recommendation system using a ‍ small set of features (12) derived from existing data sources (age/gender predictions from ⁤advertising, sales data). They started with 158 potential data points. Real-time Adaptation: ⁢Unlike customary batch processing, the system generates recommendations on-demand and adapts instantly to⁣ new user data, including contextual factors like time ⁣of⁣ day and day of the week.
Integration with Existing System: The new model seamlessly integrates with Schibsted’s existing content recommendation system (“Curate”), combining machine learning insights with editorial control.
Addressing Data Scarcity: Schibsted recognized they have less ‍user data than competitors (like app and social media companies) and focused on “selling⁣ the strawberries you have at hand” – maximizing the value of the data they do have. This ⁣is notably important for anonymous users.
Notable Results: The new model, when tested on the front page, led to a 75% average increase in subscription sales. Results were described as “staggering” despite limited data.
Technical Challenges: Building a scalable⁢ online feature store was a major software engineering hurdle,‍ as no suitable off-the-shelf solutions existed⁣ at the time.
* Future Development: The ⁤team⁣ is now working to expand the model ⁢to optimize content for anonymous users to increase engagement,not just sales,and adding more data sources.

In essence, the text highlights a successful case study of leveraging limited data, real-time ⁢processing, ⁣and smart integration to achieve substantial business results in content recommendation.