Content Recommendations for Streaming

The problem

MediaStream had 50,000 titles but users struggled to find content they liked. Their basic "because you watched" recommendations weren't driving engagement. Users were churning because they couldn't discover content that matched their tastes.

What we built

We built a recommendation system that combines collaborative filtering, content analysis, and contextual signals. The system understands not just what users watched, but how they watched it (binge-watched, abandoned, rewatched). It generates personalized carousels, search rankings, and email recommendations.

Tech stack

Results

User engagement increased 67% as measured by hours watched per user. Content discovery through recommendations grew from 15% to 45% of all views. Monthly churn decreased by 2.3 percentage points, worth $15M annually in retained revenue.