Spotify Discover Weekly: Machine Learning as Product

In 2015, Spotify quietly shipped a personalized playlist that landed in users' apps every Monday morning. Within months it had driven billions of streams, and people had started rearranging their week around it. The company was sitting on enormous amounts of listening data, but the real challenge wasn't building a smarter algorithm. It was turning invisible back-end data into something a user could open, feel, and look forward to. A small team combined three signals into a 30-song mix that felt handpicked but was entirely machine-generated.

For founders and operators, this case sharpens a question most companies never ask of their own data: whether the advantage hiding in your infrastructure could become a feature users love rather than an operational edge nobody sees. It probes how a product can compound, getting better the more it's used. What made this one of Spotify's most impactful launches, and the loop that powered it, are inside the app.