I joined Spotify in 2008 to focus on machine learning and music recommendations. It's easy to forget, but Spotify's key differentiator back then was the low-latency playback. People would say that it felt like they had the music on their own hard drive.

Erik Bernhardsson

Drawing on experience building ML at Spotify, the author argues that machine learning is rarely a key differentiator in a product's first five years. Most ML is an 'enhancer' rather than an 'enabler' — it squeezes incremental gains on top of already-established products. For ML to be a sustainable competitive advantage, companies need to pair it with non-ML moats: unique proprietary data, sticky enterprise contracts, large user bases, or hard-to-replicate integrations. Pure ML techniques commoditize quickly as research gets published, but institutional knowledge from years of production iteration and human capital remain durable assets.

When machine learning matters

Most machine learning is sprinkles on the top