Spotify

Spotify is a music and podcast streaming service that turned a shared catalogue of tens of millions of tracks into something that feels privately assembled for each listener. The through-line across its stories is that personalisation is not a feature layered on top of the product but the mechanism of the product itself: the same model that decides what to recommend also decides what a brand new user sees in their first session, which is why the algorithm doubles as the growth engine.

Personalisation: one platform, hundreds of millions of versions of it

The problem. A single catalogue treats every listener identically, leaving the work of finding something to love entirely to the user. At Spotify's scale, that generic front door served no one well: hundreds of millions of people arrive with different artists, moods, and habits, and a neutral container of songs cannot bend to any of them. The question was whether to stay a warehouse of music or become something that actively understood each person.

The approach. Spotify built personalisation into the core surfaces of the app (Home, playlists, recommendations) driven by machine learning, including a multi-armed bandit framework on Home that balances exploration against exploitation. Music is a forgiving medium to learn from: as VP of Personalization Oskar Stål noted, people take roughly 20 seconds to decide how they feel about a song, and around half a trillion events (searches, listens, likes) flow through Spotify each day, giving the system fast, frequent feedback to refine on.

How it solved it. The result was that the catalogue stopped being shared. As Spotify Engineering put it in December 2021, "We may have a single platform with 381 million different users, but it may actually be more accurate to say there are 381 million individual versions of Spotify, each one filled with different Home pages, playlists, and recommendations." The same logic extends even to podcasts, which are harder because it takes longer to learn whether a listener likes a given show.

Acquisition Economics: the free tier as the acquisition channel

The problem. Convincing people to pay for music was hard when the audience was still on the fence about paying at all. At its 2018 direct listing, Spotify pointed to a market of roughly 1.3 billion payment-enabled phones where only about 12% subscribed to any music service. Buying those users through paid marketing would have been expensive and slow against that much hesitation.

The approach. Instead of spending to acquire paying customers directly, Spotify made the free, ad-supported tier the top of the funnel and used personalisation plus deliberate friction (shuffle-only on mobile, ad interruptions, no offline playback) to pull free listeners toward Premium. New users pick a few favourite artists at signup, so the recommendation engine has signal to make the very first session relevant rather than serving up an empty library.

How it solved it. The freemium funnel converted users Spotify had acquired essentially for free: the company reported at its 2018 investor day that about 60% of its paying subscribers had started on the free tier, and executives framed the model with the line "the more you play, the more you're gonna pay." Because those converters entered through the free product rather than paid ads, the effective acquisition cost for the paying base stayed extraordinarily low.