Applying Spotify’s Recommendation Experience to New Markets: a Promising Model?
Spotify has had a very successful last couple of years, counting more than 60 million subscribers and 140 million monthly active users in 2017. Part of their success has been largely due to the launch of 3 new products: Discover Weekly, Release Radar & Daily Mix. They are playlists that require no effort from the user to get access to, as they are directly provided through the home dashboard.
More importantly, they all share the common trait of being specifically tailored to each individual on the platform. And that is what users get absolutely fond of. The recommendations are very accurate and provide users with new and enjoyable material, updated every week. They deliver “the right music at the right time”. What is the secret behind this recommendation model, that appears to be attracting and retaining so many people? We will describe the different ingredients that make up for a robust yet simple model, based on a colossal amount of behavioural data and community-generated playlists. Because of its success, one may raise the question of replicability of the aforementioned model to other markets. What if we could tailor clothing, travel or food recommendations based on a similar pattern?
We will try to apply Spotify’s model to those contexts and assess potential opportunities.
A robust yet simple model of recommendation…
The model relies on two key components: content relationships and user relationships.
Content relationships
Links are being made between similar content, based on simple collaborative filtering and clustering.
User relationships
Simultaneously, Spotify establishes relationships between users through recorded behaviour on the platform. This behaviour is translated into a “user profile”.
Those profiles are then compared one to another among the 60 Million subscribers to define “closest neighboors”, or people with similar musical tastes.
Content/User Matrix
Both relationship sets are then brought together into a matrix, leading to the recommended playlists
… applicable to other markets
The question was raised in the context of a class at Carnegie Mellon University called “Tools for UX Design”. As I was presenting Spotify’s recommendation model, a student working on a clothing project wondered if such model could be adapted for such a market. I decided to try and find a way to leverage the same logic and apply it to the clothing sales business:
While we observe an increased complexity of behaviour in the context of clothing, we can imagine that this model could be successfully implemented. Tracks become pieces of clothing, playlists outfits and Social Media provides the platform for behaviour recording.
Closing Thoughts
While Spotify’s recommendation model is simple, it provides a strong added value to its platform, bringing the brand well ahead of its competitors. Other markets might find it interesting to tap into their model and seek for opportunities in building structurally similar recommendations. Food, travel and other markets are yet to be explored…