Spotify Unwrapped

Spotify users rise as we look at the science of the streaming service’s $13.34 billion CAD (9.66 EUR) revenue and how they keep us coming back. Founded in Sweden in 2008, the application is renowned for connecting artists to billions of listeners, making them the #1 choice in music streaming (yes, above Apple Music). Not only this, but the service has mastered customized playlists, new music discovery, and a year-end gift users celebrate like a national holiday. Much of this success can, and quite literally has been calculated by the advanced algorithms behind the scenes. Read along as we investigate the black box that we will see is not quite so opaque. 

The Basics 

The nitty-gritty of the algorithm itself is hard to define, though some key elements have been extracted over time. It is reported a song played for more than 30 seconds registers a positive rating within your account, and from an artist’s perspective, an average user interaction of 1.8 plays per song, is recognized to be of high merit. This recognition is then used to generate further engagement with similar songs, artists, and genres via categorized offerings such as “Friday Night, right?” or “Country Crying” playlists — obviously very user-specific. Keep these metrics in mind as the final weeks of listening count down ahead of your 2022 Spotify wrapped… more on that later. 

Acquiring the Algorithm 

We can assume the basics of the algorithm; if you like one song, you may like a similar one. But how does Spotify define like and how does Spotify define similar? The company has strengthened their understanding of users through multiple business acquisitions ⁠—just one stop in uncovering the fundamentals of the algorithm. 

Acquisitions include $100 million USD music intelligence start-up, Echo Nest (2012), a company which came equipped with over 1 billion bits of musical data. This data, as boasted by Echo Nest CEO Jim Lucchese, is nearly infinite. With Spotify’s most engaged users joining the platform more than 20 days per month at an average of one hour per session, there is both a wide and deep set to play with. In playing with this dataset—the company, Echo Nest, helps generate what Spotify has coined as “Taste Profiles”. Classifiable attributes are as unique as hotness, mainstreamness, and even danceability. This feeds into a relative scoring system which ranks your attribution to each category against other users’. The technology also leverages common data analysis tool—clustering. This is an intuitive conceptualization which gathers like music (or users, artists, playlists, etc.) based upon closeness when positioned on a multi-dimensional attribute map. Whether “This is Adele” or “Hype Workout Mix” is what you are in pursuit of, you can imagine how this technology would aid in clustering songs together.  

Next up, the acquisition of data science consultancy Seed Scientific (2015). In bringing specialized data analysis intel in-house, this acquisition played a large role in understanding consumer usage habits, independent of music taste. Joining Spotify from a landscape of commercial, public, and social algorithmic understanding, Social Seed extracts math and science insights for cold-fact and actionable direction. Notably, this team was servicing Beats Music (owned by Apple) as well as many other recognizable Fortune 500 brands at the time of acquisition. The start-up now provides their sought-after services exclusively to Spotify.  

Lastly, Sonalytic (2017). An audio detection software similar to the likes of Shazam, and an acquisition in which many of the details were not shared. Interestingly, the Sonalytics capabilities are two-fold. Applied to recommendation generation, similar songs can be identified with the technology and matched with users listening preferences. Moreover, the technology can be used to recognize snippets in remixes, sampling, or covers to credit copyright owners where credit is due. Unrelated to the core mission of the algorithm, this functionality helps to extend trust and loyalty in both artists and listeners.  

How Users Interact with the Algorithm 

One of the most public facing elements of Spotify’s algorithm, making their back box white, is the cherished Spotify Wrapped. Since 2016, Spotify has released to its subscribers the colourful project which summarizes listening habits such as listening minutes, top tracks, top artists, even single song replay count—and in increasingly creative ways. The release has become a pop-culture centerpiece shared by users and artists, and even artists have recognized fans thanks to the release (the ultimate gift). Most recently, the 2021 Spotify Wrapped reaped 90 million consumer interactions.  

Through an algorithmic lens, what this service does is it removes a digital divide. Where users once potentially could not understand the data being collected from their profile, it is now served to them on a silver platter. Captured by one Twitter user, “Spotify is the only tech company to figure out how to successfully rebrand "we've been tracking you" to "isn't this fun".” In doing so, Spotify leverages the idea that being aware of an algorithms functions and scope is a digital power when in the hands of the consumer. In a study performed by the Routledge group, it was uncovered the strong correlation between algorithm awareness and (positive) attitude towards the algorithm (where p<0.001).  

Even in a more human-like application, we can observe the success of Spotify Wrapped. Year-over-year, friends compete to have the highest number of listening minutes, to be in the smallest percentage of their favourite artist’s listeners, have the most obscure spread of top genres, the goals go on. Ultimately, Spotify has mastered consumer engagement, all by facing algorithmic superpowers, outwards.