Bringing machine learning to music – Spotify’s vision to win the hearts and ears of humanity

Is Spotify singing to us what we want to hear?

Bringing machine learning to music – Spotify’s vision to win the hearts and ears of humanity

Every Monday, millions of Spotify users find a new playlist waiting for them on Spotify’s Discover Weekly, a curated list of 30 songs that Spotify’s machine learning algorithm predicts each user will love based on the users’ past listening behaviors [1]. Gone are the days of vinyl records; machine learning has fundamentally altered the way users listen to music, and competition to create ever-sophisticated algorithms that provide more personalized recommendations is intensifying. Enter Spotify, one of the world’s leading on-demand music-streaming services, which boasts 87 million subscribers, 191 million monthly active users, over 40 million tracks, and operations in 65 countries [2].

Trends in the streaming music landscape

Spotify is a company in the online streaming music space. Acquiring consumers depends on delivering the right personalized recommendations to each user. Songza, launched in 2007, is often credited as one of the earliest companies to define the space, which based its music curation process through highly labor-intensive manual curation which was then voted up or down by users [3]. Since then, more sophisticated entrants such as Spotify and Pandora applied machine learning techniques which allowed them to deploy algorithms that analyzed both the audio and lyrics of songs to create personalized recommendations for their users.

Spotify’s strategy

In the short term, Spotify’s strategy is to develop unparalleled personalized music curation services using machine learning. It does this by mixing three machine learning techniques to create an accurate recommendation engine, namely collaborative filtering models, natural language processing models, and audio models, which analyze user behavior, lyrics, and raw audio tracks respectively.

Exhibit 1: Spotify’s Machine learning algorithm [1]
Enabling this innovation is Spotify’s agile approach to its organization structure, which is highly informal. Working groups called “squads” work in loose structures to encourage creative design thinking. In addition, the company mandated that 10% of all employee time is spent on “hack time”, whereby employees were asked to test and come up with new ideas for implementation [4].

In the medium term, Spotify still needs to achieve profitability. Spotify’s main expense is its payments to rights holders, normally publishing labels, for licensing the usage of their songs. Spotify’s current sources of revenue are based on a two-tier model of free and paid subscribers [4]. It may consider levers such as rethinking Premium pricing or lengthening the ad time of its Free offering, but it will need to do so in a way that does not alienate its user base.

Exhibit 2: Free vs Paid offerings—Spotify and rival Pandora [4]
Other considerations

Spotify will need to consider its reliance on published content. Netflix, which used machine learning to provide personalized recommendations for streaming video, managed to begin making profits by adopting a strategy of vertical integration, whereby it began to create its own content and reduce its dependence on licensing content from film and television studios [5]. Given how Spotify has yet to achieve profitability, Spotify’s management may be wise to consider how the company can reduce its reliance on licensed content, though in the short-run the high upfront costs of developing original content may further harm Spotify’s profitability.

Yet another issue is the recent trend towards artist-owned streaming music services. Artists, who increasingly feel that they are getting slimmer portions of the revenue generated from their music, have turned to creating their own B2C solutions that cut out intermediaries such as Spotify. For example, Tidal, launched under the leadership of artist Jay Z, recruited Beyonce, Rihanna, Kanye West, among others [6]. If artists reach out to their fans directly instead of relying on platforms such as Spotify, then Spotify’s value proposition of using machine learning to deliver what users want to hear will be materially diminished since the content can no longer be offered.

Finally, there has been a trend towards artists being encouraged to sign exclusive deals. For example, Apple Music, which is quickly growing to become one of Spotify’s biggest competitors, signed an exclusive deal with the hip-hop artist Drake, formerly Spotify’s biggest artist of the year, to launch content only available via Apple Music [7]. Not only does this directly limit Spotify’s ability to stream more content, but it also indicates that Spotify may have to pay a higher price to attract and retain artists towards its platform.

Open questions

Some are skeptical of Spotify’s long-term ability to remain competitive. In particular:

  1. Despite Spotify’s innovative machine learning solutions, the company remains unprofitable, primarily due to the high cost of licensing of its content. How do you think Spotify can change its business model to achieve profitability?
  2. Given the entrance of large tech companies in this space, such as Apple Music, how can Spotify maintain a technical edge in its personalized recommendation service?

(800 words)

References

[1] Medium. (2017). How Does Spotify Know You So Well? | [online] Available at: https://medium.com/s/story/spotifys-discover-weekly-how-machine-learning-finds-your-new-music-19a41ab76efe [Accessed on 11 Nov 2018].

[2] Spotify. (2018). Spotify Investors page | [online] Available at: https://investors.spotify.com/home/default.aspx [Accessed on 12 Nov 2018].

[3] Wired. (2015). Songza is dead, but it lives on within Google Play Music | Available at: https://www.wired.com/2015/12/songza-is-dead-but-it-lives-on-within-google-play-music/ [Accessed on 11 Nov 2018].

[4] Govert Vroom and Isaac Sastre, Spotify: Face the Music (update 2018), IESE Business School Case. (2018). Available at: https://hbr.org/product/spotify-face-the-music/IES473-PDF-ENG [Accessed on 12 Nov 2018].

[5] Wired. (2017). Netflix profits up 56% as original content splurge pays off | Available at: https://www.wired.co.uk/article/netflix-2016-earnings-revenue-original-shows [Accessed on 12 Nov 2018].

[6] Billboard. (2015). It’s Official: Jay Z’s Historic Tidal Launches With 16 Artist Stakeholders | Available at: https://www.billboard.com/articles/news/6509498/jay-z-tidal-launch-artist-stakeholders [Accessed on 12 Nov 2018].

[7] Music Business Worldwide (2016). Apple Music’s biggest swipe at Spotify yet: Drake’s exclusive new album | Available at: https://www.musicbusinessworldwide.com/apple-musics-biggest-swipe-at-spotify-yet-drakes-exclusive-new-album/ [Accessed on 12 Nov 2018].

 

Previous:

Predictions for Patton: Using AI for Maintenance of the Bradley Fighting Vehicle

Next:

Teladoc about Machine Learning

7 thoughts on “Bringing machine learning to music – Spotify’s vision to win the hearts and ears of humanity

  1. Thanks for sharing! I’m a big Spotify user and enjoyed your read. Another angle I think Spotify can take to be competitive is to focus more on the social element of the platform like the ability to follow friends’ playlists. However, if it were to do so, I think it would need to invest heavily and try to exclusively partner with major platforms like Facebook. Otherwise, Apple would likely catch up quickly.

  2. Awesome post. I am a big fan and user of Spotify — actually I almost did my essay on the same topic. One thing Spotify could make to improve their algorithms is finding a way to “filter out” the cases when you are asked by a friend to play a random song — which you may not like — but ends up impacting the intelligent feed Spotify generates for you. For example, I mostly listen to rock music but when working out I like to switch to more electronic stuff. More drastically, sometimes I use Spotify to play songs to my kids when driving my car. As a result, I get this random “recommendations” from Spotify that have nothing to do with the music I actually like. I though maybe their algorithms could identify these outliers and avoid messing my cautiously curated recommendations list from Spotify…

  3. Enjoyed reading this! I actually also wrote about Spotify so I enjoyed reading a different take on it. In your write-up, you acknowledge the key concern of the financial viability of the company, which I largely ignored in my exploration of the topic. With the rising competitive pressures that you clearly acknowledge above, Spotify will be forced to continue to invest capital in its product innovation, most notably its machine learning capabilities. For this reason, you are right to be concerned about profitability. Thus far, Spotify has taken a strict stance against in-app advertising historically, but as I think about the future of the product, it may be one of the only ways to make the product financially viable. I wonder if there is a way for Spotify to leverage its deep knowledge of the consumer to post adds that are so accurate that they don’t “annoy” the consumer and detract from the listening experience.

  4. I had a fun time reading this post! On your two questions:
    1) I believe machine learning could be a huge contributor to Spotify’s revenue. I would explore Spotify’s content mix against licensing costs. A lot of the big data analysis around Spotify mentions that most of the top played songs are not actually Billboard Hot 100 hits but more dated songs (e.g. Mr Brightside, released 2004). There’s an opportunity to do revivals for popular songs that may not have as exorbitant licensing fees, as they’re more dated. Another revenue stream could potentially be providing consulting services to media and entertainment companies based on their data findings.

    2) One edge that Spotify currently has and could exploit further is its integration into more technology ecosystems. A simple but strong example is that whenever people identify a song on Spotify or Shazam, there is an option to add the song on Spotify. I would tie Spotify to more points of consumer contact around music discovery to keep relevance and active usage. Unlike big tech companies, Spotify does not have adjacent hardware and software, and thus relies on partnerships which could be the quickest way to remaining indispensable.

  5. Great essay! I think a smart way forward for Spotify could be to invest more in the podcast or even documentary space, which should make it a broader stop for entertainment. An additional avenue is investing in hosting more music videos, or becoming a platform for younger artists.

  6. I agree that their business model might be unsustainable over time. I would say that they need to go to the offline world. I believe that music is now a commodity and everyone can get it cheap and quick, so one way to monetize would be to create experiences exclusive to their customers, such as music festivals, ticket selling platforms and souvenirs. The upside on the strategy is that they know who likes what and where they live, so it seems extremely easy to target the right clients.

  7. Loved, loved, loved this post! I also like your catch phrase – “Is Spotify singing to us what we want to hear?”

    I think you raise an interesting question – given the entrance of large tech companies in this space, such as Apple Music, how can Spotify maintain a technical edge in its personalized recommendation service? I think to push the question further, we have to ask how does Spotify create a sticky customer? Can the recommendation service enhance or deter the process of locking in customers? I think that it can – through the point you raised about copying the Netflix strategy and vertically integrating so that it can use its understanding of customer music preferences to create its own content.

Leave a comment