Spotify: Jammin’ Away with Machine Learning
How Spotify uses machine learning to create personalized experience
The music industry owes its survival to streaming services, which contributed 65% of U.S. music industry revenues in 2017 [1]. The current market leader in number of subscribers, Spotify relies heavily on machine learning to create a highly personalized product for music consumption. Moving forward, their critical advantage will continue to be their ability to use their evolving dataset of customer preferences to refine and build a customer-centric product. Machine learning is central to Spotify’s core product development, as each customer interacts with content that evolves constantly based on customer behavior and feedback.
Spotify has shown its commitment to cutting-edge machine learning through a series of acquisitions of startups with advanced AI search and content recommendation algorithms [2]. In the short term, Spotify will continue to use these capabilities to personalize playlists and content for their 83 million paying subscribers [3]. Spotify uses machine learning based recommendation engines to package and elevate the most relevant content for each user with a variety of radio and playlist products, including Daily Mix, Discover Weekly, Release Radar, and Spotify Radio [4]. The result is essentially a curated “news feed” of content for each user based on their historical use as well as behavior of users with similar profiles. Spotify can use this personalized content to drive users towards specific artists, based on both user preference as well as Spotify’s strategic goals.
According to Spotify Insights, part of their goal with personalized, algorithmic playlists is to drive an increase in artist diversity for customers, which correlates with an increase in the number of minutes users spend on Spotify [5]. This capability likely has benefits down the road as Spotify may be able to drive users towards artists that Spotify is prioritizing for strategic reasons like cost. Beyond the consumer, Spotify is beginning to create products to share insights with other stakeholders in the music ecosystem. For example, Spotify recently released an analytics tool for music publishers focused on new opportunities and customer trends [6]. In the short term, Spotify should continue to create data driven products that assist other players in the ecosystem with music and lyric idea generation and development to supplement core processes in music sourcing, development, and promotion across the value chain. Spotify has hinted at these aspirations: “We want to be the R&D department for the entire music industry,” said Gustav Söderström, Spotify’s chief R&D officer, “We don’t think the industry has ever had an R&D department before — and we’re it. That’s our mission [7].”
As the music industry evolves, traditional record labels stand to lose much of their influence as streaming players dominate the customer interaction and distribution. Right now, record labels like Sony and Universal still retain significant bargaining power, as they own the content of today’s most popular artists. Spotify has historically paid just over 50% of its revenue to major record labels and does not directly own the rights to any content [8]. However, Spotify is quietly shifting into more direct partnerships with artists and it seems likely that in the future music landscape, fewer artists would need record labels. With this future in mind, machine learning could very much enhance Spotify’s strategic direction with both artist sourcing and product development. With machine learning, Spotify will have greater forecasting visibility into future musical trends and could use this to identify and develop new artist talent. Using their existing dataset of customer preferences and trends, Spotify should identify unaffiliated new artists with potential that they can invest in. Current record labels will be at a disadvantage without Spotify’s data stream as Spotify moves to fill the sourcing and marketing role that record labels have traditionally held.
Additionally, Spotify has an opportunity to further expand its role in the music supply chain process, by using machine learning to supplement or replace the traditional music development process. Last year, Spotify faced controversy that it was promoting AI composed music in its ambient and chill playlists [9]. While Spotify has denied giving preference to AI composed music, they recently hired a pioneer in AI composed music to lead Spotify’s Creative Technology Research Lab [10], as well as a Chief Content Officer [11]. It seems that content development, whether through machine learning or direct artist partnership would allow Spotify to control more of the music supply chain and reduce dependency on record labels in the future.
The two open questions that I have for my classmates on this topic is:
- How will Spotify retain a strategic edge against Apple Music? Besides a head start, how are they differentiated, particularly given Apple’s ubiquity in consumer’s lives?
- What role do you see machine learning playing in the composition of music? How do you think artists and songwriters will use machine learning to create music?
(Word Count: 781)
Sources:
[1] 2017 Year-End Industry Revenue Report. Recording Industry Association of America. March 22, 2018. https://www.riaa.com/riaa-releases-2017-year-end-music-industry-revenue-report/ Accessed November 2018.
[2] Jon Russell. “Spotify buys AI startup Niland to develop its music personalization and recommendations” May 18 2018. https://techcrunch.com/2017/05/18/spotify-buys-ai-startup-niland/ Accessed November 2018.
[3]“Number of paying Spotify subscribers worldwide from July 2010 to June 2018.” Statista. June 2018. https://www.statista.com/statistics/244995/number-of-paying-spotify-subscribers/ Accessed November 2018.
[4] “Our Spotify Cheat Sheet: 4 Ways to Find Your Next Favorite Song.” For the Record, Inside Spotify. November 2 2018. https://newsroom.spotify.com/2018-11-02/our-spotify-cheat-sheet-4-ways-to-find-your-next-favorite-song/ Accessed November 2018.
[5] David Erlandsson. “Listening Diversity Increases Nearly 40% on Spotify.” Spotify Insights. November 2 2018. https://insights.spotify.com/us/2017/11/02/listening-diversity-spotify/ Accessed November 2018.
[6] “Introducing Spotify Publishing Analytics in Beta.” Spotify Newsroom. November 8 2018. https://newsroom.spotify.com/2018-11-08/introducing-spotify-publishing-analytics-in-beta/ Accessed November 2018.
[7] Cherie Hu. “Spotify Wants to be the ‘R&D Department for the Entire Music Industry-What Does That Actually Mean?” Billboard. April 26 2018. https://www.billboard.com/articles/business/8376561/spotify-rd-department-entire-music-industry-netflix Accessed November 2018.
[8] Ben Sisario. “A New Spotify Initiative Makes Record Labels Nervous” New York Times. September 6 2018. https://www.nytimes.com/2018/09/06/business/media/spotify-music-industry-record-labels.html Accessed November 2018.
[9] Tim Ingham. “Spotify Denies it’s Playlisting Fake Artists. So why are all these fake artists on its playlists?.” Music Business Worldwide. July 9 2017. https://www.musicbusinessworldwide.com/spotify-denies-its-playlisting-fake-artists-so-why-are-all-these-fake-artists-on-its-playlists/ Accessed November 2018.
[10] Tim Ingham. “Spotify’s Scientist: Artificial Intelligence Should be Embraced, Not Feared, By the Music Business” Music Business Worldwide. January 22 2018. https://www.musicbusinessworldwide.com/spotifys-scientist-artificial-intelligence-should-be-embraced-not-feared-by-the-music-business/ Accessed November 2018.
[11] Ben Sisario. “Spotify, Nodding to Broader Ambitions, Hires Chief Content Officer” New York Times. June 26 2018. https://www.nytimes.com/2018/06/26/business/media/spotify-dawn-ostroff.html Accessed November 2018.
As an avid Spotify user, I really enjoyed learning about how Spotify is using machine learning. To answer your second question, I think there is a delicate balance between delivering music that consumers are known to like hearing versus “new” music. Incorporating too much machine learning into the composition of music could potentially stifle artists’ creativity, which would prevent consumers from discovering new genres and songs that they otherwise might start liking. Another risk is changing consumer preferences with regard to music – I would consumers prefer different types of music as they age so machine learning based on prior preferences may not be as relevant on a go forward basis.
Spotify totally changed how I consumer music and I am an avid user of their playlists. The essay poses interesting questions about their competitiveness, but I think that, unless there is a massive shift in the competitive landscape, Spotify will continue to be the market leader in music streaming due to the switching costs associated with changing services. Music streaming is based on playlists, either ones created by the listener or the service. As long as Spotify continues to offer an enormous music library and give its listeners the ability to discover new music via its playlists, the switching costs of having to recreate their own playlists are too high for the average user.
Thanks for the great read. I love how Spotify uses machine learning to cut out the middle man – record labels. By empowering artists with analytics and a platform to share their music, the artists themselves have more creative control over the music that they produce, resulting in a more authentic representation of their work.
I believe machine learning will have a greater impact on electronic dance music than on other genres. While a machine might be able to be trained on the logic behind what makes music sound good, the musical output is still very much limited by its ability to actually produce an analog sound (eg. guitar strings, percussion etc.) Other than in EDM, current attempts at re-creating analog sounds have been relatively unsuccessful. At best, the machine might be able to use samples cut from existing bits of music (eg. breakbeats).