Leveraging Machine Learning at Spotify
Will Discover Weekly and other sources of differentiation allow Spotify to survive?
Machine learning as Spotify’s differentiator
With 159 million active users streaming 25 hours per month, Spotify is currently the largest music-streaming service in the world [1]. While the industry is growing in both users and streaming engagement, the space remains extremely competitive with big-name players like Apple, Amazon and Pandora. As of summer 2018, Apple Music surpassed Spotify in terms of paid subscribers in the US, and the gap expected to widen by this Christmas [2].
Spotify’s ‘Discover Weekly’ functionality, a personally curated playlist provided every week to subscribers, however, allows users to discover new music and is a key differentiator which allows Spotify to remain competitive. To deliver this unique and personalized experience, Spotify leverages three machine learning approaches in predicting what tracks go into a subscriber’s playlist: “Collaboration filtering (what you listen to vs. other listeners to), natural language processing (analyzing content on blogs and websites on the internet) and audio analysis (analyzing data behind each music track)” [2]. This level of personalization is “tough to copy and Spotify’s scale gives it a distinct advantage” over competitors [1].
Strategic management decisions
In the short-term, Spotify is making strategic acquisitions in the machine learning space in order to further develop their recommendation engine and algorithms. With their 2017 acquisition of Niland, for instance, Spotify will use “API-based product and machine learning to provide its users with better search and recommendations to help them discover music they like” [3]. Earlier in 2017, they also acquired MightyTV, a service for recommending TV and films that users might enjoy based on their tastes [4], and Sonyaltyic, the makers of an audio detection technology that can identify “songs, mixed content and audio clips […] and aid in music discovery” [5]. These acquisitions allow Spotify to push innovation in their recommendation algorithms and continue delivering the best Discover recommendations to subscribers.
In the medium-term, Spotify plans to use machine learning to alternatively help artists and labels with music and content creation. As the space of AI-assisted music has witnessed a “dramatic growth in ability and output” in recent years through “projects like Google’s Magenta and IBM’s Watson that feed musical rules into machines and teach them to mimic human creativity,” Spotify has also started to make long-term investments in machine learning for music creation. Within the last year, Spotify brought on Francois Pachet, a French professor and artificial intelligence-focused researcher, to head their new Creator Technology Research Lab. Through the lab, the company intends to create tools which will help artists in their creative process. As Fast Company notes, an end goal is a system or tool “that can auto-generate music in a given style and serve as a sort of creative companion for artists” [6]. These tools, for example, could identify upcoming trends in music and help artists incorporate these trends in new content.
Other potential opportunities
As I reflect on broader recommendations for Spotify management, I believe there is an opportunity in the short-term for the company to use streaming data to make predictions on broader preferences a subscriber may have (what brands they may like, which fashion trends they may be into, etc.). A custom luxury menswear startup, Eison Triple Thread, is an example of a company which uses music tastes to better understand their customers. The company recently introduced an app that recommends clothes based on users’ Spotify data. As their CEO mentions, you “can infer a lot from people’s music choices. The algorithm sifts through a user’s Spotify data and pairs music genres and favorite artists with styles […] the customer eventually is served pieces that reflect his personality as well as personal style” [7].
Furthermore, I believe there is an opportunity to use machine learning to identify patterns among subscribers and sell these insights to artists and labels. Are subscribers listening to my music and is now a good time to go on tour? If so, could I tour with another artist (e.g., an artist that my fans also like)? What cities and concert venues am I most likely to be profitable? These key questions could be answered by pairing streaming data and machine learning predictions.
In the more medium-term, as Spotify further develops the AI-based tools to identify music trends, I recommend that the company also use machine learning to identify up-and-coming artists before they are signed to the large recording labels (where Spotify currently pays out more than 70 percent of its monthly sales to rights holders [8]). Spotify could potentially license these artists directly, removing the large labels from the value chain and capturing a greater share of profits.
Two questions for the group
- What are some other ways Spotify can fend off other players in this competitive space?
- Do you foresee any major risks in selling your Spotify usage data to brands?
Word count: 798
[1] The Motley Fool. “What Are Spotify’s Competitive Advantages?: It streams the same songs as everyone else, so how is it different?” https://www.fool.com/investing/2018/03/01/what-are-spotifys-competitive-advantages.aspx, accessed November 2018.
[2] Digital Music News. “Apple Music Just Surpassed Spotify’s U.S. Subscriber Count” https://www.digitalmusicnews.com/2018/07/05/apple-music-spotify-us-subscribers-2/, accessed November 2018.
[3] Credera. “How Data Is Creating Better Customer Experiences at Spotify” https://www.credera.com/blog/technology-solutions/data-creating-better-customer-experiences-spotify, accessed November 2018.
[4] Forbes. “The Amazing Ways Spotify Uses Big Data, AI And Machine Learning To Drive Business Success” https://www.forbes.com/sites/bernardmarr/2017/10/30/the-amazing-ways-spotify-uses-big-data-ai-and-machine-learning-to-drive-business-success/#343149bc4bd2, accessed November 2018.
[5] TechCrunch. “Spotify acquires content recommendation startup MightyTV” https://techcrunch.com/2017/03/27/spotify-acquires-content-recommendation-startup-mightytv/, accessed November 2018.
[6] TechCrunch. “Spotify acquires audio detection startup Sonalytic” https://techcrunch.com/2017/03/07/spotify-acquires-audio-detection-startup-sonalytic/, accessed November 2018.
[7] FastCompany. “Why Did Spotify Hire This Expert In Music-Making AI?” https://www.fastcompany.com/40439000/why-did-spotify-hire-this-expert-in-music-making-ai/, accessed November 2018.
[8] Racked. “This Menswear Startup Will Recommend Clothes Based on Your Spotify Data” https://www.racked.com/2018/7/30/17617664/eison-triple-threads-spotify-data/, accessed November 2018.
[9] Bloomberg. “Spotify to Musicians: Let Us Be Your Label” https://www.bloomberg.com/news/articles/2018-11-09/spotify-to-musicians-let-us-be-your-label/, accessed November 2018.
As I have Spotify blasting in the background while I write this comment, I can say that I’ve discovered a lot of amazing music through my Discover Weekly playlist. To stay ahead of the competition in this space, I agree that Spotify should leverage streaming data to predict a subscriber’s broader preferences. On top of this, the reverse of this strategy can work as well, with a subscriber’s social media preferences, shopping habits, and other data incorporated into Spotify’s machine learning algorithms for music recommendations.
As I was reading your post as I was hoping you would recommend that Spotify becomes a record label and cuts the traditional record labels from their value chain (I am happy you did at the end). I think Spotify has all the tools they need to make it happen: they have access to musicians, they have big data on customer behavior and they have developed machine learning algorithms to facilitate the process. I like the possibility of Spotify partnering with musicians and leverage on its subscriber data to give the artist more insights as to how to improve their music, how to make their tours more successful and how to make their content creation efforts faster and cheaper. Companies like Google are already experimenting with complimenting artists’ creative process using machine learning algorithm to fight off “writer’s block” and speed up the rate at which they produce music[1].
On the music recommendation bit, I agree with Greatest of All TOM in that Spotify hasn’t really capitalized on its content suggestion capabilities that are powered by machine learning. I suspect it is because Spotify wants to provide subscribers with space for them to make decisions on which music to listen and not feel the company is pushing for an agenda. In my personal experience, I almost never explore the “Discover Weekly” category because I know my taste and I doubt that Spotify will be able to deliver me the components that I am looking for in my music.
I would encourage you to read my posting on Netflix which talks about machine learning and how they are using it to improve their content acquisition and suggestion processes as well as how they are allowing subscribers to pick the ending for one of its shows to outsource innovation and learn from their behavior while they are most actively using the product.
[1] Dar, Pranav. 2018. “Google Is Making Music With Machine Learning – And Has Released The Code On Github”. Analytics Vidhya. https://www.analyticsvidhya.com/blog/2018/03/google-making-music-help-machine-learning/.
Thanks for a great read. I foresee that if Spotify sells their user’s listening data to third parties, there will be a huge PR blowback. What people listen to are often a personal reflection of their tastes and sometimes a reflection of their identity. Users will certainly not take kindly to having bits of their personal lives auctioned off to the highest bidder. That being said, I think there is value in making aggregated data available to guide artists in creating music that is well-received.