Spotify: Can machine learning drive content generation?
Spotify is using machine learning and artificial intelligence to explore both content curation and creation.
Introduction: Spotify
Spotify is a wildly popular music streaming platform that has become ubiquitous among it’s millennial user base. With more than 190 million monthly active users, and 87 million premium paid users, Spotify is growing quickly [1]. As millions of users access the music platform every day, Spotify has access to an unprecedented amount of data about music preferences of the masses. Through the use of machine learning and artificial intelligence, Spotify has begun to utilize this data to provide better services to both the content creators and consumers.
Why Machine Learning?
Having reached a critical mass of users, Spotify is now using machine learning to capitalize on the quantity to make sure their users stick around. They are continuously collecting information about what a user listens to, how often, and when they save a song to a playlist. Using all of this information, Spotify is able to implement a few different methods of machine learning in order to predict what songs a user will like with uncanny accuracy [2]. These methods are critical to product development and process improvement at Spotify, because it is their main differentiator. Pandora, Apple Music, and Amazon Music are just a few of its competitors, but none have the same amount of user data generated on a daily basis [3].
The ability to provide impeccable recommendations as well as unique content like the “Discover Weekly” and “Release Radar” is a significant pull for new users, as well as a critical for the stickiness of the product to existing users [4]. As Spotify “learns” about a user’s preferences, the personalized playlists improve, and users are incentivized to continue the use of the product. The service Spotify provides beyond just access to music is hinged upon their ability to show users new music that is curated for their taste. This is why they keep coming back for more.
Spotify has been making active efforts to stay agile and well equipped for the advances made in machine learning in recent years. In Spotify’s short term, they have acquired many data and machine learning-focused companies. For example, they acquired Niland in 2017, which is a machine learning product that helps provide better search and recommendation power. They also acquired MightyTV for its content recommendation service and Sonalytic, an audio detection service [5]. They are using this strategy to keep abreast with the latest innovation in this space.
For their medium-term strategy, they have been making talent acquisition a priority by seeking out data scientists and statisticians that will be a source of future innovation. Francois Pachet, a science and AI music composition expert, was recently hired, signaling an interest in AI-generated music [5]. With the launch of “Spotify for Artists”, Spotify is also starting to offer more services to the other side of the music industry – content creators. This tool generates analytics and trend information that can help artists to connect more closely with their audience [6]. This is just the beginning of the applications of machine learning and big data to content generation.
Recommendations
Spotify should continue to invest time and money in both sides of their business – content curation and generation. By evangelizing this technology on both sides, they would be able to grow their user base (driving revenues) as well as develop greater relationships with artists and the music industry (their cost drivers). In the short term, Spotify should continue to make acquisitions where applicable and hire top talent. Seeking risk takers from related industries and academic will drive innovation forward.
In the medium term, Spotify should begin to explore the curation of other forms of media. Podcasts are already available on Spotify, live talk radio may be an interesting foray. One can also imagine music videos are a simple next step, giving way to longer form video content in the future. As Spotify’s artificial intelligence-generated content research continues, these other forms of media may be an interesting application of these innovations.
Open Questions
- How can Spotify capitalize on AI-generated content? Do you expect users to engage in this kind of content beyond the “novelty” phase?
- Should Spotify expand beyond music into video content? How can would this affect their place in the competitive landscape?
(711 words)
References
[1] Spotify. “Spotify Technology S.A. Announces Financial Results for Third Quarter 2018.” News release, November 1, 2018. Accessed November 12, 2018. Spotify Technology S.A. Announces Financial Results for Third Quarter 2018.
[2] Ciocca, Sophia. “How Does Spotify Know You So Well?” Medium. June 21, 2018. Accessed November 13, 2018. https://medium.com/s/story/spotifys-discover-weekly-how-machine-learning-finds-your-new-music-19a41ab76efe.
[3] Aswad, Jem, and Roy Trakin. “As Spotify Prepares to Go Public, How Do Its Competitors Measure Up?” Variety. April 03, 2018. Accessed November 13, 2018. https://variety.com/2018/music/news/as-spotify-goes-public-how-do-its-competitors-measure-apple-music-amazon-1202741460/.
[4] Pasick, Adam. “The Magic That Makes Spotify’s Discover Weekly Playlists so Damn Good.” Quartz. December 22, 2015. Accessed November 13, 2018. https://qz.com/571007/the-magic-that-makes-spotifys-discover-weekly-playlists-so-damn-good/.
[5] Marr, Bernard. “The Amazing Ways Spotify Uses Big Data, AI And Machine Learning To Drive Business Success.” Forbes. October 30, 2017. Accessed November 13, 2018. https://www.forbes.com/sites/bernardmarr/2017/10/30/the-amazing-ways-spotify-uses-big-data-ai-and-machine-learning-to-drive-business-success/#74a3aa8a4bd2.
[6] Constine, John. “Sound Lessons from Spotify: How Technology Can Power a Business.” TechGenix. January 22, 2018. Accessed November 13, 2018. http://techgenix.com/lessons-from-spotify/.
Not surprisingly, Spotify already expanded beyond music and into video. And not only song music videos, but also music-related documentaries, “behind the scenes” and a “best advice” channel featuring essential mantras from the biggest artists in the industry. In my opinion, this better positions them as the authority in music content and was a logical step forward given the infrastructure, platform, and resources they already had in place. Who knows if video will be a successful play, but at least they’ll be able de further collect data from users and learn from it.
It is very interesting to see how Spotify play with their huge datasets. I totally agree that there’s a lot of untapped value to be extracted from these datasets. I also agree that Spotify should make a fast move at acquiring talents. However, I do think that there’s a big gap between the required and the current capability of the AI technology today. I have come across earlier the AI-generated poems that are based on Shakespeare’s poems. The sample is shown below:
“With joyous gambols gay and still array,
no longer when he ‘twas, while in his day
at first to pass in all delightful ways
around him, charming, and of all his days.”
(source: https://www.digitaltrends.com/cool-tech/ai-generates-shakespearean-sonnets/)
I would argue not many people notice that this poem isn’t computer-generated and thus the AI can be said to have done quite a good job. However, music creation is much more complex than writing a poem. I’m not sure HOW and WHEN the AI capability would reach that point, the point that they can write lyrics, produce the background music, and direct the tempo.
Therefore, giving different possible initiatives and limited resources, Spotify should prioritize initiatives to be executed. I would suggest using 2×2 metric of impact to company and ability to win. In this case, AI might have high impact on company, yet low ability to win given current state of technology. Thus, if there are other inititives with high impact and ability to win such as moving towards video content. Spotify should probably pursue those initiatives before AI investment on content creation.
Great choice of a popular consumer application of learning models! Spotify’s Discover Weekly feature is an amazing example of machine learning’s relevance in the real world. They have capitalized on the complexity of the data and used algorithms to successfully synthesize musical preferences for users. Their success seems to rest in what you discussed as the continual loop of data feedback – more listens, more preferences, better suggestions. What’s great about Spotify is its simplicity and relevance, this is where I think introducing video content could be a tricky proposition. While Spotify is well positioned to introduce music videos along with songs, at what point will its new content cross over into the social media category with its own set of challenges?
As a daily user of Spotify, I think this piece really hits home for me. I’ve been curious and also impressed by the company’s use of AI to create personalized recommendations and playlists. The next step of using AI to select artists and potentially even create content is one that fascinates me.. On one hand, given the similarities between different artists and themes on the US Top 40 lists one might argue that there might be a formula behind the next great earworm hit. On the other hand, I look at the careers of music producers and agents like Jimmy Iovine and Dr. Dre and the often displays of human judgment they had to make to select controversial and innovative artists to back, from Eminem, No Doubt, U2, Trent Reznor, artists who often times were unlike anything or anyone that had come before.
I hope that Spotify can one day find a happy middle if they decide to move into the record label space, where machine learning and AI can predict an artists viability in the market, but final decision on who to produce lies with the human agent. This would allow for the truly innovative and not derivative artists an opportunity to be discovered and promoted.