Spotify may know you better than you realize

Spotify, the largest on-demand music service, uses big data, artificial learning, and machine learning techniques to deliver a unique and personalized music listening experience.

It’s crazy to think that we used to pay for music downloads. When Apple introduced the iPod and iTunes, it was normal to pay $0.99 for a track you really wanted. Then came along Pandora, Songza, etc., allowing users to stream music. In Songza’s case, a team of “music experts” would curate playlists that they thought “sounded good” and users would then listen to those curated playlists, regardless of each listener’s individual music taste. On the other hand, Pandora’s strategy allowed for more differentiation because it manually tagged attributes of a song so that users could pick and choose tags and filter to make playlists that they liked. However, the real winner here is Spotify, which uses big data, artificial learning, and machine learning techniques to deliver a unique and personalized music listening experience.

Spotify, launched in 2008, is the largest on-demand music service in the world with more than 150 million active users. Spotify operates under a freemium business model and earns revenue through paid subscription fees and advertisements to third parties. As of 2018, 45% of its users pay to use the premium subscription plan. Spotify is most definitely a data-driven company, using data in pretty much any part of the organization. Daily, Spotify users create 600 gigabytes of data that the company uses to perfect its algorithms and machines to improve customer experiences and extrapolate insights. In addition, Spotify crawls the web constantly to look for blog posts and other pieces of text about music to understand what people are saying about specific artists and songs, as well as which other artists and songs are being discussed alongside them.

There are many ways in which Spotify uses data to create value for the organization and its customers. Some of them are as follows:

  1. Recommended playlists – Spotify offers its users playlists that have been curated algorithmically, including music that the user already knows as well as music that the user may be unaware of. Because Spotify has so much data on users’ listening habits, it can also curate playlists depending on different weather conditions.
  2. Discover tab – Every week, Spotify users can find a fresh new playlist of two hours waiting for them called Discover Weekly. It’s a custom playlist that includes primarily new music from a user’s favorite artists, but also introduces recommended artists based on listening history (i.e. number of times listened to certain track, whether user saved track to their own playlist, whether user visited artist’s page after listening). As shown in the figure below, many Spotify users find this to be a useful feature.

  1. Spotify “Insights” – Spotify also has a page where they list important findings or interesting insights derived from the data. For example, in the article “How Students Listen 2017”, they performed a study to see which universities had the highest percentage of party playlists and which universities had the highest amount of time spent on Spotify.

In addition to these more permanent features, Spotify also uses its data for other fun and interesting things. For example, in 2013, Spotify used streaming data to predict the winners of the Grammy Awards by analyzing users’ listening habits to determine the popularity of the music. In the end, 4 out of 6 predictions made by Spotify were correct. Furthermore, Spotify has used its data analyses to shape its marketing campaigns in different areas of the world. For example, if a particular marketing campaign worked well in Chicago, it could run a similar campaign in Den Haag, Netherlands because based on music preferences, Den Haag is more similar to Chicago than Austin, New York, etc.

On February 28th, 2018, Spotify filed for listing on the New York Stock Exchange and shares recently began trading on April 3rd. With an ever-growing presence in many countries and continuously growing customer base, more data will be created, meaning better recommendations and predictions, and it will be interesting to see how Spotify continues to push technological boundaries in an industry that can be classified as “art”.

 

 

 

 

Sources:
[1] Spotify Insights page: https://insights.spotify.com/us/

[2] “Spotify buys Seed Scientific” http://fortune.com/2015/06/24/spotify-data-acquisition/

[3] Spotify Labs page: https://labs.spotify.com/

[4] “How Does Spotify Know You So Well?” https://medium.com/s/story/spotifys-discover-weekly-how-machine-learning-finds-your-new-music-19a41ab76efe

[5] “Spotify predicts Grammy winners using song and album streaming data” https://www.digitaltrends.com/music/spotify-data-predicts-grammy-winners/

Previous:

Correlation Ventures: Taking the “Gut Feel” out of VC

Next:

Winning the Super Data Bowl: Eagles Fly on 4th-Downs

Student comments on Spotify may know you better than you realize

  1. In another example of how Spotify is cleverly using data, they recently announced the feature “Line-In,” which crowdsources data on the music library. It asks users to confirm and/or edit tags used to bucket songs, on everything from genre, languages, mood, to explicitness. This tool brings together our modules on crowdsourcing and data in a unique way. By leveraging the crowd, Spotify accesses more data and gains a better understanding of how their users are interacting with the platform and with music. This could offer a huge leg up against other music streaming services, such as Apple or Pandora.

  2. Thanks for your post, RPARK. Spotify knows us very well as you mentioned, and that data is extremely valuable to record labels. Although Spotify’s relationship with their suppliers seems bitter-sweet (they are major investors, yet they’re also the reason why the service is still unprofitable), the Big Three record labels couldn’t do their job if it weren’t for all the data that Spotify provides them with (e.g. where fans are listening and thus where to tour, which trending song to pick as the next single, how to time album releases, etc.). So interesting to think about the weird relation between them!

  3. This is an interesting post. As an avid Spotify playlist user, I am always amazed by how accurately their curated playlists fit my needs. As Spotify accumulates more data on their consumers, they will continue to lead the music streaming market. I am interested to see how they will use this data to revolutionize the music listening experience as technology continues to improve. Thanks!

  4. Thank you for your post!
    As you mention, given the nature of their business, Spotify needs to find ways to use data to improve the customer experience in order to be able to differentiate. The amount of data they have and hence the amount of user engagement and input is core to their competitive advantage. My question is therefore their capacity to continue using data in differentiated ways. There are many on-demand music server providers, and some with significant scale, and thus I wonder how easy it is for them to replicate Spotify’s use of data analytics. In this regard, I find playlists and recommendations based on data as good ways to raise barriers to exit, but I fail to see them work as effectively in the future.

  5. Great post! I agree that Spotify is the market leader for music recommendations. With the amount of data it has on human preferences, I wonder if there is a way for them to try to predict what type of songs would become hits. This is similar to what Netflix is trying to do with movie production.

  6. Thanks for an interesting post! Inspired by JZ’s comment on crowd-sourcing and also the example of ReCAPTCHA, if Spotify could truly mobilize and empower users to label the music by themselves, it would save Spotify huge efforts and time and that Spotify could now focus only on verifying user labeled data instead of doing the work from ground up.

  7. Thanks for this great post! I’m also thinking of the use of data generated and analyzed by Spotify in feeding back the musicians and help them develop better music works? Maybe it’ll be interesting to see Spotify break down the most popular music pieces in different dimensions, rhythm flows, beats, words of lyrics, etc. to see if there’s any shared insights that we could learn from these popular songs. As a result, the musicians could have a better idea of what specific music styles are loved by people

  8. Great post! I have become a heavier Spotify user with time, but have yet to turn into a paid subscriber. Many people I know including myself still pay for Apple Music, despite knowing that Spotify’s music variety far exceeds that of Apple Music’s. I believe Spotify has a huge opportunity for growth and can’t help but wonder what their strategy will be moving forward. Perhaps it should focus on communicating its differentiating factors more effectively to the customers, such as the mentioned “leader for music recommendations”, variety and interface.

  9. Amazing blogpost. As a Spotify user, I can certainly attest to the “creepiness” (in a good way) of Spotify knowing me so intimately. To diversify their existing subscription-based business model, I am wondering if Spotify can expose their state-of-the-art recommendation system as an enterprise cloud offering (e.g., in the form of a RESTful API) that might potentially generalize to another non-music domains and verticals as well?

    Thank you so much again for the wonderful contribution!

  10. Thanks Raina. Are you suggesting that Spotify’s machine learning capability helped it beat Pandora and other competitors? How much do you attribute to machine learning and how much to other factors such as the business model?

Leave a comment