Streaming platform and strategic recommendation bias

Marc Bourreau, Germain Gaudin
2021 Journal of Economics and Management Strategy  
We consider a platform that carries content from two upstream content providers and presents personalized recommendations to participating customers. We focus on streaming platforms in media markets, where users pay a subscription fee to join the platform but no usage fee, and consume a mix of content originating from each provider. We characterize the bias in the userspecific recommendations offered by the platform when one content provider charges lower royalties than the other. We establish
more » ... hat if consumers are sufficiently insensitive to bias, the recommendation system allows the platform to credibly threaten upstream providers to steer consumers away from their content, which reduces their market power. We also investigate the effects of vertical integration by the platform and show the robustness of our results to nonlinear (personalized) streaming services. | INTRODUCTION Media streaming platforms, such as Spotify, Pandora, or Deezer in the music industry, provide their customers with access to a broad range of content from various providers. To attract new consumers or increase their retention rates, streaming platforms typically set up sophisticated recommendation systems, which offer personalized recommendations to users. Personalized recommendations can be based, for instance, on users' past behavior (i.e., previous purchases or consumption), as well as on the information obtained through surveys or feedback (likes vs. dislikes, or ratings). 1 In turn, subscribing customers rely heavily on platforms' recommendations to consume content. For instance, in 2016, Spotify announced that its personalized playlists ("Discover Weekly") were used by 40 million of its users (out of a total of 100 million users). 2 Recommendation systems can also serve a different, more strategic purpose for streaming platforms. When a platform controls an integrated recommendation system, it can easily shift away from a situation where recommendations are used merely to increase customer retention or usage, and instead, consider the overall profitability of the service that is recommended. In particular, when a platform pays royalties to content providers, it may have an incentive to bias its recommendations to steer consumers away from the most expensive content and towards the cheapest one. 3 For instance, in the music streaming industry, the online radio company Pandora revealed that it manipulates its recommendation algorithm to increase or decrease the frequency at which a music title is played based on the ownership of the sound recordings. 4 In 2014, Pandora engaged in a special agreement with the indie-label coalition Merlin, whereby Merlin would accept reduced royalty rates in exchange for increased performance of its titles. As put by Pandora, "the Merlin agreement provides that as Pandora increases its performances of covered recordings-i.e., as Pandora 'steers' toward Merlin-label recordings and away from competing recordings-its effective rate drops. [...] Pandora has precisely
doi:10.1111/jems.12452 fatcat:eusfot5ycbcnphn4m6xc7536be