Recommender Systems and Supplier Competition on Platforms

Abstract

Digital platforms can offer a multiplicity of items in one place. This should, in principle, lower end-users' search costs and improve their decision-making, and thus enhance competition between suppliers using the platform. But end-users struggle with large choice sets. Recommender systems (RSs) can help by predicting end-users' preferences and suggesting relevant products. However, this process of prediction can generate systemic biases in the recommendations made, including popularity bias, incumbency bias, homogeneity bias, and conformity bias. The nature and extent of these biases will depend on the choice of RS model design, the data feeding into the RS model, and feedback loops between these two elements. We discuss how these systemic biases might be expected to worsen end-user choices and harm competition between suppliers. They can increase concentration, barriers to entry and expansion, market segmentation, and prices while reducing variety and innovation. This can happen even when a platform's interests are broadly aligned with those of end-users, and the situation may be worsened where these incentives diverge. We outline these important effects at a high level, with the objective to highlight the competition issues arising, including policy implications, and to motivate future research.