Why platforms bias recommendations: advertising competition among sellers

We study an e-commerce platform's choice among three personalised product recommendation strategies for its organic slot: random recommendation, unbiased recommendation, and biased recommendation, when sellers compete for a sponsored advertising slot through bidding. The three strategies differ in the extent of recommendation mismatch, which is captured by the divergence between a consumer's ideal product type and the recommended product type. We develop a game-theoretic model in which two sellers offering horizontally differentiated products compete on price and bid for the sponsored ad slot. The platform earns revenue from two sources, sales commissions and advertising, and determines which recommendation strategy to adopt. The choice of recommendation strategy generates a trade-off between these two revenue sources: greater recommendation mismatch reduces commission revenue by lowering recommendation conversion, but increases advertising revenue by intensifying bidding competition. Counterintuitively, we show that the biased recommendation can be optimal for the platform when the commission rate is low and product substitutability is high. In such cases, recommendation mismatch induces higher bids from sellers and allows the ad-slot winner to charge a higher product price, while softening price competition relative to the other recommendation strategies. We further consider mismatch-induced engagement and unplanned purchases, which favour biased recommendation.