The Long Tail Effect of Personalized Rankings

We study the extent to which consumers benefit from personalized product suggestions offered by online retailers. Using data from a large-scale randomized experiment conducted by a major online retailer, we show that personalized product rankings induce users to purchase a greater variety of items relative to non-personalized bestseller-based rankings. To study whether users benefit from this shift in demand, we estimate an empirical choice model and use it to measure the effect of personalization on consumer surplus. Our model captures flexible taste heterogeneity by combining a standard consideration model with a latent factorization approach from computer science. This combination enables us to estimate users' tastes from data on both search histories and personalized rankings, while also recognizing that user histories are confounded by the prominence of displayed items. We estimate users' tastes and show that personalized rankings increase consumer surplus by 30% for the average user. This effect arises primarily because personalization makes users with niche tastes more likely to discover relevant niche items. Without personalization, the retailer would only offer 25% of its current assortment because most of these niche items would never find their audience.