Impact of Recommender Systems on Sales Volume and Diversity

We investigate the impact of several different recommender algorithms (e.g., Amazon.com's “Consumers who bought this item also bought”), commonly used in ecommerce and online services, on sales volume and diversity, using field experiment data on movie sales from a top retailer in North America. Sales volume refers to the number of products purchased or amount of money spent by individual consumers, while sales diversity refers to the market share distribution or concentration of purchased products (e.g., niche item vs broad-appeal item). For sales volume, we show that different algorithms have differential impacts, with some widely used algorithms having no impacts. Purchase-based collaborative filtering (“Consumers who bought this item also bought”) causes a 25% lift in views and a 35% lift in the number of items purchased over the control group (no recommender) for those who purchase. In comparison, View-based collaborative filtering (“Consumers who viewed this item also viewed”) shows only a 3% lift in views and a 9% lift in the number of items purchased, albeit not statistically significant. Lastly, Recently Viewed (“Items you have recently viewed”) shows a 5% decrease in views and a 12% decrease in the number of items purchased, albeit not statistically significant. For sales diversity, we find that collaborative filtering algorithms cause individuals to discover and purchase a greater variety of products but push each individual to the same set of popular titles, leading to concentration bias at the aggregate level. For managers and policy makers, our results inform personalization, filtering, and recommender strategy in e-commerce and website design. Academically, this paper adds to the emerging but controversial topic of the impact of recommenders on sales volume and diversity for a particular set of algorithms with a field experiment.