Recommender Systems: Friend (of Choice) or Foe? A Large-scale Field Experiment in Online Shopping Platforms

Abstract

Recommender System (RS) algorithms are increasingly involved in online decision-making, helping consumers quickly screen large assortments by presenting the most appealing products first. However, while purporting to facilitate choice, RS algorithms might be inadvertently hindering it due to the highly attractive choice-sets they produce which might contribute to choice overload. I conduct the first randomized controlled field experiment examining the effect of a RS algorithm on choice overload in an online shopping context. In this experiment with 23,165 consumers in two online retailers, I find that a considerably altered RS algorithm that presented consumers with only the most appealing product results followed by results of diminished attractiveness did not affect various measures of consumer choice. The revised algorithm neither mitigated nor aggravated choice overload, as measured by likelihood of purchase/add-to-cart, number of products viewed and session time, suggesting that such algorithms could be designed more effectively. This article proposes that studying consumer psychology phenomena in the unique setting provided by RS algorithms can result in both rich scientific insights and more helpful algorithms.