Is It Beneficial to Recommend Differently Priced Products? Experimental Evidence From Online Recommender Systems

Abstract

Conventional item-based collaborative filtering recommender systems display recommended products (RPs) on a focal product's (FP's) page based on their co-views and co-purchases, but do not explicitly consider their price differences. While the latitude of price acceptance theory suggests that consumers may avoid purchasing products with significantly different prices from their reference prices, the effect of displaying differently priced RP on the FP's page remains unclear. On the one hand, consumers may notice the price difference between the RP and FP more because of their joint display on the FP's page due to distinction bias. On the other hand, consumers may infer the relationship between RP and FP from other consumers’ choices in collaborative filtering recommendation systems (observational learning), reducing the salience of their price differences. We examine the impact of price differences between the RP and FP on product sales using a randomized field experiment on a US apparel retailer's website. We find that recommending a differently priced RP on the FP's page reduces its purchase probability by 21%. The effect of price differences is more pronounced for closely related products, indicating that consumers are less likely to purchase differently priced RPs when they are closely related to the FP. Our estimates also reveal an asymmetric effect when the RP price is lower versus higher than the FP price. Since price differences between the FP and RP were not randomly assigned in our field experiment, we validate our results by exogenously varying the price differences of RPs on FPs’ pages in an online experiment on Amazon MTurk. The retailer could gain an additional 4% in total product sales by replacing differently priced recommendations with similarly priced ones, as per the causal forest. Our findings provide implications for online sales strategies and recommender system design.