AI-driven personalized pricing may not help consumers

The autonomous nature and adaptability of artificial intelligence (AI)-powered pricing algorithms make them attractive for optimizing pricing strategies in dynamic market environments. However, certain pricing algorithms may learn to engage in tacit collusion in competitive scenarios, resulting in overly competitive prices and potentially harmful consequences for consumer welfare. This has prompted policymakers and scholars to emphasize the importance of designing rules to promote competitive behavior in marketplaces.

In a new study published in the SSRN preprint server, researchers at Carnegie Mellon University investigated the role of product ranking systems on e-commerce platforms in influencing the ability of certain pricing algorithms to charge higher prices. The study’s findings suggest that even absent price discrimination, personalized ranking systems may not benefit consumers.

“We examined the effects of personalized and unpersonalized ranking systems on algorithmic pricing outcomes and consumer welfare,” explains Param Vir Singh, Carnegie Bosch Professor of Business Technologies and Marketing and Associate Dean for Research the Tepper School of Business, who coauthored the study.

Because of the number of options available, online consumers face a difficult, time-consuming search process. This has led to the emergence of online search intermediaries (e.g., Amazon, Expedia, Yelp), which use algorithms to rank and provide consumers with ordered lists of third-party sellers’ products in response to their queries. These intermediaries reduce search costs and boost consumer welfare by helping consumers find suitable products more efficiently.

In this study, researchers examined two extreme scenarios of product ranking systems that differed in how they incorporated consumer information for generating product rankings. The first system, personalized ranking, used detailed consumer information to rank products based on predicted utility for each individual. The second, called unpersonalized ranking, relied solely on aggregate information across the entire population, resulting in an inability to customize the rankings for individual consumers.

Researchers used a consumer demand model characterized by search behavior, in which consumers searched sequentially to learn about the utilities they could obtain from various products. In this model, the ranking algorithm suggested by the intermediary affects the order in which consumers evaluate products, with each evaluation incurring a search cost.

Consumers engage in optimal search and purchase behavior to maximize their utility. Hence, an intermediary’s specific ranking system can steer demand in different ways, which have important implications for pricing outcomes.

“We compared these two systems to provide a clear understanding of the pricing implications of personalization in ranking technologies, specifically reinforcement learning (RL) algorithms, for AI-powered pricing algorithms,” says Liying Qiu, a Ph.D. student at Carnegie Mellon’s Tepper School, who led the study.

“Studying RL pricing algorithms in the context of realistic consumer behavior models is challenging due to the complexity of the dynamics they create, but by setting up controlled simulated environments, we were able to examine how these algorithms evolve and interact over time experimentally,” notes Yan Huang, Associate Professor of Business Technologies at Carnegie Mellon’s Tepper School, who coauthored the study.

Personalized ranking systems, which rank products in decreasing order of consumers’ utilities, may encourage higher prices charged by pricing algorithms, especially when consumers search for products sequentially on a third-party platform. This is because personalized ranking significantly reduces the ranking-mediated price elasticity of demand and thus incentives to lower prices.

Conversely, unpersonalized ranking systems lead to significantly lower prices and greater consumer welfare. These findings suggest that even without price discrimination, personalization may not necessarily benefit consumers since pricing algorithms can undermine consumer welfare through higher prices. Thus, the study highlights the crucial role of ranking systems in shaping algorithmic pricing behaviors and consumer welfare.

The study’s results remained the same across various values of RL learning parameters, different value of outside goods, different types of reinforcement learning algorithms, and multiple firms in the market.

“We conclude that the effectiveness of personalized ranking in improving the match between consumers and products needs to be carefully evaluated against its impact on consumer welfare when pricing is delegated to algorithms,” suggests Kannan Srinivasan, Professor of Management, Marketing, and Business Technology at Carnegie Mellon’s Tepper School, who coauthored the study.

The findings offer insights for policymakers and platform operators responsible for regulating the use of pricing algorithms and designing ranking systems. It is essential to consider the design of ranking systems when regulating AI pricing algorithms to promote competition and consumer welfare.

The study also has implications for consumer data sharing. Increased consumer data sharing may not always result in improved outcomes, even in the absence of price discrimination, since personalized ranking, empowered by access to more detailed consumer data, facilitates algorithms to charge higher prices. The negative effect of higher product prices can outweigh the positive impact of improved product fit, leading to a decline in consumer welfare.