Person:

Israeli, Ayelet

The authors investigate whether sellers treat consumers differently on the basis of how well informed consumers appear to be. They implement a large-scale field experiment in which callers request price quotes from automotive repair shops. The authors show that sellers alter their initial price quotes depending on whether consumers appear to be correctly informed, uninformed, or misinformed about market prices. The authors find that repair shops quote higher prices to callers who cite a higher benchmark price and that women are quoted higher prices than men when callers signal that they are uninformed about market prices. However, gender differences disappear when callers mention a benchmark price for the repair. Finally, the authors find that repair shops are more likely to offer a price concession if asked to do so by a woman than if asked by a man.

An inherent risk of algorithmic personalization is disproportionate targeting of individuals from certain groups (or demographic characteristics such as gender or race), even when the decision maker does not intend to discriminate based on those “protected” attributes. This unintended discrimination is often caused by underlying correlations in the data between protected attributes and other observed characteristics used by the algorithm (or machine learning (ML) tool) to create predictions and target individuals optimally. Because these correlations are hidden in high dimensional data, removing protected attributes from the database does not solve the discrimination problem; instead, removing those attributes often exacerbates the problem by making it undetectable and, in some cases, even increases the bias generated by the algorithm.

We propose BEAT (Bias-Eliminating Adapted Trees) to address these issues. This approach allows decision makers to target individuals based on differences in their predicted behavior—hence capturing value from personalization—while ensuring a balanced allocation of resources across individuals, guaranteeing both group and individual fairness. Essentially, the method only extracts heterogeneity in the data that is unrelated to protected attributes. To do so, we build on the General Random Forest (GRF) framework (Wager and Athey 2018; Athey et al. 2019) and develop a targeting allocation that is “balanced” with respect to protected attributes. We validate BEAT using simulations and an online experiment with N=3,146 participants. This approach can be applied to any type of allocation decision that is based on prediction algorithms, such as medical treatments, hiring decisions, product recommendations, or dynamic pricing.

Does the adoption of descriptive analytics impact online retailer performance, and if so, how? We use the synthetic difference-in-differences method to analyze the staggered adoption of a retail analytics dashboard by more than 1,500 e-commerce websites, and we find an increase of 4%–10% in average weekly revenues post-adoption. We demonstrate that only retailers that adopt and use the dashboard reap these benefits. The increase in revenue is not explained by price changes or advertising optimization. Instead, it is consistent with the addition of CRM, personalization, and prospecting technologies to retailer websites. The adoption and usage of descriptive analytics also increases the diversity of products sold, the number of transactions, the numbers of website visitors and unique customers, and the revenue from repeat customers. In contrast, there is no change in basket size. These findings are consistent with a complementary effect of descriptive analytics that serve as a monitoring device that helps retailers control additional martech tools and amplify their value. Without using the descriptive dashboard, retailers are unable to reap the benefits associated with these technologies.

This paper investigates empirically the effect of market power on dynamic pricing in the presence of inventories. Our setting is the auto retail industry; we analyze how automotive dealerships adjust prices to inventory levels under varying degrees of market power. We first establish that inventory fluctuations create scarcity rents for cars that are in short supply. We then show that dealers' ability to adjust prices in response to inventory depends on their market power, i.e., the quantity of substitute inventory in their selling area. Specifically, we show that the slope of the price-inventory relationship (higher inventory lowers prices) is significantly steeper when dealers find themselves in a situation of high rather than low market power. A dealership with high market power moving from a situation of inventory shortage to a median inventory level lowers transaction prices by about 0.57% ceteris paribus, corresponding to 32.5% of dealers' average per vehicle profit margin or 145.6 on the average car. Conversely, when competition is more intense, moving from inventory shortage to a median inventory level lowers transaction prices by about 0.3% ceteris paribus, corresponding to 90.9, or 20.2% of dealers' average per vehicle profit margin. To our knowledge, we are the first to empirically show that market power affects firms’ ability to dynamically price.

Past customer spending in a category is generally a positive signal of future customer spending. We show that there exist “canary categories” for which the reverse is true. Purchases in these categories are a signal that customers are less likely to return to that retailer. We demonstrate the robustness of this finding at two retailers. We propose an explanation for the existence of canary categories and then develop a stylized model that illustrates four contributing factors: the probability a customer finds their favorite brand, customers’ willingness to substitute brands, the cost and attractiveness of visiting other stores, and expectations about future brand availability. We use both field data and experiments to investigate these factors. The findings suggest that canary categories exist (at least in part) because store assortments are not completely adjusted to local preferences. An implication is that canary categories are endogenous to each retailer; the same category may be a canary category at one retailer and a destination category at a competing retailer.