Person:

Ferreira, Kristine

We consider a price-based network revenue management problem where a retailer aims to maximize revenue from multiple products with limited inventory over a finite selling season. As common in practice, we assume the demand function contains unknown parameters, which must be learned from sales data. In the presence of these unknown demand parameters, the retailer faces a tradeoff commonly referred to as the exploration-exploitation tradeoff. Towards the beginning of the selling season, the retailer may offer several different prices to try to learn demand at each price (“exploration” objective). Over time, the retailer can use this knowledge to set a price that maximizes revenue throughout the remainder of the selling season (“exploitation” objective). We propose a class of dynamic pricing algorithms that builds upon the simple yet powerful machine learning technique known as Thompson sampling to address the challenge of balancing the exploration-exploitation tradeoff under the presence of inventory constraints. Our algorithms prove to have both strong theoretical performance guarantees as well as promising numerical performance results when compared to other algorithms developed for similar settings. Moreover, we show how our algorithms can be extended for use in general multi-armed bandit problems with resource constraints, with applications in other revenue management settings and beyond.

Problem definition: We consider the problem of demand learning and pricing for retailers who offer assortments of substitutable products that change frequently, for example, due to limited inventory, perishable or time-sensitive products, or the retailer’s desire to frequently offer new styles. Academic/practical relevance: We are one of the first to consider the demand learning and pricing problem for retailers who offer product assortments that change frequently, and we propose and implement a learn-then-earn algorithm for use in this setting. Our algorithm prioritizes a short learning phase, an important practical characteristic that is only considered by few other algorithms. Methodology: We develop a novel demand learning and pricing algorithm that learns quickly in an environment with varying assortments and limited price changes by adapting the commonly used marketing technique of conjoint analysis to our setting. We partner with Zenrez, an e-commerce company that partners with fitness studios to sell excess capacity of fitness classes, to implement our algorithm in a controlled field experiment to evaluate its effectiveness in practice using the synthetic control method. Results: Relative to a control group, our algorithm led to an expected initial dip in revenue during the learning phase, followed by a sustained and significant increase in average daily revenue of 14%–18% throughout the earning phase, illustrating that our algorithmic contributions can make a significant impact in practice. Managerial implications: The theoretical benefit of demand learning and pricing algorithms is well understood—they allow retailers to optimally match supply and demand in the face of uncertain preseason demand. However, most existing demand learning and pricing algorithms require substantial sales volume and the ability to change prices frequently for each product. Our work provides retailers who do not have this luxury a powerful demand learning and pricing algorithm that has been proven in practice. History: This paper has been accepted as part of the 2021 Manufacturing and Service Operations Management Practice-Based Research Competition. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2022.1080.

We consider the product-ranking challenge that online retailers face when their customers typically behave as “window shoppers.” They form an impression of the assortment after browsing products ranked in the initial positions and then decide whether to continue browsing. We design online learning algorithms for product ranking that maximize the number of customers who engage with the site. Customers’ product preferences and attention spans are correlated and unknown to the retailer; furthermore, the retailer cannot exploit similarities across products, owing to the fact that the products are not necessarily characterized by a set of attributes. We develop a class of online learning-then-earning algorithms that prescribe a ranking to offer each customer, learning from preceding customers’ clickstream data to offer better rankings to subsequent customers. Our algorithms balance product popularity with diversity, the notion of appealing to a large variety of heterogeneous customers. We prove that our learning algorithms converge to a ranking that matches the best-known approximation factors for the offline, complete information setting. Finally, we partner with Wayfair—a multibillion-dollar home goods online retailer—to estimate the impact of our algorithms in practice via simulations using actual clickstream data, and we find that our algorithms yield a significant increase (5–30%) in the number of customers that engage with the site.

Problem definition: We seek to provide an interpretable framework for segmenting users in a population for personalized decision making. Methodology/results: We propose a general methodology, market segmentation trees (MSTs), for learning market segmentations explicitly driven by identifying differences in user response patterns. To demonstrate the versatility of our methodology, we design two new specialized MST algorithms: (i) choice model trees (CMTs), which can be used to predict a user’s choice amongst multiple options, and (ii) isotonic regression trees (IRTs), which can be used to solve the bid landscape forecasting problem. We provide a theoretical analysis of the asymptotic running times of our algorithmic methods, which validates their computational tractability on large data sets. We also provide a customizable, open-source code base for training MSTs in Python that uses several strategies for scalability, including parallel processing and warm starts. Finally, we assess the practical performance of MSTs on several synthetic and real-world data sets, showing that our method reliably finds market segmentations that accurately model response behavior. Managerial implications: The standard approach to conduct market segmentation for personalized decision making is to first perform market segmentation by clustering users according to similarities in their contextual features and then fit a “response model” to each segment to model how users respond to decisions. However, this approach may not be ideal if the contextual features prominent in distinguishing clusters are not key drivers of response behavior. Our approach addresses this issue by integrating market segmentation and response modeling, which consistently leads to improvements in response prediction accuracy, thereby aiding personalization. We find that such an integrated approach can be computationally tractable and effective even on large-scale data sets. Moreover, MSTs are interpretable because the market segments can easily be described by a decision tree and often require only a fraction of the number of market segments generated by traditional approaches. Disclaimer: This work was done prior to Ryan McNellis joining Amazon. Funding: This work was supported by the National Science Foundation [Grants CMMI-1763000 and CMMI-1944428]. Supplemental Material: The online appendices are available at https://doi.org/10.1287/msom.2023.1195 .