Revenue Management: From Theory to Revenue

An EC'22 Workshop - Boulder, CO, USA, July 15, 2022

Organizers: Jason Bryant, Valyn Perini, and Andrew Vakhutinsky

The workshop is intended to serve as a forum for discussing analytics of hospitality revenue management and how it can be applied to the open problems facing hotel operators.

We plan a hybrid workshop consisting of several presentations followed by the panel discussion

Invited Talks


Discrete Choice Models and Applications: Modeling, Estimation, and Optimization

Ruxian Wang*, Ovunc Yilmaz, Zifeng Zhao, Andrew Vakhutinsky, Jorge Rivero Perez

Abstract: Modeling customers’ choice behavior has been an active research area for several decades. In this talk, we review the classic discrete choice models that are widely used in studying choice behavior for customers faced with multiple substitutable options. In addition, many other choice models have also been proposed to capture new features that arise in the choice process, such as network effects, consideration set, sequential choice, and bounded rationality. Pricing is a widely-used marketing strategy to attract consumers and win the market competition. Assortment management is viewed as another effective retailing strategy. In the assortment problems, sellers decide which products should be carried in their stores or presented to the arriving consumers. To implement these discrete choice models in practice, a critical step is to calibrate the models using real data. We provide the general estimation procedure for discrete choice models using sales data and discuss how to develop algorithms to deal with the issues of choice modeling or data availability.

Dr. Ruxian Wang is a Professor at Johns Hopkins University, Carey Business School. Before returning to academia, he worked at Hewlett-Packard Company for several years as a research scientist. He received Ph.D. from Columbia University. His research and teaching interests include operations management, revenue management, pricing, discrete choice models, and data-driven decision-making. His articles appeared in several flagship journals in his field, such as Management Science, Manufacturing & Service Operations Management, Operations Research, Production, and Operations Management. He received Meritorious Service Awards from Management Science, Manufacturing & Service Operations Management, and the Outstanding Service Award from the Production and Operations Management Society (POMS). He is currently serving the Production and Operations Management (POM) journal as a senior editor.


Strategic Behavior for Hotel Standby Upgrade Programs: Empirical Evidence and Pricing Implications

Ovunc Yilmaz*, Mark Ferguson, Guangzhi Shang, Pelin Pekgün


Many hotels have recently started to offer standby upgrades (i.e., availability-based, discounted premium room upgrades) after the completion of booking to replace traditional front-desk upselling during check-in. However, customers, in particular loyalty members, may become knowledgeable about standby upgrades through repeated interactions, and act strategically, i.e., initially choose a standard room with the expectation of being offered a premium room discount through standby upgrades. Consequently, while enjoying the benefits of this program, hotels may face the potential cannibalization of premium sales due to such strategic behavior and need to adjust their pricing accordingly. Using a major hotel chain's 16-month booking and standby upgrades data, we empirically investigate the existence and extent of strategic customers in the context of standby upgrades. We develop a maximum likelihood estimator to estimate the percentage of customers who are strategic and find evidence of strategic behavior in three (out of eight) hotels examined. Considering both a weak-form and a strong-form strategic behavior, our estimates suggest that 12% to 42% of the loyalty customers act strategically in these three properties. We then propose a new pricing policy to help hoteliers maximize their premium room revenues from direct bookings and standby upgrade requests. This policy recommends a discounted full price, but also a higher standby upgrade price for loyalty customers, which can bring a revenue improvement of up to 19% over a policy ignoring the strategic behavior and 34% over a policy assuming that all customers are strategic -- two reasonable benchmarks without an estimate of the fraction of strategic customers.

Ovunc Yilmaz is an assistant professor of Operations at the Leeds School of Business, University of Colorado Boulder. His area of interest is Revenue Management and Pricing, particularly innovative RM applications in the airline, hotel, and event industries. His articles have appeared in several flagship journals such as Manufacturing & Service Operations Management and Journal of Operations Management.

Before joining Leeds in 2020, Yilmaz was an assistant professor at the Mendoza College of Business at the University of Notre Dame. He completed his Ph.D. studies at the Moore School of Business, University of South Carolina. He also has an MS degree in Operations Research from the University of North Carolina at Chapel Hill and a BS degree in Industrial Engineering from Koc University.


Optimal Upselling of Hotel Rooms: A Discrete-Choice Approach

Natalia Kosilova, Aydin Alptekinoglu*, Andrew Vakhutinsky


Upselling is a sales technique used to motivate a customer to purchase a more expensive option than the one that the customer initially chooses. When carefully implemented, it can considerably improve revenues in the hospitality industry. We propose a simple and tractable framework based on the Multinomial Logit Model (MNL) that employs the information about the customer’s initial choice to better understand their idiosyncratic reaction to an upsell offer, and compute the optimal upselling strategy of a firm providing hospitality services. We first analyze the case where the firm upsells a single product to a customer. We characterize the optimal upsell price for any product the firm may choose to offer as an upsell item. We also generalize our framework to allow the firm to offer a portfolio of upsell offers simultaneously and develop the customer choice probabilities for this case. We estimate the generalized model on a real dataset provided by our industry partner, Nor1 (now part of Oracle’s Hotel Business Unit), which provides upselling solutions in the hospitality industry. We demonstrate that our model adequately explains consumer behavior. The contribution of our paper is two-fold. First of all, to the best of our knowledge we are the first to consider an upsell problem with multiple products. Second, we provide a tractable, convenient, and readily implementable framework to calculate the optimal upsell strategy (which product to offer and at what price). Since our framework is based on MNL, the model parameters required for computing the terms of the optimal upsell offer can be estimated by well-known techniques that practitioners already use, and the application of our method appears straightforward.

Aydın Alptekinoğlu is a Professor of Supply Chain Management and Robert G. Schwartz University Endowed Fellow in Business Administration at Penn State’s Smeal College of Business. He holds a Ph.D. in Operations Management from the UCLA Anderson School of Management. His broad research interest is in three dimensions of product strategy: Variety, Price, and Availability. That is, what products to offer, at what price, and when and how to deliver them or make them available? The major emphasis of his research program has been on the first dimension: product variety management, which he broadly views as including retail assortment planning, mass customization, product competition, new product introduction, and “old” product deletion (aka stock-keeping unit rationalization). As a natural outgrowth of his research on product variety management, he has recently developed a strong interest in the analytics of choice and returns – often a consequence of choice mismatches that emanate from lack of variety. He is currently teaching a capstone course to Smeal seniors majoring in supply chain management, and a Ph.D. course on the tools (e.g., discrete choice theory) and topics (e.g., product variety management) that fall under operations-marketing interface research. He is also serving as the Director of Research at Penn State’s Center for Supply Chain Research.


Coffee Break 25 min.


The Impact of Pricing on Cancellations in the Hotel Industry

Xiao Huang, Gloria Urrea, Dan Zhang*

John Molson School of Business, Concordia University

Leeds School of Business, University of Colorado Boulder,

Abstract: Hotels that offer flexible cancellation terms often witness a high number of canceled bookings, many close to the planned arrival dates. Thus far, a common assumption in the industry is that canceled bookings are driven by random and exogenous factors beyond customers' control. However, in transaction data obtained from a hotel company, cancellation rates increase substantially with the booking price. This points to the possibility that customers' cancellations may also be driven by the hotel's endogenous pricing decisions. In this study, we investigate the causal relationship between prices and cancellation rates for hotel bookings. Using survival analysis, we find support for a positive effect of booking price on cancellation rate. The analysis also provides evidence for customers' forward-looking behavior in the hotel industry. These findings are robust across several alternative model specifications, including a matching approach. In addition, we conduct a counterfactual analysis of the revenue impact. Numerical experiments suggest that ignoring the effect of pricing on cancellations may result in a significant loss of revenue.


Assortment and Price Optimization under MNL Model with Price Range Effect

Stefanus Jasin, Chengyi Lyu*, Huanan Zhang, Andrew Vakhutinsky


In this paper, we study the assortment and price optimization problems under the Multinomial Logit (MNL) model with the price range effect, where the utility of a product is affected by the relative position of its price with respect to the highest and the lowest prices in the offer set. This model is motivated by the so-called Range Theory popularized in the behavioral economics and psychology literature. It addresses the limitation of a single-point interpretation of reference price, which ignores the impact of all other distributional information. We investigate the pure assortment problem, the pure pricing problem, and the joint assortment and pricing problem under the MNL model with price range effect. For each model, we first identify the structure of the optimal policy, and then we propose tractable algorithms that either output the optimal solution in polynomial time or admit a Fully Polynomial-Time Approximation Scheme (FPTAS). We also use real and synthetic data to demonstrate the improvement in the goodness of fit of the MNL model with the price range effect and the efficiency of our proposed algorithms.


Panel Discussion: Hospitality Industry as an Opportunity for Developing New Analytical Solutions.

Valyn Perini, Jason Bryant, and Pelin Pekgün

Dr. Pelin Pekgün is an associate professor of management science in the Darla Moore School of Business at the University of South Carolina. She also serves as the Faculty Director for the Master of Science in Business Analytics program. Her research interests include applications of management science and operations research in pricing and revenue management, supply chain management, marketing/operations interface, and humanitarian operations. Prior to joining academia, she led the operations research team in North America at JDA Software’s Pricing and Revenue Management Group, where she worked on various projects in retail, hospitality, passenger travel, and other leisure industries. Her work with the Carlson Rezidor Hotel Group on stay night price optimization was a finalist for the 2012 INFORMS Franz Edelman Award. She currently serves as an Associate Editor for Manufacturing & Service Operations Management, INFORMS Journal on Applied Analytics and Decision Sciences, and as the VP of Membership and Professional Recognition on the INFORMS Board of Directors. Dr. Pekgün holds M.S. and Ph.D. degrees in industrial and systems engineering from Georgia Institute of Technology, and B.S. and M.S. degrees in industrial engineering from Bogazici University in Istanbul, Turkey.

Jason Bryant,

Some might call Jason a pathological entrepreneur: “he just can’t help himself!” Jason has founded more than ten start-ups in his career, a number that have had successful exits and others that were great learning experiences. Currently, Jason is Vice President Nor1 - Oracle Hospitality, as a result of Oracle's acquisition of Nor1 at the end of 2020. Jason founded Nor1 in 2005, and with this experience has discovered the greatest opportunity for any business is to finally realize applicable, repeatable results from artificial intelligence efforts. Much has been sold, yet not delivered regarding AI/ML in most industries. This is going to change dramatically and rapidly because the tools and approaches have matured and continue to evolve at an unprecedented pace. Jason earned a Bachelor of Science degree from the University of Michigan and currently sits on a number of non-profit boards.


Lunch Break 110 min.


Threshold multinomial logit model and its application to demand prediction

Ruxian Wang, Zifeng Zhao*, Chenxu Ke


This paper proposes and studies the (random) threshold multinomial logit model, a discrete choice model with threshold effects. The newly proposed model incorporates the classical multinomial logit model and the recently proposed threshold Luce model as special cases. Under the threshold multinomial logit model, consumers first form their (heterogeneous) consideration set: If an alternative with significantly low utility is dominated by another one, it will not be included in the consideration set. The threshold multinomial logit model can alleviate the independence of irrelevant alternatives (IIA) property and allow more flexible substitution patterns. We characterize the optimal strategy and develop efficient solutions for the associated assortment optimization problems. Moreover, we develop maximum likelihood-based estimation to calibrate the proposed threshold model and further establish its statistical properties such as asymptotic consistency and normality under mild conditions. An efficient EM algorithm is also developed to handle the scenario with incomplete sales data. Our numerical study on synthetic and real datasets shows that the new model can improve the goodness of fit and prediction accuracy of consumer choice behavior, which suggests the threshold effect should be taken into account in firms’ decision-making.

Zifeng Zhao is an Assistant Professor of Business Analytics at the Mendoza College of Business, University of Notre Dame. He received B.S. in Financial Risk Management from the Chinese University of Hong Kong, M.S. in Computer Science, and Ph.D. in Statistics from the University of Wisconsin-Madison. His research focuses on advancing statistical and machine learning methods for solving business problems in revenue management and pricing. His interests include copula-based dependence modeling, functional data analysis, portfolio optimization, and large-scale forecasting. His research has been funded by the National Science Foundation and has been published in top journals such as Journal of the Royal Statistical Society - Series B, Journal of Machine Learning Research, Journal of Econometrics, and INFORMS Journal on Computing. He was awarded an Honorable Mention for the American Statistical Association Zellner Thesis Award in 2019 and was a finalist for the INFORMS Workshop on Data Science best paper award in 2020. Dr. Zhao also worked as a research intern at the core Data Science team at Google and Target.


Federated Learning for the Hospitality Industry: A case of study for Oracle HGBU Nor1 upsell Cloud Service

Jorge L Rivero Perez, Andrew Vakhutinsky, Kirby Bosch, Recep Bekci


Preserving data privacy by processing the clients' data independently has gained paramount importance in today’s data-driven solutions. Federated Learning (FL) is a distributed training paradigm allowing multiple clients to train a global model without sharing their data. The clients not only contribute to but also benefit from a much larger data set to train the model. Horizontal Federated Learning (HFL), or sample-based Federated Learning, is a special FL case arising when data sets share the same features but are sampled from the non-intersecting spaces. In our work, we employ HFL to model demand in the Hospitality Industry scenario with multiple properties which have similar features but are booked by mostly different guests. First, we present an HFL training setting and discuss how it can be extended as a framework to cover multiple use cases. Second, we train and evaluate HFL models on several Oracle Nor1 Cloud Service data sets containing the customer responses to upselling hotel reservation upgrade offers. We discuss the results of our computational experiments comparing the performance of several models including HFL as well as independent and hierarchical models.

Jorge L Rivero Perez is a Data Scientist at Oracle HGBU. He received his B.S. in Informatics Engineering and his MBA from The University of Cienfuegos, Cuba. He holds a Ph.D. in Artificial Intelligence and Machine Learning from the Central University of Las Villas, Cuba, in collaboration with the University of Coimbra, Portugal, funded by the European Commission under the Erasmus Mundus Action 2 Programme, Eureka SD. He is a former lecturer and researcher at the aforementioned universities. His research focused on Zero-Shot Learning with Instances-based inference on data stream environments, publishing books and papers in Journals and Conferences such as the IEEE International Joint Conference on Neural Networks (IJCNN), the International Neural Network Society (INNS), and the International Conference On Neural Information Processing (ICONIP) among others. Currently, he focuses on Optimization and Machine Learning Methods for solving Hospitality Industry problems in revenue management.


Demand estimation with missing data on covariates: Imputation-based approach

Sanghoon Cho, Danhyang Lee, Andrew Vakhutinsky*


1. Department of Management Science, Darla Moore School of Business, University of South Carolina

2. Department of Information Systems, Statistics and Management Science, Culverhouse College of Business, University of Alabama

3. Oracle Labs


Discrete choice modeling has been widely used in academia and industry to understand the demand of customers for a product. When collecting data for demand estimation, however, it is often found that only purchased product records are available in practice. In the hotel industry, for example, booking records such as purchased rooms, purchased room prices, and purchased room features are only observed, while prices of other alternative rooms available when a purchase is made (that is, prices of alternative rooms in a choice set) are not recorded. In this case, discrete choice models are not applicable unless prices and other features of alternative rooms in a choice set are imputed. One may use historical data, which also consists of purchased records but was collected during a previous time period, to impute those alternative room features. However, without adjusting for heterogeneity between the current and historical data due to the time lag, naïve imputation based on historical data could bring another source of bias. In this study, we propose a semiparametric model approach based on joint modeling for demand patterns and features of products in a choice set from the two different data sources. Imputation for missing data is embedded in the process of estimating the model parameters, which adjusts for the bias associated with such partially missing information. We apply our proposed method to a real hotel transaction dataset provided by Oracle Hospitality Global Business Unit.

Call for Papers


Topics of interest include:

  • Predictive Demand Modeling: Theory and Implementation

  • Price Optimization

  • Hotel Operations Management

Important Dates

  • Submission deadline: June 15, 2022

  • Notifications: June 22, 2022

Submission Instructions

  • By email to, cc:

  • The email should contain the title of the paper, the list of authors, the email address of the corresponding authors, the venue at which the paper appeared or will appear (if applicable), and the paper in PDF format

  • Decisions will be sent out to the corresponding author by email

Submissions will be evaluated on their relevance to the workshop, academic merit, and the potential for impact.