Informs Annual Meeting 2017

TE22

INFORMS Houston – 2017

TE23

3 - Resource Allocation Decisions to Manage Material Convergence Merve Ozen, Research Assistant, University of Wisconsin-Madison, 1513 University Avenue, Madison, WI, 53706, United States, mozen@wisc.edu, Ananth Krishnamurthy The large flow of donated items, both solicited and unsolicited, immediately after a major disaster, lead to the material convergence problem. Inefficient resource allocation can cause significant wastage of donated goods as well as deprivation of critical items for victims affected by the disaster. Using transient queuing models, we model this phenomenon to quantify material convergence and assess the impact of staffing strategies that improve efficiency. 4 - A Preposition and Allocation Model for Disaster Relief Inventories with Explicitly Considering Demand Risk Feng Li, PhD, University of Science and Technology of China, NO. 96, JinZhai Road, Baohe District, Hefei, Anhui Province, 230026, China, lfeng90@mail.ustc.edu.cn In humanitarian operations management, the relief organizations always have incomplete information on the beneficiaries. As a result, there would exist an estimation error between the estimated demand and real demand for relief supplies. In this paper, we propose a preposition-allocation model for disaster relief inventories by considering a demand risk parameter. We will investigate the impacts of demand risk on decisions of humanitarian logistics and human sufferings of beneficiaries. A cooperative strategy is also developed to cover the demand risk and to reduce human sufferings. Further, the proposed model will be demonstrated using the real case of Rammasun Typhoon in Hainan Island. 342D Novel RM Models for Loyalty Programs Sponsored: Revenue Management & Pricing Sponsored Session Chair: Ang Li, PROS, Inc., Houston, TX, 77002, United States, ali@pros.com 1 - Allocation Priority Tiebreakers for Frequent Flyer Seat Allocation Optimization Shigeru Tsubakitani, PROS, Inc., 43360 Hillpark Street, South Riding, VA, 20152-2597, United States, stsubakitani@pros.com Three approaches for breaking ties of alternative solutions for a frequent flyer seat allocation optimization model are discussed. The tiebreakers help generate solutions more desirable for airline companies by eliminating randomness and allocating frequent flyer seats more evenly among flights when there are many $0 bids exist in a market. 2 - Which Customers are More Valuable in a Dynamic Pricing Situation? Anton Ovchinnikov, Queen’s School of Business, 143 Union Street, Kingston, ON, K7L.3N6, Canada, ao37@queensu.ca, Jue Wang We consider a firm that dynamically price its inventory and examine whether customers who purchase at higher prices indeed add higher marginal value to the firm. We present modeling and computational results which are calibrated on a unique data set from a major airline. We show that in some situations a customer’s value can be independent of the price paid. 3 - Predicting Customer Retention using Cart Models Pratik Mital, Senior Consultant, Operations Research, Revenue Analytics, Inc., 300 Galleria Parkway, Suite 1900, Atlanta, GA, 30339, United States, pmital@revenueanalytics.com Predicting customer retention is an important problem in airlines, cruise lines, and any other industry where demand has a tendency to go higher than capacity. This allows the planners to optimally book more than the capacity, depending upon how many customers are expected to cancel. The margin for error in predicting customer retention can be quite low, since a wrong prediction can lead to either a denied boarding or an empty seat cost. The following are the key areas that will be covered during the talk: 1. Overview of customer retention and CART models 2. Customer attributes that can be utilized in the models 3. A real-world example 4. Accuracy metrics to evaluate the accuracy of the models TE22

342E Joint Session RMP/APS: Data-driven Revenue Management Sponsored: Revenue Management & Pricing Sponsored Session Chair: Will Ma, willma353@gmail.com Co-Chair: David Simchi-Levi, Massachusetts Institute of Technology, Cambridge, MA, 02139, United States, Dslevi@mit.edu 1 - Data-driven Dynamic Pricing with Model Misspecification and Endogeneity Effect He Wang, Georgia Institute of Technology, Atlanta, GA, 30332- 0205, United States, he.wang@isye.gatech.edu, Milashini Nambiar, David Simchi-Levi We study a dynamic pricing problem with contextual information where the seller may assume an incorrect demand model. The seller sequentially observes demand, estimates model parameters, and then chooses price. In this setting, model misspecification can cause price endogeneity, which in turn leads to inconsistent estimation of price elasticity and bad pricing decisions, a phenomenon known as the “spiral-down effect” in revenue management. To address the endogeneity effect we propose a “Random Price Shock” algorithm that dynamically generates independent price shocks to estimate price elasticity while maximizing revenue. 2 - Learning with Abandonment Sven Peter Schmit, Stanford University, 729 Escondido Rd, Apt 224, Stanford, CA, 94305, United States, schmit@stanford.edu, Ramesh Johari Consider a seller that repeatedly sets a price for a subscription product to the same customer. The seller would like to raise the price to the maximum the customer is willing to pay, but if it exceeds this value the customer abandons forever. Therefore, the learner has to be conservative in its policy.We propose a general thresholded learning problem that models such scenarios. The goal is to maximize the sum of discounted rewards based on a activity level set at discrete times. However, the process stops if the activity level exceeds an unknown threshold. We show that simple policies perform well under fairly general conditions. 3 - Data-driven Promotion Planning with Multiple Forms of Cannibalization Hanwei Li, MIT, Cambridge, MA, United States, hanweili@mit.edu, Will Ma, David Simchi-Levi, Jinglong Zhao We study the problem of promotion planning given data from a distributor which sells multiple products through multiple retailers. While planning these promotions, three cannibalization effects must be taken into consideration: across products, across retailers, and across time. We also analyze the effects of competing SKU’s from other distributors. 4 - Continuous Assortment Optimization with Incomplete Information Arnoud den Boer, University of Amsterdam, Postbus 94248, Room Number F3.33, Amsterdam, 1090 GE, Netherlands, A.V.denBoer@uva.nl We study the classical assortment optimization problem with multinomial-logit type choice probabilities and finite capacity, but with a continuous instead of discrete set of products. We discuss how to determine an optimal assortment, how to estimate the choice probabilities from data, and how to apply these in an online-learning setting. 342F Topics in Revenue Management, Choice Modeling and Assortment Optimization Sponsored: Revenue Management & Pricing Sponsored Session Chair: Vineet Goyal, Columbia University, Columbia University, New York, NY, 10027, United States, vgoyal@ieor.columbia.edu 1 - Revenue Management with Dynamic Customer and Assortment Selection Adam Elmachtoub, Columbia University, 500 W. 120th St., New York, NY, 10027, United States, adam@ieor.columbia.edu, Vineet Goyal, Roger D. Lederman We provide algorithms that select customers dynamically, each with their own choice model, in order to maximize revenue from a limited supply of inventory. We propose constant factor approximations when assortments may be chosen dynamically, and for a constrained setting where the set of offered products can only decrease over time. TE24

418