2015 Informs Annual Meeting

TB45

INFORMS Philadelphia – 2015

TB43 43-Room 103A, CC Choice Modeling and Assortment Optimization Sponsor: Revenue Management and Pricing Sponsored Session Chair: Vineet Goyal, Columbia University IEOR department, 500 West 120th Street, 304 Mudd, New York, NY, 10027, United States of America, vg2277@columbia.edu 1 - Approximation Algorithms for Dynamic Assortment Optimization Models Ali Aouad, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Bldg. E40-149, Cambridge, MA, 02139, United States of America, aaouad@mit.edu, Danny Segev, Retsef Levi We study the joint assortment and inventory management problem, where demand consists in a random sequence of heterogeneous customers. Although the problem is hard in general, we provide the first polynomial time algorithms that attain constant approximations, for variants proposed in previous literature as well as more general choice models. In addition, our algorithms provide practical means for solving large-scale instances and for incorporating more realistic contraints. 2 - Capacity Constrained Assortment Optimization under the Markov Chain Based Choice Model Chun Ye, Columbia University IEOR department, 500 West 120th Street, Mudd 315, New York, NY, 10027, United States of America, cy2214@columbia.edu, Danny Segev, Vineet Goyal, Antoine Desir We consider a capacity constrained assortment optimization problem under the Markov Chain based choice model proposed by Blanchet et al. We first show that even severely-restricted special cases are APX-hard. We then present a constant factor approximation for the general problem. Our algorithm is based on a “local- ratio” method that allows us to transform a non-linear revenue function into a linear function over appropriately modified item prices. 3 - Assortment Optimization under a Random Swap Based Distribution over Permutations Model Antoine Desir, Columbia University IEOR department, 500 West 120th Street, Mudd 315, New York, NY, 10027, United States of America, ad2918@columbia.edu, Vineet Goyal, Danny Segev We consider a special class of distribution over permutations model based on modeling the consumer preferences by a random number of random swaps from a small set of fixed preference lists. This model is motivated from practical applications where preferences of “similar” consumers differ in a small number of products. We present polynomial time approximation schemes for capacity constrained assortment optimization problem under the random swap based distribution over permutation model. 4 - Design of an Optimal Membership Promotion Policy with Experiments Spyros Zoumpoulis, Insead, spyros.zoumpoulis@insead.edu, Duncan Simester, Artem Timoshenko Deciding what customer to target with what type of membership promotion is among the most important decisions that wholesale clubs face. We use the results of a large-scale membership promotion field experiment involving multiple types of membership promotions to propose various promotion policies, each relying on a different algorithm for customer segmentation. We then evaluate the performance of the proposed policies as implemented in a large-scale field test. Machine Learning in Operations Sponsor: Revenue Management and Pricing Sponsored Session Chair: Srikanth Jagabathula, NYU, 44 West Fourth Street, New York, United States of America, sjagabat@stern.nyu.edu 1 - Prediction vs Prescription in Data-driven Pricing Nathan Kallus, MIT, 77 Massachusetts Ave., E40-149, Cambridge, MA, 02139, United States of America, kallus@mit.edu, Dimitris Bertsimas We study the problem of data-driven pricing and show that a naive but common predictive approach leaves money on the table. We bound missed revenue relative to the prescriptive optimum, which we show is unidentifiable from data. We provide conditions for identifiability and appropriate pricing schemes. A new hypothesis test shows that predictive approaches are practically insufficient while parametric approaches often suffice but only if they take into account the problem’s prescriptive nature. TB44 44-Room 103B, CC

2 - The Big Data Newsvendor: Practical Insights from Machine Learning Gah-Yi Vahn, Assistant Professor, London Business School, Sussex Place, Regent’s Park, London, NW1 4SA, United Kingdom, gvahn@london.edu, Cynthia Rudin We study the newsvendor problem when one has n observations of p features related to the demand as well as demand data. Both low- and high-dimensional data are considered. We propose Machine Learning (ML) and Kernel Optimization (KO) approaches, and derive tight bounds on their performance. In a nurse staffing case study we find that the best KO and ML results beat best practice by 23% and 24% respectively. 3 - Applying Machine Learning to Revenue Management at Groupon David Simchi-levi, Professor, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA, 02139, United States of America, dslevi@mit.edu, Alexander Weinstein, He Wang, Wang Chi Cheung We propose a new data-driven pricing algorithm for online retailers, which learns customer demand from online transaction data. Our method first generates multiple demand functions using a clustering algorithm, and then learns on the fly which demand function is more likely to be correct. We will also discuss some field experiment result through collaborating with Groupon, a large daily deal website. 4 - Demand Forecasting when Customers Consider, Then Choose Ying Liu, Stern School of Business, New York University, 44 West 4th Street, KMC 8-154, New York, NY, 10012, United States of America, yliu2@stern.nyu.edu, Srikanth Jagabathula We consider the problem of demand forecasting when customers choose by first forming a consideration set and then choosing the most preferred product from the consideration set. The consideration set is sampled from a general model over subsets. We propose techniques to estimate such models from purchase transaction data. Revenue Management for Marketing Sponsor: Revenue Management and Pricing Sponsored Session Chair: John Turner, Assistant Professor, University of California, Irvine, Room SB2 338, The Paul Merage School of Business, Irvine, CA, 92697-3125, United States of America, john.turner@uci.edu 1 - Scheduling of Guaranteed Targeted Display Advertising under Reach and Frequency Requirements Ali Hojjat, University of California Irvine, Paul Merage School of Business, Irvine, CA, 92697, United States of America, hojjats@uci.edu, John Turner, Suleyman Cetintas, Jian Yang We propose a novel mechanism for the scheduling of guaranteed targeted advertising in online media. We consider a new form of contract in which advertisers specify the number of unique individuals (reach) and the minimum number of times (frequency) each individual should be exposed. We further integrate a variety of new features such as desired diversity and pacing of ads over time or the number of competing brands seen by each individual. We perform extensive numerical tests on industry data. 2 - Transaction Attributes and Customer Valuation Michael Braun, Associate Professor Of Marketing, Southern Methodist University, 6212 Bishop Blvd., Fincher 309, Dallas, TX, 75275, United States of America, braunm@smu.edu, Eli Stein, David Schweidel We propose a model of customer value and marketing ROI that incorporates transaction-specific attributes and unobserved heterogeneity. From this model, one can estimate an upper bound on the amount to invest in retaining a customer. This amount depends on the recency and frequency of past customer purchases. Using data from a B2B service provider, we estimate the revenue lost by the firm when it fails to deliver a customer’s requested level of service. 3 - Auctions with Dynamic Costly Information Acquisition Negin Golrezaei, Auctions With Dynamic Costly Information Acquisition, University of Southern California, Bridge Memorial Hall, 3670 Trousdale Parkway, Los Angeles, CA, 90089, United States of America, golrezae@usc.edu, Hamid Nazerzadeh We study the mechanism design problem for the seller of an indivisible good in a setting where buyers can purchase the additional information and refine their valuations for the good. This is motivated by information structures in online advertising where advertisers can target users using cookie-matching services. For this setting, we propose a rich class of dynamic mechanisms, called Sequential Weighted Second-Price, which encompasses the optimal and the efficient mechanisms as special cases. TB45 45-Room 103C, CC

299