Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TB14

Linear Program and show it can be solved to optimality efficiently using Pareto optimal Benders cuts. For large problems, we consider a random forest approximation of only a subset of trees and prove analytical guarantees that this gives rise to near optimal solutions. The error of the approximation decays exponentially as the number of trees increases. We propose heuristics that optimize over smaller forests rather than one large one and showcase their performance on two case studies: a property investment problem and a jury selection problem. 2 - Picking Multi-item Orders in the Warehouse of a Large Online Retailer: Data, Models and Simulations Chongli Daniel Chen, Massachusetts Institute of Technology, Cambridge, MA, 02142, United States, Mehmet Tolga Cezik, Retsef Levi, Georgia Perakis We describe theoretical and data-driven work in collaboration with an online retailer. In a warehouse, multi-item orders take up capacity on a wall until all items are picked. To avoid exceeding capacity, the retailer must balance picking efficiency with completing orders quickly. We model a novel variant of the Traveling Repairman Problem to account for this tradeoff. We propose a picking policy and determine conditions under which it achieves a greater percentage decrease in order cycle time than the percentage increase in makespan compared to the optimal makespan path, when items lie on a line or grid. We build a data- driven simulation and test our policy, reducing wall utilization by 38%. 3 - Joint Pricing and Production; A Distributionally Robust Approach Qinshen Tang, National University of Singapore, Singapore, Singapore, Georgia Perakis, Melvyn Sim, Peng Xiong In this paper, we consider a two-period joint pricing and production problem for multiple fashion items. We take distributionally robust approach. By clustering the historical data, we construct a scenario-wise ambiguity set with the support and moment(s) information of the demand. We then reformulate the problem as a tractable MIP problem. Through an extensive numerical study, we get some interesting managerial insights. In particular, we compare policies that arise from different decision rules using various methods for the single (as well as two) product(s) problem. 4 - The Practice of Bundling Under Competition Araz Khodabakhshian, UCLA, Los Angeles, CA, 90024, United States, Guillaume Roels, Uday S. Karmarkar Bundles are common in many industries, but there are also several examples of firms that choose not to bundle when faced with competition. We study competitive bundling under two models of quantity and price competition. Our analysis shows that even with full symmetry assumptions, competition can give rise to non-symmetric equilibrium strategies among the firms. We study duopoly, as well as, oligopoly bundling and examine the implications of bundling on the entry game. We show equilibrium results given different customer utilities and firm cost structure, covering the full range of strategies for a two product market. Our results have implications for a wide range of service and product industries. Sustainability in Food and Agricultural Supply Chains Sponsored: Manufacturing & Service Oper Mgmt/Sustainable Operations Sponsored Session Chair: Yanchong Zheng, Massachusetts Institute of Technology, Cambridge, MA, 02139, United States Co-Chair: Somya Singhvi, MIT, Cambridge, MA, 02139, United States 1 - Knowledge Sharing and Learning Among Smallholders in Developing Economies: Implications, Incentives, and Reward Mechanisms Shihong Xiao, HKUST, Rm5569, IELM Dept., Clear Water Bay, Kowloon, Hong Kong, 999077, Ying-Ju Chen, Christopher S. Tang NGOs and governments are advocating various knowledge-sharing platforms for farmers to exchange farming techniques. Putting altruism aside, we examine the overall economic implications for heterogeneous farmers to share their private knowledge voluntarily with others under (implicit) competition. By analyzing a multi-person sequential game, we provide a plausible explanation of why and conditions under which knowledge sharing can be beneficial even when each farmer’s profit depends on the total output. We find that the voluntary shared level is inadequate in maximizing farmer welfare, and propose a quota-based reward mechanism that can entice farmers to share more knowledge. n TB16 North Bldg 127B

n TB14 North Bldg 126C Behavioral Service Operations Sponsored: Manufacturing & Service Oper Mgmt/Service Operations Sponsored Session Chair: Ilan Lobel, New York University, New York, NY, 10012, United States Co-Chair: Yash Kanoria 1 - Managing Customer Churn via Service Mode Control Jiaqi Lu, PhD Student, Columbia Business School, New York, NY, 10027, United States, Yash Kanoria, Ilan Lobel Customer churn is an important issue for service firms. They are expensive to acquire and often leave quickly if disappointed with the service. We formulate an optimal control problem for a firm that needs to dynamically choose between service modes with different risk-reward profiles, faced with a customer that is likely to leave if unhappy with recent experiences. Our results reveal when the firm should deviate from a myopic strategy: prefer a low risk service mode if the customer is currently not a flight risk but may become one if unhappy with the next few services; prefer a high risk service mode if the customer is currently at a flight risk but may no longer be one if happy with the next few services. 2 - Rewards or Discounts: Improving Fast Food Chain Operations Rim Hariss, MIT, Georgia Perakis, Yanchong Zheng We empirically examine the joint effects of rewards program and price discounts on customer purchase behavior in a fast food chain setting. We use an instrumental variable approach to reliably estimate their joint impact on customer spent and company profit. Leveraging this empirical relationship between reward redemption and purchase on discount, we develop a parsimonious customer choice model to dynamically predict purchase behavior, which allows us to effectively evaluate alternative designs of rewards program and quantify their impacts on profitability. Insights from this analysis are offered to the fast food chain company to help with current efforts on redesigning its rewards program. 3 - Inconvenience, Liquidity Constraints, and the Adoption of Off- Grid Lighting Solutions Bhavani Shanker Uppari, INSEAD, Singapore, Singapore, Serguei Netessine, Ioana Popescu, Rowan Clarke, Manuel Barron, Martine Visser A significant proportion of world’s population does not have access to electricity. Solar-based solutions are usually unaffordable due to consumers’ poverty. There are alternative business models relying on rechargeable light bulbs that are sold at a subsidized price and require regular payments for recharges. We investigate the viability of these recharge-based models under poverty. In collaboration with a firm in Rwanda, we collected the bulb usage data from randomized experiments wherein the price and the bulb capacity were varied. We also build a structural model that incorporates the light consumption dynamics, and use it to evaluate theoretically-preferred changes to the existing model. 4 - Modeling Customer Response to Service Quality Variability with Implications for Pricing Jordan Tong, University of Wisconsin, Gregory A. DeCroix Given the same expected quality, customers tend to demand less when quality is more variable. How should firms price their service in light of this behavior? In a simple setting of repeat service under stationary but variable quality, we show that it is not necessary to model customers with a risk-averse utility function in order to generate a revenue penalty for quality variability. Instead, a simple and behaviorally robust learning-from-experience formulation - even with risk neutrality - can lead to a quality variability penalty. We study the structure of the optimal pricing policy under this relatively understudied mechanism to generate insight into pricing strategy under quality variability. n TB15 North Bldg 127A New Problems in Service Operations Sponsored: Manufacturing & Service Oper Mgmt/Service Operations Sponsored Session Chair: Georgia Perakis, Massachusetts Institute of Technology, Cambridge, MA, 02142-1347, United States 1 - Optimizing Objective Functions Determined from Random Forests Max R. Biggs, Massachusetts Institute of Technology, 77 Massachusetts Ave NE49-40497, Cambridge, MA, 02139, United States, Rim Hariss We study the problem of optimizing a tree-based ensemble objective with the feasible decisions in an arbitrary polyhedral set. We model it as a Mixed Integer

280