INFORMS 2021 Program Book

INFORMS Anaheim 2021

TB33

month engagement with two emails separated by a half-month interval increased the likelihood of subscriber retention by 7.4% five months after the experiment started and decreased the subscriber churn odds by 26.3%. Meanwhile, we find that the same engagement increased a subscriber’s per-period service consumption by 8.8%. Our study highlights that email engagement is a double- edged sword—it increases both customer retention and service consumption, and it may decrease profitability when the increased operating cost to serve retained customers outweighs the benefit of customer retention. TB35 CC Room 210A In Person: E-commerce logistics General Session Chair: Reem Khir, Georgia Institute of Technology, Atlanta, GA, United States 1 - Dynamic Containerized Consolidation in Urban Parcel Logistics Sara Kaboudvand, ISyE Georgia Tech, Atlanta, GA, 30318, United States, Benoit Montreuil Hub-based network structures are common in urban parcel logistics for better freight consolidation and economies of scale. However, sorting every parcel at (intermediate) hubs requires significant investment in real-estate, human, and machine resources and imposes extra waiting and processing times. Such re- sorting can be bypassed by smartly encapsulating parcels that share common service features and a subsequent destination. In this study, we formally describe the problem of dynamically consolidating parcels into containers of potentially different sizes, then present and compare four highly scalable heuristic policies. We provide empirical results and a set of sensitivity analyses using an agent- oriented discrete-event megacity logistics simulator developed by Georgia Tech’s Physical Internet Center. 2 - Learning-based Online Decision-making in Multi-order Picking Environments Jana Boerger, Georgia Tech, Atlanta, GA, United States, Marlin Wolf Ulmer, Benoit Montreuil Promising their customers fast deliveries, retail companies and logistics providers need efficient warehouse processes, especially for the resource consuming order picking. Orders are unknown and are streaming in in real-time. Pickers perform repeated picking trips throughout the day. With a picking trolley or cobot, they move through the warehouse to pick items according to a pick list of ordered items. As multiple orders can be picked concurrently, controls need to carefully balance the trade-off between consolidation and timely fulfillment. Aiming for smart balancing, we present a reinforcement learning based decision-support algorithm and evaluate it through simulations based on real world data. 3 - Dynamic Parcel Consolidation and Containerization in Hyperconnected Logistic Hubs Nidhima Grover, Graduate Research Assistant, Georgia Institute of Technology, Atlanta, GA, United States, Benoit Montreuil In Physical Internet-based hyperconnected logistics, parcels are consolidated into modular containers and routed through a network of logistic hubs so that they remain together for a long portion of their multi-hub journey. At hubs where the parcels’ joint travel is completed, the containers are opened, and parcels are sorted for final delivery or re-consolidated for the next part of their inter-hub journey. This research focuses on the optimization of parcel consolidation at a hub, considering each parcel’s sequence of hubs in the path, arrival time, target time of departure, and dimensions. We develop an optimization model that minimizes handling cost such that delivery time, consolidation target, and other operational constraints are met. We present preliminary computational results that demonstrate the increase in performance due to effective consolidation. 4 - An Optimization Model of U-Shape Kitting Cell Configuration Design Wencang Bao, Georgia Institute of Technology, Atlanta, GA, United States Kitting is an effective part-feeding mode to supply varied parts to highly customized assembly lines, however, there are less optimization models of kitting cell design have been investigated yet. Considering material handling and space cost, we focus on U-shape configuration, build an optimization model to determine: 1. the size of the kitting cell, 2. the storage type of each part and 3. the location of each part. To deal with the computational inefficiency, we propose an Upper-Lower-Bounds (ULB) strategy to accelerate the solving process via deciding the max and min number of parts in each storage type. Experiments show that our model can get a significantly lower cost than current benchmarks. Finally, some heuristic algorithms are discussed. 5 - The Value of Limited Adaptability for Workload Balance in Logistics Operations Reem Khir, Georgia Institute of Technology, Atlanta, GA, United States, Alan Erera, Alejandro Toriello This talk presents a flexible assignment balancing problem with a focus on parcel sort systems critical for modern e-commerce operations. The idea is to use simple and practical recourse strategy that allows sort systems to be reconfigured once

TB33 CC Room 209A In Person: Learning and Optimization in Decision Making General Session Chair: Xiaoyue Gong, MIT, Cambridge, MA, 02139-4301, United States 1 - Chasing Convex Bodies Optimally Mark Sellke, Stanford University I will explain our recent understanding of the chasing convex bodies problem posed by Friedman and Linial in 1993. In this problem, a player receives a request sequence K_1,...,K_T of convex sets in d dimensional space and moves online into each requested set. The player’s movement cost is the length of the resulting path. Chasing convex bodies asks to find an online algorithm with cost competitive against the offline (in hindsight) optimal path. This is equivalent to a competitive analysis view on online convex optimization. Obtaining any finite competitive ratio for this problem was open until 2018. We give an optimal algorithm based on an object from classical convex geometry known as the Steiner point. 2 - Simultaneous Learning of Consumer Preference Over Different Markets. Fabricio Previgliano, University of Chicago, Chicago, IL, United States We study the ranking and selection problem faced by a company that wants to identify the most prefered product among a finite set of alternatives when consumer preferences are unknown over different markets that may have similar characteristics. The company is able to sequentially display a subset of products to different customers on each market and ask them to report their top preference over the displayed set. The objective of the firm is to design a display policy that minimizes the expected number of samples needed to identify a top product on each market with a fixed high probability. 3 - Dynamic Planning and Learning under Recovering Rewards Feng Zhu, MIT. IDSS, Boston, MA, United States, David Simchi-Levi, Zeyu Zheng Motivated by emerging applications in promotions and recommendations, we introduce a general class of multi-armed bandit problems that satisifies: (i) at most K out of N different arms are allowed to be pulled in each time period; (ii) the expected reward of an arm immediately drops after it is pulled, and then non- parametrically recovers as the idle time increases. To maximize expected cumulative rewards over T time periods, we propose and prove performance guarantees for a class of “Purely Periodic Policies”. For the offline problem when all model parameters are known, our proposed policy obtains an asymptotically tight approximation ratio that is at the order of 1-O(1/K1/2). For the online problem when the model parameters are unknown and need to be learned, we design an Upper Confidence Bound (UCB) based policy that has O(NT1/2) regret against the offline benchmark. TB34 CC Room 209B In Person: Service and Technology General Session Chair: Yiwei Wang, University of California-Irvine, Irvine, CA, 92617, United States 1 - An Empirical Examination of Food Waste in the Food Service Industry Yu Nu, Cornell University, Ithaca, NY, United States, Karan Girotra, Elena Belavina Roughly one third of food produced globally for human consumption is wasted each year, which has been a major contributor to carbon emissions. This paper studies the value of Al-enabled monitoring in the food service settings. Using synthetic control method, we analyze the staggered adoption of Winnow (new measurement tech of food waste) by more than 130 food service sites in the UK, and find an average reduction of 18%-20% in daily food waste around one week post-adoption. We explain site-heterogeneity in terms of their treatment effects through covariates of interest, and further examine the mechanisms of action that lead to the effect of Winnow, with a special focus on the changes in critical fractile. 2 - Does Customer Email Engagement Improve Profitability? Evidence From a Field Experiment of a Subscription-based Service Provider Yiwei Wang, University of California-Irvine, Paul Merage School of Business, Irvine, CA, 92617, United States, Lauren Xiaoyuan Lu, Pengcheng Shi This paper analyzes the outcome of a field experiment conducted by a large U.S. car wash chain, which offers subscription services to consumers and employs an RFID-based technology to track subscriber service events. We find that a one-

108