Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TC14

n TC14 North Bldg 126C Optimization, Analysis, and Modeling of Service Systems Sponsored: Manufacturing & Service Oper Mgmt/Service Operations Sponsored Session Chair: David Goldberg, Cornell University, Ithaca, NY, 14850, United States 1 - On the Design of Service Systems when Servers are Strategic Ragavendran Gopalakrishnan, Postdoctoral Associate, Cornell University, Ithaca, NY, United States We study the impact of strategic server behavior on the optimal design of a service system, which involves the choice of system configuration (pooled with one central queue or dedicated with multiple parallel queues), the routing policy (e.g., Random routing or Fastest Server First), and the staffing policy (how many servers to hire). In this talk, we present key results from jointly optimizing the routing policy (over a large class of rate-based routing policies) and the system configuration (pooled vs. dedicated) at equilibrium and discuss implications for optimizing the staffing policy. We also identify an interesting trade-off between the system performance and server utility at equilibrium. 2 - Asymptotic Optimality and Heuristics of Base-stock Policies for Perishable Inventory Systems Jinzhi Bu, The Chinese University of Hong Kong, Hong Kong NT. Shatin, Shatin, Hong Kong, Xiting Gong, Xiuli Chao In this paper, we study the asymptotic properties of the performance of the base- stock policies for a classical perishable inventory system with fixed lifetime. We prove that the best base-stock policy is asymptotically optimal with large product lifetime, large demand size, and large unit penalty cost. Moreover, the optimality gap of the best base-stock policy decays exponentially fast to zero in the lifetime and demand size. We also construct a class of simple heuristic base-stock policies with the same asymptotic properties as the best base-stock policy. Finally, we conduct a numerical study to show the effectiveness of the best and heuristic base-stock policies. 3 - Dual Sourcing and Smoothing under Non-stationary Demand Time Series: Re-shoring with Speedfactories involve a complete re-shoring of demand. Using a breakeven analysis we investigate the lead time, demand, and cost characteristics that make dual sourcing with a SpeedFactory desirable compared to off-shoring to a single supplier. We extend the celebrated order-up-to replenishment policy to settings where capacity costs exist and demonstrate their excellent performance. We highlight the significant impact of autocorrelated and non-stationary demand series 4 - Queue Joining Decisions when there is a Prerequisite Condition for Receiving Service Tim Huh, University of British Columbia, Vancouver, BC, Canada, Mona Imanpoor Yourdshahy, Steven Shechter We consider an M/M/1 queueing system in which a customer requires some prerequisite conditions to be met prior to receiving service. We investigate whether an individual arriving to this system should join the queue at that time, wait to join at some future time, or leave the system. We formulate the problem as a Markov decision process and show how the structure of the optimal policy depends on in-queue and out-of-queue waiting costs, the arrival and service rates, as well as the time until the prerequisite condition is satisfied. We present the structural results of an individual’s optimal policy. 5 - New Product Diffusion in Closed-loop Supply Chains Emre Nadar, Bilkent University, Department of Industrial Engineering, Bilkent University, Ankara, 06800, Turkey, Baris Emre Kaya, Kemal Guler We study the sales planning problem of a manufacturer who sells new and remanufactured versions of a product over a finite life cycle. We develop a dynamic model in which demand arrives according to a slightly modified Bass diffusion process and end-of-use product returns required for remanufacturing are constrained by the earlier sales. We show the optimality of partial demand fulfillment in certain time periods when innovators significantly contribute to the diffusion process or an unmet demand is likely to be backlogged to be satisfied with a remanufactured product. Our findings suggest that curbing the initial sales may be desirable for remanufacturable products in fast-clockspeed industries. Jan A. Van Mieghem, Professor, Northwestern University, Evanston, IL, United States, Robert Boute, Stephen Disney We investigate the emerging trend of near-shoring a small part of the global production back to local SpeedFactories. The short lead time of the responsive SpeedFactory reduces the risk of making large volumes in advance, yet it does not

n TC15 North Bldg 127A

Theory and Pratice of Revenue Management Sponsored: Manufacturing & Service Oper Mgmt/Service Operations Sponsored Session Chair: He Wang, Georgia Institute of Technology, Atlanta, GA, United States 1 - Dynamic Pricing for Parking Maokai Lin, Smarking Inc., San Francisco, CA, United States, Yingxiang Yang Parking is an industry with 30 billion dollar annual revenue in the US, and more than 100 billion dollar worldwide. With the popularization of smartphones, more and more people search and reserve parking spaces online, leading to great opportunities for dynamic pricing and revenue management. At Smarking, we work with our industry partners to dynamically change prices online in real-time for multiple parking locations in Boston and Chicago. In this talk, we will introduce patterns in real-world parking data, approaches we use for dynamic pricing, and results we achieve. 2 - A Re-solving Heuristic with Uniformly Bounded Loss for Network Revenue Management Pornpawee Bumpensanti, Georgia Institute of Technology, 755 Ferst Drive NW, Atlanta, GA, 30332, United States, He Wang We consider a network revenue management problem. The goal is to find a customer admission policy that maximizes expected revenue over a fixed finite horizon. We study a class of re-solving heuristics. These heuristics periodically re- solve the deterministic linear program (DLP), where random customer arrivals are replaced by their expectations. We find that frequently re-solving the DLP produces the same order of revenue loss as one would get without re-solving. However, by re-solving the DLP at a few selected time points and applying thresholds to the customer acceptance probabilities, we design a new algorithm that has a revenue loss bounded by a constant that is independent of the horizon length. 3 - Assortment Optimization for Parallel Flights under a Multinomial Logit Choice Model with Cheapest Fare Spikes Yufeng Cao, Georgia Institute of Technology, Atlanta, GA, United States, Anton J. Kleywegt, He Wang The classical multinomial logit (MNL) choice model does not capture the phenomenon that airline customers tend to choose the cheapest fare classes. We study an assortment optimization problem for parallel flights under an extended spiked-MNL model, which introduces a separate attractiveness parameter for the cheapest available fare class on each flight. We show that the corresponding optimal assortment policy selects revenue-ordered assortments. We also propose static booking limit heuristics based on deterministic approximations of the problem. We evaluate different assortment policies in numerical experiments using both synthetic and real-world data provided by an airline partner. 4 - Online Ad Allocation with Temporally Asymmetric Learning Hanzhang Qin, Massachusetts Institute of Technology, 100 Memorial Drive, 8-10 A, Cambridge, MA, 02142, United States, David Simchi-Levi, Xinshang Wang, Yang Sen, Shenghuo Zhu We study an online resource allocation problem in which heterogeneous customers arrive at time-varying rates. Customer purchase probabilities are unknown and need to be learned in an online fashion. We devise online algorithms that reserve resources for future customers when (i) the total arrival rate of each customer type is known but (ii) the purchase probabilities of different customer types are learned at time-varying speeds due to non-stationary arrivals. We prove performance guarantees for the proposed algorithms. We test the algorithms using huge industrial datasets.

312