Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SC18

n SC17 North Bldg 127C Supply Chain Management at JD.com Sponsored: Manufacturing & Service Oper Mgmt/Supply Chain Sponsored Session Chair: Zuo-Jun Max Shen, University of California Berkeley, Berkeley, CA, 94720-1777, United States Co-Chair: Di Wu, JD.com, Mountain View, CA, 94043, United States 1 - An Approximation Algorithm for Joint Replenishment-and- Allocation Problem Meng Qi, University of California, Berkeley, 1911A Berkeley Way, Berkeley, CA, 94704, United States, Rong Yuan, Di Wu, Max Shen We study the joint replenishment-and-allocation inventory management problem in the context of JD.com’s two-echelon inventory network. Different from the traditional two-echelon system, inventories at the upper-level warehouse are used to fulfill customer orders for a major city as well as transshipped to warehouses that used to fulfill orders for small cities and rural area. We formulate the problem as a multiple period decision problem and derived an approximation algorithm with worst-case bound. We demonstrate the effectiveness of the proposed algorithm through an extensive numerical study. 2 - Matrix Factorization with Missing Data for Improving Accuracy of Probabilistic Demand Forecasting Di Wu, JD.com, 675 E. Middlefield Rd, Mountain View, CA, 94043, United States, Xiaoyue Li, Yi Pan Demand forecast, as a classical time series problem, is crucial for e-commerce supply chain management. The performance of sales forecast has significant impact on revenue and operation cost. At the same time, missing or incomplete sales data commonly exists in the historical data which causes traditional time series models to be less effective. This paper describes an algorithm developed based on matrix factorization that specifically designed to handle missing-value issues and demonstrates its usage for demand forecast for JD.com. 3 - Understanding the Value of Fulfillment Flexibility in an Online Retailing Environment Yehua Wei, Boston College, 140 Commonwealth Avenue, Chestnum Hill, MA, 02467-3809, United States, Levi DeValve, Rong Yuan, Di Wu We propose a two-echelon inventory fulfillment model for a large online retailer, containing a large central distribution center (DC), and N smaller local DCs, to study the value of fulfillment flexibilities. Motivated by practical constraints, we propose a class of threshold fulfillment policies to effectively incorporate flexibilities at DCs. Based on our threshold policy, we estimate the savings on shipping costs and lost sales cost for a more flexible fulfillment structure considered by the online retailer. 4 - Omni-Channel Online Fulfillment Threshold Policies Xin Chen, UIUC, 216C Transportation Building, 104 S. Mathews Avenue, Urbana, IL, 61801, United States, Ebrahim Arian, Rong Yuan, Di Wu We consider inventory models in which online demands are fulfilled by physical stores. We focus on the threshold type of policies to control the amount of online demands to be accepted and compare their performances.. The potential impacts of our models/policies are illustrated on JD.com’s 7fresh business. 5 - Product Placement Optimization Using Parametric Cuts Titouan Jehl, University of California - Berkeley, Berkeley, CA, United States, Yuhui Shi, Di Wu, Max Shen E-Commerce companies use forward distribution centers (FDC) that are closer to customers distance-wise to fulfill orders in a timely manner. However, due to the capacity, it is impossible to store every item vended in the FDC. If an customer order contains items not stored as inventory, it comes in several packages (called order split), leading to a worse consumer shopping experience and higher operations costs. The proposed method in this research reduces the number of split orders by building an item-order graph and solving for the optimal assortment problem using parametric cut over the graph, which can reflect consumers’ co-purchase behaviors.

n SC18 North Bldg 128A Operations Planning in Supply Chain Sponsored: Manufacturing & Service Oper Mgmt/Supply Chain Sponsored Session Chair: Sean Zhou, Chinese University of Hong Kong, Shatin N. T, Hong Kong 1 - Multilocation Newsvendor Problem: Centralization and Inventory Pooling Sean Zhou, Chinese University of Hong Kong, 12 Chak Cheung Street, Cheng Yu Tung Building, Shatin N. T, Hong Kong, Chaolin Yang, Zhenyu Hu We study a multilocation newsvendor model with a retailer owning multiple retailing stores, each of which is managed by a manager who decides its order quantity for filling random customer demand of a product. The store managers and the retailer are all risk-averse while the managers are more risk-averse than the retailer. When there is inventory pooling, if the store managers are sufficiently more risk-averse than the retailer or the demands are very heavy- tailed, we find that inventory pooling brings less benefit than centralization. 2 - Optimal Inventory Control on a Multi-Item Inventory Model with Partial Pricing Strategies Youyi Feng, Zaragoza Logistics Center, C/ Bari 55, Edificio Nayade 5 Plaza, Zaragoza, 50197, Spain We study an inventory and pricing control on the multi-item inventory system with K = K0 + K1+ K2 products, among which there are K0 complementary products whose prices are controlled by the firm. Other K1 products and K2 products are complement and substitute to either of those K0 products, respectively. Demand functions for these products are linear with additive independent noises. We characterize optimal joint partial pricing and inventory control on all the products and compute optimal policy with an efficient algorithm. In addition, we demonstrate the firm views the economic relations between the products differently from what the customers do. 3 - Dynamic Substitution Policy for Selling Multiple Products under Supply Uncertainty Chengzhang Li, Purdue University, West Lafayette, IN, United States, Qi Feng, Mengshi Lu, J. George Shanthikumar We study a firm selling multiple substitutable products over a selling horizon of multiple periods. The firm faces replenishment cycles of fixed lengths with uncertain supply. In each period, when the random demands materialize, the firm may choose one product to meet the demand for another at a cost. Extending the notion of stochastic linearity, we show that the replenishment problem is concave via transformation. We design an efficient algorithm to determine the allocation policy that delivers close-to-optimal performance. We also show that restricting substitutions between products with adjacent characteristics can yield a benefit close to allowing full substitution among all products. 4 - Effects and Mitigation of Natural Hazards in Retail Networks Jorge Garc a Castillo, Massachusetts Institute of Technology, Cambridge, MA, United States, Jarrod D. Goentzel Knowing the impact of natural hazards is crucial to invest in mitigation. We cross sales and inventory from a retail network with emergency management data to quantify the variability in product and financial flows. We propose a two-stage multi-period inventory flow stochastic program to plan investment in buffer stock or real options contracts with suppliers that execute in declared emergencies. We show that a risk averse approach reduces worst-case cost by 15% while increasing average cost by 2%. We close by showing how various risk profiles shape the nature of investments.

67