This dissertation addresses decision problems of distributing inventory to stores in fashion retail supply chain. We consider a setting that fashion brands operate multiple offline stores and distribute a number of different products to the stores in the beginning of every selling season. Determining the distribution quantities of new products to stores is one of the key operational decision makings that affect the fashion brand's revenues. In most of the brands, however, the decision is made based on planning managers' experience and intuition. Moreover, not much research has been conducted on the fashion distribution problems.
In this dissertation, we discuss how we define and model the fashion distribution problems and evaluate our optimization solutions that can significantly improve the current practice. We mainly consider three types of fashion distribution problems: (1) assort-packing and distribution, (2) visual-based distribution, and (3) initial distribution ratio. For each problem, we develop and analyze quantitative optimization models to be ultimately applied to an operational decision support system in practice. Assort-packing distribution involves a fashion-specific logistics strategy called assort-packing for operational efficiency. We studied the optimization modeling and solution algorithm for finding optimal assort-pack configurations and their distribution to stores. Visual-based distribution seeks to ensure a visual variety of a set of products to be distributed to each store. We defined the problem and the fashion variety measures and proposed a solution approach that combines deep learning techniques and optimization modeling. Initial distribution ratio problem is to decide how much percentage of total quantity to distribute to stores in the beginning of season while the remainder is retained in the warehouse for replenishment after observing the actual demand. We studied the relationship between the initial distribution ratio and the total profit and analyzed the optimality condition.
The dissertation makes five contributions: (1) we derive the problem complexity and develop an efficient solution algorithm for assort-packing distribution that outperforms the best-known benchmark algorithm for industry-size instances; (2) we describe the validation of our assort-packing distribution solution by sales simulation and on-site pilot experiments and its implementation into the internal system of the company's flagship brand, Kolon Sport, contributing to the 8% increase of the sales in fall and winter seasons of 2015; (3) it is the first study to define and model the visual-based distribution based on its current practice; (4) it is the first attempt in suggesting a solution approach incorporating an artificial intelligence approach and an optimization approach for fashion distribution; and (5) we analytically model the initial distribution ratio problem and derive its properties that brings some managerial insights.