Optimization-based Decision Support System for Retail Sourcing

Some of the biggest challenges in the retail sourcing lie in predicting demand for a new article and making purchase decisions such as quantity, source, transportation mode and time of the order. Such decisions become more complex and time consuming as the number of SKUs and suppliers increase. The thesis addresses the issue of managing retail sourcing using forecasting and optimization based decision system developed for Zara, a leading fast-fashion clothing retailer. We started with an existing pre-season demand forecasting method that uses POS data from a comparable older article to forecast demand for a new article after adjusting for stock-outs and seasonality. We developed and compared various forecast updating methods for accuracy and found that an exponential smoothing-based model, modified to accommodate for changes in level few steps ahead, resulted in highest accuracy using Cumulative Absolute Percentage Error (CAPE). Next, we implemented a profit-maximizing optimization model to produce explicit sourcing decisions such as quantity, time and source of orders. The model takes in distributional forecasts, supply constraints, holding cost, pricing information and outputs explicit sourcing decisions mentioned above. A prototype for forecasting and optimization code is ready and currently being evaluated to secure approval for a live pilot for Summer 2013 campaign sourcing. Thesis Supervisor: Stephen Graves Title: Professor of Management Science and Engineering Systems, MIT Sloan School of Management Thesis Supervisor: David Simchi-Levi Title: Co-director, Leaders for Global Operations Professor of Civil Engineering and Environmental Engineering and Engineering Systems