"Sales forecasting for luxury retail stores: a strategy to integrate in-store and e-commerce demand"

The aim of the following thesis is to improve the management of the items expeditions in luxury retail business, especially taking care of the interactions between online orders and physical outlets. In order to do this, statistical methods will be used to improve sales forecasting in order to understand the needs of item for multiple stores during the year. This thesis project takes origin from an order manager re-platforming project, specifically the configuration of Salesforce Order Manager (SOM) for an Italian luxury brand. Methodologically, the analysis employs regression modelling in order to achieve maximal precision and pick up on finer sales trends. The models employ traditional regressions augmented with seasonal and monthly dummy variables meant to identify recurring patterns of retail fluctuation. Subsequent refinements introduce time-series techniques, in this instance, AR(1) and AR(2), to identify autocorrelated patterns typical of weekly or monthly sales. One of the major pieces of analytical work is the segmentation of stores into two broad categories (touristic and non-touristic) with different patterns of sales volatility and each requiring very specific model formulations per subset. The final model combined the AR(2) model applied on the difference between years of the same weeks with store-specific and week-specific terms, in order to reduce residual variance and provide a better forecast than the other models tested. In this way it was possible to determine with a certain accuracy the need of any shop at any time of the year. As it will be shown, the forecast has been made up to 3 weeks in advance with respect to the target week, a time span that would allow any shop to have the time to be filled up for the target period. With this knowledge at hand, it would be possible to change the rules about expedition orchestration of online order, avoiding to take items from shops that needs them in store, especially in periods of high flow of tourists.