Sales Forecasting and Discount Optimization for a Fashion Outlet Retail Network: Predictive Modeling and Clustering.

In today’s highly competitive retail landscape, the ability to forecast sales trends and optimize discount strategies plays a central role in supporting managerial decision-making. This thesis investigates these aspects in the context of a European Fashion Outlet Retail Network, where sales dynamics are shaped by seasonality, irregular restocking, and the interplay between discounts and consumer demand. The study relies on a dataset covering sales transactions from 38 outlet stores across Europe, comprising 8,490 distinct items sold between March 2022 and March 2025. The research was developed along two main directions: the construction of reliable sales forecasting models and the design of discount optimization strategies. To this end, the work followed a structured approach. First, the dataset was analyzed to highlight key patterns and potential limitations. Clustering techniques were then applied to segment both stores and products into more homogeneous groups, distinguishing between average and top performers and further refining product categories into meaningful sub-clusters. This segmentation provided the foundation for subsequent modeling efforts. The forecasting task combined classical statistical methods and machine learning algorithms, with particular focus on SARIMAX, Prophet, and Gradient Boosting (XGBoost). Extensive fine-tuning and evaluation revealed that SARIMAX performed best for average-performing stores, while XGBoost achieved superior results in top-performing stores, characterized by more complex dynamics. A key explanatory variable consistently identified across clusters was total customer entries, which proved more influential than price in determining sales outcomes. Building on these insights, discount optimization was explored through a new cluster-based approach, where each product category represents a cluster. Representative ones—jackets, sneakers, and ballerinas—were analyzed, and the results indicated optimal discount levels of 20% for jackets and sneakers and 20–25% for ballerinas, depending on store performance. Overall, the thesis contributes both methodologically and managerially, demonstrating how clustering combined with advanced forecasting techniques can improve decision-support systems in outlet retailing. The findings not only provide robust benchmarks for forecasting but also offer actionable guidance for discount planning, emphasizing the strategic importance of customer inflow and store segmentation in shaping commercial performance.