Spatial variable modelling for sales estimation of retail stores

This thesis introduces a geospatial-driven framework to estimate the untapped sales potential of retail stores, moving beyond traditional sellout prediction models. While conventional approaches often rely on historical sales data only, they overlook spatial and environmental variables that may reveal hidden growth opportunities. Current marketing strategies tend to focus on top-performing stores, reinforcing success at the expense of underperforming stores, which often have unexpressed potential. The proposed approach integrates a diverse range of geospatial and socio-demographic data, including Points of Interest (POIs), mobility patterns, land use data, and additional territorial variables. The core of this research lies in creating descriptive vectors for each retail store, encapsulating its unique spatial context, and estimating its unexpressed sales potential. The research was conducted using data from Lombardy, Italy, encompassing over 3000 retail stores, with detailed spatial information. Results demonstrate that many retail stores have potential for growth, with some stores capable of increasing their sales performance by up to +33%.