Post by Nauman Riaz

As part of the Analytics for Business Lab at Politecnico di Milano, my team and I worked on a real-world customer analytics project using transactional data from Coop Consorzio Nord Ovest, comprising over 5.6 million retail transactions. Our objective was to understand customer purchasing behaviour and develop models to predict future purchases while generating insights for inventory planning and marketing. Throughout the project, we: • Performed extensive EDA and data cleaning : recovering 251K missing product labels, handling refunds, outliers, and store closures to ensure data integrity. • Built an RFM-inspired customer segmentation framework using K-means clustering • Modelled customer purchasing behaviour using a Gamma-Poisson approach (Zhang & Seetharaman, 2017), estimating interpurchase times at both customer and product-category level. • Developed point prediction models using linear regression on top of the Gamma-Poisson output. • Built a probabilistic model incorporating customer activity status to rank customers by likelihood of purchase in the next 30 days , enabling priority-based marketing actions. • Developed a weekly demand forecasting model at the product-category level using HistGradientBoosting with cluster dummies, achieving an R² of 0.941. • Evaluated all models using MAE, RMSE, Weighted MAPE, and Adjusted R², translating findings into concrete business recommendations. One of my biggest takeaways was that successful analytics isn't just about building predictive models. It is about understanding the data, validating assumptions, and communicating insights that support better business decisions. A sincere thank you to my teammates and Professors Lucio Lamberti & Piercesare Secchi for their support and valuable feedback throughout this project.