An empirical analysis of data-driven supply chain optimization in retail and logistics

This study investigates the integration of big data and artificial intelligence (AI) to optimize decision-making in interconnected retail and logistics operations. While previous research often treats these domains in isolation, this paper proposes a unified framework for sales forecasting, delivery risk classification, and customer segmentation. Using the DataCo SMART Supply Chain dataset over 180,000 records (n=180,519), we apply machine learning techniques including Random Forest and XGBoost. To ensure methodological rigor, we address common pitfalls in supply chain analytics by removing data-leakage variables to establish realistic performance baselines. The revised models achieve robust predictive performance for delivery delays (AUC = 0.97) and profit estimation (R² = 0.99), correcting the inflated metrics observed in unconstrained models. Furthermore, we employ K-Means clustering to identify four distinct customer segments for targeted strategy development. Finally, we propose a theoretical extension of the Technology Acceptance Model (TAM) to account for supply chain complexity, offering avenues for future empirical validation. This research contributes a reproducible workflow for vetting AI models before deployment in live supply chain environments. • Demonstrate near-perfect forecasting accuracy in retail and logistics analytics • Reveal four profitable customer segments using clustering techniques • Provide data-driven recommendations for pricing and logistics decisions • Extend acceptance models with supply-chain-specific variables • Validate actionable insights for integrating retail and logistics analytics