Data-Driven Inventory Control For Large Product Portfolios: A Practical Application of Prescriptive Analytics

Abstract

Motivated by the real-world inventory management problem of a large network of pharmacies, this paper proposes and studies a practically relevant Prescriptive Analytics approach for data-driven dynamic inventory control of large portfolios of interrelated products. We extend existing research on weighted Sample Average Approximation by integrating a `global learning' model that effectively exploits cross-learning opportunities within the product portfolio. The results of an extensive numerical evaluation on real-world data suggest that our approach outperforms relevant benchmarks—in particular, models that rely on `local learning' strategies where weight functions are trained separately for each product. The numerical results also allow us to derive important practical and structural insights regarding the value of contextual information in our global learning framework.

Keywords: Dynamic inventory control, prescriptive analytics, machine learning, cross-learning

JEL Classification: C44, C61