Drug Inventory Control: Human Decisions Versus Deep Reinforcement Learning

We investigate whether and how deep reinforcement learning (DRL) can be exploited for managing inventory systems with a specific reference to perishable pharmaceutical products. A real-world case study is formulated as a Markov decision process, where states, actions, and rewards are defined. We then developed a DRL agent based on the Proximal Policy Optimization algorithm and compared its performance with a human decision-maker with several years of experience. Our findings reveal that the DRL agent outperforms the human policy by 11%, optimizing storage space and leading to growing profitability. Such incremental improvements can translate into substantial value for pharmaceutical companies operating in complex scenarios, and patients also stand to benefit. Finally, the study highlights the strategic advantage of integrating DRL into inventory management business operations, particularly for its ability to estimate uncertainty and manage corresponding supply chain risks.