Effective spare parts inventory management is a critical function that balances the trade-off between two conflicting objectives: maximizing service levels and minimizing ownership costs. This challenge is particularly sensitive in the service industry, where the shift towards “servitization” has fundamentally altered the business landscape. Service providers and Original Equipment Manufacturers (OEMs) have expanded their business models to include after-sales services, such as maintenance activities and spare parts inventory management. While these services generate a steady stream of revenue for the service providers, they also bind them contractually to provide high service levels for geographically dispersed fleets of equipment across a diverse customer base. Under these performance-based contracts, equipment downtime translates directly into financial penalties, potentially hindering the extra profits driven by this business model. More importantly, failure to adhere to high service standards could lead to severe reputational damage. Meticulous planning of maintenance activities is crucial for increasing asset uptime, thereby ensuring the achievement of the required service levels and building long-term customer satisfaction. However, the execution of these activities relies critically on the availability of spare parts. In this context, the rapid advancement of sensor technologies and the Industrial Internet of Things has enabled the widespread adoption of Condition-Based Maintenance (CBM). This proactive maintenance strategy utilizes real-time monitoring to assess the actual health of equipment and determine precisely when maintenance is necessary. By leveraging these health prognostics, CBM provides valuable Advance Demand Information (ADI) regarding future spare parts demand. The potential of real-time information for improving spare parts control is widely acknowledged and has recently been an active area of research. However, existing work typically addresses either single-machine systems or multi-machine systems with deterministic lead times. This research addresses a more complex and industrially relevant challenge by considering systems with multiple machines, stochastic lead times, and batch ordering. We first introduce a Proactive Base Stock Policy (ProBSP) for a multi-machine system and stochastic lead times. The ProBSP leverages degradation data to order spare parts in advance, thereby minimizing average stock levels while ensuring the fulfillment of contractual service agreements. The ProBSP involves two decision variables: an initial stock level and a degradation order threshold. The inventory starts with an initial stock level, and a part is ordered proactively every time the degradation of a machine exceeds the order threshold. To assess the ProBSP’s performance, a Discrete Event Simulation is employed, and a simulation-based optimization algorithm is developed to find the policy’s optimal parameters. We explore and prove the structural properties of the ProBSP and leverage them to develop an intelligent algorithm that optimizes the policy’s decision variables efficiently. We conduct extensive numerical experiments to gain insights into the performance of ProBSP. These experiments reveal that the ProBSP reduces the stock levels by 68% on average compared to the Base Stock Policy, which does not leverage degradation data, while achieving the required service level. While the results of the ProBSP are promising, a fair comparison against policies that leverage degradation data for similar systems is not possible because the ProBSP is the first policy to consider such systems. Consequently, we establish a comprehensive framework for developing and benchmarking degradation-aware spare parts ordering policies for systems with multiple machines and uncertain lead times. To enhance the practical relevance of this framework, we explicitly incorporate batch ordering, driven by its economic and logistical value, as well as inventory capacity constraints. This framework enables the development and benchmarking of policies using different solution methodologies. Subsequently, we develop policies using two different methodologies: policy search and look-ahead policies. We adopt the modified Base Stock Policy and the ProBSP as a first solution method. These policies fall under the category of policy search as their decision rule is an analytical function with parameters that require optimization. Additionally, we formulate the problem as a Markov Decision Process and adopt the corresponding optimal policy as a look-ahead policy. However, the continuous nature of degradation values results in an infinite state space, rendering exact solutions intractable. Moreover, the derivation of analytical expressions for transition probabilities necessitates simplifying assumptions that can constrain the model’s practical applicability. Using our framework, we develop degradation data-driven policies using three state-of-the-art Deep Reinforcement Learning (DRL) algorithms. These algorithms approximate the optimal policy through trial and error, interacting with the simulation environment to estimate the value function or the optimal policy directly. However, the extensive hyperparameter tuning required by these algorithms constitutes a significant barrier to their practical adoption. Consequently, given the demonstrated benefit of integrating domain knowledge in the learning process, we integrate insights from the modified BSP and the ProBSP into the algorithms. This integration not only yields robust performance without the need for instance-specific tuning but also leads to policies that outperform the heuristic baseline. In particular, the Deep Controlled Learning algorithm enriched with insights from the ProBSP outperforms the ProBSP by up to 15%. Furthermore, we investigate the explainability and scalability of these methods in an attempt to facilitate their acceptance in industry. These policies exhibit a more intricate and dynamic behavior for ordering spare parts orders when compared to the ProBSP. The structure of the DRL-based policies highlights how these methods can adapt their decision-making strategies to the structural characteristics of the problem. It is expected that this increased transparency enhances the trust and acceptance of data-driven methods, paving the way for their adoption in industrial settings. In conclusion, this dissertation offers a comprehensive solution to the open challenge of effectively leveraging degradation data in spare parts inventory control for multi-machine systems with stochastic lead times. By combining Operational Research methods with Machine Learning techniques, this thesis highlights the potential of leveraging degradation data in spare parts decision-making. Our work has resulted in three synergistic contributions: the formulation of a novel ProBSP, the development of a comprehensive framework, and the generation of policies using DRL augmented by domain knowledge. Our results demonstrate that leveraging degradation data in spare parts inventory management yields savings of up to 70% compared to traditional methods. These findings suggest that the integration of degradation data into the spare parts decision-making process is of strategic value for service providers and OEMs, enabling them to fully realize the profit potential of their after-sales services by ensuring excellence in service while minimizing inventory costs.