Prescriptive Process Monitoring Under Resource Constraints: A Reinforcement Learning Approach

Prescriptive process monitoring methods seek to improve business process outcomes by triggering interventions at runtime based on an intervention policy. Reinforcement learning has been proposed as an effective method to learn these policies. However, existing approaches assume that there are unlimited resources available to execute the interventions, which is unrealistic in practice. This paper argues that, under resource constraints, intervention decisions should consider not only the necessity, timeliness, or effectiveness of an intervention but also the current level of resource utilization and the uncertainty of the available estimates of necessity, timeliness, and effectiveness. Allocating scarce resources to interventions with uncertain outcomes may lead to suboptimal policies. Accordingly, the paper proposes a prescriptive process monitoring method that integrates reinforcement learning with conformal prediction techniques, allowing for uncertainty-aware and resource-constrained intervention decisions. An evaluation on real-life datasets shows that reinforcement learning agents that incorporate uncertainty estimates converge faster toward policies with higher intervention gains. Thus, in the presence of resource constraints, uncertainty estimates should be considered as an integral component of prescriptive process monitoring methods.