AI-enhanced business process simulation for decision making support

Organizations face constant change driven by internal and external factors such as competitive pressure, regulatory shifts, and technological advancements, requiring continuous adaptation to improve efficiency, service quality, and overall performance. Effective decision-making in this context is critical, but relying solely on managerial intuition is often insufficient for complex systems. Business Process Simulation (BPS) has therefore emerged as a valuable decision-support tool, allowing analysts to systematically evaluate potential changes and their impacts within a controlled, risk-free environment. Indeed, BPS enables rapid assessment of numerous hypothetical scenarios, a practice commonly known as what-if analysis, which strengthens and simplifies the decision-making process. The key challenge lies in defining a simulation model that accurately represents the real business process while adequately capturing its inherent complexity. In practice, however, simulation models are often oversimplified and may fail to fully reflect real world dynamics and variability. In order to address the limitations of BPS, this thesis investigates the integration of AI models into BPS. In particular, the use of AI models helps overcome unrealistic and oversimplified assumptions in reproducing process behavior. Conversely, integrating BPS techniques into AI approaches enables the incorporation of a global process perspective, capturing dependencies and interactions among multiple process instances. Therefore, this integration aims to leverage the respective strengths of the two components while mitigating their weaknesses. The thesis starts from the definition of a hybrid simulation model capable of integrating multiple predictive models at runtime across different perspectives, which enables a more accurate representation of real-world processes. Further building upon this foundation, it investigates a process-based optimization approach integrated with the hybrid simulation model. This integration aims to overcome existing limitations, particularly the lack of consideration of the overall process as a multi-instance system and the limited applicability of current approaches to real world processes. The results highlight the adaptability and configurability of the hybrid simulation model, demonstrating its ability to assume different roles, ranging from actively identifying optimal solutions to supporting decision makers in selecting among alternative scenarios. This versatility is achieved through the design of an advanced simulation model that integrates process-based AI optimization techniques while maintaining a comprehensive evaluation framework capable of capturing all process perspectives, multi-instance dynamics, and the inherent stochasticity of business processes.