Leveraging Process Flexibility to Improve Process Outcome: From Descriptive Analysis to Actionable Insights

Business Process Management (BPM) focuses on managing processes to create value for organisations and their customers. As organisations strive for agility, processes have been increasingly designed to be flexible, enabling variations in execution to respond effectively to varying customer needs and a dynamic environment. This flexibility enhances responsiveness to uncertainties and allows for personalised process executions tailored to different contexts and customer requirements. However, these benefits can only be realised if flexibility is properly leveraged. Otherwise, the complexity introduced by flexibility makes it harder to determine the best execution path leading to the best possible process outcome. A possible solution is offered by Predictive Process Monitoring (PPM). PPM addresses these issues by forecasting future states of ongoing processes, helping organisations mitigate risks and make informed decisions. For flexible processes, where multiple paths may be viable depending on the context, accurate predictions provide clarity and enable better strategies. Recent advancements in Artificial Intelligence (AI) have introduced sophisticated predictive models, such as deep neural networks, to PPM. While these models are powerful, their “black-box” nature often makes them challenging to interpret. This lack of transparency limits their effectiveness, as stakeholders not only require predictions but also actionable insights that explain the factors leading to the predicted outcome and recommend what could be done differently to improve process outcomes in current or similar future cases. The main research objective of this thesis is to develop a data-driven framework that leverages process flexibility aiming to support the improvement of process outcomes within flexible and highly variable processes. To leverage process flexibility, we aim to identify process behaviours through the discovery of relevant process patterns. Understanding these process behaviours allows for a better examination of how flexibility in process execution affects outcomes through descriptive analysis. Unlike end-to-end process models, which can be complicated and overwhelming in flexible settings, process patterns represent subsets of these models that capture specific behaviours in a simpler, more manageable way. These process patterns can help in two key ways: 1) by providing an understanding of process behaviours that are otherwise hard to grasp in end-to-end models and 2) by revealing relationships between process behaviours and their outcomes. The research begins by identifying frequent process patterns, aiming to encode recurring execution behaviours for predictive models. This enhances process-aware predictions, improves model performance, and establishes a foundation for process-aware recommendations. Building on this foundation, an interactive approach is introduced to discover process patterns affecting process outcomes based on multiple areas of interest, enabling users to gain deeper insights on past execution behaviours affecting the process outcome. Besides this descriptive analysis, the discovered patterns enrich the prediction models as process-related features, ultimately enhancing outcome prediction. To deliver actionable recommendations within the proposed framework, the thesis introduces innovative approaches that uncover the reasons behind predicted process outcomes, focusing on the process patterns, encompassing all process perspectives, including control-flow relationships and process context. By identifying and analysing these patterns, they explain predicted outcomes, enabling users to understand the structural and contextual factors influencing process outcomes. The thesis offers varied insights by explaining specific cases in detail and providing a broader understanding of the entire process, enabling users to comprehend outcomes at both granular and holistic levels. By offering clear and transparent explanations, these frameworks support improved decision-making in dynamic and flexible environments. The thesis directly addresses real-world challenges by applying the proposed methods to real-world datasets, including case studies from the healthcare domain, a field particularly well-suited for this research due to its inherent complexity. High variability in patient characteristics and treatment processes often makes it difficult to establish standardised pathways. The proposed methods in the thesis demonstrate the ability to identify critical areas for adjustment and provide actionable insights that empower treatment personalisation in healthcare processes. This practical application underscores the relevance of research in addressing complex real-world problems.