Doctoral defence: Kateryna Kubrak "Towards user-centered prescriptive process monitoring systems“

On 13 June 2025 at 14:15 Kateryna Kubrak will defend her doctoral thesis „Towards user-centered prescriptive process monitoring systems“ to obtain the degree of Doctor of Philosophy (in Computer Science).

Supervisors:
Assoc. Prof. Fredrik Payman Milani, University of Tartu
Assist. Prof. Alexander Udo Nolte, Eindhoven University of Technology (The Netherlands), Carnegie Mellon University (USA)
Prof. Marlon Dumas, University of Tartu

Opponents:
Prof. Pnina Soffer, University of Haifa (Israel)
Assoc. Prof Adela del Río Ortega, University of Seville (Spain)

Summary
In today's business world, staying ahead of the competition means making smarter decisions faster. But what if there was a way to guide those decisions in real-time? Prescriptive process monitoring (PrPM) systems can analyze ongoing cases of a business process and recommend actions that improve outcomes. Consider a loan application process. A PrPM system can recommend offering multiple loan offers to a customer instead of a single one. Why? Because the data says that giving customers more choices increases the likelihood they'll accept one. By providing such recommendations, PrPM can help process workers make decisions that directly impact business outcomes. However, PrPM is often underutilized.

Most PrPM research has focused on the efficiency and accuracy of the techniques, paying less attention to the usability of recommendations. This gap limits the practical application of PrPM. This thesis aims to bridge the gap by exploring how the outputs of PrPM techniques can be presented to end users. First, we elicit a framework to categorize existing PrPM techniques. With this, we pave the way for understanding the variety of PrPM outputs and their relevance to end users. Second, we explore how to design usable interfaces for PrPM outputs, such as recommendations for loan officers in open loan applications. Based on the analysis of PrPM techniques outputs and user interviews, we propose a web-based tool, Kairos, which presents recommendations for ongoing cases. In addition, user evaluation helps formulate a set of best practices for designing usable PrPM interfaces. Third, we aim to improve the understandability of PrPM recommendations by proposing a method using Large Language Models (LLMs) to provide explanations for the outputs. This method was implemented in Kairos and evaluated by users, providing insights into the benefits and challenges of using LLM-based systems to enhance the explainability of PrPM outputs.