Dr. Kiran Rouven Busch Receives KLU Best Dissertation Award

In his dissertation, “Machine Learning for Intelligent Business Process Management: Resource Allocation, Anomaly Detection, and Remaining Time Prediction,” Kiran Rouven Busch explores the potential of large language models (LLMs) and reinforcement learning to analyze and improve business processes.

The first part of his dissertation focuses on resource allocation using process mining and machine learning to make multiple business processes running simultaneously more efficient. “At present, individual processes tend to be analyzed and optimized in isolation. In companies, however, employees are involved in several different processes. If resources are concentrated in one area, they are no longer available elsewhere.” Busch developed an algorithm trained using reinforcement learning that efficiently allocates shared resources across multiple business processes running in parallel.

Semantic Anomaly Detection

The second part of the dissertation examines anomaly detection based on semantic rather than statistical interpretation. Once an error has become systematically established and occurs repeatedly, it may no longer be identified as an anomaly. “We trained our model on a wide range of processes from an SAP dataset so that it can recognize anomalies based on their meaning and context,” Busch explains.

By “we,” Busch is referring to himself and his KLU supervisors, Prof. Dr. Henrik Leopold and Prof. Dr. André Ludwig. “I am very grateful for the support I received at KLU. Four years is a long time, and I was not only supported academically, but also with a great deal of empathy whenever things did not go quite as planned.”

This collaboration not only resulted in a dissertation awarded summa cum laude but also led to the founding of the startup Qiola. “Professor Leopold, my wife, Professor Diana Sola, and I developed a platform that enables companies to model and execute both rule-based and autonomous processes based on established standards, while also improving them using the latest research.”

"I advocate a hybrid approach"

The third part of the dissertation addresses the prediction of remaining processing times in ongoing business processes. Once again, Busch developed an algorithm that takes parallel processes into account, enabling more accurate planning and timely intervention. “If, for example, a company knew when a potential customer was likely to leave its website, it could display a ‘20% discount’ banner shortly before that point,” the 33-year-old explains, illustrating one of the potential benefits of his AI research.

While Busch sees considerable potential in modern technologies, he also remains mindful of the associated risks. “We should absolutely automate processes, but autonomous AI agents should only be used where they genuinely make sense,” he emphasizes. Busch distinguishes between rule-based automation as a deterministic technology and agentic AI as an autonomous and stochastic form of automation. “It becomes risky when we no longer understand how systems work. That is why I advocate a hybrid approach that keeps humans as the ultimate decision-making authority.”
 
Photo credit: KLU/Christin Schwarzer