Integrating Large Language Models and Optimization in Semi- Structured Decision Making: Methodology and a Case Study

Semi-structured decisions, which fall between highly structured and unstructured decision types, rely on human intuition and experience for the final choice, while using data and analytical models to generate tentative solutions. These processes are traditionally iterative and time-consuming, requiring cycles of data gathering, analysis, and option evaluation. In this study, we propose a novel framework that integrates Large Language Models (LLMs) with optimization techniques to streamline such decision-making processes. In our approach, LLMs leverage their capabilities in data interpretation, common-sense reasoning, and mathematical modeling to assist decision makers by reducing cognitive load. They achieve this by automating aspects of information processing and option evaluation, while preserving human oversight as a crucial component of the final decision-making process. Another significant strength of our framework lies in its potential to drive the evolution of a new generation of decision support systems (DSSs). Unlike traditional systems that rely on rigid and inflexible interfaces, our approach enables users to express their preferences in a more natural, intuitive, and adaptable manner, substantially enhancing both usability and accessibility. A case study on last-mile delivery system design in a smart city demonstrates the practical application of this framework. The results suggest that our approach has the potential to simplify the decision-making process and improve efficiency by reducing cognitive load, enhancing user experience, and facilitating more intuitive interactions.