Use of artificial intelligence for business decision-making

The Decision Model and Notation (DMN) standard provides a unified framework for the representation of business decisions, with decision tables constituting its central construct. Despite the advantages offered by the manual creation of DMN decision tables from textual descriptions in terms of transparency and analysis, the process is time-consuming and prone to errors. This results in a discrepancy between the textual rules and the formal structures required for automation. The proposed framework integrates Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG), facilitating the automated conversion of natural language descriptions into decision tables. The framework utilises a modular pipeline comprising schema injection, retrieval of relevant domain-specific data, prompt composition, generation, and validation. This design guarantees that the resulting decision tables comply with DMN's rigorous XML-based specification while also capturing the contextual meaning of the original rules. A significant challenge in relying exclusively on LLMs is their proclivity to hallucinate or deviate from the required format. The integration of RAG has been demonstrated to mitigate these issues by grounding generation in reliable sources, enhancing both factual accuracy and structural consistency. The experimental results demonstrate that the LLM-RAG configuration significantly enhances the precision and robustness of generated DMN decision models, producing valid and contextually appropriate outputs with minimal human intervention. The advancement of automation in the field of decision modelling has the potential to enhance the efficiency, transparency and adaptability of decision-making processes. The findings indicate a pathway towards scalable, explainable, and verifiable decision automation across diverse business domains.