Large Language Models in Decision Support Systems for Operations Research: Domain-Focused Perspectives

Large Language Models have recently attracted significant attention in Operations Research (OR), particularly for their potential to enhance Decision Support Systems. However, existing studies often conflate linguistic reasoning with optimization capability, leading to unclear positioning of LLMs within OR workflows. This survey provides a systematic, domain-focused review of LLM-enabled OR-based DSS, emphasizing their role as interaction, reasoning, and orchestration layers rather than optimization solvers. Using a structured literature search and domain-based classification, we analyze applications across scheduling, production planning, healthcare, energy systems, and supply chains. Using a structured literature search and domain-based classification, we analyze applications across scheduling, production planning, healthcare, energy systems, and supply chains. The review highlights recurring architectural patterns, evaluation practices, and domain-specific risks, with particular attention to scalability, reliability, and ethical concerns in decision-critical settings. We further propose a solver-centric integration framework that preserves the mathematical rigor of OR while improving usability and interpretability through LLM support. Finally, key research challenges and future directions are identified, focusing on validation, benchmarking, human-in-the-loop design, and governance. This survey aims to provide a grounded foundation for responsible and effective adoption of LLMs in OR-based decision support.