Large Language Models as Decision Support Tools in Procurement and Supplier Selection

Abstract

Large language models (LLMs) have emerged as transformative technologies in enterprise decision-making, offering unprecedented capabilities in natural language processing (NLP), information synthesis, and predictive analytics. This review examines the application of LLMs as decision support tools in procurement and supplier selection processes. Procurement professionals face increasingly complex challenges in supplier evaluation, risk assessment, contract analysis, and market intelligence gathering. Traditional decision support systems (DSS) often struggle with unstructured data and nuanced linguistic information inherent in supplier communications, contracts, and market reports. LLMs, powered by transformer architectures and trained on vast corpora, demonstrate remarkable abilities in understanding context, extracting insights from diverse data sources, and generating actionable recommendations. This paper systematically reviews recent developments in LLM applications for procurement, analyzing their capabilities in supplier risk assessment, performance prediction, contract compliance monitoring, and market trend analysis. We examine how models such as GPT-4, Claude, and domain-specific variants are being integrated into procurement workflows to enhance decision quality, reduce processing time, and improve supplier relationship management. The review identifies key technical challenges including data privacy concerns, hallucination risks, integration with existing enterprise resource planning (ERP) systems, and the need for domain-specific fine-tuning. We also explore emerging research on multi-agent LLM systems for collaborative procurement decisions, retrieval-augmented generation (RAG) for supplier database queries, and interpretability mechanisms that enhance trust in LLM-generated recommendations. Our analysis reveals that while LLMs offer substantial benefits in processing efficiency and insight generation, successful implementation requires careful consideration of organizational readiness, data governance frameworks, and human oversight mechanisms.