RETRIEVAL-AUGMENTED GENERATION ARCHITECTURE FOR AUTOMATED DECISION-MAKING AND STRATEGIC AUDITING IN CORPORATE INFORMATION SYSTEMS

Abstract

The rapid adoption of Generative Artificial Intelligence (GenAI) in corporate information systems has created new opportunities for automated decision-making and strategic auditing. However, conventional Large Language Models (LLMs) may generate inaccurate, outdated, or unsupported information, creating concerns regarding reliability, transparency, accountability, and evidence verification. This study proposes a trustworthy Retrieval-Augmented Generation (RAG) architecture that integrates enterprise information retrieval, evidence validation, provenance tracking, explainability, LLM reasoning, and human oversight to support evidence-based corporate decision-making and strategic auditing. The study adopts a design science and quantitative research approach and examines the perceptions of 250 respondents with diverse professional expertise in finance, auditing, information systems, management, and risk management. The proposed framework considers RAG Architecture Characteristics, Evidence Traceability, Explainability, Human Oversight, Automated Decision-Making Effectiveness, and Strategic Auditing Effectiveness as key constructs. The findings indicate high perceptions of RAG Architecture Characteristics (Mean = 4.12), Evidence Traceability (Mean = 4.18), Explainability (Mean = 4.12), and Automated Decision-Making Effectiveness (Mean = 4.21). Human Oversight recorded the highest mean (4.42), highlighting the importance of expert validation, accountability, and intervention in high-impact AI-supported decisions. Strategic Auditing Effectiveness recorded a mean of 4.23, while Hallucination Reduction also achieved a mean of 4.23. Furthermore, RAG Architecture Characteristics demonstrated significant positive relationships with decision outcomes, with Decision Relevance showing the strongest relationship (β = 0.68, R² = 0.520, p < 0.001). The study concludes that evidence-grounded, human-supervised RAG can strengthen reliable, explainable, and auditable enterprise AI systems.