Automated Root Cause Analysis in SAP Landscapes Using Large Language Models and Operational Telemetry

Abstract

Enterprise SAP landscapes are increasingly complex, distributed, and mission-critical, making rapid and accurate root cause analysis (RCA) essential for operational resilience. Traditional monitoring approaches rely on rule-based correlation, threshold alerts, and deterministic workflows, which struggle to handle cross-layer dependencies, unstructured log data, and cascading failures. This paper proposes an automated RCA framework that integrates Large Language Models (LLMs) with operational telemetry collected across SAP application, database, and infrastructure layers. The proposed architecture combines telemetry ingestion, semantic enrichment, vector-based retrieval, and LLM-driven reasoning to identify probable root causes and generate human-readable explanations. A hybrid reasoning model integrates deterministic graph-based correlation with probabilistic scoring and contextual LLM analysis to improve diagnostic accuracy while mitigating hallucination risks. Experimental evaluation across representative SAP failure scenarios including database lock contention, RFC timeouts, and background job failures demonstrates significant improvements in root cause identification accuracy and reduction in Mean Time to Resolution (MTTR) compared to traditional rule-based systems. The findings suggest that LLM-augmented AIOps systems can enable cross-domain reasoning, enhance explainability, and reduce alert fatigue in SAP operations. This work contributes a scalable architectural blueprint and evaluation methodology for deploying AI-driven RCA in enterprise SAP environments.