Integrating ETL Workflows with LLM-Augmented Data Mapping for Automated Business Intelligence Systems

Abstract

The exponential growth of heterogeneous data sources in modern enterprises has intensified the complexity of Extract, Transform, and Load (ETL) workflows and exposed the limitations of rule-based data mapping approaches used in traditional Business Intelligence (BI) systems. Schema drift, semantic inconsistencies, unstructured data integration, and frequent source system changes demand adaptive, context-aware mapping mechanisms capable of operating at scale. Recent advances in Large Language Models (LLMs) present a transformative opportunity to augment ETL pipelines with intelligent data interpretation, semantic alignment, and automated transformation logic generation. This review paper examines the integration of LLM-augmented data mapping within ETL workflows to enable automated, resilient, and self-optimizing Business Intelligence systems. It synthesizes current research and industry practices on LLM-driven schema inference, ontology alignment, metadata enrichment, and natural language–assisted transformation design. The paper further analyzes architectural patterns for embedding LLM services into ETL orchestration layers, including prompt-driven mapping engines, human-in-the-loop validation frameworks, and feedback-based learning loops.
Key challenges such as model hallucination, explainability, data privacy, governance compliance, latency constraints, and operational cost are critically evaluated. The review also explores performance evaluation metrics, enterprise deployment considerations, and emerging trends toward autonomous BI platforms. By consolidating interdisciplinary insights from data engineering, artificial intelligence, and analytics governance, this paper provides a comprehensive reference framework for researchers and practitioners seeking to modernize ETL-driven BI systems through LLM-enabled automation.