Large Language Models for Explainable and Self-Service Business Intelligence in Large Enterprises
Business intelligence (BI) in large enterprises refers to the systematic process of collecting, integrating, analyzing, and interpreting organizational data to support strategic and operational decision-making. Traditional enterprise BI systems rely heavily on predefined dashboards, static reports, and manually authored Structured Query Language (SQL) queries, which ensure governance and reliability but require technical expertise and limit flexibility for non-technical users. As data volumes and analytical demands increase, these approaches struggle to provide accessible, interactive, and explainable analytics at scale. Recent advances in large language models (LLMs) have introduced new possibilities for natural language-driven analytics; however, existing LLM-based BI solutions often lack semantic grounding, governance enforcement, and explanation fidelity, rendering them unsuitable for deployment in large enterprise environments. This study proposes a semantic-constrained LLM architecture leveraging Generative Pre-trained Transformer 4 (GPT-4) to enables explainable and self-service BI while preserving enterprise analytical correctness and control. The approach integrates GPT-4-based natural language understanding and explanation generation with an enterprise semantic layer and symbolic query planning, ensuring alignment with enterprise-defined metrics, schemas, and aggregation rules. The proposed system is evaluated using the AdventureWorks enterprise data warehouse. Experimental results demonstrate a query correctness of 94.6% and semantic validity of 97.8%, indicating reliable interpretation of natural language analytical requests. Explanation faithfulness reaches 95.2%, and explanation quality achieves an average score of 4.5 out of 5, supporting transparent and interpretable analytics. A self-service success rate of 92.1% further confirms effective support for non-technical users with interactive response times. This study paves the way for the adoption of LLM-enabled, explainable, and governance-aware BI systems in large enterprises, enabling scalable and trustworthy natural language analytics.
Paper
Full text
Large Language Models for Explainable and Self-Service Business Intelligence in Large Enterprises
Semantic Scholar · 2026
Abstract
Business intelligence (BI) in large enterprises refers to the systematic process of collecting, integrating, analyzing, and interpreting organizational data to support strategic and operational decision-making. Traditional enterprise BI systems rely heavily on predefined dashboards, static reports, and manually authored Structured Query Language (SQL) queries, which ensure governance and reliability but require technical expertise and limit flexibility for non-technical users. As data volumes and analytical demands increase, these approaches struggle to provide accessible, interactive, and explainable analytics at scale. Recent advances in large language models (LLMs) have introduced new possibilities for natural language-driven analytics; however, existing LLM-based BI solutions often lack semantic grounding, governance enforcement, and explanation fidelity, rendering them unsuitable for deployment in large enterprise environments. This study proposes a semantic-constrained LLM architecture leveraging Generative Pre-trained Transformer 4 (GPT-4) to enables explainable and self-service BI while preserving enterprise analytical correctness and control. The approach integrates GPT-4-based natural language understanding and explanation generation with an enterprise semantic layer and symbolic query planning, ensuring alignment with enterprise-defined metrics, schemas, and aggregation rules. The proposed system is evaluated using the AdventureWorks enterprise data warehouse. Experimental results demonstrate a query correctness of 94.6% and semantic validity of 97.8%, indicating reliable interpretation of natural language analytical requests. Explanation faithfulness reaches 95.2%, and explanation quality achieves an average score of 4.5 out of 5, supporting transparent and interpretable analytics. A self-service success rate of 92.1% further confirms effective support for non-technical users with interactive response times. This study paves the way for the adoption of LLM-enabled, explainable, and governance-aware BI systems in large enterprises, enabling scalable and trustworthy natural language analytics.