Enterprise analytics aims to make organizational data accessible for decision-making, but non-technical users still face barriers with traditional BI tools and Text-to-SQL systems. According to a new research paper published on arXiv, recent Text-to-SQL approaches based on Large Language Models (LLMs) promise natural language access to structured data, yet they fall short in enterprise settings where analytics pipelines rely on governed APIs rather than raw databases.
These governed APIs encapsulate complex business logic to ensure consistency, auditability, and security, but delegating mathematical or aggregation logic to an LLM introduces reliability and compliance risks. To overcome this, the paper presents Analytic Agent, an LLM-based agentic system that translates natural language intents into secure interactions with enterprise analytics APIs.
The Governance Problem in Enterprise Analytics
In practice, enterprise analytics environments enforce strict governance over data access and computation. Raw database queries are often prohibited; instead, pre-defined APIs expose curated metrics and dimensions. Text-to-SQL systems that generate SQL directly cannot operate under such constraints, as they bypass governance layers. The paper notes that "delegating mathematical or aggregation logic to an LLM introduces reliability and compliance risks." Analytic Agent addresses this by acting as an intermediary that understands the governance policies and interacts only via permitted API calls.
How Analytic Agent Works
Analytic Agent is designed as a multi-step reasoning system with policy-aware orchestration. It interprets user goals from natural language, validates permissions against enterprise access control policies, executes governed queries through the appropriate APIs, and generates compliant visualizations. The system does not generate raw SQL; instead, it maps intents to API endpoints and parameters, ensuring that all analytics operations stay within the governed framework.
The paper describes the agent as "policy-aware" and capable of handling complex user intents through a chain of reasoning steps. This eliminates the need for end users to understand the underlying data schema or API documentation.
Evaluation on Real Enterprise Use Cases
The researchers evaluated Analytic Agent on 90 real enterprise use cases constructed by domain experts. The evaluation tested the system's ability to reliably interpret user goals, validate permissions, execute governed queries, and generate compliant visualizations. According to the paper, the system demonstrated robust performance in translating natural language into secure API interactions without requiring raw database access. The use cases covered a variety of analytics scenarios, though the paper does not specify industry verticals.
"While recent Text-to-SQL approaches based on Large Language Models (LLMs) promise natural language access to structured data, they fall short in enterprise settings where analytics pipelines rely on governed APIs rather than raw databases."
This finding underscores a critical gap in current LLM-based analytics tools: they are designed for open-ended data access, not for governed environments where every query must pass through business logic layers.
Implications for Enterprise Technology Leaders
For CTOs and technology procurement leaders, Analytic Agent represents a shift toward secure, governed self-service analytics. By keeping analytics within the API layer, enterprises can maintain audit trails, enforce data access policies, and reduce the risk of unintended data exposure. The system could be integrated into existing business intelligence stacks, enabling natural language queries without sacrificing governance.
The paper's authors—Gundeep Singh, Parsa Kavehzadeh, Jing Xia, Xue-Yong Fu, Julien Bouvier Tremblay, Md Tahmid Rahman Laskar, Vincent Lum, and Shashi Bhushan TN—have open-sourced the paper under a Creative Commons license, inviting community feedback and collaboration. While no pilot deployments or commercial partnerships are mentioned, the research provides a framework that could be adopted by enterprises developing internal analytics platforms.
As enterprises increasingly adopt LLMs, the need for governed interfaces will grow. Analytic Agent offers a blueprint for how to balance accessibility with control—a key challenge for any organization handling sensitive data.
The full paper is available on arXiv under the identifier 2605.21027.
