Advancing ESG Intelligence: An Expert-level Agent and Comprehensive Benchmark for Sustainable Finance

TL;DR

This paper introduces ESGAgent, a hierarchical multi-agent system with specialized tools for comprehensive ESG analysis, and presents a benchmark to evaluate such systems using corporate sustainability reports.

Contribution

The paper develops ESGAgent, an advanced agentic framework for ESG analysis, and creates a detailed benchmark to assess multi-level ESG understanding and reporting capabilities.

Findings

ESGAgent achieves 84.15% accuracy on atomic question-answering tasks.

ESGAgent outperforms state-of-the-art LLMs in professional ESG report generation.

The benchmark effectively evaluates complex, multi-step ESG analysis capabilities.

Abstract

Environmental, social, and governance (ESG) criteria are essential for evaluating corporate sustainability and ethical performance. However, professional ESG analysis is hindered by data fragmentation across unstructured sources, and existing large language models (LLMs) often struggle with the complex, multi-step workflows required for rigorous auditing. To address these limitations, we introduce ESGAgent, a hierarchical multi-agent system empowered by a specialized toolset, including retrieval augmentation, web search and domain-specific functions, to generate in-depth ESG analysis. Complementing this agentic system, we present a comprehensive three-level benchmark derived from 310 corporate sustainability reports, designed to evaluate capabilities ranging from atomic common-sense questions to the generation of integrated, in-depth analysis. Empirical evaluations demonstrate that ESGAgent outperforms state-of-the-art closed-source LLMs with an average accuracy of 84.15% on atomic question-answering tasks, and excels in professional report generation by integrating rich charts and verifiable references. These findings confirm the diagnostic value of our benchmark, establishing it as a vital testbed for assessing general and advanced agentic capabilities in high-stakes vertical domains.