An Agentic AI Framework for ESG Disclosure Monitoring and Governance Evidence

Abstract

Environmental, social, and governance (ESG) disclosure is increasingly important for sustainability assessment, compliance review, and managerial decision-making. However, existing AI-based ESG applications are often task-specific and provide limited support for longitudinal indicator harmonisation, missing-data handling, adaptive rule updates, and auditable decision trails. This study develops and evaluates an Agentic ESG Integration Framework (AEIF), a four-agent decision-support architecture for ingesting disclosures, mapping and aligning ESG indicators, detecting anomalies, and generating governance-oriented decision trails. The study harmonises 43 ESG indicators across 480 firm-year reports from 24 listed firms in four countries and six sectors over 2005–2024. AEIF is compared with a rules-based baseline using 60 retrospective monitoring scenarios and missing-data stress levels of 10%, 20%, and 30%. AEIF reduced mean compliance-update latency from 47.2 to 12.4 days and remediation time from 16.2 to 3.5 hours. Precision increased from 0.62 to 0.89 and recall from 0.58 to 0.86. At 30% missingness, detection accuracy was 84% versus 42% for the baseline. The findings indicate that governed agentic workflows can improve ESG monitoring while supporting comparability, traceability, human review, and assurance readiness.