Automated ESG Compliance Auditing using Retrieval- Augmented Generation and Semantic Vector Search: A Computational Economics Approach

ESG (Environmental, Social, Governance) is a framework used by investors and companies to assess an organization’s sustainability, ethical impact, and long-term performance along with financial metrics. Given that only larger companies are required to report ESG metrics, corporate annual reports, which are mandatory for market listed companies can be used to estimate the ESG scores of a company. However, such reports are often long, fraught with technical jargon, and inconsistent narrative language. To address this challenge, this paper proposes an artificial intelligence based automated auditing methodology using the RAG (Retrial-Augmented Generation) framework. Here, both disclosures and compliance queries are encoded as semantic vector embeddings and information are retrieved based on semantic similarity rather than keyword matching. A large language model then processes the collected evidence into a set of relevant disclosure categories, finally culminating in a proposed ESG score to assess firms and compare them with their peers. The proposed framework is expected to improve the scalability of ESG disclosure analysis and also enable cost effective automated verification of ESG reports using other reports and material published by the firm under scrutiny.