AI-Driven ESG Analytics for Sustainable Investment in U.S. Non-Profits: Integrating LLMs and Causal Modeling for Policy-Enhanced Resilience

Abstract

This study develops an integrated analytical framework that combines large language models with causal inference methods to strengthen sustainable investment decision-making in U.S. non-profit organizations. Drawing on advances in natural language processing, the proposed system applies an LLM-based architecture capable of interpreting regulatory texts, extracting domain-specific ESG signals, and synthesizing policy-relevant insights. Complementing this linguistic capability, the framework incorporates causal modeling-particularly Difference-in-Differences estimation-to identify the impact of policy changes on environmental, social, and governance performance factors. Together, these tools provide a structured foundation for supporting responsible investment strategies that align with mission-driven objectives. The model design also includes a multilayered feedback mechanism for continuous refinement, multilingual accessibility, and multi-format output generation, enabling diverse nonprofit stakeholders to access interpretable ESG results. The findings suggest that the integration of LLM-driven analytics with empirical causal evaluation enhances transparency, improves resilience in policy-sensitive contexts, and supports equitable governance practices. This research contributes to emerging scholarship on AI-enabled sustainability systems while offering practical implications for organizational strategy in the nonprofit sector.