Building a Multimodal AI-based Authenticity Assessment Pipeline for ESG Investment Decision-Making: Detecting Greenwashing and Automatically Extracting Sustainability Indicators Using Satellite Imagery, Supply Chain Data, and LLMs

As ESG (Environmental, Social, and Governance) investment emerges as a core element of corporate evaluation, companies' superficial 'greenwashing' practices pose serious risks to investors. This article proposes an innovative approach to quantitatively assess the ESG authenticity of companies, effectively detect greenwashing, and automatically extract necessary sustainability indicators by building a multimodal AI pipeline that integrates satellite imagery, supply chain data, and state-of-the-art Large Language Models (LLMs). This will be a game-changer, addressing the opacity and reliability issues of ESG data and elevating the quality of investment decision-making to a new level.

1. The Challenge / Context

Today, ESG investment has moved beyond simply choosing 'good companies' to become an essential factor in actual financial performance and risk management. However, the biggest problem with current ESG data lies in its lack of reliability and consistency. Companies often provide information in a way that distorts or exaggerates their actual environmental impact or social responsibility, which is referred to as 'greenwashing' or 'ESG washing'. Since traditional ESG evaluations are primarily based on self-reported data from companies, this self-reporting method is fundamentally vulnerable to greenwashing.

Furthermore, manually extracting meaningful sustainability indicators from vast amounts of unstructured data (news articles, social media, corporate reports, etc