Sustainalyze: An LLM-Driven Approach for Transparent Sustainability Benchmarking

The current economic system is undergoing a challenging transformation to become more sustainable. To achieve this, many organizations are legally obliged to publish sustainability reports in order to show their impact on the environment. Analyzing sustainability reports and benchmarking against competitors is a time-consuming and complex task for organizations. This study presents Sustainalyze, a software solution for conducting transparent sustainability benchmarking at the intersection of Large Language Models and Natural Language Processing. Based on our findings, this study offers an overview of the software architecture and an evaluation of the methods used for the automatic extraction of sustainability Key Performance Indicators from publicly available reports. The software achieves a mean extraction accuracy of 74\%, with accuracies reaching up to 85\% for certain values when processing reports containing up to hundreds of pages. These results demonstrate the potential of AI-driven tools in streamlining sustainability report analysis, facilitating more efficient benchmarking, and supporting progress towards net-zero emission targets.