Innovating ESG and sustainable business practices through Generative AI

This thesis explores the transformative potential of Generative Artificial Intelligence (GenAI), particularly Large Language Models (LLMs), in advancing corporate environmental sustainability. The increasing importance of ESG (Environmental, Social, and Governance) topics and the complexities of current reporting frameworks, characterized by heterogeneity and unstructured data, present significant challenges for stakeholders seeking consistent and comparable information. This research addresses how GenAI can overcome these hurdles, thereby enhancing transparency and accountability in sustainability reporting. The methodology is structured in three distinct phases. First, a systematic literature review, leveraging GenAI itself, identifies the current landscape of GenAI applications in corporate environmental sustainability. This review classifies existing research into key thematic clusters, including applications for improving business operations, enhancing sustainability reporting and ESG analysis, and addressing the environmental impact of GenAI itself. Second, the thesis empirically assesses GenAI's capability in evaluating corporate alignment with the EU Taxonomy. This phase specifically compares the accuracy and efficiency of Retrieval Augmented Generation (RAG) and Long Context processing techniques in extracting relevant information from non-financial disclosure documents, demonstrating the superior performance of Long Context models in this task. Finally, a case study illustrates the practical application of GenAI in supporting ESG researchers. Focusing on Italian transport infrastructure companies, this case study showcases how LLMs can automate data extraction, identify trends and patterns, and generate actionable insights from complex sustainability reports, facilitating cross-company comparisons and identifying opportunities for best practice adoption. The key findings confirm that GenAI is a powerful tool with the potential to significantly contribute to the field of sustainable business practices. Our results provide valuable insights into the nascent but rapidly accelerating literature on GenAI and environmental sustainability. From a practical standpoint, the demonstrated efficacy of LLM-based tools in assessing EU Taxonomy alignment suggests a promising avenue for more scalable and consistent regulatory monitoring, potentially reducing the resource burden for companies, especially SMEs. For academia, this research lays a foundational understanding of GenAI's impact on corporate sustainability and outlines fertile ground for future research, including exploring other GenAI models and refining methodological approaches. Ultimately, this work advocates for the integration of GenAI into ESG auditing and reporting frameworks to foster a more transparent, accountable, and sustainable business environment. Despite the promising potential highlighted, this work also acknowledges several limitations. A critical aspect for future research involves developing robust methodologies for validating the reliability and accuracy of AI models used in this domain, ideally against a "golden dataset" to ensure replicability of results. Furthermore, while Generative AI offers significant advantages, the continuous evolution of this technology necessitates a strategic approach: creating bespoke, dedicated models for ESG analysis may not be cost-effective given the rapid pace of innovation and the constant emergence of new, powerful out-of-the-box solutions. Instead, grounding existing highly-capable LLMs with specific ESG information appears to be a more viable and future-proof strategy, allowing continuous leveraging of their advanced features. The ability to ground information with the source of the answer is also crucial for maintaining transparency and trustworthiness in AI-generated insights.