Conference 2: Universities’ Transport Study Group (UTSG). Dublin 2025.

Enhancing ESG Data Collection: LLM Approaches for Airline Emissions KPIs

Luis MARTÍN-DOMINGO^{1, 2}; Jaime B. FERNÁNDEZ¹; Marina EFTHYMIOU¹; Muhammad Intizar ALI¹

1: Dublin City University; 2: Ozyegin University;

ABSTRACT

Extracting environmental Key Performance Indicators (KPIs) from airline sustainability reports is critical for evaluating environmental performance and ensuring regulatory compliance in the European aviation industry.
Manual extraction from lengthy, unstructured reports is time-consuming and prone to inconsistency. This research systematically explores the capabilities of state-of-the-art Large Language Models (LLMs)—namely GPT-4.0, o3-mini, and Deepseek R1—for automating the extraction of emissions-related KPIs from the 2023 sustainability disclosures of 16 publicly listed European airline groups.

Leveraging the Perplexity platform, the study compares manual expert extraction with automated LLM-based methods, examining different models, prompt engineering techniques, and input data formats. Findings reveal that LLM extraction accuracy is highly sensitive to the specificity of prompt instructions. Unstructured document extraction without targeted guidance resulted in poor performance, whereas prompts explicitly referencing KPI terms improved accuracy from less than 30% to over 70%. The structure of the source data also played a significant role, with HTML sources enabling more accurate extraction than PDF files. Although challenges remain in harmonizing data formats and extracting granular KPI values, results show that LLMs can significantly enhance the efficiency of environmental, social, and governance (ESG) data collection when prompt design and data standardization are emphasized. The paper offers practical recommendations for integrating LLMs into ESG analytics workflows and outlines future research avenues for advancing automation in sustainability reporting.

Keywords: Airlines, LLMs, Sustainability, KPIs, GHG Emissions