An AI-assisted workflow for emissions data preparation in sustainability reporting

Abstract

Sustainability reporting requires accurate, traceable, and well-formatted emissions data. However, this information is often fragmented across PDF and Excel documents, creating manual effort for collection, restructuring, checking, and transfer. This thesis investigates how an AI-assisted workflow can support emissions-data preparation. Following a Design Science Research Methodology approach in collaboration with an international energy company, the research combines a literature review, qualitative workflow analysis, prototype development, and user evaluation. The study characterises emissions-data preparation as a workflow problem shaped by fragmented tools, repeated checks, and limited traceability. To address this, a prototype was developed using rule-based preprocessing, schema-guided LLM extraction, and human-in-the-loop review. The workflow maps extracted information into predefined reporting fields, allows users to edit values, and exports the reviewed data into the required reporting format. Evaluation results indicate that practitioners found the workflow useful for reducing repetitive preparation work, while still considering human review necessary before reporting. This work shows that AI can support emissions-data preparation when used in a controlled, review-based workflow.