Abstract
Climate change is causing increased precipitation and water intrusion in buildings, leading to a rise in building damages in Norway. Insurance companies hold extensive data on such damages through claims processing. In collaboration with SINTEF, Gjensidige explored how this data can be structured and analysed to better understand damage causes, severity, and links to building types and locations. Under a data processing agreement, SINTEF accessed 17,000 anonymized climate-related damage reports. Sensitive details were removed, and the reports were organized using advanced text extraction and categorization methods, including Large Language Models. The analysis revealed patterns in when and where damages occur, identified vulnerable building components, and examined how damage severity varies by building type and age. It also uncovered weaknesses in assessor templates, providing a basis for improving data collection and documentation. This work demonstrates the potential of leveraging large text-based datasets from the insurance industry to support data-driven decision-making in the building sector. Future efforts should expand the dataset to include all insurers, integrate weather and geographical data, and enhance reporting templates and standardization to improve data quality and interoperability across stakeholders.