Real estate underwriting with AI: climate risk, valuation models and the new rules

FEMA's Risk Rating 2.0 changed National Flood Insurance Program pricing to reflect each property's individual flood risk, using factors such as distance to water and cost to rebuild, instead of relying mainly on flood zones ^[1]. Commercial datasets such as First Street's publish flood, fire, heat and wind risk scores for individual properties ^[2]. Underwriting that ignores this data is now pricing risk others can see.

Automated valuation models estimate value from sales, listings and property data. In 2024 six US agencies, including the CFPB and FHFA, finalised a rule requiring mortgage originators and secondary market issuers that use AVMs to adopt quality controls: confidence in the estimates, protection against data manipulation, avoiding conflicts of interest, random testing and compliance with nondiscrimination laws ^[3]. It took effect on 1 October 2025.

For commercial property, the time sink is documents: leases, rent rolls, operating statements, environmental and inspection reports. Language models can extract rents, terms, options and obligations into a standard model, with a citation to each clause, so the analyst checks rather than types.

• Extract lease terms and flag non-standard clauses.
• Reconcile the rent roll against leases and bank statements.
• Summarise inspection and environmental reports by severity.
• Attach climate and flood scores to every asset in a portfolio.

Found this useful?

Share the key finding on X.