Mortgage decisioning for an AI lending agent

Decision: Banking — Mortgage underwriting. Pattern: AI agent decision layer

Use case: An AI lending agent handles the mortgage application conversation and gathers everything needed to assess it, but the approve, refer, or decline decision itself is delegated to Rainbird — the agent acts only on the outcome it gets back, never on its own judgement of the application. See the UK Mortgage Underwriting & Affordability accelerator for a ready-built starting point.

End-user: Mortgage underwriter, reviewing agent-assisted applications that Rainbird has referred

The decision: Whether a mortgage application should be approved, referred for manual underwriting, or declined — based on stressed affordability, credit history, and the lender's property policy.

The knowledge graph: Encodes the lender's affordability stress-testing rules, credit policy, and property criteria, so every application is judged against the same standard regardless of which channel or agent it came through.

Integration: The AI agent conducts the application conversation and assembles the supporting data — income, outgoings, credit file, property valuation — from the applicant and connected systems. Once assembled, that data is passed to Rainbird as a single tool call over MCP or API. Rainbird returns approve, refer, or decline with its reasoning, and the agent acts strictly on that outcome: progressing an approval, handing a referral to a human underwriter, or communicating a decline. At no point does the agent make or override the underwriting decision itself.

Evidence: Every decision returns the affordability stress test result, the specific credit and property policy checks applied, and a rationale for the outcome — giving the underwriter, and the applicant where required, a clear explanation for every approve, refer, or decline.