Conversational underwriting

The design principle from the A2A Core Patterns standard is look up, do not ask. In motor insurance, the vehicle registration and driving licence number key the majority of the risk picture from authoritative databases: vehicle specification from DVLA, licence history from MyLicence and IIADD, claims history from CUE, conviction records from MyLicence. In home insurance, the property address, resolved to a UPRN, keys property characteristics, rebuild cost from BCIS, and flood risk from the Environment Agency and Flood Re. In each case, the agent retrieves and confirms rather than interrogating the customer for information already recorded in an authoritative source.

The assume-infer-confirm pattern governs how enriched data is treated. An enriched value is never treated as warranted by the customer until the customer has explicitly attested it. The agent presents enriched facts - 'we have found your vehicle is a 2021 Ford Focus, 1.5L diesel - is that correct?' - and marks each field in the audit record as either enriched or attested. No material fact is relied upon for a bind unless it is attested, regardless of how reliable the enrichment source is considered to be.

Conversational underwriting is the most visible demonstration of why an AI agent outperforms a traditional web form: the form asks the customer twenty questions; the agent asks for two canonical identifiers and confirms the rest. The motor standard is the exemplar of this approach, but the same principle applies across all lines defined in the A2A protocol.