Insurance Application

There is no formal insurance application. Agents call in submissions verbally or email free-form descriptions of the risk. The underwriter scribbles notes on a legal pad and prices from memory. When someone asks 'where is the Smith submission?' the answer is 'check my email or ask Janet — she took the call.'

None — AI cannot process submissions because no structured application records exist in any system.

Applications arrive as emailed PDFs, faxed ACORD forms, and broker portal uploads in different formats. The underwriting assistant manually re-keys applicant name, address, coverage limits, and loss history into the policy admin system. Each line of business uses a different supplemental form. Finding a specific application means searching email by agent name or scrolling through a shared drive folder.

AI could potentially OCR the ACORD forms but cannot reliably extract structured risk details because form layouts vary and supplemental pages are inconsistent. Manual re-keying remains necessary.

Applications submit through a standard portal with required fields for each line of business. The system enforces consistent collection — property applications require construction type, square footage, and protection class; auto applications require VIN, driver details, and garaging address. All applications are stored in the underwriting workbench. But supplemental narratives, loss control reports, and broker notes sit as attachments rather than structured fields.

AI can auto-populate rating worksheets and flag incomplete applications. Cannot perform holistic risk assessment because supplemental information (loss control reports, broker narratives) exists only as unstructured attachments.

The insurance application is a fully structured record with all risk characteristics captured as discrete fields. Construction details, occupancy classifications, driver histories, fleet compositions, and coverage specifications are all queryable. An underwriting manager can query 'show me all pending commercial property applications in Florida with frame construction and limits above $5M' and get an accurate, instant answer.

AI can perform automated risk assessment, triage applications by complexity, and route straight-through submissions for auto-binding. Cannot yet integrate real-time external risk signals because application records are point-in-time snapshots.

The insurance application is schema-driven with formal entity relationships. Each application links to its risk score, loss history, property imagery, cat model output, and third-party enrichment records as structured references. An AI agent can ask 'what are all the open applications where the risk score exceeds the guideline threshold and the property imagery shows roof age over 15 years?' and get a precise, computed answer.

AI can perform fully automated underwriting for standard risks — scoring, guideline compliance, pricing, and binding without human intervention. Complex or referred risks surface with complete context for human underwriter review.

Insurance applications are living records that evolve in real-time. As the applicant fills out the portal, risk characteristics stream to the underwriting engine. Third-party enrichments append automatically. The application, risk score, cat model output, and guideline evaluation co-evolve as a single processing stream. There is no 'submission' event — the underwriting decision emerges continuously as information arrives.

Fully autonomous application processing. AI manages the entire intake-to-decision pipeline in real-time without manual intervention for standard risks.

Ceiling of the CMC framework for this dimension.