Underwriting Guideline

There are no documented underwriting guidelines. Risk acceptance and declination criteria live entirely in senior underwriters' heads. When a new underwriter asks 'should I write this restaurant in a flood zone?' the answer depends on which senior underwriter they ask. Pricing authority varies by personality rather than formal delegation.

None — AI cannot apply underwriting rules because no documented guidelines exist.

Underwriting guidelines exist as Word documents and spreadsheets maintained by senior underwriters. Each line of business has its own guide in a different format. Some guidelines are detailed (auto physical damage has a 12-page rate manual), others are sparse ('use judgment for this class'). The guidelines were last updated two years ago — newer risk types like cyber and cannabis are not covered. Finding the right guideline for a specific risk requires knowing which file to open.

AI could reference the guideline documents as context, but cannot consistently apply them because the format is inconsistent, the coverage is incomplete, and the interpretation requires institutional knowledge.

Underwriting guidelines follow a standardized format across all lines of business. Each guideline defines acceptable risk characteristics, declination triggers, required documentation, and pricing authority levels. Guidelines are stored in a central repository, accessible to all underwriters. Annual reviews update the guidelines. But the guidelines are still narrative documents — 'frame construction in coastal zones requires senior underwriter approval' — rather than machine-interpretable rules.

AI can reference guidelines as context for recommendations but cannot automatically enforce them. A human must interpret whether a specific submission meets the narrative criteria described in the guideline.

Underwriting guidelines are structured as machine-readable rules in a rule engine. Each rule defines the condition (risk characteristics), the action (accept, decline, refer), and the authority level. Rules link to the specific risk parameters they evaluate. An underwriting manager can query 'show me all guideline rules that reference coastal exposure for commercial property' and get a precise list of applicable rules with their parameters.

AI can automatically evaluate submissions against underwriting guidelines, generate accept/decline/refer recommendations, and explain which specific rules drove the recommendation. Straight-through processing is possible for risks that cleanly pass all rules.

Underwriting guidelines are schema-driven with versioned rule sets. Each rule has formal metadata — effective date, expiration date, approval authority, and performance metrics (how many risks this rule has accepted/declined and their loss outcomes). An AI agent can evaluate 'score this submission against Guideline v4.2 and compare the result to v5.0-beta to identify which rule changes affected the outcome.' Guidelines are programmable objects.

AI can perform autonomous guideline management — evaluating rule effectiveness, recommending rule modifications based on loss outcomes, and testing proposed changes against historical submissions. Human oversight limited to strategic guideline governance.

Underwriting guidelines are living rules that evolve continuously. Market condition changes trigger automatic guideline adjustments within defined parameters. Loss experience feedback recalibrates rule thresholds. Regulatory changes propagate to affected rules immediately. The guideline set adapts continuously to the current risk environment.

Fully autonomous guideline management. AI maintains and evolves underwriting rules in real-time based on market conditions, loss experience, and regulatory requirements.

Ceiling of the CMC framework for this dimension.