Risk Score

There is no formal risk score. The underwriter evaluates risk based on experience and gut feel. Pricing decisions rely on personal judgment about whether a risk is 'good' or 'bad.' When asked 'why did you price this at $50K?' the answer is 'that felt right given my experience with similar accounts.'

None — AI cannot automate or assist with risk scoring because no scoring methodology or historical scores exist.

A risk scoring worksheet exists but varies by underwriter. Some use a point system they developed personally. Others use the carrier's legacy rating algorithm that nobody fully understands. Scores are not consistently recorded — the underwriter calculates a score mentally, makes a pricing decision, and moves on. Comparing risk scores across underwriters or time periods is unreliable because the methodology is not standardized.

AI could replicate individual underwriters' scoring patterns from historical decisions, but the models would be inconsistent across underwriters and unexplainable to regulators.

A standardized risk scoring model exists and is applied consistently. Scores are recorded for every submission. The methodology defines clear factors (loss history weight, industry hazard class, years in business) and produces a numeric score on a defined scale. Underwriting management reviews score distributions monthly. But the model is static — it was built three years ago and has not been recalibrated against actual loss experience.

AI can apply the existing scoring model consistently and flag submissions that deviate from expected score ranges. Cannot refine the model because the factors and weights are not transparent or modifiable.

Risk scores are generated by a well-documented model with transparent factor definitions and published weights. The model recalibrates quarterly using actual loss experience. Scores link to the specific input variables that drove them — an underwriter can see 'this score is elevated because the loss ratio exceeded 80% in two of the last three years and the industry class carries a 1.4 hazard factor.' Regulators can audit the scoring methodology end-to-end.

AI can generate risk scores autonomously, explain scoring decisions to underwriters, and recommend model adjustments based on emerging loss patterns. Full transparency enables regulatory compliance for AI-assisted scoring.

Risk scores are schema-driven with machine-readable factor definitions and API-accessible scoring endpoints. Model versions are formally managed with A/B testing capabilities. An AI agent can query 'score this submission using Model v3.2, then re-score using v4.0-beta and compare the factor-level differences.' Scoring models are parameterized objects that AI can evaluate, compare, and optimize programmatically.

AI can perform autonomous model management — testing new scoring approaches, evaluating accuracy, and recommending model updates. Fully automated scoring with human oversight limited to model governance.

Risk scores are living metrics that recalculate continuously as new signals arrive. A change in the insured's credit score, a new claim filing, or updated telematics driving behavior immediately adjusts the risk score. There is no 'scoring event' — the risk score is a continuous function of all available input signals, always reflecting the most current risk assessment.

Fully autonomous risk scoring. AI maintains continuously current risk assessments that adapt in real-time to changing conditions.

Ceiling of the CMC framework for this dimension.