Build and own a Risk Selection Specialist Model.
Should we accept, decline, refer, or conditionally accept this risk at underwriting, and at what price?
Overview
Improves risk selection accuracy at point of sale, reducing adverse selection into the book.
Who owns this decision
Chief Underwriting Officer, VP Underwriting
Why owned AI may fit
Underwriting appetite, pricing adequacy, and risk tolerance are carrier-specific and tied to the carrier's own loss experience and rating plan.
What this Specialist Model actually does
Decision: Should we accept, decline, refer, or conditionally accept this risk at underwriting, and at what price?
This is a larger system with 4 specialist model candidates and 2 supporting or external components.
Who uses this Specialist Model
- Chief Underwriting Officer
- VP Underwriting
- Head of Product
- Chief Actuary
Where the specialist model sits
This is a larger system with 4 specialist model candidates and 2 supporting or external components.
Classifier/pricing model trained on historical underwriting decisions and loss outcomes
Specialist Model
Structured extraction of application and inspection data into underwriting features
Specialist Model
Vision model
Specialist Model
Vision model for property/asset inspection imagery where applicable
More Specialist Models in Insurance →
Specialist LLM
Specialist Model
Specialist LLM for underwriting rationale narrative and evidence citation on referred risks
More Specialist Models in Insurance →
Rules engine encoding regulatory rating plan and eligibility constraints
Supporting component
How Oumi connects to your stack →
Senior underwriter review queue
Supporting component
Senior underwriter review queue for referred and low-confidence risks
How Oumi connects to your stack →
How to evaluate and improve it
Evaluation
- Loss ratio on accepted risks versus prior underwriting process
- Decline/accept consistency across underwriters and regions
- Price adequacy versus actual loss experience
Feedback signals
- Underwriter override of the recommended decision
- Loss experience on bound policies versus predicted risk
- Renewal retention and rate adequacy outcomes
- Referral outcome (underwriter accepted, modified, or declined the recommendation)
- Regulatory rate/form filing feedback affecting acceptable risk criteria
- Claims emerging shortly after binding (early indicators of poor selection)
Model candidates to evaluate
- modernbert-large
- qwen3-vl
- qwen3.5-9b
These are candidates, not performance claims. Run them against your evaluation set.


