Engineering Blog · Post #48

AI-Recommended Declinations: Risk Factor Analysis with Underwriter Final Decision

InsightUW's AI analyzes 7 risk factors, scores confidence, and recommends decline — but the underwriter always has the final word.

The Problem

Auto-decline rules catch the obvious cases — blocked clearance, extreme loss ratios, out-of-appetite limits. But many borderline submissions require judgment: a marginal appetite score combined with clearance issues and elevated claims frequency. These submissions sit in queue while underwriters manually assess each one. The AI can analyze faster — but the UW needs to retain decision authority.

The InsightUW Approach

InsightUW's AI recommendation engine evaluates multiple risk factors simultaneously, assigns weights and severity, and recommends a disposition. But unlike auto-decline rules that fire automatically, AI recommendations require explicit UW approval before any action is taken.

graph TD subgraph Analysis["AI Risk Factor Analysis"] F1["Appetite Score < 40 = high (w:25)"] F2["Clearance Issues blocked/issue = high (w:20)"] F3["Loss Ratio > 80% = high (w:25)"] F4["Claim Frequency > 5 claims = med (w:15)"] F5["Requested Limit > $25M = med (w:10)"] end subgraph Decision["Decision Logic"] Score["Total Weight Score + Confidence %"] DEC["weight >= 40 → Decline weight 20-39 → Refer weight < 20 → Proceed"] end subgraph UW["UW Final Decision"] Card["AI Recommendation Card Risk factors + confidence + next-best-actions"] Approve["UW Approves Decline → Email draft generated"] Reject["UW Rejects → Submission continues"] end F1 --> Score F2 --> Score F3 --> Score F4 --> Score F5 --> Score Score --> DEC DEC -->|"decline/refer"| Card Card --> Approve Card --> Reject

Risk Factor Scoring

Factor	Trigger	Weight	Severity
Low Appetite Score	< 40	25	high
Marginal Appetite	40-60	10	medium
Clearance Issues	blocked or issue_found	20	high
Adverse Loss Experience	loss_ratio > 80%	25	high
Elevated Loss Ratio	60-80%	10	medium
High Claim Frequency	> 5 claims	15	medium
High Requested Limit	> $25M	10	medium

Confidence Scoring

Base confidence starts at 50% and increases with each risk factor. High-match-score factors add 15-20%, medium add 5-8%. Maximum confidence caps at 95%.

The AI Recommendation Card

A purple-bordered card appears on the submission detail showing:
- Recommendation badge (DECLINE in red or REFER in amber)
- Confidence score (e.g., 85%)
- Risk factors — each as a row with severity badge, factor name, detail, and weight
- Next-best-actions — what to do after the decline
- Approve Decline (red button) — generates email draft
- Reject & Continue UW (green button) — dismisses recommendation

LOB-Specific Example

D&O for Solaris Energy Holdings:

AI Analyze is clicked. The engine evaluates:
- Appetite score: 35 → Low Appetite (weight 25, high)
- Clearance: issue_found → Clearance Issues (weight 20, high)
- Total weight: 45 → DECLINE recommended
- Confidence: 85%

UW Sarah Chen reviews the card, sees both risk factors, but notes the insured is expanding into renewable markets. She clicks "Reject & Continue UW" — the recommendation is logged but the submission proceeds.

What This Means for Underwriters

AI assists, UW decides — recommendations are never executed without explicit UW approval
Multi-factor analysis — considers appetite, clearance, loss history, claims, and limits simultaneously
Confidence transparency — UW sees exactly why the AI recommends decline and how confident it is
Speed — AI analysis runs in milliseconds, surfacing borderline risks that would take hours to assess manually
Auditable — every recommendation, approval, and rejection logged with UW name and notes

What's Next

Next: Override Auto-Declinations — The Exception Approval Workflow

Want to see AI-powered decline recommendations in action? Request a demo.