Automating Underwriting Decisions with AI: From 72-Hour Reviews to 4-Minute Approvals
Challenge
Manual underwriting reviews took 48–72 hours and required three senior underwriters per commercial policy. Each application involved cross-referencing multiple external data sources and applying manual rating logic. As submission volumes increased, the underwriting team became a bottleneck, forcing the business to turn away new opportunities due to limited processing capacity.
Solution
We built an AI-powered underwriting engine that ingests structured application data, third-party risk signals (credit bureaux, Companies House, property databases), and historical claims data to generate a risk score and coverage recommendation in under four minutes. The platform includes a rules-based decision layer for regulatory and product constraints, and a referral workflow that routes complex cases to underwriters with pre-populated risk insights.
Results
Policy approval time reduced from 72 hours to 4 minutes for 78% of applications. Straight-through processing reached 78% with no increase in loss ratio. Underwriter capacity was reallocated to complex, high-value cases, increasing monthly processing volume from 400 to 640 applications with the same team. Broker NPS increased from 34 to 67.
The Underwriting Bottleneck
Global Shield processed approximately 400 commercial property applications per month. Each required manual review by senior underwriters: analysing application data, querying external datasets, applying actuarial tables, and drafting terms.
Average turnaround time was 48–72 hours. During peak periods, queues formed, and brokers placing time-sensitive business often chose faster competitors. The bottleneck was not demand, it was processing speed.
Data and Risk Architecture
Data ingestion pipeline
Real-time API integrations with credit bureaux, company registries, property datasets, and internal claims systems. All inputs normalised into a unified risk feature layer.
Risk scoring model
Gradient boosting model trained on 8 years of historical policy and claims data. Key features include industry sector risk, geographic exposure, financial strength, claims history, and property attributes. Model performance improved significantly over legacy actuarial approaches (AUC 0.87 vs. 0.71).
Decisioning engine
Model outputs feed a rule engine that applies underwriting guidelines, regulatory requirements, and risk thresholds to produce decisions: approve, refer, or decline, with recommended pricing and coverage terms.
Human-in-the-Loop Workflow
Approximately 22% of applications are routed for manual review. Underwriters receive a pre-populated review pack including:
- Model risk score and key drivers
- Relevant third-party data
- Suggested decision
Review time reduced from 2.4 days to 35 minutes per case.
Operational Impact
The underwriting process shifted from sequential manual review to real-time decisioning:
- Instant approvals for standard risks
- Faster turnaround for complex cases
- Increased throughput without increasing headcount
The Outcomes
Decision time
72 hours → 4 minutes (straight-through cases)
Straight-through processing rate
0% → 78%
Monthly applications processed
400 → 640 (same team)
Loss ratio
62% → 61.4%
Business Impact
Faster underwriting improved broker conversion and reduced lost deals during peak periods. Underwriters were redeployed to complex risk assessment rather than repetitive processing. The platform enabled growth without increasing operational risk or team size.
Result
Global Shield moved from a capacity-constrained underwriting model to a scalable, real-time decisioning system, increasing throughput and competitiveness without impacting loss performance.