Insurance
Intelligent Fraud Detection at Scale
Customer Profile
One of India’s largest life insurance companies processing approximately 20 lakh claims annually across term, endowment, and ULIP products.
Challenge
The insurer faced a cost-viability problem with AI-powered fraud detection. Their existing system combining rules and ML models achieved 71% detection but with a 40% false positive rate that overwhelmed the Special Investigations Unit (SIU).
A promising LLM-based proof-of-concept reached 89% detection accuracy, but at ₹4.5 crore annual cost—failing to meet the CFO’s 10:1 ROI requirement against estimated ₹100 crore in annual fraud losses.
Key challenges:
- High false positives causing investigator fatigue and delayed legitimate claims
- Unsustainable AI costs making advanced detection economically unviable
- IRDAI pressure for transparent fraud detection with audit trails
- Regional patterns missed by models trained primarily on metro data
Solution
Rotavision implemented a “Trust Cascade” architecture—a five-level intelligent routing system that matched detection complexity to claim value and risk profile:
| Level | Method | Cost/Claim | Volume |
|---|---|---|---|
| 1 | Rules Engine | ₹0.008 | 68% |
| 2 | ML Models | ₹0.08 | 22% |
| 3 | Single AI Agent | ₹0.65 | 7% |
| 4 | Multi-Agent Panel | ₹2.00 | 2% |
| 5 | Adversarial Review | ₹3.60 | 1% |
Routing Logic:
- Low-value claims (< ₹5 lakh) capped at Level 2
- High-value claims (> ₹25 lakh) received full cascade review
- Regional risk factors triggered enhanced scrutiny
- Historical claimant patterns influenced routing
Implementation
The 16-week implementation focused on cost optimization without sacrificing accuracy:
Phase 1: Pattern Analysis (Weeks 1-4)
- Analyzed 3 years of claims data for fraud patterns
- Identified regional variations in fraud typologies
- Mapped investigator decision patterns for model training
Phase 2: Cascade Design (Weeks 5-8)
- Designed routing logic based on claim characteristics
- Built confidence thresholds for level escalation
- Created feedback loops for continuous improvement
Phase 3: Deployment (Weeks 9-14)
- Deployed Guardian for accuracy monitoring across all levels
- Integrated Orchestrate for multi-agent coordination
- Built SIU dashboards with explainable alerts
Phase 4: Optimization (Weeks 15-16)
- Fine-tuned routing thresholds based on live data
- Trained SIU team on new investigation workflows
- Established IRDAI-compliant audit trails
Results
| Metric | Before | After | Change |
|---|---|---|---|
| Detection Rate | 71% | 94% | +32% |
| False Positive Rate | 40% | 12% | -70% |
| Cost Per Claim | ₹0.65 | ₹0.19 | -71% |
| Annual Detection Cost | ₹1.5Cr | ₹46L | -69% |
| Fraud Prevented | ₹85Cr | ₹113Cr | +₹28Cr |
Business Impact
- Investigator productivity increased 3x with focused, high-confidence alerts
- 127 auto-generated rules within six months from AI-detected patterns
- IRDAI audit passed with commendation for transparency
- CFO approved expansion to health insurance claims
What’s Next
The insurer is extending the platform to:
- Health insurance claims with hospital network verification
- Motor insurance with image-based damage assessment
- Agent fraud detection for distribution channel integrity
Rotavision is powered by Rotascale’s globally-proven AI trust platform.
Ready to Achieve Similar Results?
Let's discuss how Rotavision can help your organization.
Schedule a Consultation