National General Insurance Provider
Claims analytics and fraud detection saving 12% in annual claims leakage.
All case studiesA top-10 general insurance provider in India processing over 80,000 claims annually engaged Cydez Technologies to build a claims analytics and fraud detection platform. The insurer had no systematic fraud detection capability — suspicious claims were identified only through the intuition of experienced adjusters, leading to an estimated 12% annual claims leakage. Legitimate claimants suffered from the same extended processing timelines as fraudulent ones, damaging customer satisfaction and Net Promoter Scores.
The platform was built on Azure cloud infrastructure with Databricks for data engineering and ML model training. Five years of historical claims data — covering motor, health, property, and travel insurance lines — was used to train gradient boosting and neural network models for fraud scoring. The system integrates with the core policy administration system via MuleSoft, scores every incoming claim in real time, and routes high-risk claims to a dedicated investigation queue while fast-tracking low-risk claims for automated settlement.
Over 50 Power BI dashboards provide claims portfolio visibility, fraud trend analysis, adjuster performance metrics, and regulatory reporting. The platform also includes a network analysis engine that identifies fraud rings by mapping relationships between claimants, service providers, and witnesses across claims. Within the first year, the fraud investigation hit rate improved from 15% to 62%, and overall claims processing time was reduced by 45%.
- Claims analytics platform on Azure/Databricks
- ML fraud detection across 4 insurance lines
- Real-time claim scoring and routing engine
- Network analysis for fraud ring detection
- 50+ Power BI dashboards
- MuleSoft integration with policy admin system
A general insurance provider processing 80,000+ claims annually had no systematic fraud detection capability. Suspicious claims were identified only through manual review by experienced adjusters, leading to significant leakage. Claims processing was slow, and legitimate claimants suffered from the same delays as fraudulent ones.
Cydez built a claims analytics platform on OCI with ML-based fraud scoring models trained on 5 years of historical claims data. The platform integrates with the core policy administration system via MuleSoft, scores every incoming claim in real time, and routes high-risk claims to a dedicated investigation queue. Oracle Analytics Cloud dashboards provide claims portfolio visibility, fraud trend analysis, and adjuster performance metrics.
How we delivered this project
Discovery
Analyzed 5 years of historical claims data across motor, health, property, and travel lines. Interviewed 30 experienced adjusters to capture fraud indicator heuristics. Mapped the claims lifecycle and integration points with the policy administration system.
Design
Designed the Azure/Databricks architecture for data engineering and ML pipelines. Created feature engineering specifications with 120+ fraud indicator variables. Designed the real-time scoring API and claims routing logic. Planned 50+ Power BI dashboard layouts.
Development
Built Databricks data pipelines for claims data ingestion and feature engineering. Trained gradient boosting and neural network fraud models. Developed the real-time scoring API on Azure Functions. Built MuleSoft integration flows and 50+ Power BI dashboards with RLS.
Launch
Deployed in production with a 90-day shadow scoring period (scoring all claims but not routing). Validated model accuracy against adjuster decisions. Switched to live routing after achieving 62% investigation hit rate. Trained 150 claims staff across 4 regional offices.
What we built
Real-Time Fraud Scoring
Every incoming claim scored in real time using ML models trained on 5 years of historical data. High-risk claims automatically routed to investigation queue. Low-risk claims fast-tracked for settlement.
Network Analysis Engine
Graph-based analysis identifying fraud rings by mapping relationships between claimants, service providers, witnesses, and repair shops across claims. Visual network exploration for investigators.
Multi-Line Coverage
Separate fraud models optimized for motor, health, property, and travel insurance lines. Each model trained on line-specific fraud patterns with 120+ feature variables.
Claims Portfolio Dashboards
50+ Power BI dashboards covering claims volume, severity, reserve adequacy, cycle time, adjuster workload, and fraud trends. Row-level security by region, branch, and line of business.
Adjuster Performance Analytics
Individual adjuster performance metrics including cycle time, accuracy, customer satisfaction, and fraud detection effectiveness. Workload balancing recommendations.
Regulatory Reporting
Automated generation of IRDAI regulatory reports including claims ratio, fraud statistics, and settlement timelines. Audit trail for all claim decisions and model scores.
Measurable outcomes
- 12% reduction in annual claims leakage from fraud detection
- Average claims processing time reduced by 45%
- 80,000+ claims per year scored in real time
- Fraud investigation hit rate improved from 15% to 62%
Built with
Want similar results?
Tell us about your project and we'll explore how Cydez can deliver comparable outcomes for your organisation.
Start a Conversation