$45M
Overpayments recovered
20x
More claims flagged
60%
Detection accuracy
25%
Flagged claims are systemic issues
The challenge
Rooting out overpayments and containing costs
A leading US health payer wanted to curb Medicare and Medicaid overpayments. The healthcare system loses over $200B to fraud, waste, and abuse (FWA) annually. Comparing similar claims is essential for identifying irregular claims or aberrant billing practices, but made more difficult by disconnected data and complex member-provider patterns.
Key challenges
Healthcare system vulnerable to fraudulent and wasteful claims
Inconsistent detection of coverage eligibility
Limited capacity to compare similar claims
Pressing need for scalable cost containment
The solution
Leveraging unsupervised learning for overpayment detection
Improved claim accuracy
Identified systemic overpayments
More successful detections
Flagged bogus billing patterns
Automated validation
Amplify detection accuracy
Prioritized low-complexity claims
Minimize manual reviews
Implementation approach
1
Focus areas
Define FWA drivers
Develop hypotheses
Identify impact
2
Segmentation
Automate framework processes
Conduct clinical validations
Cost containment protocols
3
Four-step unsupervised learning framework:
Define FWA drivers and high-impact areas
Map 50+ hypotheses for internal & external data
Create homogenous segments with unsupervised methods
Evaluate anomalies in dosage, pricing, or coverage to flag overpayments
The impact
Transformed claims recovery through intelligent insights
Financial gains
$45M
Overpayments uncovered
75% dosage/unit errors
25% systemic issues
Reduced fraud
Process efficiency
60%
Detection accuracy
Vs. 10–20% baseline
Targeted leads for foragers
Reduced manual rechecks
Strategic results
20x
More claims flagged
Prioritized by impact
Fewer overlooked aberrancies
Boost payer confidence
Looking ahead
Real-time alerts
Instant fraud detection halts overpayments at submission
Fraud risk index
Developing a global scoring system for claims risk
Ethical AI protocols
Developing transparent, bias-free fraud detection
