Cross agency data fusion and predictive intelligence to reduce fraud across U.S benefits programs

Fraud and improper payments across U.S. federal benefits programs impose substantial fiscal and social costs, with the Government Accountability Office estimating annual fraud-related losses of $233–521 billion and cumulative improper payments exceeding $2.7 trillion since fiscal year 2003. Traditional "pay-and-chase" recovery models have proven insufficient, prompting growing interest in preventive strategies that leverage cross-agency data fusion and predictive intelligence. This paper presents a qualitative comparative analysis of four dominant institutional and technical models used to integrate data and apply predictive analytics for fraud reduction: centralized federal data hubs, federated and privacy-preserving architectures, vendor-managed identity and fraud detection systems, and state-level or program-specific predictive analytics. Each model is evaluated against five criteria derived from federal oversight guidance and academic literature: legal and institutional authority, data governance and privacy protection, technical architecture and scalability, analytic effectiveness, and transparency and accountability. The analysis demonstrates that no single model adequately balances these competing demands. Centralized hubs offer statutory grounding but limited analytic flexibility; federated approaches strengthen privacy at the cost of governance complexity; vendor systems raise accountability concerns; and program-specific models lack cross-jurisdictional reach. The paper proposes a hybrid framework combining centralized authoritative checks, federated analytics for sensitive data, and program-level predictive modeling under unified governance with human-in-the-loop decision-making. Policy recommendations address statutory authorization, oversight structures, model validation, vendor accountability, and workforce capacity to support responsible, equitable, and scalable fraud prevention.