Eliminating Fraud waste and abuse – solving a hundred-billion-dollar problem for Americans

What should federal health and benefits agencies do in this moment?  In this two-part series I’ll share Accenture’s perspective on the current moment, how we got here, the value of using a proven commercial best practice and how agentic AI is the game changer federal agencies need.

What would an additional $50 billion in the federal healthcare budget accomplish?

That’s the potential windfall if agencies deploy generative artificial intelligence (GenAI) and agentic AI (automated systems) to address the improper payments that currently escape prevention and recovery efforts. Fraud, waste, and abuse (FWA) in the federal healthcare system have proven stubbornly resistant to traditional solutions, with recovery rates estimated to be below 4%.*

The FWA challenge has taken on new urgency in 2025 as the White House presses for across-the-board budget cuts. Many agencies have invested in automation projects since 2020, when Congress passed the Payment Integrity Information Act to mandate tougher monitoring practices for FWA. Still, implementation has been sluggish, and results have not met expectations. Government-wide, agencies are estimated to still lose more than $168 billion each year to improper payments. A GAO report** cites that healthcare agencies are responsible for two-thirds of that loss, or about $108 billion annually.

In industry, financial services institutions are already improving their data and AI capabilities to combat fraud, waste, and abuse (FWA). With similar investments, government programs could reduce FWA by 30 to 50%—that equates to upwards of $50 billion that could be saved across healthcare and other public health services. These savings would come from reducing incorrect coding, duplicate billing, medically unnecessary services, and enhancing eligibility adherence and identification of fake claims including those with false identities.

So what would a $50B windfall mean?  In effect, federal leaders could self-fund data modernization and AI implementation by drastically reducing improper payments.

 How did we get here?

As the saying often goes…slowly and then all of a sudden.

Several scenarios led to this moment, principally the following four, and for each there is a compelling path forward to solve.      

1.        System fragmentation

The healthcare claims processing landscape is scattered across numerous adjudication organizations nationwide—with over eight regions and 16 different contracts managing the process. This system fragmentation leads to data silos and inconsistent methods for policy adjudication.  

The way forward: Adopt a shared cloud- based data platform that is engineered to deploy agentic workflows, inclusive of large language models and graph networks tailored to improper payment business and operational workflows, to read data, and make inferences from diverse beneficiary, provider, and population data.

2.        Manual clinical validation

Of the roughly 5 to 6 billion claims processed by U.S. federal health agencies each year, a limited fraction is verified with clinical data through an AI or machine learning system. Without automated clinical validation in place, in many cases examiners must manually request each medical record they want to review from healthcare providers, adding drag to claims processes and likely overlooking suspicious correlations between the claim and clinical condition.

The way forward: Modernize adjudication workflows with specialized AI agents, tools, and data feeds to reduce the cost, risk, and tedium of fraud detection and prevention. Leverage health information networks to capture medical records for AI-based review of health claim alignments. Advance policies and data sharing agreements to enable efficient use of clinical records in payment determinations, such as extending TEFCA to require the return of clinical data for payment use cases.

3.        Outdated infrastructure

Many federal health payment processors are still operating on legacy systems built before the recent advances in cloud services, data analytics and AI components, and LLMs. By using outdated systems they are not elegantly able to make the best use of modern technologies, like AI. 

The way forward: Invest in a cloud-based infrastructure with an extensible architecture that supports scaled deployment of AI. Fit- for-purpose data platforms like Databricks, Palantir, Google Cloud Platform, Snowflake, Azure, Oracle Health, and AWS can accelerate data-driven results and value. These platforms now natively support agents, optimized LLM selection and use, and responsible, secure AI workflows integrated with user interfaces that empower more intelligent payment integrity processes.

4.        Sophisticated fraud schemes

Fraudulent actors are employing emerging technologies like generative AI to create increasingly convincing false claims. They use large language models to generate fake medical bills and identities with unprecedented fidelity. Many detection systems rely on rules-based syntax and weak identity verification methods that cannot handle complex cases and put agencies at a perpetual disadvantage.

The way forward: Develop and deploy advanced AI models that can match the sophistication of fraudulent actors, using complex reasoning and continuous learning to stay ahead. Employing methods such as graph analytics indexed to LLMs tuned to find FWA patterns in big healthcare data surfaces potential FWA. Applying AI agents trained in social network analysis can identify patterns of suspicious referrals and other relationships in healthcare networks predictive of fraud. Employing privacy-preserving record linkage (PPRL) fosters the use of previously restricted identifiable data across federal health datasets in combination without exposing personally identifiable information of patients and beneficiaries, enabling better at-scale use of multi-agency data.

This is an incredibly exciting moment in history to bring a new arsenal of FWA tools to bear that advance the best use of federal dollars for promoting health and wellness rather than enriching bad actors at the taxpayers’ expense. Up next, a roadmap to combating fraud waste and abuse with data and AI at the core.  If you would like a copy of our recently published paper outlining the path to eliminating fraud waste and abuse just reach out, I’m happy to send it to you. 

*M. Bourdon, E. Benoit, J. Sonin (2024). Fraud, Waste, and Abuse in Healthcare. Retrieved from https://www.goinvo.com/vision/fraud-waste-abuse-in-healthcare/ (Accessed June 28, 2025).

**U.S. Government Accountability Office. (2025, March 11). Improper Payments: Information on Agencies' Fiscal Year 2024 Estimates (GAO-25-107753). Retrieved from https://www.gao.gov/products/gao-25-107753