Strengthening healthcare program integrity with AI
Improper payments cost the Centers for Medicare and Medicaid Services (CMS) an estimated $100 billion in 2023 alone. That number is enormous, but it’s only a small piece of the actual fraud, waste, and abuse happening each year in the American healthcare system.
CMS, FBI, and DOJ routinely identify suspicious billing patterns among healthcare providers. In fiscal year 2023:
- CMS uncovered 1,137 new leads and worked on 1,994 existing investigations.
- FBI opened 658 new healthcare fraud investigations but still had 3,440 cases pending at year’s end.
- DOJ opened more than 770 new civil healthcare fraud investigations but still had 1,147 cases pending at year’s end.
The logjam isn’t identifying potential fraud cases but rather investigating them and taking timely action. The scale of this work needs innovative approaches that can protect federal resources and restore public trust. Using AI can speed up and strengthen investigations so fewer suspected cases of fraud, waste, and abuse go unexamined.
The bottleneck: Volume outpaces capacity
Federal investigators receive far more leads than staff can thoroughly review. Manual analysis, fragmented data systems, and limited technical bandwidth slow prioritization and case development.
As a result, potentially recoverable funds may go unexamined—and systemic vulnerabilities may persist. Agencies need to accelerate the process of prioritizing the cases that are most likely to recover funds.
Key definitions in Medicare and Medicaid program integrity
What is an improper payment?
Improper payments include over- or under-payments made for a Medicare or Medicaid beneficiary. In some cases, improper payments are those that should not have been made at all. These can be the result of mistakes (e.g., using the wrong medical code) or fraud (e.g., billing for services not rendered).
What’s the difference between fraud, waste, and abuse?
CMS generally defines fraud, waste, and abuse as follows:
Fraud
Intentional deception or misrepresentation by a person or entity with the knowledge that it could result in an unauthorized benefit or payment under a health care program. In Medicare and Medicaid, this includes knowingly billing for services not provided, upcoding to increase reimbursement, or falsifying information to get paid.
Waste
Practices that overuse, misuse, or consume resources inefficiently, resulting in unnecessary costs to Medicare or Medicaid programs — without intent to deceive. Waste generally stems from poor administrative or clinical decision-making and does not necessarily violate law.
Abuse
Provider or beneficiary practices that are inconsistent with sound fiscal, business, or medical practices and that lead to unnecessary costs, reimbursement for services that aren’t medically necessary, or services that don’t meet professionally recognized standards. Abuse does not require intentional deception, but it results in improper costs or payments.
How AI can enable proactive fraud, waste, and abuse investigations
- Connect different data sets to uncover unusual or suspicious behavior.
- Process more potential leads faster so fewer cases are overlooked
- Give medical reviewers the context they need to make connections without heavy manual work or long waits for technical support
- Identify system-wide weaknesses, such as repeated upcoding in certain services or suspicious billing patterns in certain locations
- Develop prevention strategies, including smarter pre-payment checks and data-driven billing limits
- Automate routine tasks and remove delays by using input from users
Realizing these gains at scale, however, depends not just on the technology itself but on how agencies govern data, manage risk, and embed explainability and security into everyday investigative workflows.
Considerations for leveraging AI for fraud, waste, and abuse investigations
Safeguard data quality
Before using AI tools in fraud, waste, and abuse investigations, investigators must ensure their data is high quality and suitable for the task. They can do this by using modern data integration methods such as Zero-ETL and low-code or no-code pipelines, automated and rule-based workflows, and AI-supported integration tools.
Address ethical and regulatory concerns
To prevent the misuse of AI tools, investigators must build strong governance into the data lifecycle from the beginning, not as an afterthought. Governance should be applied consistently to data, models, prompts, and results. This includes setting safety standards, ensuring ethical use, protecting privacy, detecting misuse, and monitoring systems over time. These safeguards should be part of everyday staff workflows, not treated as optional add-ons.
Incorporate data lineage and explainability
Investigators must understand why AI tools make certain decisions to successfully pursue fraud, waste, and abuse cases. Data lineage maps show how information moves through AI systems, while explainability helps clarify why the system reached specific conclusions. Together, these tools give investigators confidence to trust and act on the system’s findings.
Keep security at the forefront
All AI use must follow CMS Data Use Agreements, HIPAA rules, and applicable Federal Information Processing Standards. Data should be encrypted both where it is stored and while it is being transferred. Role-based access and authentication should be used, and all code and data processing should take place within secure systems.
What AI-enabled fraud investigations look like in practice
At ICF, we've piloted an AI-enabled investigation tool designed to accelerate case review while maintaining transparency, auditability, and human oversight. The approach helps investigators rapidly explore complex Medicare fee-for-service claims patterns, prioritize high-risk cases, and complete administrative tasks that traditionally consume significant analyst time.
In pilot testing, the capability reduced investigation turnaround time by more than 80%, completing key administrative and analytic steps at least six times faster than existing processes. By compressing review cycles, investigators were able to spend more time on judgment, validation, and enforcement—rather than data preparation and manual analysis.
A critical design principle of the pilot was explainability. The system surfaces clear, traceable indicators for why claims or providers are flagged, supporting defensible decision making and alignment with federal AI governance expectations. Investigators can interact with the system using plain language questions, lowering technical barriers and expanding access to advanced analytics across program integrity teams.
In 2023, CMS reported $14.9 billion in fraud, waste, and abuse recoveries. Improving investigative speed and throughput—without increasing staffing—can materially expand agencies’ ability to identify, prioritize, and act on potential fraud across Medicare and other federal health programs.
Related pages