As fraud in government programs becomes a central political issue, states are turning to AI and data analytics for solutions; however, technology alone won't work without addressing fundamental coordination and structural challenges
Key insights:
-
-
-
Fraud prevention presents systems challenges — In the current advanced tech-enabled environment, AI-driven tools in the public sphere need human coordination and oversight to be employed properly.
-
The intangible cost of fraud is public trust — When public programs that are designed to support vulnerable populations fall victim to exploitation, confidence in the ability of government-administered programs falters.
-
The full scope of fraud is unknown — Many improper payments from government programs aren’t criminal. It’s that we need better data, clearer definitions, and a stronger understanding around what fraud truly is.
-
-
As public programs and the scale of fraud become the subject of national and state political discourse, more state governments and public sector agencies are evaluating the potential of data analytics and AI tools to curb fraud. As these institutions are realizing, effective fraud prevention is easier said than done, and investment in new detection tools will fail to keep pace with fraud if they don’t address structural, cultural, and coordination challenges.
Early fraud detection
Early detection of fraudulent activity, aided by AI tools and systems, may still be the most effective way to deter future problems. For example, California Community Colleges, the largest higher education system in the United States, currently serves more than two million students and has deployed an AI tool in two-thirds of its institutions in order to detect fake students. With open access for enrollment, the system is a magnet for fake students who apply, enroll, and fill seats in virtual classes that real students should be occupying, while fraudulently collecting federal and state student aid.
In 2024, it was estimated that nearly one-third (31%) of financial aid applicants at California Community Colleges were fraudulent, resulting in approximately $13 million in state and federal aid dollars being disbursed to fake students.
California Community Colleges employed a three-phased approach seeks to catch fraudsters who may slip through at the time of application, when they register for courses, and when they apply for financial aid. The system engaged in cross-agency collaboration with the California Department of Motor Vehicles and deployed a mobile ID system to authenticate student identity. Further, its data analytics looked at factors such as students’ IP addresses, time zones, age, and contact information to flag those patterns that could indicate a fake applicant. An AI tool analyzed course registration patterns, as well, identifying whether applicants have illogical or unusual patterns in the courses they are taking.
Then, system educators integrated human icebreaker activities and assignments into their virtual courses early on, requiring students to submit an introductory video, for example. This allows educators to cut non-participating students (presumed to be fake) before financial aid is disbursed.
These early detection tools, paired with human judgment, showed how a proactive approach can stop fraud before funds are lost.
Gaming government systems
While fake students illustrate small-scale exploitation in California, provider fraud is where large dollar amounts and case complexity arise. When fraudsters illegally obtain Medicaid funds for services they never rendered, for example, individuals in need of services suffer.
Minnesota’s now-shuttered Housing Stabilization Services program intended to help move individuals who were experiencing housing insecurity into transitional and then permanent housing solutions. According to a report from the U.S. Department of Justice, the well-intentioned program enriched sophisticated fraudsters, who formed business entities and falsified employee hours, reimbursement claims, and patient identities — even going so far as to manufacture false case notes as a precaution against their records ever being audited. Not surprisingly, illegally gained reimbursements were used to fund high living expenses, luxury shopping, and cars.
Similar fraudulent providers have been charged by the U.S. Attorney’s Office for the Northern District of Texas as part of the National Health Care Fraud Takedown. Four individuals fraudulently billed around $20 million to federally funded programs and other insurers.
In another case that showed that fraudsters sometimes can come from inside the house, a group of Indiana Medicaid officials were ruled against by a federal judge in a whistleblower lawsuit alleging that four insurers and six health systems routinely, improperly billed the state’s Medicaid program. Part of the reason for the judgment was because the disputed claims were still paid by the Indiana Medicaid program, despite it being aware of alleged issues. In this case, oversight gaps and a consistent pattern of not flagging improper payments revealed a structural weakness within the state office.
Different approaches for data analytics
Some agencies are employing different methods to leverage advanced tech to help in the fight against fraud. For example, the Louisiana Department of Health is using AI to scrutinize Medicaid recipients and their eligibility. A proprietary AI and data analysis tool developed at the University of Louisiana at Lafayette will allow the Louisiana Department of Health to share data with the state’s Office of Motor Vehicles. By analyzing whether individuals have duplicate licenses in other states, their eligibility to receive benefits in Louisiana may be rescinded.
Focusing on a different tack and target, the Center for Medicaid and Medicare will deploy a six-year pilot program of the Wasteful and Inappropriate Service Reduction Model (WISeR) across six states: New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington. The pilot program will specifically target low-value services with little to no clinical, evidence-based benefits and will expedite review of those services that are at a higher risk for provider fraud, waste, or abuse. This heightened scrutiny of providers seeking Medicaid reimbursement is in alignment with recommendations from the U.S. Government Accountability Office around program integrity.
These varying approaches raise a difficult question: Is it better to risk inefficiency by targeting providers, or it better to risk inequity by targeting recipients?
Understanding the measures of fraud, waste, and abuse
The total cost of fraud is difficult to calculate, as there are countless incidents of fraudulent reimbursement requests, overbilling, or unnecessary medical treatment that cannot be counted. Two data measures that we have to understand to truly gauge the efficacy of public health systems’ financial health are the payment error rate and the dollars recovered through fraud controls.
Improper payments are those payments that fail to meet statutory, regulatory, or administrative requirements. They may be for non-eligible services, be inappropriately or inaccurately coded, or may exceed program maximum amounts — but their common denominator is that they represent funds that were misspent or out of step with fund guidelines.
Improper payments are calculated and reported to Congress annually across all federal healthcare programs. The dollar recovery rate calculates the amount of inappropriate reimbursements recovered from fraudulent actors each year, usually through the pursuit of civil or criminal damages. However, it’s important to remember that not all improper payments are lost to fraud. For example, improper payments in 2022 within the Medicaid program were most often tied to missing appropriate documentation for individuals receiving care.
Understanding these definitions determines how we measure success, design large government systems, and allocate enforcement dollars across states. Such preventative measures, especially now aided by AI and other advanced tech, will help the next generation of fraud detection professionals who will come to rely on the tools and platforms that we design now.
And as more state governments and public sector agencies seek to leverage AI tools and platforms, they would be wise to focus on efforts that collect and analyze real-time participant data and incorporate ethical AI oversight, while balancing an investment in prevention as well as prosecution.
You can learn more about the challenges that government agencies face today here
