Localities and States Are Turning to Data Analytics to Catch Fraudsters—and It's Working

When a health care provider submitted a request for $8,002,021 to New York’s Medicaid program in October, it raised eyebrows among state auditors, who, just a few years ago had started scouring government databases for suspicious public assistance transactions.

Flagged as an abnormally large invoice, the state denied the payment and investigated the claim. It turned out that the vendor had inadvertently made a typo that combined the amount of the payment—$800—with the year—2021.

“We are stopping those [kinds of] payments before they occur,” Tina Kim, deputy comptroller for the Division of State Government Accountability, said. “We prevent a lot of fraud from actually occurring. … It’s important to a state that you prevent a fraud from happening by being always vigilant.”

For New York and for an increasing number of states, that vigilance has come in the form of data analytics, the process of using computers to collect huge stores of data and flag abnormalities, and then assigning humans to identify which transactions are legitimate and which are suspicious. Often, an analysis of the data can pinpoint waste and inefficiencies that lead to policy changes.

Over the past couple of decades, state and local governments have collected data to keep track of everything from tax returns to Medicaid, public assistance and unemployment insurance payments, to pension checks, to employee hours worked. As more governments have converted to digital data collection systems, the need for and number of analysts to interpret that data has grown.

“We’ve been doing data analytics for decades, even before computers,” David Eaves, a public policy lecturer at the Harvard Kennedy School, said. “Governments could use census data to figure out where to position schools, where to position certain services. What has changed is the quantity of data that we have and the computational power and the sophistication of tools at our disposal.”

That wealth of data led auditors in Oregon, originally assigned to parse the long list of those on public assistance for deceased recipients, to dive deeper into potentially improper payments. After finding more than 1,000 deceased recipients who had received a collective $6.8 million in payments, the auditors uncovered 384 inmates on public assistance and a high-dollar lottery winner who continued to collect public assistance payments for 16 months after hitting the jackpot.

Next, they used the data to identify five stores that apparently were paying recipients of $100 electronic benefits cards—granted to low-income residents to buy food and household necessities—$50 per card, and then submitting $100 claims to the state.

The auditors’ strategy: Find stores whose average sales amounts were far higher than other merchants; identify customers who drove many miles from home to shop at that store and spent large sums during each visit; and, eventually, view the stores’ security footage, which revealed that many customers who reportedly spent $100-plus per transaction were leaving empty handed.

“We knew there was something going on when we were looking at the data,” Kathy Davis, a senior auditor with the Oregon Secretary of State Audits Division, said of one tiny, rural market. Audit manager Ian Green added, “Over time, the fraud grew. … This is one of the lessons we’ve learned: If they get away with it for a period of time, the word spreads in the community … that they could sell their benefits at that market.”

The investigation by local prosecutors and the U.S. Office of Inspector General that followed resulted in the prosecution of five merchants for fraud; the banning of 59 Supplemental Nutrition Assistance Program recipients for life and 40 for one year; more than $525,000 in court-ordered restitution; and more than $1.7 million in savings from potential future fraud that the investigation averted.

“What the value of data analytics is, is we’ve got this haystack of data and we’re looking for these little needles in there,” Green said. “That’s what analytics lets us do when we work with law enforcement to prosecute the fraudsters.”

States and Localities ‘Primary Targets of Fraudsters’

Shaun Barry, director of fraud and security intelligence for government and health care for data analytics vendor SAS, said city, county and state governments “are now the primary targets of fraudsters.”

Because different levels of some multitiered government assistance programs are administered by different agencies, fraudsters are finding ways to sneak past the gatekeepers, Barry said. Couple that with the addition of benefits like telemedicine and changes in unemployment insurance rules during the pandemic, and “there are murky areas where it’s not easy for governments to communicate well because there’s no one person responsible for monitoring” all of the spending.

Plus, he said, governments have been eager to quickly respond to people with a pandemic-related need for government assistance, resulting in less diligent oversight that has invited organized crime to try to engage in government fraud en masse as it never has before.

“The fraudsters are continuing to escalate in an arms race,” Barry said. “They’re smarter, much more sophisticated. We’re currently in a period … where the fraudsters have the advantage. They act faster in stealing the money than the government can stop them.”

His solution: Collecting more data; making online identity verification more personal; and borrowing data analytics practices from banks and credit card companies, which had a head start on governments when it comes to combatting the identity theft that Barry calls “the root of all of these fraud schemes.”

Monitoring Employees and Residents

Still, even government insiders sometimes participate in fraud, he said. County governments have turned to data analytics to monitor fraud and abuse by employees and residents.

Rita Reynolds, chief information officer for the National Association of Counties, said governments can use data to spot abnormalities in the frequency with which employees access confidential county records. 

“If I’m a county staff person and I don’t usually log on after hours and all of a sudden there’s a log-on from the individual at night, that raises a red flag,” Reynolds said. “Why is the person doing that?”

Increasingly, she said, employees are accessing government data with the intention to sell it. 

The same is true of hackers posing online as legitimate contractors to request payments. “Data analytics can identify where the email came from,” Reynolds said. “If it’s coming from Africa, then you know that the contractor isn’t real.”

Counties also are using data analytics to keep track of incoming payments for fishing and hunting licenses.

“If someone comes in and pays with cash and you issue the license, the [employee] can keep the money,” she said. “Those are potential areas where you would want to watch for abuse and fraud.”

In Kansas, state auditors also are using data to keep an eye on government employees.

Matt Etzel, principal auditor for the Kansas Legislative Division of Post Audit, said the three auditors on his team designed a “home-grown solution” to spot potential fraud in employees’ use of state-issued credit cards, which workers swipe to pay for gas, travel expenses and supplies.

Using a scoring system that ranks transactions and programs for their potential for fraud, auditors flagged the Department of Children and Families’ foster care program, whose employees used their state credit cards to purchase a number of Walmart gift cards.

It turned out that the employees gave the cards to foster families to tide them over while the state replaced the computer system that typically generated checks for those payments.

“We didn’t find any cases of fraud, but we did identify risk,” Etzel said, noting his team recommended a change in policy to outline how employees could make more secure payments if a similar situation arises in the future.

In New York, on the other hand, the state Comptroller’s Office used similar analytics to identify $100 million in improper payments to special education vendors who billed the state for personal expenses like furniture for their homes and gas for their personal cars. The investigation led to 13 arrests, nine convictions and $21 million in court-ordered restitution.

Data collection and analysis allowed auditors to spy suspicious vendors and abnormal payment requests, Kim said.

Other Uses for Electronic Data

But governments are using data analytics for more than fraud detection. A New York audit of automated highway tolls, for example, revealed more than $56 million in “leakage” over four years—payments lost because cameras were unable to read the license plates of many cars using E-ZPass. That finding led the comptroller to recommend changes to the state Department of Motor Vehicles. 

In Syracuse, N.Y., the Office of Accountability, Performance and Innovation used data to determine the most efficient way to divide work crews when they mow the grass in city parks.

“Maybe if you did it differently, you could have two fewer people,” Eaves suggested. “That could save the city a lot of money.”

While the use of electronic data analytics can save or recover millions of taxpayer dollars, it’s not cheap to get started.

Barry estimated that a basic fraud detection system could cost a few hundred thousand dollars, while a more advanced setup might top $1 million.

A low-tech, traditional file cabinets-and-paper system, of course, costs much less, Barry conceded, but he added, “You’re not going to catch the really good” fraudsters.