This story was originally published by Data-Smart City Solutions.
Governments cannot afford to treat fraud as an inevitable cost. Tax evasion alone costs the United States Federal Government more than $400 billion each year, to say nothing of the numerous other types of fraud that happen at the federal, state, and local levels. Lower income for government means fewer resources for vulnerable populations and higher taxes on those who have followed the rules all along.
Realizing the immense price of letting fraud run rampant, many governments have sought to leverage the power of data and analytics to identify, predict, and then prevent instances of fraud. However, governments must remain careful to protect the benefits of eligible residents while identifying fraudsters. If they do so, governments can save public funds that can be distributed to those that need benefits the most.
Thankfully, most people don’t commit fraud. This means that to spot dishonest behavior, governments can analyze claimant data to look for patterns of behavior that diverge from the norm.
Under the direction of Finance Commissioner David Frankel, between 2009 and 2013, the New York City Department of Finance improved efficiency in finding fraudulent tax returns by 40 percent. Using data from city, state and federal tax records, along with other data such as business licenses, the department created a model that can identify patterns in New York City businesses. By focusing on businesses that were outliers in taxes paid, the city is able to better identify businesses that fail to pay taxes or pay too little.
As a result, when the city sends auditors to assess a business, they are more likely to find fraud. Between 2009 and 2013, the department reduced the portion of audit cases closing without change from 37 percent to 22 percent without deploying one more auditor.
The State of Maryland has had similar success in identifying tax fraud with analytics. Using historical records of individuals’ tax returns, the state’s Bureau of Revenue Estimates created a model that can spot anomalies and assign a risk score to every return. If the risk score is high enough, the bureau flags the return for review. In order to uncover larger tax fraud conspiracies, the system also analyzes patterns of anomaly across returns, looking for suspicious activity coming from a particular address or preparer.
Thanks to this program, Maryland has made its fraud identification efforts 12 times as effective: previous years averaged 110,000 audits to find fraud in five to 10 percent of cases, while now half as many audits find fraud 60 percent of the time. Maryland now recovers $40 million yearly as opposed to an average of $10-20 million before implementing the analytic approach. And, the state has been able to use its auditing resources more judiciously, reducing pressure on public employees.
Governments can use this process of identifying outliers to spot nearly any type of fraud. They must merely determine what normal patterns of behavior look like in a certain area, then flag activities that diverge from that norm. For example, in order to detect fraud in the Supplemental Nutrition Assistance Program (SNAP), the Department of Agriculture’s Food and Nutrition Services targets residents that travel unusually far to use their SNAP benefits, indicating that they may be seeking out fraudulent retailers that will exchange cash for SNAP benefits. The same applies to unusual behavior in affordable housing, unemployment insurance, healthcare, and any number of other areas.
Identifying fraud after it has already happened allows governments to recoup lost funds, but what if they could stop fraud from happening in the first place? The State of New Mexico has sought to do just that, pairing analytics with insights from behavioral science to discourage residents from committing unemployment fraud, which costs state governments more than $4 billion yearly.
Like other efforts, New Mexico’s initiative starts with identifying residents most likely to commit fraud. In addition to examining traditional variables like past wage reporting and overpayments, the state’s Department of Workforce Solutions (DWS) and Deloitte partnered to create a tool that can analyze resident clickstream data as they file claims online. According to Joy Forehand, Deputy Cabinet Secretary at the New Mexico Department of Workforce Solutions, “the model looks at behavior within the system like clicking back or stopping halfway through an application and picking it up a few days later as potential indicators of false reporting.”
Yet unlike other programs, instead of targeting these residents for audits, the state seeks to discourage residents from committing fraud in the first place by deploying three nudges—strategies that use insights from behavioral science to influence behavior. By introducing these interventions during the online insurance application process, the state hopes to reduce false reporting on claims.
The first intervention aims to improve honest reporting of residents’ reasons for unemployment. Because only those laid off are eligible for unemployment, residents have an incentive to bend the truth about why they are unemployed. While New Mexico already had a system in place to verify employment separation reasons—sending a letter to former employers asking them to confirm eligibility—often employers did not respond right away or had some disagreement with the claimant. Therefore, to streamline the process and encourage residents to provide the honest answer up front, New Mexico instituted a procedural change to show claimants a copy of the verification letter sent to their employers, making the process more transparent and real to applicants. Since DWS started displaying this letter, applicants are 10 percent less likely to have been paid benefits and 15 percent more likely to have self-reported something that delays payment until additional investigations are complete.
The second strategy targets residents during the weekly check-ins they must make to confirm eligibility. The system asks residents “Did you work this week,” and if they have, the amount they were paid is deducted from their unemployment payments. To improve honest reporting, DWS now requires residents to certify their employment information with their initials and displays a pop-up message to claimants that analytics have determined are at high risk of fraud. The most effective message reads, “99 out of 100 people in <your county> report their earnings accurately. If you worked last week, please ensure you report these earnings.” Thanks to this message, claimants are almost twice as likely as the control group to report earnings.
The third intervention targets residents in the process of reporting their work search activities, of which claimants are required to complete two each week to receive benefits. DWS implemented a commitment mechanism, asking claimants to specifically outline what actions they will take each week as well as showing them what they had planned to do the previous week and how they performed on those goals. According to DWS’s internal research, claimants who plan certain work search activities are more likely to find a job and find jobs more quickly.
Taken together, these initiatives have led to a 40 percent reduction in unemployment fraud in the state without the need for additional staff or resources, said Forehand. These savings not only free up more funds for eligible residents, but also ultimately benefit ineligible residents who may have filed dishonestly without the nudges. “Overpayment can lead to withholding of future benefits and a number of other problems for claimants,” said Forehand. “So this program helps them, too.”
Perhaps most importantly, these insights are useful not only for combating unemployment insurance fraud, but also for preventing fraud in a number of service areas at the local level. When applying for affordable housing, requesting food assistance, or filing taxes, residents must complete applications attesting to their income, household size, employment history, and a variety of other pieces of information. By repurposing interventions like displaying letters sent to former employers, requiring residents to initial attestations, and creating pop-ups that use social norms, cities too can improve compliance.
Lawmakers must be careful to structure their anti-fraud efforts in a way that benefits, rather than harms, those residents eligible for a service.
In 2012, the State of Illinois decided to address Medicaid fraud, which when combined with Medicare fraud costs federal and state governments just shy of $100 billion annually. The state knew that there were likely many people enrolled that were not eligible, but had neither the time nor the resources to address the problem previously. Illinois hired a private contractor to identify residents who might not be eligible and make recommendations for whose benefits to cancel.
Within a year, the state had canceled benefits for nearly 150,000 people whose eligibility could not be verified and saved an estimated $70 million. However, a short time later, the state encountered a strange phenomenon: 20 percent of those who had been kicked off Medicaid re-enrolled.
How did 20 percent of ineligible residents suddenly become eligible for Medicaid? Well, they were in fact never ineligible at all. Rather, the contractor Illinois hired had falsely marked a large number of eligible residents as ineligible. The contractor had given residents as little as 10 days to respond when asked for more information to prove their eligibility, and many simply did not respond in time.
This difficulty is not unique to Illinois nor Medicaid fraud, but rather reveals a problem inherent in the model governments often use to take on fraud—hiring a private contractor to develop and implement fraud detection models. Usually, governments will only continue contracts for fraud identification when the cost savings exceed the amount paid to the contractor. While seemingly logical, this model creates an incentive for contractors to target as many ineligible residents as possible, which results in practices like short response windows that strip eligible residents of their services.
According to Roy Mitchell, executive director of the Mississippi Health Advocacy Program, governments can avoid depriving eligible residents of critical benefits by looking for anomalies in the behavior of health care providers like doctors and pharmacists, such as an unusually high number of claims, especially for expensive equipment. By targeting fraudulent providers, governments can identify larger chunks of fraud at a time without jeopardizing legitimate benefits.
Governments can also make their anti-fraud efforts more effective by building the in-house analytics capacity to predict and prevent fraud. When governments create their own fraud analytics programs, there is less incentive to remove as many residents as possible from a benefit, as third parties are no longer fighting to maintain their government contracts by proving their cost-effectiveness. Granted, building analytics capacity within government is also expensive, but often involves a large fixed cost and low marginal costs, creating an economy of scale through which analytics becomes more cost-effective each year. As a result, governments have an incentive to continue these programs regardless of the number of claimants targeted.
And creating an in-house fraud analytics program can also be less expensive from the outset. While Illinois is still early in its Medicaid fraud detection efforts, AFSCME Council 31—the union that represents the state’s workers—estimated that having state employees do the work would have saved Illinois an additional $18 million a year.
Local governments should apply these same considerations when tackling fraud in areas like affordable housing or food assistance. The goal is not to kick as many people as possible off benefits, but rather to identify fraudulent claimants so that eligible ones can receive the benefits they need. And, as many city governments already have analytics programs, it may be even more cost-effective for municipalities to apply their in-house analytics capacity to tackling fraud.
These examples illustrate important lessons for governments interested in using analytics to reduce fraud. For one, while analytics allows governments to identify instances of fraud that cost taxpayers billions, identification is not enough. Governments then need to act on these insights, not only by auditing high-risk claimants, but also putting in place measures to prevent fraud before it happens.
Moreover, analytics programs must be carefully structured to avoid depriving eligible residents of benefits. If governments hire a contractor to kick-start their fraud analytics programs, they must ensure the contractor gives residents ample time and resources to prove their eligibility before taking them off a benefit. In many cases, building analytics capacity within government is the more effective option.
When developed with these considerations in mind, fraud detection and prevention programs can go a long way towards redistributing public funds from fraudsters to the residents that need government services the most.