One Way to Save Money, Reduce Fraud and Employ People Faster
New Mexico has a unique program that combines behavioral economics and predictive analytics.
During the height of the Great Recession, when 10 percent of workers were out of a job, unemployment insurance pumped $155 billion into the pockets of laid-off workers. Today, with unemployment at less than 5 percent, the state-administered systems that distribute such benefits receive less attention. Even so, they still pay out hefty sums in benefits -- $32.9 billion in 2016. They also pay out hefty sums improperly.
Unemployment insurance has one of the highest error rates among state benefits programs, worse than Medicaid, the Supplemental Nutrition Assistance Program or Rental Housing Assistance. In fiscal 2015, the program made $3.5 billion in improper payments, an error rate of 10.7 percent, according to the U.S. Department of Labor.
A big reason why unemployment insurance performs so poorly is outdated technology. Thanks to a lack of funding at the state and federal level, fewer than half of the states have modernized their computer systems. They struggle to reduce errors because they can’t perform tasks such as cross-matching the names of claimants with birth and death records to make sure the person is alive and lives in the state where he or she is claiming benefits.
One state that has modernized its unemployment insurance technology is New Mexico, which overhauled computer systems in 2011 and now cross-matches benefit claimants. As a result, New Mexico reduced unemployment insurance fraud, often in the form of identity theft or fictitious employers, by 60 percent, saving $10 million between 2012 and 2013.
But when the savings started to level off, the state realized it had to do something different if it wanted to get at the biggest source of improper payments. “The reality is that 95 percent of improper payments have to do with reporting errors,” says Joy Forehand, deputy cabinet secretary in the state Department of Workforce Solutions. “It’s not just fraud; it’s incorrect information from employers and individuals.”
To reduce improper payments, the agency assembled a team of lawyers, communications specialists, economists and technology experts to create a unique program that uncovers trends and patterns in applicant behavior to predict when an error might occur. The program then “nudges” the person at the right point in the process to help solicit an accurate response and avoid making a reporting error. Forehand describes the technique as a combination of predictive analytics and behavioral economics. The secret sauce includes carefully written and well-tested pop-up messages that appear during the online application process and encourage people to answer in a way that doesn’t lead to incorrect responses.
One problem the state kept encountering, for example, was people misreporting their earnings. So when applicants got to the question about income, a pop-up appeared and stated: “9 out of 10 people in your county report their earnings accurately.” Results show that because of the pop-up a quarter of claimants were more likely to report their actual income, compared to little change in behavior when the pop-up simply displayed the law and penalties for breaking it.
“This combination of nudge messaging and analytics allows the state to save money by stopping overpayments before they occur,” says Jennifer Thornton, a data manager at the Pew Charitable Trusts. Thornton studied what New Mexico has done and says it’s one of the few examples in the country where behavioral analytics has been used as a strategy to reduce error rates in a benefits program.
So far, New Mexico has been using its analytics system for over a year and has seen drops in reporting errors and fraud that are expected to generate an additional $1.9 million in savings annually. The same software and nudging techniques are also helping steer the unemployed back to work at faster rates. The agency is now turning its analytics efforts on employers who misclassify individuals as independent contractors rather than as employees.
Other states are beginning to notice. At least 20 are testing or using some kind of predictive analytics in their unemployment insurance program, according to the National Association of State Workforce Agencies. New Mexico, though, is the only state to currently apply behavioral analytics to the program.