Iowa Workforce Development (IWD) is utilizing sophisticated big data analytics in the cloud to target fraudulent unemployment compensation claims. The new initiative, which began in March, uses publicly available data sets in conjunction with IWD data to generate potential fraud leads.
Gary Bateman, chief information officer for IWD, said the agency implemented the solution after recognizing the rising potential for fraud given increasingly sophisticated computer systems.
“Advanced technology had made it easier and more efficient for people to conduct unemployment insurance (UI) and other types of fraud,” said Bateman. “So we began looking for new tools to help us mitigate that threat and protect taxpayer dollars.”
Bateman said IWD considered building its own solution or purchasing an existing analysis program, but determined both routes would be expensive and time consuming. It decided instead to issue an RFP.
“We didn’t specify a solution in the RFP, only that we wanted something that could help us analyze data and detect fraud,” explained Bateman.
IWD ultimately chose Folsom, Calif.-based Pondera Solutions Fraud Detection as a Service (FDaaS), a Google-powered, cloud-based analytics solution that analyzes claims and program participants, quickly sifting through massive data sets to identify and alert agencies to problems that require follow up investigation.
Managing UI claims is a major public service administered by the state. Last year, Iowa processed 190,000 claims, paying out approximately $432 million in benefits in 2012 and 2013.
Bateman said the fact that FDaaS is cloud-based was a huge plus for IWD.
“Because it’s a service, we’re not responsible for the hardware or the software – we simply provide the data to Pondera and they analyze it and send us back results in a dashboard fashion,” he said. “It’s easy to use but didn’t require significant capital outlay to get it started or for maintenance.”
FDaaS analyzes incoming claims using Google’s predictive modeling, data mining and matching, geospatial, and search engine technologies to investigate and help prevent potential fraud before it occurs. And because FDaaS uses machine learning and geospatial technology, it continues to “learn” over time, uncovering new and emerging methods of fraud.
Pondera CEO Jon Coss said FDaaS was built by investigators for investigators, and explained that the company was launched in late 2011 with a singular focus: to leverage new technologies such as Google’s Web analytics and processing power to combat fraud, waste and abuse in large public sector programs.
“Fraud detection systems for government have been very difficult to use in the past, but when cloud and Google came along, the ability to throw massive resources at this problem, in addition to machine learning and web-based analytics, seemed to hold great potential for combating the issue.”
Coss said the process begins by taking historical data on businesses, claimants, and actual payments or transactions and then configuring Pondera’s existing system for the specific client.
“That means building models, adjusting flag thresholds and alerts, and utilizing geospatial maps,” he said. “Using historical data allows us to configure our systems for each individual client and mock live runs, so when we go live with real-time data, we’ve already tuned our models. That makes it much more accurate right out of the box.”
Coss said they also include third-party data in their analysis to validate the information businesses and claimants provide.
Working with IWD, Pondera collects data on a weekly basis. Pondera then analyzes the data overnight and any anomalies or emerging clusters or patterns are flagged for further investigation. IWD receives a dashboard of those flagged files, and its then up to them to further investigate the issue. Anything they decide to take action on must first be independently verified, said Bateman, so the data Pondera provides is simply an “indication of potential fraud risk.”
“We take those hits and we have fraud investigators go out and verify a case independently,” he explained. “Based on the results of the investigation, we then take the appropriate action.”
Bateman said he was initially surprised at the number of hits they received.
“When we first began using the system, our parameters were too broad,” he said. “We got so many hits that it was impossible to look at it all. We had to start tweaking it to be more precise. But even with narrower parameters we still have a lot of hits, so now we rank those and go after the most egregious ones first.”
Today, IWD continues to receive more hits than their current investigators can examine, so they have begun adding investigators to their staff. They have also reconfigured their system a bit so that more of the fraud-indicating criteria are examined and flagged as claims are filed rather than further along in the process. In that manner, they further decrease their chances of potentially paying a fraudulent claim and then attempting to reclaim the funds – a costly and time-consuming endeavor. Approximately two percent of claims are either fraudulent or wasteful, according to IWD, although that number is expected to climb with the use of Pondera. The annual fee for using Pondera has not been determined yet, according to IWD.
Coss said FDaaS is currently being used in other states to detect Medicaid fraud, and there are implications for other areas, including SNAP, public housing and tax.
“For Medicaid, we use Google geospatial technology to visually display potential anomalies, patterns, clusters, etc.,” said Coss. “Things like out-of-state claimants, people traveling long distances to see a provider or to fill a prescription, suspicious procedures performed given a particular diagnosis, unusual billing activities, etc. Those instances will generate an alert that a case may need to be watched or investigated.”
Bateman said the fact that FDaaS uses push analytics has been enormously helpful for IWD.
“We didn’t have learn any advanced analytics tools, and then go in and search for problems ourselves,” he said. “We just supply our data and the system does all the work and pushes the results to us.”
Coss said they also tune the system every quarter so that it’s constantly learning and improving.