More than 600,000 Louisiana residents receive food stamps, costing the federal government $600 million annually, and some believe at least $24 million of that is collected and used fraudulently.
To reduce these losses, the state implemented an investigative information system that tracks false claims and helps lead to the prosecution of offenders.
And Louisiana is not alone -- other states, such as Georgia and Texas, have turned to software systems to help track and audit potential fraud in areas such as welfare, food stamps, child care and corporate tax noncompliance to recover some of the money lost.
Looking to take a bite out of its fraudulently collected $24 million, the Louisiana Department of Social Services (DSS) is in the initial phases of implementing a system that mines data and then graphs it with GIS to show investigators where to look for fraud. First, business intelligence software -- Information Builders' WebFOCUS system -- detects patterns inconsistent with normal commerce. Then ESRI GIS technology develops maps that show trends. Investigators say at that point it's pretty easy to detect unusual patterns.
"We're looking for the obvious to stick out," said Raymond Pease, assistant director of the Fraud and Recovery Section of the Louisiana DSS Office of Family Support. "And it usually does."
Through years of investigating fraud, DSS staff knows what to look for. How they look for it, however, has changed drastically with the new system, which is programmed to report suspicious data.
The system calculates, for instance, how far a benefit recipient travels to use food stamp benefits, which are allotted via electronic benefits transfer (EBT), a method of delivering government benefits to recipients. Louisiana recipients use the Louisiana Purchase card, which lets them access benefits at point-of-sale machines much like a credit card.
"We're going to look at clusters of stores within a certain area that a person could go to, their favorite store and why that may be inconsistent with a store even closer to them, like a large chain store, which would have cheaper prices," Pease said.
Such activity may signify that a benefit recipient is selling his or her benefit card for cash to a merchant at less than face value, instead of using it to buy food. The merchant then redeems the card for full value with the state.
Transactions that occurred before or after normal business hours may be another sign of illegal activity, which Louisiana's system can also track. Previous to the system, investigators worked from tips that certain stores were illegally trafficking EBTs.
Now investigators detect those stores themselves by looking at data displayed as clusters on maps, which are shown on a "digital dashboard" that can be accessed on a laptop or PC. When the software is programmed to look for something specific, it can flag areas where suspicious patterns are detected. Users also can feed data to the dashboard in an attempt to find trends that were previously undetected.
"We can do some reports that kind of let [evidence] float to the top by defining certain parameters," Pease said. "Or we can be more specific."
The browser-based system calculates data in seconds, said Sherwood Lemoine, an internal management consultant in the Louisiana DSS IT department. "We have some very sophisticated signatures of fraud, and that's based on some very complicated measures we built that enable you to look at the data."
Pease said his department created a measure that signifies when someone spends their entire monthly benefit amount in one transaction. "That's a fairly complicated thing to write, but with the system, as fast as you can talk, you can have a red light popping up on a PC that tells you you've got a problem," he said.
When a user clicks on the visual dashboard, the system displays the information according to the data model requested, such as every person who used all their EBT money at once, at a certain location or both. A benefit recipient spending all of his or her money on a single purchase is a red flag because most recipients try to stretch the money for the month.
Louisiana implemented the system in late 2004, at a cost of about $700,000. The state plans on expanding the application to include use by other agencies for other management tasks, Pease said. "When we started this application, we said it can't just be a fraud application."
The system, for instance, could be used to detect problems with child-care providers by displaying clusters of complaints that point to certain child-care centers.
Georgia recovered more than $6 million in fraudulent food stamp money and $1.5 million in child-care overpayments for fiscal 2005 -- and did so in two days with the help of its new Investigative Services Information System (ISIS) from HCL Technologies.
The state's Office of Investigative Services (OIS), within the Department of Human Resources (DHR), handles approximately 4,000 benefits fraud cases every year. Prior to ISIS, investigating those cases was a tedious, paper-heavy process that could take two or three weeks just to get started.
Although shortening the investigative process, eliminating reams of paper and saving manpower were lauded by Georgia officials, those aren't the only benefits ISIS provides for the state. The system also speeds up the fraud detection process, and it automatically spots possible fraudulent food stamp transactions through unusual spending patterns, such as repeated purchases or the spending of unusual dollar amounts. The Web-based system also detects child-care overpayment to day-care centers and welfare fraud.
Georgia's previous mainframe system was basically a reporting application that collected data. When an investigation request was received, the known data was keyed into the system on a case-by-case basis, and then letters were drafted by a secretary and sent out -- one to the person being investigated and another to the federal government. Then if an investigation was deemed warranted, the secretary drafted another round of letters and sent them out to the accused and to the district attorney. An investigator was then assigned to the matter based on caseload and location.
The new system does all that automatically. It creates letters with the case information and spits them out in minutes, and selects a caseworker based on geography and caseload. Within 15 minutes, the assignment is in the caseworker's e-mail inbox. Prior to ISIS, cases requiring supervisory approval were handled with paper, but ISIS knows which cases to send to a supervisor electronically.
Before ISIS, agencies requested fraud investigations by mailing a paper referral, investigators could only handle about 15 cases each per month, and cases older than a year or two could be lost because they were shipped in boxes to archives in another building.
With the new system, the OIS receives requests online, investigators can handle 20 to 30 cases per month -- almost twice their previous workload -- and it's easy to review cases that are years old.
The system finds key data for an investigator, including benefit payment amount and client demographics, and captures it into one file. Previously, investigators searched the mainframe system separately for different pieces of information, then printed out each search separately on paper.
The new process also is considerably shorter because of the system's investigative nature, which looks for unusual patterns of commerce. "We've got these profiles of particular traffic patterns," said Judy Bullard, assistant director of the OIS. "It can do in two days what it took us three to four weeks to do [before]."
The system also includes an electronic benefit component that helps detect unusual patterns associated with food stamp commerce. "ISIS will automatically identify possible fraudulent transactions based on certain things, such as repeated purchases within a small amount of time," Bullard said. The patterns have been analyzed and previously determined to indicate a high incidence of fraud, she said.
ISIS is still in its infancy -- the first phase was rolled out in September 2004 -- but eventually it will interface with multiple agencies, such as Georgia's Department of Labor, Department of Revenue and Department of Human Resources.
Interfacing with those state agencies could further refine the process, making data that much easier to obtain. When investigating a mother receiving food stamps, for example, an investigator may need to know whether she is married or living with the father, and if so, whether she is employed, since those factors may affect her eligibility. A search of vital statistics would determine whether the woman is married, and a search of the Department of Labor would determine whether the father or husband is employed.
The OIS has invested approximately $750,000 in the system thus far, and hopes to obtain more money to continue adding to it. Like many of these types of applications, it's not an off-the-shelf product, but must be adapted to an agency's or state's processes and procedures.
If the OIS encounters a case of fraud totaling more than $20,000, the agency establishes a claim, investigates and prosecutes. That's not the case in most other states, according to Bullard, who said it's more common for caseworkers to do the legwork, then hand the case over to law enforcement.
Trouble in Texas
Texas recouped $437 million since implementing a system that flags businesses shorting the government on tax payments
The Audit Division of the Texas Comptroller of Public Accounts created a database in 1999 that consolidated information from internal sources, such as tax returns and taxpayer records, and external sources, such as wage records from the Texas Workers Compensation Commission and results from prior audits.
The agency then used analytics software from SPSS Inc. to create an in-depth scoring system that helps target tax audits. The scoring system analyzes the more than 750,000 business tax accounts in the state and returns a score on each. The score indicates the statistical probability of noncompliance and allows the 450 or so auditors to hone in on those businesses with the most potential to yield a claim.
The system compares each business to its peers in terms of tax deductions, and investigators look at the percentage of deductions on total sales. In the restaurant business, for example, all sales are taxable, so it would be unusual for a restaurant to have a high proportion of deductions, according to Daniele Micci-Barreca of Elite Analytics, which consulted with the state and SPSS on the implementation.
"We continue to learn from the past," said Micci-Barreca. "Using data-mining technology, we're able to extrapolate from past audits and learn which type of taxpayer attributes correlates to the outcome of an audit."
The system analyzes each business in five different areas: business information, such as taxpayer code; business type and location; sales tax filings from recent years; other tax filings, primarily franchise tax; reported employee and wage information; and prior audit outcomes.
The system searches for businesses with unusual patterns -- restaurants claiming write-offs when everything they sell is typically is taxable, for example -- and assigns them a high score. Businesses with prior audits that yielded claims also earn high scores.
Micci-Barreca cautioned that the scores are statistical predictors and not always indicators of noncompliance.
"Two people may audit the same taxpayer and come up with two results," he said. "The other thing is, data tells you only part of the picture. The taxpayer data or bookkeeping may not be accurate. You cannot exclusively rely on the system because you're still limited to the data you have."
Texas measured the system's success by examining the correlation between higher scores and audits that yielded claims. About 85 percent of audits performed on taxpayers with high scores resulted in new tax claims. By comparison, audits on taxpayers with average scores yielded claims in about 50 percent of cases.
Micci-Barreca said there is no mandate to audit those with a high score. "It's just an additional tool; it's just more data." Still, results from the system are a good indication it's worth the nearly $5 million investment thus far.