Business Intelligence Best Practices - BI-BestPractices.com

Collaboration. Communication. Community.

 
 
 Printer-friendly
 E-mail to friend
ADVERTISEMENT
Louisiana Battles Food Stamp Fraud
Every year in the state of Louisiana, at least $25 million is lost to food stamp fraud. A new business intelligence tool, however, is cutting down on cash-for-food-stamp trades.
Editor’s note: This case study was conducted before Hurricane Katrina struck.
Every year in the state of Louisiana, at least $25 million is lost to food stamp fraud. A new business intelligence tool, however, is cutting down on cash-for-food-stamp trades.

CaseStudyLouisianaBattlesFoodStampFraud

Louisiana’s food stamp program is worth $1 billion a year; about 600,000 people receive benefits annually. According to the most recent nationwide figures available, food stamp trafficking accounts for 2.5 cents of every benefit dollar issued, or $25 million per year in Louisiana alone. Some experts, however, think fraud may account for up to 4 cents of every benefit dollar.

Combating that fraud is the Fraud and Recovery Section, part of the Office of Family Support in the state of Louisiana’s Department of Social Services, based in Baton Rouge. The department investigates food stamp trafficking in conjunction with local, state, and federal agencies. Recently it adopted an emerging (yet somewhat unusual) law enforcement tool: business intelligence software coupled with geographic information system (GIS) software.

The groundwork for the system was laid in 1997, when the federal government mandated all states move to electronic benefit transfer (EBT) instead of paperbased food stamps. (The federal government contributes about $600,000 per year to Louisiana’s food stamp benefits program.)

Now investigators had access to EBT information—data in electronic form—for investigating fraud. This information aided investigators’ efforts, yet scalability quickly became a problem, especially during an investigation conducted from 1999 to 2000 at some large stores in New Orleans and Baton Rouge. “We ran into difficulty getting information on the high volume of transactions that we were trying to track,” says Raymond Pease, assistant director of the Fraud and Recovery Section. The department’s existing query tool, Borland’s Paradox, couldn’t handle the data load. So after researching new software, the department adopted WebFocus, a product from Information Builders Inc., based in New York City.

With WebFocus, investigators gained new reporting capabilities, yet they also encountered new problems. “We were doing basic data mining and getting information that was usable, but it maybe wasn’t as usable as we wanted to have it,” explains Pease. “We were still spending many, many man hours arranging this group of recipients with this particular retailer. So it literally took weeks to get a full picture of the investigation that we were trying to do.”

Ultimately, using WebFocus helped investigators complete the investigation and successfully prosecute a number of individuals who were trafficking in food stamps. Yet the approach would need tweaking to be used on a regular basis. “We stepped back from this investigation and said we just have to have a better way of getting this information in a more timely manner,” notes Pease.

To help, last year he rolled out GIS software from Redlands, Calif.-based Environmental Systems Research Institute (ESRI), the world’s largest GIS company, which has annual revenues of about $500 million. ESRI offers software that plugs into WebFocus to render data geographically. (Note this isn’t creating a map or generating routing information, such as what a Google Maps or MapQuest does, but rather mapping multiple data points.)

The ability to plot transactions on a map dramatically cut the time investigators needed to discern potential fraud. In fact, says Pease, “right now we’re using it in a major investigation, and what took weeks to do before” his department can now accomplish in just a day or two. Initial training on the software, he notes, is only about an hour per investigator, then they’re running their own queries and reports.

Business Intelligence Goes Geographic

Louisiana is at the forefront of a trend to analyze business intelligence information geographically. To date, there have been other notable examples of GIS and business intelligence combinations. For example, some police departments use such tools to determine whether there’s a geographic pattern to vandalism or other crimes, or to see where to deploy officers. Beyond law enforcement and some government uses, however, GIS is typically only employed for physical asset management or retail applications—for example, to track warehouse stock distribution or analyze customers’ shopping patterns.

Why isn’t the use of combined business intelligence and GIS software more widespread? One historical barrier was that users needed real technical savvy; no off-the-shelf products would do it. “It just hadn’t been on the radar screen of the business intelligence vendors. They were focused on improving their own solutions,” notes Dan Vesset, research director of analytics and data warehousing for IDC in Framingham, MA. Older databases also weren’t designed to easily store geographic information.

Over the past five years, however, things have changed. For starters, many of today’s databases, including those from Oracle and IBM, have spatial extenders, “so you can easily have the latitude and longitude as a dimension in the database,” says Vesset. Second, thanks to greater competition—in particular from Microsoft software and Oracle databases—business intelligence vendors are now scrambling “to come up with the next big thing.” While GIS isn’t actually that next big thing, he notes, “it’s a big one.”


To offer GIS technology to users, business intelligence vendors are working with GIS software vendors. “The point here is that the business intelligence vendors themselves will not build it on their own,” says Vesset. “It’s a level of expertise that they don’t have themselves.”

Use of GIS has the potential to become widespread. “Ideally, we want to see the geographic component in almost every business intelligence application, because geography is always a component, whether it’s people located somewhere or goods within a supply chain application,” says Vesset. One catalyst for increased use could be radio frequency identification (RFID) tags. “Once the market moves to more RFIDs, there will be the opportunity to have more precise tracking by geographic location.” Even so, the RFID market is still immature, and Vesset recommends interested companies focus on other, more immediate GIS opportunities.

Mitigating Human/Computer Limitations

Geographical mapping can help overcome one big, basic human/computer interaction problem. Humans have a hard time holding more than about seven pieces of information in our heads at once, as was revealed 50 years ago in a groundbreaking essay by research psychologist George Miller (“The Magical Number Seven—Plus or Minus Two: Some Limits on Our Capacity for Processing Information”).


What does that mean for visual displays of information? “It’s very difficult, if you have a list of numbers, to see if you have a trend; but as soon as you graph it, you can see,” notes Vesset. Where geographic correlations are possible, mapping offers similar potential, “because that’s how you can quickly and easily pinpoint what’s most important or direct a person’s attention to it.”

That’s just what Louisiana’s investigators found. “We’ve developed signatures of fraud,” explains Pease. One potential sign of fraud, for example, is an unusual number of whole-dollar transactions—when benefits are bought for cash, it tends to be for round dollars, and frequently in $100 or $200 increments. (In a typical fraudulent transaction, the store pays just 50 cents on the dollar and pockets the rest.) So investigators analyze EBT data using these signatures, then view results geographically.

Looking for whole-dollar transactions was a start, but investigators continue to refine their fraud indicators. For example, analyzing how far recipients go to redeem benefits illegally has also been quite effective. “Let’s say this recipient lives here, and is going to this dishonest retailer which is five miles away,” says Pease. “So you have a 10-mile roundtrip from the recipient’s home, and they’re passing this many legitimate stores on the way to do an illegal transaction.” Passing so many other, more convenient stores to visit could indicate trafficking. By studying large numbers of transactions, investigators can identify such trends and pinpoint suspicious activity at a store.

In short, GIS gives investigators additional fraud-busting tools. “I’m just astounded what the map brings to all this—spatial analysis, as they say,” says Pease. “I always want to look at the map now.” Having such information means being able to get a “more complete picture of what’s going on.”

Data Warehouse Required

For newcomers to business intelligence who want geographical decision support comes a warning: don’t expect to complete geographic analysis on day one. “You probably don’t start with GIS in the business intelligence context,” notes IDC’s Vesset. “You would need to be fairly mature in your use of data warehousing and business intelligence.”


That was the case for Louisiana’s Fraud and Recovery section: the person who led the business intelligence rollout also had the foresight to create a data warehouse. “It was probably the most important thing that anyone did…to say, ‘We can mine data from this and it will help us in the future.’ I don’t think a lot of people understood the importance of that at the time,” says Pease. “If we’d tried to start building a data warehouse at the beginning of this, using this application, it would have been a lot more difficult.”

Data-quality concerns dog many business intelligence projects, and GIS is no exception. Pease says it’s a challenge he also faces. Currently, all EBT data comes from JPMorgan, which collects it from the approximately 160,000 stores nationwide that accept food stamps, then shares it with government agencies. Louisiana brings EBT data into its data warehouse, which currently contains about 150 million rows of data. Even so, "we don't scrub any of our data at all. There's just a process where we geocode the information," notes Pease. To do that—by creating map points based on address information—he uses a geo-coding engine from Dutch company Tele Atlas NV.

Geo-coding software can also note when addresses in the production system aren't accurate, meaning they can't be matched to a real address, so no map coordinates can be drawn. So far the department hasn't begun trying to clean such data, but it plans to do so eventually. Ideally, says Pease, that will happen by pushing addresschecking to the front lines—the personnel who work with consumers of state services. For example, when a state worker enters a recipient's information, including address, into the system, the system could verify that the information matches an accepted U.S. Postal Service address, or else return an error.

Funding GIS

There are many potential applications for business intelligence and GIS software, yet especially where government agencies are concerned, new technology funding is an ongoing problem. Louisiana, of course, is using its tool in an investigative context, and the return on the investment for the software includes recouping food stamp benefits lost to fraud, as well as allowing investigators to pursue more investigations.

Even so, when Louisiana started using GIS, "we said that it couldn't be just about food stamp fraud. It had to be a tool that could be used by our department in a number of ways," says Pease. For example, "we quickly got to see that this application might be useful for data modeling from a social services standpoint." Another likely use is for members of Congress who want reports on how the state's social services are consumed. Social services data modeling can improve service quality. All that's needed is a data model with a client (such as a food stamp recipient), a server (such as the social services agency), and electronic data (such as EBT), plus the ability to capture data into a data warehouse. Then it can be analyzed using GIS to spot unknown relationships. "The warehouse is the key," says Sherwood Lemoine, an internal management consultant at Louisiana's Department of Social Services. "When you're attacking poverty in Louisiana, you need to put all that stuff on the map—the food stamp system, the labor system, the transportation system, [and] the child care system. GIS gives you the ability to do that."


Funding for such additional ventures remains uncertain, but here's the vision: building those kinds of crossservices maps helps create a better "awareness of a population that you're serving, and offices that are around them—where they're located—so you can better serve your recipient population," says Pease. For example, where's the best place to put a food stamp office, given recipients' transportation options? Or, where would the state ideally open new day-care centers, given where consumers of that service live and work?

Analyzing those kinds of relationships, of course, requires accessing evergreater amounts of information, but there's a return on the investment. "You start seeing things you've never seen before," says Lemoine. Ostensibly, "there's no relationship between a lot of these programs, but there is a relationship when you put a lot of them on the map."