Big Data

How the Post Office's 'eye in the sky' fights fraud

yellow mailbox

A big-data repository and geographic analytics are helping the Postal Service detect and prevent scams. (Stock image)

Scamming the U.S. Postal Service isn't what it used to be.

Whether it's stealing mail, filing fraudulent worker's compensation claims, or financial and contract fraud, would-be cheats run up against the Inspector General's virtual eye in the sky -- mountains of data to predict and detect likely cases of fraud and identify the individuals who might be responsible. Dubbed RADR – short for Risk Assessment Data Repository – the system provides investigators and auditors a continuously updated overview of potential risks, and opportunities to launch investigations.

The basic RADR interface is a map of the U.S., with red, orange, yellow, and green circles indicating levels of risk across four areas – health care fraud, financial fraud, contract fraud, and mail theft. Red means high risk, green means low risk. It launched in October 2011. The OIG had been using analytics models before then, but the reports were delivered as spreadsheets. Auditors liked the spreadsheets, but old-school investigators did not, said Bryan Jones, director of the Countermeasures and Performance Evaluation (CAPE) group at the OIG, which runs the analytics effort.

Investigators are seeing improved results by using the analytics. Cases opened using tips generated by RADR risk profiles take, on average, 30 percent fewer hours to investigate, and lead to higher rates of recovery – almost $1 million per case. It's hard to attach a firm dollar figure to the effort: the OIG count lumps together cost avoidance in worker's compensation cases and restitution in contract and financial fraud cases. Jones, however, estimates that RADR pays for itself between three and four times over on an annual basis.

RADR is also interesting because it provides an example of a non-classified use of predictive analytics and data driven investigative techniques in action. While efforts by the intelligence community to make use of analytics tools have made headlines in recent days, the picture of how they work is necessarily incomplete. By comparison, RADR is an open book.

On the health care side, RADR provides insight to investigators on potential cases of fraud in worker's compensation claims. The data that fuels the model comes from investigations that have been successful. This provides the model with information on the type of claims that are most likely to involve fraud, whether because of the type of injury, the duration of recovery time, the length of service of a claimant and hundreds of other variables, explained Adam Lundberg, director of the Decision Analytics division of IHS, which built the heath care fraud model. The predictive model is applied to every open worker's compensation case, and each case is assessed with a risk score. Investigators trolling for cases on their beat can get a list of the most likely fraud risks, as determined by RADR, and access detailed information on the claimant, down to their home address and Social Security number.

RADR screenshot

A screen shot of the Postal Service's RADR system, showing areas at higher or lower risk of fraud. Click here for larger image.

Once a case is opened, the inquiry proceeds along traditional lines, with evidence gathered through documents, interviews and surveillance. The results of those investigations can be used to refine the model, to account for changes in fraudulent behavior and to minimize instances of false positives in risk profiling. RADR isn't continuously updated this way, because the OIG doesn't own the data, said Jones, and the process of extracting, transforming, and loading the data is manually intensive.

The mail theft model is among the newer aspects of RADR. The OIG uses data from Netflix, Gamefly and the mail-order pharmacy service of the Veteran's Administration. Investigators can drill down on reports of missing DVDs, video games or prescription drug orders at a granular level, following the path of purloined post from the source, to sorting facility, to postal routes. While it's based on a limited dataset, the mail theft model is paving the way for a time when the OIG has more information about mail theft. "When we can ingest that data from the postal service, we'll be able to measure and identify more types of mail than movies, games, and prescription drugs," Jones said.

The OIG uses RADR to generate leads on contract fraud and employee embezzlement. Gerhard Pilcher, vice president of Elder Research, which designed the financial fraud models, said that while the system doesn't have enough historically data to be predictive, it does provide signals that can tip investigators proactively possible avenues of inquiry. "If you collectively get the wisdom of people in the past and build that into a model across different dimensions in a continuous fashion, anomalies pop out without questions having to be asked," Pilcher said.

Elder also constructed a model for OIG to determine whether the USPS is overpaying for any of its 20,000-plus leased properties. The USPS contacts leases in five-year terms, so a good percentage are up for renewal each year. The Post Office looked at its commercial real estate contracts as an area of potential cost savings, and sought to trim about 10 percent from the price of each lease. What analytics showed is that the USPS was paying market rates or below on many of its leases, and that it was unfeasible to seek an across-the-board reduction. This insight was gleaned by incorporating commercial data on market conditions into the RADR model. The OIG is planning to enhance RADR by linking it to its complaints database. Individuals report mail theft in a variety of ways – by telephone, online, and by mail. The OIG also gets tips on health care fraud from individual. These are dues to be indexed and searchable in the RADR tool by the end of FY 2014. The hope is that investigators will be pushed to use the system by unifying the complaint database with RADR.

"Culturally, the inspector general community has been reactive," Jones said. "You wait for a phone to ring or wait for your boss to assign a case. Now we're putting something in front of an investigator or auditor that gives them the ability to be proactive. They can initiate a case based on what they see."

Jones hopes the success of RADR will help inspire other agencies and IG offices to adopt the same model. "Analytics isn't as much of a voodoo science in government anymore. More people are interested, more people are trying it," Jones said. He's glad to see this trend partly for selfish reasons -- he joked that he'd like to have the opportunity to learn from the mistakes of other efforts, rather than making them himself. "I would like to say we had a grand master plan," Jones said. "But honestly we've bumped and felt our way through."

About the Author

Adam Mazmanian is a staff writer covering Congress, the FCC and other key agencies. Connect with him on Twitter: @thisismaz.

Who's Fed 100-worthy?

Nominations are now open for the 2015 Federal 100 awards. Get the details and submit your picks!

Featured

Reader comments

Please post your comments here. Comments are moderated, so they may not appear immediately after submitting. We will not post comments that we consider abusive or off-topic.

Please type the letters/numbers you see above