Treasury Expands Data Analytics to Combat Fraud, Improper Payments

Treasury officials are expanding fraud detection efforts across federal programs, using shared data services, analytics and machine learning tools to help agencies identify improper payments before funds are disbursed.

“We’re going from a model where every agency is essentially on their own protecting their own funds, making sure that their payments are going to the right people, to a model where we are providing a government-wide service to protect federal resources,” said Bureau of the Fiscal Service Financial Integrity Executive Director Justin Marsico last week during a Data Foundation event in Washington, D.C. “It changes the role the Treasury has in the government and reinforces our ability to be stewards of federal funds.”

The effort follows a series of executive orders that called for a national strategy to combat fraud, waste and abuse in federal benefit programs, increase data sharing across agencies and centralize payment oversight under Treasury.

A Government Accountability Office report released earlier this year cited Treasury’s Do Not Pay program as an example of how agencies are using analytics and machine learning to identify fraudulent and improper payments. It uses multiple layers of screening, including eligibility checks, identity verification and pre-disbursement reviews designed to identify suspicious payments before funds leave government accounts.

Treasury officials said gaps remain in the data available to identify many types of improper payments. There are many causes of improper payments for which data does not exist. Some data sources, such as the Social Security Administration’s Death Master File, help agencies identify payments issued to deceased individuals and are updated daily. However, payments made to deceased individuals represent only one category of improper payments.

“We’re working to solve that problem by analyzing why fraud occurs, why improper payments occur, what the data sources are that address those and bringing them into our analytics and our models,” said Marsico.