DE25 - Digital Transformation as a Shield: GAO’s Strategy to Combat Fraud, Waste, and Abuse

This interview was filmed on location at The Helix, Booz Allen’s Center for Innovation in Washington, D.C., as part of the event DE25: Driving Outcomes through Data. The program features top technology leaders from the public and private sectors sharing insights on cloud transformation, agentic AI, fraud prevention, and data governance. Through a series of dynamic conversations, the program captures how agencies are aligning digital infrastructure with mission needs to deliver real results for the American people. Watch the full show.

At a recent Fed Gov Today event, Jared Smith, Chief Statistician and Director of the Center for Statistics and Data Analytics at the U.S. Government Accountability Office (GAO), offered a sharp and data-driven perspective on how digital transformation is evolving the fight against fraud, waste, and abuse in government programs.

Smith began with a historical lens, noting that fraud detection in government isn’t new—dating back to the 1980s when large datasets were first used to uncover anomalies. But with modern tools and a more sophisticated understanding of data, the approach today is markedly different. “As data became cheaper to store and easier to manage, the bar lowered,” Smith said. “We can now look for fraud in ways that weren’t possible even a decade ago.”

A key theme in Smith’s comments was integration—specifically, the integration of fraud detection into the operational workflow of agencies. Rather than leaving fraud identification to isolated skunkworks teams, Smith emphasized the need to “bake” fraud detection into everyday government processes. “The people on the ground need tools they can actually use,” he said. “Not just something thrown over the fence by a lab.”

He also highlighted the GAO’s Fraud Risk Management Framework, a playbook many agencies now use to assess and respond to fraud threats. But despite progress, he noted that some agencies still operate with the mindset that fraud is someone else’s problem. “We encourage agencies to seriously assess their risk profiles and take actionable steps,” he urged.

An example of a practical step is the growing use of the Treasury’s ‘Do Not Pay’ list, which agencies are integrating into their payment systems to prevent improper disbursements. According to Smith, success depends on more than just tech—it requires cultural and policy shifts within agencies to truly mitigate risk.

Discussing the interplay between policy and technology, Smith remarked that “you need both.” While data lakes and AI tools have enormous potential, they’re ineffective without a clear understanding of where fraud risks lie. “It’s not a case of ‘build it and they will come,’” Smith said. “Direction matters.”

Smith also made the case for tailored approaches: different agencies have different fraud profiles, data complexities, and operational hurdles. Yet, certain solutions—like the Treasury’s tools—can offer scalable benefits across the board.

On the innovation front, Smith touched on the transformative potential of large language models (LLMs). He illustrated a hypothetical scenario involving suspicious narrative data—like an 85-year-old launching a lawn service in the winter—and how LLMs could flag such inconsistencies. Although such tools are still maturing, Smith views them as promising assets in the future of fraud detection.

Still, he warned against overreliance on buzzwords. “At their core, these are data analytics tools,” he reminded. “The real innovation is our newfound ability to interpret unstructured data in meaningful ways.”

In closing, Smith encouraged agencies to think long-term. “Start with a fraud risk strategy and a data foundation,” he advised. “That’s what will allow you to take advantage of emerging tools when they’re ready for primetime.”

Key Takeaways

• Integrated Fraud Detection: Agencies should embed fraud tools directly into workflows, rather than treating them as external analytical exercises.

• Tech + Policy: Effective fraud prevention requires both technological tools and strategic risk assessment frameworks.

• Future Potential: Large language models could play a key role in interpreting unstructured data to identify potential fraud, though hurdles like cost and reliability remain.