Better data quality translates into better AI models. It’s that simple, government experts say. But, of course, data governance itself is a multifaceted topic. Here are some areas in which experts suggest agencies focus their governance efforts.
Encourage Sensible, Secure Data Sharing
To take full advantage of AI, various agency programs and offices must share data where it makes sense, creating more robust datasets. Technically, secure data sharing is no longer a major technical problem, thanks to advances in data management platforms. And a March 20, 2025, executive order should help overcome any cultural resistance, said Sujit Mohanty, General Manager, Public Sector Field Engineering, at Databricks.
While focused on addressing fraud, waste and abuse, the order calls for agencies to eliminate barriers to data collaboration. “The biggest thing that’s probably occurred amongst agencies is really just getting the overall guidance that it’s OK to actually share information,” Mohanty said.
Reduce Risk Through Better Data Governance
As agencies increase their use of AI, they need to think about the risks involved: Will a given system create financial or reputational risk? Or will it bring material risk, such as reduced security or ineffective operations? To address those threats, agencies must take a hard look at their data governance practices, said David Hendrie, Principal, Federal Civilian Sales, at SAS.
Agencies should take the same kind of systematic approach to risk management that the commercial banking sector adopted under the Sarbanes-Oxley Act of 2002, which aimed to enhance the accuracy of financial reporting, he said. “Tie [the risk] back to the quality of data, tie it back to the process and the people who are doing the modeling itself,” Hendrie said.
Use Data From the Edge to Train AI Models
Your most valuable data is not necessarily in the data center. Instead, it resides at the enterprise’s edge, in applications and systems your users have deployed in the field. The challenge, said Ken Rollins, Chief Technology Strategist, AI and Digital Engineering, at Dell Technologies, is finding a way to use that data to train your AI models.
One promising approach is federated learning, he said. Rather than copying and sending that data to the core, you can train the model at the edge and send it to the data center, where it can be integrated with models generated at other edge locations and used to create a new, consolidated model. “There are companies that we are working with that do that solution really well,” Rollins said.
Don’t Overlook Your Legacy Data
An agency’s most valuable data isn’t always its most recent. That’s a problem when an agency buys a modern data management platform that works only with modern data systems and not the highly customized applications that have been around for decades, said Mia Jordan, Digital Transformation Executive, Public Sector, at Salesforce.
This is especially important as agencies face changes in the workforce. Jordan, a former government CIO, said it can take months for new employees to learn to navigate all the data they might use to train AI models. Using data integration tools, such as Salesforce’s Data Cloud, “can really help bring that information into view and bring those insights quickly into view,” she said.
Use AI to Prep Data for AI
Government experts often say that in most cases, agencies have all the data they need to train AI models. The challenge is figuring out what data they need for a given application and how to find it. As it turns out, AI itself can help with that task, said Mary Strain, AI and Machine Learning Strategy Leader, at AWS.
AI has a remarkable facility for combing through vast troves of databases and understanding the connections between different elements. This can highlight connections that data scientists might have missed. “I think there are lots of ways we can accelerate these really cumbersome and time-consuming but critically important processes that actually make your organization ready for what’s next,” Strain said.
Invest in Data Governance Up Front
The U.S. Department of Defense’s Advana program is a case study in how all the work that goes into establishing good data governance at the start of a program will pay off in the long run, said Dan Tucker, Senior Vice President, Data and AI Engineering, at Booz Allen Hamilton.
As DoD’s enterprise data and analytics environment, Advana provides users with data from more than 400 business systems, plus a variety of tools and services. It addresses key issues around data lineage and provenance, access to APIs, semantic descriptions, and other governance must-haves. “They’ve done a really nice job [of data governance] on that program, and I think it’s because they built that in early on,” Tucker said.
This article appeared in our report, “How to Deliver on the Promise of AI.” To read more about how governments are putting AI into action, download it here:
