Federal agencies have proven they can launch artificial intelligence tools. But are they helpful?
Organizations must learn how to answer that question if they want to succeed with AI, David Chou, vice president of solutions architecture at Maximus told Federal News Network for our Forward-Thinking Government series.
“There’s a lot of time spent trying to figure out, ‘What should we do?’ ” he said. “But the most important question is: Is this better than it was six months ago?”
Chou said agencies are getting better at launching AI projects, but many are still learning how to measure success. Beyond implementation, organizations should track workforce adoption, repeat usage and improvements in service delivery to determine whether AI is delivering meaningful results.
AI as smart assistant
Building a new application or a workflow, collecting data on how it’s used and compiling the findings can now happen over the course of weeks, rather than months. One key metric is how widely a tool has been adopted. Another is whether that adoption has actually improved the work the AI was designed to assist.
Take contact centers, for instance.
“You can measure a known — how well you’re addressing the calls that are coming in — and you can measure that against the AI version,” Chou said. “If it’s within four points, then you’re getting there. … Now, you’re able to handle a larger volume but maintain the quality.”
Or imagine the customer service benefits of introducing AI into the building permit process. Using AI to sift through paperwork and letting a human review its progress, rather than doing the work themselves, can cut processing time to a couple of minutes per permit application, Chou said.
“You can actively work through a whole queue of backlogs,” he said. “They’ll be like, ‘Wow, permitting used to take me two months. Now, it took me like a day.’”
The human necessity of AI validation
Employees can get stuck when they’re suddenly expected to shift from producing work to evaluating AI-generated outputs, taking on the mentally taxing and sometimes unfamiliar task of fact-checking a bot. That means finding where its data came from and verifying whether that source itself is accurate.
This practice is critical as it will ultimately define responsible AI use, Chou said.
“AI is very good at explaining,” he said. “But at the end of the day, these outputs are going to potentially touch systems that are mission critical for the nation and for citizens, and so you need to be able to check its answer.”
Data ontologies may help ease that mental load.
Think of them like a dictionary and a map, Chou said. The dictionary defines each data field in an organization, and the map lays out how data is connected. Together, they show AI how to interpret and reason over the information provided.
“That will be valuable to your organization for decades,” he said. “That means you can adapt any tool, AI or otherwise, in the future.”
Keeping the right eyes on the prize
But creating an ontology doesn’t eliminate the need for human eyes. Real people still have to make sure AI behaves, Chou advised.
Once a program is given the rules and restrictions for carrying out its work, someone needs to track whether it’s providing correct answers that fall within the prescribed parameters. If not, the supervisor needs to tell the AI it’s out of bounds, he said.
“It’s that learning loop that you’re enforcing between you and the AI that makes it better and makes the guardrails better,” Chou explained. “It requires consistently working at it every single day. It is probably the most important work you can do, but it is definitely hard work.”
Discover more ways to use technology to reimagine how your agency meets its mission in our Forward-Thinking Government series.
Copyright © 2026 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.