AI in State Government: Moving From Pilots to Operational Impact

There may be a lack of clear vision from the start. “Sometimes, state or locals can say, ‘We should do a pilot on AI,’ and not really know what the clear use case is. What is the clear problem that they’re trying to solve?” Dwyer says. Without that, “you can’t build an effective business case for governmentwide adoption.”

While government is fast learning how to launch AI pilots, “the harder question is, what happens when the pilot works?” says Karl Kopper, agency information security officer at the California State Transportation Agency (Caltrans).

“A pilot can often be funded as a one-time initiative, but moving into production requires ongoing funding, infrastructure, support, cybersecurity, data management, licenses, monitoring and people who are responsible for operating and improving the solution,” he says.

All these need to be factored in at the start. “One lesson I would share with other state leaders is to plan for production when you plan the pilot,” Kopper says.

The State Government AI Initiatives Leading the Way

Despite the hurdles, many states already are seeing the benefits of AI, from constituent-facing chatbots and fraud detection to legislative and policy support.

READ MORE: How cities can test small before scaling AI and data projects.

California

California has a statewide GenAI assistant called Poppy. Employees can use it to draft documents and reports, summarize and analyze large amounts of data or complex materials, and research policy rules and guidance using trusted California sources.

The state also has seen AI come to life in transportation. “We have been very intentional about focusing on AI use cases that solve real operational problems rather than implementing AI simply because the technology is available,” Kopper says.

Caltrans has moved two early GenAI efforts from pilot into production. Road Safety Insights helps teams analyze large volumes of safety-related information and identify patterns faster. Traffic Mobility Insights applies AI to help staff better understand congestion and operational conditions.

“The goal is to help our engineers and safety professionals spend less time searching through information” and more time solving problems, Kopper says.

Maryland

Maryland has ramped up AI using commercially available productivity tools, says Senior AI Advisor Michael Boyce. The state has rolled out Google Gemini and Microsoft Copilot to various agencies, “and we’ve seen adoption by about 55% of employees,” he says.

“One agency was able to use our Gemini tools to save about $400,000 in what otherwise would have been a vendor contract. They were just able to do the work themselves using these tools,” he adds.

The state also has developed a chatbot that supports the SUN Bucks nutrition assistance program “to make it easier for folks to understand how to apply for and access those benefits,” Boyce says. “We’ve had over 30,000 chats, and in 97% of cases, the chatbot is able to resolve the question or inquiry.”

Colorado

Colorado piloted Gemini with more than 150 users from across different agencies. “Because of our emphasis on accessibility, we made sure to include employees with disabilities, and that proved to be one of the most valuable decisions we made,” says former CIO David Edinger. “One participant with vision impairment shared how Gemini transformed her work life.”

After the pilot, Gemini went statewide across 18 agencies, initially drawing some 2,000 users, and adoption is growing steadily.

“But we’ve been deliberate about how we roll it out,” Edinger says. “Employees request access, managers approve it, and users complete responsible AI use training before they get started.”

Georgia

Georgia “is equipping our teams with the tools and knowledge to thoughtfully incorporate AI into our daily work” through AI licensing and targeted training, according to the Annual State IT Report.

For example, state has launched an AI chatbot explicitly designed for the secure government environment. It allows users to interact with and query their private data securely, without data ever leaving the state environment. In addition, the state has implemented AI-assisted accessibility audits, developed with Code for America, as part of work supporting 18 agencies on ADA Title II compliance.

EXPLORE: How states use AI, cloud and data to modernize constituent services.

Key Operational Considerations for AI in State Government

To take AI from pilot to production, state IT leaders need to address a number of key operational considerations, including data quality, upskilling and human oversight.

“There are several things that have to come together before AI can become an operational capability. First is quality data,” Kopper says. “AI cannot compensate for data that is inaccessible, inconsistent, poorly governed or not trusted.”

Workforce is another key factor. “We need to invest in AI literacy and upskilling, but we also need our business experts involved,” he says. “Some of our best use cases have come from bringing technology teams together with the people who actually perform the work every day.”

Change management also ranks high. “You cannot simply deploy a tool and expect adoption,” Kopper says.

And, of course, there’s human oversight. “Particularly in government, AI should support people, not replace accountability. For decisions affecting safety, public services, funding or individuals, there needs to be an appropriate level of human review.”

Click the banner below to receive the latest state and local insights to your inbox.

The Center for Democracy and Technology has addressed this head on with an AI governance checklist and operational requirements, created in partnership with the GovAI Coalition.

“The first area that we really stressed for governments to prioritize was public transparency and stakeholder engagement. That can include things like improving public awareness and understanding,” Dwyer says.

“The second area that we stress is accuracy and reliability,” she says. “Governments can adopt actual acceptable AI use policies or guidelines, so that employees know, ‘How can I use AI, and how can I not?’”

IT leaders also need to focus on governance and coordination by establishing a state government AI strategy. That means “not just having the IT people take it, but actually talking to other senior leaders in government, actually doing things like appointing a chief AI officer or an equivalent senior leader,” Dwyer says.

In Maryland, for example, state legislation on AI “lays out at a high level what the responsible governance process looks like,” Boyce says. “We have an AI subcabinet that governs at the very highest levels how we use AI, and we’ve issued responsible AI policies and implementation guidance for that.”

DISCOVER: How states secure the cloud era with identity-first strategies.

How State IT Leaders Can Scale AI Responsibly

As agencies scale their AI pilots, they need to keep responsibility front and center.

That means, for example, “dedicating more resources to the high-risk use cases,” Dwyer says: “Say you’re using AI to determine who gets access to a public benefit. That’s obviously an area where you need to have more scrutiny, more folks with eyes on the outcomes of those systems.”

In addition, IT can look at implementing regular audits, she says, “making sure that you’re checking repeatedly: Is this working as intended? Is it having some sort of bias or discriminatory outcome?”

Scaling responsibly doesn’t need to stifle innovation. “It should create a clear path that allows good ideas to move forward safely,” Kopper says. “We have been working toward that by creating a structured approach for identifying, evaluating and prioritizing AI use cases.”

AI governance also cannot belong only to IT. “Cybersecurity, privacy, legal, data procurement, HR and agency leadership all have responsibilities,” he says. “IT can help provide the framework and technical expertise, but responsible AI has to be an enterprise effort.”