AI in state government: Moving beyond pilots to operational impact

State governments are transitioning from experimental AI pilots to operational deployments that deliver measurable improvements in public service delivery. 

Our Chief Technology Officer, Mike Raker, joined Doug Robinson, Executive Director of the National Association of State Chief Information Officers (NASCIO), and Will Clark from the National Conference of State Legislatures (NCSL) in a recent panel discussion that revealed both the promising applications already showing results and the governance challenges that must be addressed for successful scaling. The panel was moderated by Senator Louis DePalma of Rhode Island.

Here are five takeaways from their insightful conversation.

1. From crisis response to strategic implementation

The rapid adoption of AI-powered tools during the COVID-19 pandemic demonstrated both the potential and the necessity of strategic AI deployment in government. According to Doug Robinson of NASCIO, chatbot usage exemplified this acceleration: "In 2019, we found very little use of chatbots in state government. A year later, with COVID, we went from almost none to 36 states deploying them."

This crisis-driven adoption has evolved into more sophisticated applications. Today's AI-enhanced chatbots and virtual assistants offer higher containment rates and faster resolution times, while new generative AI capabilities are expanding into areas like document analysis and content creation.

2. Applications showing measurable results

Customer service and interaction

The most mature AI applications in state government focus on public-facing services. Enhanced chatbots now handle complex inquiries with greater accuracy, while AI-powered virtual assistants guide customers through multi-agency processes that previously required multiple touchpoints and extended wait times.

Legislative and policy support

For legislative bodies, AI tools are proving particularly valuable in information processing and analysis. Will Clark from NCSL noted that "summarization is the most promising use right now, taking lots of documents and condensing them into something useful. First drafts of policy briefs, bill ideas, and hearing transcripts can all benefit."

Automated transcription and summarization of legislative hearings enable faster policy review and improve public access to government proceedings. These applications also support staff efficiency by creating searchable, AI-powered transcripts.

Operational efficiency

Behind the scenes, state agencies are deploying AI for document summarization, policy analysis, and code generation to address developer shortages. Human resources departments use AI for automated content creation in job postings and translation services, while internal analytics help optimize operations.

Security and fraud prevention

Cybersecurity represents one of the most critical AI applications, with 75% of states now using AI for threat analysis and anomaly detection. This technology is evolving toward real-time, adaptive defense systems that can respond to emerging threats more quickly than traditional methods.

Perhaps more significantly, states are beginning to shift from reactive to proactive approaches in fraud detection. As Doug Robinson emphasized, "Fraud detection and prevention is where we need to go. We have to abandon the pay-and-chase model that has cost $220 billion in fraud in recent years."

3. Implementation barriers and governance gaps

Despite promising applications, state governments face significant challenges in scaling AI deployment effectively.

Data governance foundation

A fundamental barrier is the lack of robust data governance frameworks. Doug Robinson noted that "only about a quarter of the states have what they would describe as robust data governance. If you don't have that beforehand, you're going to have the negative consequences of ingesting poor-quality data into your models."

While 34 states have appointed Chief Data Officers, most lack the authority and resources necessary to ensure data quality across agencies. Poor or inconsistent data undermines AI effectiveness and can lead to biased or inaccurate outcomes.

Workforce development

The human element remains crucial for successful AI implementation. Mike Raker from Maximus highlighted the importance of workforce planning: "As we implement AI that changes the nature of work, I want to make sure I've got upskilling and career opportunities in place, so we maintain a good employment relationship with the 35,000 employees we support."

While AI literacy training has doubled over the past year, adoption remains uneven across agencies and skill levels.

Technology integration challenges

Existing IT infrastructure and procurement processes can impede AI adoption. Multi-year contracts and legacy systems weren't designed to accommodate rapidly evolving AI capabilities. Additionally, licensing costs for premium AI tools can strain already tight technology budgets.

Will Clark noted that "many tools just aren't quite there yet for the specific legislative environments. And if you've got a multi-year IT rollout, you can't always just add generative AI into that workflow."

4. Ethical considerations and transparency requirements

Everyone on the panel emphasized that successful AI deployment requires proactive attention to transparency and ethical considerations.

Procurement transparency

Currently, only about 25% of states require vendors to disclose when AI is embedded in their solutions. This gap limits agencies' ability to understand and govern the AI tools they're using.

Decision-making boundaries

Panelists stressed the importance of maintaining human oversight for consequential decisions. Doug Robinson was clear about key limitations: "There are certain processes — like [program or benefit] eligibility determination — where generative AI should not be making the decision. That needs to be done by humans, with AI supporting the process."

This principle is also important in areas like criminal justice, where predictive AI systems raise significant equity concerns.

Documentation and auditability

AI-generated content may fall under open records laws, requiring clear policies for retention and disclosure. There's growing interest in technical standards like C2PA-style credentials that would tag AI-generated content with origin and metadata information.

5. Principles for successful implementation

Based on their experience with AI deployments, the panelists identified several key principles for successful implementation:

Mission-driven approach

Mike Raker emphasized starting with clear objectives: "When you start on an AI journey, the first thing in any successful implementation is really clarifying what the outcome or mission is you're trying to solve with the technology. It's the application of technology to a problem, not the application of technology for technology's sake."

Ongoing Transparency: Successful implementations treat transparency as a continuous process rather than a one-time compliance requirement. "Transparency is a process, not a checkbox," Mike Raker noted.

Risk-appropriate governance

The governance approach should match the risk profile of the specific AI application. High-stakes decisions require more rigorous testing and review processes than lower-risk applications.

Collaborative Learning

Given AI technology's rapidly evolving nature, agencies, vendors, and legislative bodies must share lessons learned to accelerate safe and effective adoption.

Looking ahead

The next twelve months will likely determine which states successfully transition from pilot programs to enterprise-scale AI deployment. Success will depend on addressing governance gaps, improving data quality, and developing workforce capabilities while maintaining public trust through transparent and ethical implementation.

As governments work to harness AI's potential for better public service delivery, the focus must remain on measurable outcomes for citizens rather than technology adoption for its own sake. The states that combine disciplined governance with strategic deployment will be best positioned to realize AI's promise of more efficient, accurate, and responsive government services.

The conversation around AI in government is moving beyond whether to adopt these technologies to how to implement them responsibly and effectively. For state leaders, the question is no longer if AI will transform government operations, but how quickly they can build the governance frameworks necessary to deploy it successfully at scale.