AI in State Government (Part 8)
Richard Davis
December , 2025
AI in State Government (Part 8)
Introduction
AI’s use in state governments is rapidly increasing, with GenAI emerging as a transformative tool. As the former CIO of the state of Colorado, I know the wide potential to states from AI’s deployment. Many states have moved from experimentation to rapid operational pilots. Large-scale chatbots and virtual assistants for unemployment and benefits, document-automation, and internal coding/ops tools are now common while leaders race to add governance and controls. Many states report active GenAI pilots alongside guidance workgroups led by CIOs and Chief Counsels as they balance service gains with privacy, bias, cost, and security concerns. StateScoop
Two Short Case Studies
1. CUSTOMER-SERVICE CHATBOT. States like Texas and many others deployed web chat / FAQ assistants to handle high volume operations (unemployment questions, licensing, voter registration), cutting phone / email backlog, and providing 24/7 answers. These systems route complex or high-risk cases to humans and use human-approved responses. Early results indicate millions of handled queries and measurable reduction in call-center load. The Texas Tribune
2. OPERATIONAL TRAINING and 911 SIMULATION. Colorado has piloted GenAI for staff training (911 scenario simulations), internal policy-reference assistants, and idea-generation workshops while explicitly forbidding AI from making consequential decisions. The state frames its approach as “bullish with guardrails” — encouraging innovation but requiring oversight. Axios
See more case studies in IBM’s 2025 report on AI in State Government.
Benefits / ROI
 |
LABOR PRODUCTTIVITY. Faster first-draft documents, automated routing, and 24/7 self-service reduce staff time on repetitive tasks. NASCIO
|
|
COST AVOIDANCE. Lower call-center staffing and faster processing (unemployment, permitting) show quick payback in pilots. StateScoop
|
|
IMPROVED ACCESS. Multilingual assistants and simpler citizen journeys increase accessibility and satisfaction. Deloitte |
David Edinger, CIO of State of Colorado, says that the measurement of ROI also includes increased societal benefits through reduced crime or faster deployment of SNAP food stamps. See the video above.
Things to Watch Out for / Risks
|
BIAS and FAIRNESS. Models can replicate biased patterns. Automated decisions affecting benefits, licensing, or enforcement carry legal risk. NASCIO
|
|
PRIVACY and DATA LEAKAGE. Feeding protected data into third-party models risks exposure. Vendor contracts and data-handling rules are critical.
|
|
MISINFORMATION and ELECTION THREATS. AI makes deepfakes and targeted disinformation easier. Election offices must plan detection and response. Brennan Center for Justice |
Challenges in AI Implementation IBM
|
STATES FACE OBSTACLES in AI adoption, including procurement restrictions and underfunding for staff.
|
|
POOR DATA QUALITY hampers AI development, as seen in Massachusetts, where unclean data led to project delays.
|
|
COMPLEXITY OF VENDOR ENVIRONMENTS complicates the procurement of suitable AI solutions.
|
|
DEVELOPMENT AND MAINTENANCE OF AI SYSTEMS can be expensive, with concerns about sustainable funding. Some states have abandoned projects due to high costs. |
Practical Pointers for Implementing Generative AI in States
| 1. |
EMPOWER EMPLOYEES. Provide workforce training and approved personal AI productivity tools to allow experimentation. There is a recognized skills gap in the state workforce regarding AI tools and processes. 71% of state CIOs are training employees to address these gaps, but training budgets are often cut. Explore additional use case suggestions. Keep humans involved.
|
| 2. |
DEVELOP AND COMMUNICATE AI POLICIES. Build an AI Steering Committee to bring together perspectives from various agencies and allow for cross-agency learning. Determine governance frameworks. Establish guardrails. Communicate guidance.
|
| 3. |
START WITH LOW-RISK PILOTS that deliver clear time savings (FAQs, document summarization, intake triage). StateScoop
|
| 4. |
PERFORM INVENTORY AND CLASSIFY USE CASES by risk and impact. Require human review for consequential decisions. NASCIO
|
| 5. |
USE CONTRACTS AND DATA CONTROLS. Prohibit sending private information to uncontrolled models. Require vendor model cards and right-to-audit clauses. Texas DIR
|
| 6. |
MEASURE OUTCOMES such as time saved, accuracy, user satisfaction, and error rate. Use these outcomes to track ROI. Deloitte
|
| 7. | RUN RED-TEAM / TABLETOP EXERCISES for election-related and high-risk scenarios in order to rehearse detection, escalation, and public communication. Brennan Center for Justice |
Summary
States are running AI pilots and then implementing larger-scale AI solutions. ROI is beginning to be found. While being aware of potential risks from bias, data leakage, misinformation, and election threats, states are moving forward. They are empowering employees through training, creating AI policies and governance frameworks, and implementing pilots to understand benefits, scaling, and measurement techniques. In the end, society benefits.
