Artificial intelligence (AI) promises to help local governments streamline operations, improve services, and free employees to focus on more meaningful work. But with new tools emerging rapidly, how can public leaders distinguish genuine opportunities from expensive experiments?

At the 2026 ICMA Annual Conference Game Changer session, "Thinking Like an Investor When You Buy AI," Rachel Stern, founder and managing partner of Govtech Ventures, encouraged local government leaders to approach AI procurement like venture capitalists by identifying promising opportunities, demanding evidence, and increasing investment only when results justify it.

“You are investors, whether your title says so or not. You’re allocating capital into uncertain technology decisions with consequences that will compound for a decade.”

For local governments spending public money, the stakes demand a disciplined approach. To help guide local leaders through this process, Stern outlined five steps to help them evaluate AI solutions while protecting their resources, data, and communities.

1. Define the Problem Before Searching for Solutions.

Rather than starting with a product or a list of desired features, Stern urged local governments to identify the operational problem they need to solve. Traditional requests for proposals (RFPs) can inadvertently limit options by specifying a solution before exploring what the market can offer. With AI capabilities evolving so quickly, those specifications can become outdated before a contract is awarded.

“An RFP says, ‘Here’s the solution I’ve already decided I want. Show me you can build it.' [While, developing a] problem statement opens the field.”

Before approaching vendors, develop a solid "problem statement" by answering five questions:

• What specific operational problem needs to be solved?
• What does it cost today in staff time, dollars, or resident dissatisfaction?
• What measurable outcomes would define success?
• What budget, technology, privacy, or data constraints must a solution respect?
• What must a 90-day pilot demonstrate to justify further investment?

For example, a local government struggling to route hundreds of service requests each week could set targets for faster response times, fewer misrouted requests, and improved resident satisfaction. Vendors could then propose different ways to achieve those outcomes. A clear problem statement keeps the focus on results rather than flashy features. It can also help local governments discover solutions they might otherwise overlook.

2. Find the Right Vendors and Ask Tough Questions.

Once the problem is defined, local governments need to look beyond the companies that happen to be pitching their products. Stern recommended publishing problem statements through procurement webpages, LinkedIn, professional networks, and municipal leagues to attract a wider range of potential solutions. If only a few companies respond, that is useful information, too: the problem may need refining, or several jurisdictions may need to work together to attract vendors.

When evaluating potential partners, leaders should ask whether vendors understand local government’s budget cycles, procurement requirements, and operational realities. They should also be able to explain how their products differ from competing solutions. AI introduces additional questions for inquiring local governments to prompt potential vendors with: 

• What data trained the model? 
• How accurately does it perform on data resembling your local government’s? 
• Who controls model updates? 
• What happens to your data if you terminate the contract?

Vendors should provide clear answers, not vague promises about proprietary algorithms: A compelling sales pitch is not evidence that a tool will work in your community.

3. Test Before Committing to Scale.

Even a promising vendor must prove its solution works in practice. Stern recommended a limited, typically 60- to 90-day proof of concept before committing to a broader implementation. Before beginning such a pilot, it's important to come prepared. Designate who will oversee it and make the final decision, consult legal counsel about applicable privacy and automated-decision requirements, and determine exactly what data the vendor can access. Contracts should establish how data may be used, prohibit unauthorized use for training other clients’ models, and require deletion at the end of the pilot.

Just as importantly, define success metrics in advance. During testing, use realistic scenarios, not just the polished examples vendors demonstrate. Include ambiguous requests, conflicting information, and situations where the system should defer to a human. Assess whether employees can verify the tool’s answers, recognize errors, and use it effectively. Measure the entire workflow, not simply how quickly AI completes one task. A tool that drafts responses in seconds may offer little benefit if employees spend just as much time correcting them.

After the pilot, calculate the full cost, including training and ongoing support. Then make a deliberate decision to either scale the solution, narrow its use, revise it, pause it, or retire it. The principle is simple: invest more only when the evidence supports it.

4. Collaborate With Other Local Governments.

Smaller jurisdictions may lack the staff and technical expertise needed to evaluate AI solutions independently. Stern encouraged leaders to connect with other local governments facing similar challenges and share research, compare results, and explore joint purchasing opportunities. Collaboration can reduce duplicated effort, strengthen negotiating power, and help local governments identify risks they might otherwise miss. 

Stern pointed to the GovAI Coalition, led by San Jose, California, as an example of jurisdictions sharing AI governance frameworks, contract templates, and procurement resources. Her challenge to attendees was straightforward: find two peers facing the same persistent technology problem and start a conversation. Local governments do not have to navigate the AI market alone.

5. Build a Foundation Before Taking Bigger Risks.

Responsible AI adoption does not require starting with complex, fully automated systems. Stern described four levels of AI use: 

• A thought partner for drafting and analysis.
• An assistant for routine tasks.
• A teammate that provides information to support human decisions,
• A system that operates infrastructure with human oversight.

Each level builds on the previous one, requiring increasingly reliable data and stronger organizational safeguards.

For many local governments, the first step is understanding how employees already use AI and establishing clear policies for safe use. From there, leaders can identify appropriate applications, improve data practices, and build confidence through smaller projects. 

The Real Return on Investment

For local governments, an AI investment is about more than efficiency or keeping pace with emerging technology. It is about making deliberate choices with public resources, protecting residents’ trust, and delivering better results for the communities they serve.

As Stern emphasized, local governments’ purchasing decisions will help determine the standards AI vendors must meet. Clear performance measures, strong data protections, and a willingness to walk away from tools that fail give public leaders the power to shape a more accountable market.

By resisting the pressure to adopt AI simply because it is new, local governments can make decisions grounded in evidence rather than hype. That means staying curious about the possibilities while remaining disciplined about the results.

The smartest AI investment isn’t necessarily the most advanced technology. It’s the one that proves its value to the public.

Check out ICMAtv to see more from the 2026 ICMA Annual Conference!