What steps can government agencies take to become leaders in AI?
The first step is to identify your goals and work backwards from there.
- Define your prioritized use cases and the outcomes you want to drive. Start with two or three and keep it focused and simple. If you don’t understand your purpose, you won’t know what you’re measuring against or if you’re making progress.
- Establish governance early. Don’t wait to define guardrails for data security, safety, and observability.
- Experiment only after understanding the purpose and basic rules. Start small with specific use cases that drive results, and expand as your experiments turn into positive signals.
Why do some AI pilots succeed and others fail?
The difference between permanent experimentation and real scale is whether leaders think of AI as an operating model rather than just a technology.
It’s not just about access to tools. It requires operational discipline: clear governance, use cases tied to results, prioritization, and expansion plans. These enable you to move from experimentation to production scale. When pilots are siled and lack integrated strategy, ownership, and coordination, organizations are stuck.
How can leaders avoid AI sprawl?
Make important decisions upfront about which tools, infrastructure, data layers, and reusable capabilities can be used across your organization. Start by experimenting with these common components. When you’re ready, you can scale it on a robust platform. There are also new roles to consider, such as governance leaders and experience owners who handle workflow issues.
How can an agency with an uncertain budget make the right long-term investments?
We will focus on three strategies. First, shift the conversation from tokens to results. For example, what services do they offer? How much does it cost per transaction? Is the expansion sustainable? If a vendor can’t map usage to business outcomes, that’s a red flag.
Next, you need to understand cost predictability, not just cost efficiency. Are there any guardrails for use? What about price protection? Can you have visibility into cost accruals? Otherwise, you’re taking on endless financial risks.
Third, look for partners who not only pass on costs but also optimize them. There is a big difference between vendors that simply publish raw AI models and those that actively manage model selection. You need a partner who understands your use case and makes architectural decisions that provide predictability and a path to innovation.
