Trying AI 3 min read
Where should our local government start with AI?
Start wide with access and narrow with jobs. Why small pilots stall, and a plan for your first 30 days.
Part of the path I’m ready to try AI
Start wide with access and narrow with jobs. Give every department an approved place to use AI, get a few leaders using it in the open, and then help each department pick one job of its own. The most common mistake is the opposite: a tiny pilot group, a long evaluation, and nothing to show at the end.
Why small pilots tend to stall
A small pilot feels responsible. Ten people, three months, a report at the end. In practice it often produces the result MIT’s Project NANDA found across enterprise AI pilots in its State of AI in Business 2025 report: 95% returned nothing measurable.
There are local government reasons for that too:
- The wrong ten people. Pilot groups are usually picked by title. The person who would have found the best use, the clerk with the impossible agenda or the buyer drowning in specs, often isn’t in the room.
- AI stays optional. When most staff can’t use the approved tool, the ones who want AI keep using personal accounts. The pilot gets the cautious users and the risk stays where it was.
- Nothing to measure. A pilot without a named job ends with a survey about whether people liked it. That’s a weak case to bring to council.
What starting wide looks like
Every team already has a use on day one. Across 41 teams in GovAI, the same four requests sit in the top eight for 39 to 41 of them: rewriting for tone, restructuring, proofreading, and drafting emails. You don’t need to design those uses. You need to give people a safe place to do them.
At a typical GovAI customer, nearly 20 departments are using it. That happens when access is broad and each department is shown how AI fits its own work.
Five things that work
- Leaders go first. When a manager or director uses AI in real work and talks about it at a staff meeting, staff take the cue. When leaders stay quiet, staff assume it’s not really allowed.
- Make it about each department. Generic training falls flat. A 45-minute session built around a department’s own documents, like finance’s variance notes or planning’s staff reports, makes the use obvious.
- Let peers teach peers. Informal show-and-tell sessions spread practical ideas fast, because people trust colleagues in similar roles more than a vendor or IT.
- Keep the door open. Short, optional lunch-and-learns reach the people who won’t sign up for formal training.
- Name a go-to person. An internal AI lead, or a champion in each department, who answers questions and collects what’s working.
Your first 30 days
| When | What happens | Who owns it |
|---|---|---|
| Week 1 | Approve one AI tool for city or county work. Publish a one-page set of rules: what’s allowed, what never goes in, who reviews. | IT, with legal and HR |
| Week 1 | Open access to all staff, not a pilot group. | IT |
| Week 2 | Two or three leaders use AI on real work and share what they did at a staff meeting. | City or county manager |
| Weeks 2–3 | One short session per department, built on that department’s own documents. | Department heads, with the AI lead |
| Week 4 | Each department names one repeat job to work on next, with an owner and a reviewer. | Department heads |
By day 30 you won’t have measured savings. You’ll have something better for the next conversation: a list of named jobs, each with an owner, that you can track.
What to keep in view
Starting wide still starts with rules. The rules, the approved tool, and the review step come first. And some work, like legal advice, HR decisions, and anything under a records hold, needs more care no matter how broad access is.
The rules don’t need to be a finished policy on day one, though. That’s the next question: Do we need an AI policy before staff start using AI?