Public agencies should write a basic AI use policy before any pilot, then run a narrow, low-risk, time-boxed pilot with clear success criteria and a public explanation ready before launch. Governance first, technology second, is the sequence that survives scrutiny.
Why governance comes before technology
Unlike a private business, a public agency’s AI decisions face public records requests, community questions and audit review from day one. Agencies that buy a tool first and write policy later end up retrofitting governance under public pressure, a much harder position than building it up front.
A four-step pilot framework
- Draft a lightweight AI use policy covering data retention, public records exposure, human review and prohibited uses.
- Pick one narrow, low-risk use case. Internal drafting help or document summarization usually carries less risk than anything resident-facing.
- Set success criteria before you start. Define in writing what “working” means: time saved, error rate, staff satisfaction.
- Prepare a plain-language public explanation in case the pilot comes up in a council meeting or records request, before it does.
Procurement considerations
Most agencies can run a small pilot within existing discretionary spending thresholds, testing real value without a full RFP. When you’re ready to scale, write the procurement around clear requirements (data handling, accessibility, human oversight and exit terms) rather than a product name, to keep it defensible and competitive.
What agencies should avoid
- Piloting anything resident-facing (chatbots, eligibility screening, automated decisions) before governance and bias review are in place.
- Signing multi-year contracts based on a pilot that hasn’t finished.
- Rolling out agency-wide before staff training is ready.
Frequently asked questions
Do we need a formal AI policy before any pilot?
At minimum, a short written policy covering data handling, human review and prohibited uses. It doesn’t need to be exhaustive. It needs to exist.
How long should a first pilot run?
Most effective first pilots run 60 to 90 days with a small group of staff: long enough for real usage data, short enough to avoid a long commitment.
Can Lost Tree Consulting help with our RFP?
Yes. See our Governance & Pilot Package.
Founder & Principal, Lost Tree Consulting. Former state insurance regulator and CEO. About Heather →