An operational playbook for AI in state and local government.
Most guidance on government AI stops at principles. This project turns those principles into the instruments that decisions actually require — required steps, contract language, governance minimums, and decision frameworks — drawn from 194 documented deployments across cities, counties, and states.
Governments aren't short on AI principles. They're short on instruments.
Existing frameworks from national associations and federal agencies establish sound, high-level adoption principles. The trouble is that those principles rarely translate into the operational decisions a department head, procurement officer, or IT director has to make on a Tuesday. When actionable guidance is missing, vendors fill the vacuum — and what a vendor can sell becomes the de facto strategy.
There's a quieter problem underneath it. Staff at every level are already using public AI tools without sanction or oversight, and failure cases stay buried until they surface in litigation or the press. The jurisdictions that haven't acted don't have an AI-free environment — they have an ungoverned one.
This playbook is built to close that gap. It's operational rather than principled: a meta-analysis of 194 real implementations, distilled into tested use cases, the lessons that separated success from failure, a step-by-step roadmap, and stand-alone guides for the people who actually make these calls. Designed to be used — not filed away.
A modular, role-based toolkit.
Read it straight through, or jump to the one part you need. Select any card below to go to its deep dive.
The use cases, with lessons embedded
The six categories the deployment database identified as the most replicable and best-evidenced — three internal-facing, three constituent-facing — each with preparation, governance minimums, resistance patterns, and failure modes.
Explore use cases →What separates success from failure
People, Process, Technology, and Policy — the cross-cutting patterns that distinguished durable deployments from the ones that ended in settlements, each grounded in real jurisdictions.
Explore lessons →The implementation roadmap
From defining the problem to sustaining oversight — in the order that prevents the most expensive failures, with a gate at every step.
Explore roadmap →Stand-alone guides by role
Executive, manager, IT director, procurement officer — written to be read and distributed independently, no prerequisites required.
Explore role guides →Implementation templates
Scope statements, stakeholder memos, impact assessments, AUPs, contract addenda, pilot charters, override logs — fill in the fields and keep the record.
Explore templates →Six categories, chosen because they actually work.
The most widespread, well-tested deployments in the database — the ones replicable across jurisdiction sizes. Each pairs a clear scope with a real government that has run it.
Administrative automation — internal-facing
Document drafting & summarization
AI produces first drafts and summarizes long-form content; a human edits and approves everything before it leaves the building.
Process & workflow automation
AI sorts, flags, and routes incoming work at the intake stage; human authority is preserved at every decision point.
Internal knowledge & policy lookup
Staff ask plain-language questions and get cited answers pulled from an indexed corpus of policies, SOPs, and codes.
Public engagement — constituent-facing
311 routing & service triage
Routine requests are classified and routed automatically; humans handle the complex cases and catch errors.
Resident-facing virtual assistants
A chat assistant answers questions from approved content only, and hands off to staff when a question falls out of scope.
Language access services
Real-time translation for routine interactions, built and verified with trusted community partners in each language.
Four themes that decided the outcome.
The literature agrees AI in government should be transparent, accountable, equitable, supervised, and monitored. The unsolved problem is implementation — translating that into workflow steps that hold up under real conditions.
People
- Stakeholder engagement, before procurement
- Education & training as a launch requirement
- Human oversight built into the architecture
Process
- Workflow process mapping
- Knowing the patterns AI is actually built for
Technology
- Data readiness & feedback discipline
- Inventory of the AI you already run
- Infrastructure & interoperability
Policy
- Acceptable-use policies that live in the workflow
- Vendor selection & contracting
- Piloting with pre-specified criteria
The order matters more than the technology.
The most expensive failures weren't caused by bad tools. They were caused by steps run in the wrong order: vendor contact before problem definition, procurement before governance, deployment before training. Each step has a gate — a condition met before the next begins.
Define the problem before you talk to a vendor
Write a problem statement that names the specific workflow stage, the quantified bottleneck, and the metric that will confirm improvement — before any vendor shapes the framing.
Map your stakeholders
Name a specific person to each of five roles — problem owner, data owner, output-affected party, veto authority, and public accountability holder — before procurement begins.
Assess your readiness
Inventory the AI already running in your tools, confirm the data is ready, and assign a risk tier that drives the governance and contract requirements ahead.
Establish governance before procurement
Draft the acceptable-use policy and the workflow boundary map — the explicit line where AI output stops and human decision-making begins — as specifications a vendor must meet.
Vendor discussions and evaluation
Engage vendors on the problem statement, not a product catalog. Five non-negotiable provisions — data ownership, training prohibition, audit and override access, output liability, and update notice — make every other safeguard enforceable.
Run a real pilot
Bounded scope, success criteria set in advance, a rollback trigger, and human oversight running from day one — then a 90-day review that ends in a documented decision: scale, modify, or stop.
Ongoing oversight and adaptive review
Watch for performance drift and feedback loops; keep monthly, quarterly, and semiannual cadences on the calendar. Going live is the beginning of the oversight obligation, not the end.
Built for the four people who make these decisions.
AI implementation isn't a single decision by one person — it's a sequence of decisions by different officials at different stages. Each guide is written to be handed to the person who needs it.
Elected officials & executives
Why AI accountability is a political decision, not a technical one — the four decisions only you can make, and six questions to answer in writing before you approve anything.
County & city managers
How to sequence adoption so governance stays ahead of the tool, manage the IT-versus-program gap, and answer the three resistance patterns every deployment meets.
CIOs & IT directors
A 90-day governance infrastructure checklist, a structured vendor-evaluation method, and capacity strategies for offices without dedicated AI staff.
Procurement & legal officers
The five contract provisions every AI agreement needs — and what to do when vendors push back — plus due diligence and the public-records risks specific to government AI.
Sixteen implementation templates.
Each step and role guide points to specific templates by number. Fill in the bracketed fields, remove the instructions, and keep the completed document as part of the implementation record.
AI Scope Statement & Problem Definition
Name the workflow stage, the quantified bottleneck, and the success metric — before any vendor contact.
Workflow Integration & Boundary Map
Assign every step to AI or a human, and mark the line where AI output stops.
Embedded AI Inventory
Surface the AI already running inside the tools you pay for, plus staff "bring your own" use.
Cross-Functional Stakeholder Alignment Memo
Put a named individual in every role before procurement begins.
AI Governance Checklist for Elected Officials
The executive approval gate, turned into a signed record.
Algorithmic Impact Assessment
Score the system on rights impact and data sensitivity; assign its risk tier.
AI Contract Addendum
The five non-negotiable vendor provisions, in ready-to-insert language.
Governance-as-Process AUP
An acceptable-use policy that lives in the workflow, not in a binder.
Data Readiness Checklist
Audit, de-conflict, and assign ownership of the corpus before you index it.
AI Vendor Assessment
Structured due diligence — provenance, training data, bias testing — on the record.
Pilot Charter
Bounded scope, success criteria, the failure threshold, and the rollback procedure.
Prompting Syllabus & Cheat Sheet
Turn a tool license into a workflow staff actually use — a precondition for access.
Human-in-the-Loop Override Log
The audit trail that proves human authority and surfaces model drift early.
Public AI Registry / Inventory Entry
Public-facing accountability: name, purpose, vendor, data, oversight, review cadence.
AI Incident Response Playbook
What counts as an incident, who pulls the kill switch, and the communication protocol.
Constituent Engagement FAQ & Sentiment Survey
Disclosure language, a resident FAQ, and a feedback loop in every language served.
Jansen Weaver
Jansen is the Chief Business Officer of Peripheral, an AI advisory for private market investors, and serves as a Marine Attaché in the Marine Corps Reserves.
His blend of national-security decision-making and private-sector AI work is the lens this playbook is written through: how leaders leverage AI to make high-consequence decisions while keeping human judgment and accountability at the center.
A Hoover Institution Veteran Fellowship capstone.
The Hoover Institution's Veteran Fellowship Program gives veterans the time, platform, and scholarly resources to develop solutions to pressing challenges in the United States. Fellows undertake a capstone project aimed at real-world impact.
This playbook is that capstone — researched through a review of existing frameworks and documented deployments, then structured interviews with the government officials, regulators, and practitioners closest to the work. The goal is dissemination: getting tested, practical guidance into the hands of the state and local decision-makers who need it.
- Empowering state and local governance
- Understanding the effects of technology on economics and governance
- Answering challenges to advanced economies
Building this with the people closest to the problem.
If your organization works with state or local governments — or you're navigating an AI adoption decision right now — I'd like to hear what would make this most useful.