Skip to main content
How Can We Help?Contact

Government & Public Sector

Shorten the wait for public services, and account for every AI tool you run.

We join what your case, call-center, permitting and finance systems each know, including the older ones you cannot replace yet. Residents get answers and status without calling twice, applications reach a caseworker complete, and agencies share what the law lets them share. The same joined-up data lets you account for every AI tool: an owner, an approval and a record, inside your security plan and the accessibility rules.

Talk through your agency’s data See two public-sector results

State agencies · Cities and counties · Public health · Transportation and transit

A public building with residents queuing on its steps; a resident’s call, the application, the records two agencies hold and the register of AI tools light in turn, and with each one a person leaves the queue1234
  1. 1Front doorResident’s question answered for where they live, no transfer
  2. 2CaseworkApplication complete on arrival; a caseworker decides
  3. 3RecordsEligibility confirmed with another agency, no file shared
  4. 4GovernanceNew AI tool logged, approved and disclosed to residents
Illustrative: as each step joins up, the queue at the counter gets shorter.

Where public services stall, and what it costs residents

Four lines we find filled in at almost every agency we meet. Each one reads across: what residents see, what staff see, why it happens, and the cost to the public.

  1. Residents call because the website could not answer

    Residents see

    Long holds, and a second call to get the same answer.

    Staff see

    The same handful of questions all day, answered from memory.

    Why it happens

    The answers live in PDFs, department pages and people’s heads, not in one source that every channel draws on.

    Cost to the public

    Residents who give up on a service they qualify for, most often those with the least time, language support or internet access; and agent hours spent on repeat questions and abandoned calls.

    See Part IV: the front door

  2. Applications wait in a queue nobody can see

    Residents see

    No status for weeks, so they call, write or come in to ask.

    Staff see

    Documents keyed in again from mail, email and the counter, and cases chased by hand.

    Why it happens

    Intake is manual and the case system does not tell the web, the phone line or the resident where a case stands.

    Cost to the public

    Benefits that arrive late, and a business that cannot open or a worker who cannot start while a license is pending; for the agency, backlogs, overtime and missed timeliness standards.

    See Part IV: casework

  3. Each agency keeps its own copy of the same resident

    Residents see

    The same documents asked for by three agencies.

    Staff see

    Eligibility checked again by hand, and reports to the legislature assembled from spreadsheets.

    Why it happens

    Records cannot be matched across programs, and what the law allows to be shared is unclear or unwritten.

    Cost to the public

    Residents who prove the same thing three times, or lose a benefit over one missing paper; for the agency, duplicate verification, improper payments found late, and numbers that differ by agency.

    See Part III: counting rules

  4. AI arrives before the policy does

    Residents see

    A chatbot or a translated notice with no sign of who checked it.

    Staff see

    Public AI tools used on agency work, and AI features switched on inside existing contracts.

    Why it happens

    No inventory, no approval path, and no answer yet on records, retention or disclosure.

    Cost to the public

    Residents’ data in tools nobody approved, and decisions nobody can explain to the person they affect; then a records request, an audit finding or a news story the agency cannot answer.

    See Part VI: reviewing an AI use case

Two public engagements, entered as schedules

Each one set out the way a schedule attaches to a form: the agency, the program, what was built and what it achieved. The figures are the client’s own; the clients are described, not named.

Schedule R-1

A US state government

Public health records and verification

What was built

Health events verified without sharing the record

Proof that a health event was real had to be checked quickly by businesses and agencies without exposing anyone’s medical record. A permissioned ledger let only approved state health entities write; it held proofs and metadata, never the records themselves, and every write was time-stamped and attributable. The protected data never left the issuer.

of recorded events validated on the first attempt98%
fewer manual “is this real?” checks42%
average time to verify an eventUnder 5 s

Reported by the client.

Case study: Tamper-evident verification for a state

Schedule R-2

A major US city’s 311 service

Resident information during a public-health emergency

What was built

An AI assistant inside the live 311 call flow

As guidance changed weekly, 311 agents spent their time on the same questions, and some answers depended on where the caller lived. The assistant sat inside the existing 311 call flow, answered from a knowledge base city staff could update without code, took the caller’s ZIP to point to the right local facility, and passed anything outside its scope to an agent with the context.

What it achieved

One trusted source, localized by ZIP, during a period when misinformation spread easily.

No figures were reported for this engagement; it is cited for what it shows.

Case study: An AI assistant in a city’s 311 service

Numbers the council and the legislature ask for, defined first

The same queue can be reported three ways, and each way survives an audit. Pick a folder to see how a number can be counted, who relies on it, the decision it changes, and what to decide before it is published.

Processing time

Can be counted as

  1. From when. From the day an application arrives, or from the day it is complete.
  2. Which days. Calendar days, or business days only.
  3. Which statistic. The average, the median, or the share decided within the standard.

Who relies on it

Program directors, budget offices, and federal partners that hold the program to a timeliness standard.

Kept in

the case management system, the document intake log, the mailroom

Decision this changes

Where to redesign the workflow or add capacity, and which step to fix first.

Example with numbers

Ten applications take 10, 12, 14, 15, 16, 18, 20, 22, 60 and 93 days. The median is 17 days, the average 28, and 80% are decided within 30 days.

Decide before it is published

From arrival or completion, which days, and which statistic goes in the report?

Backlog

Can be counted as

  1. Everything open. Every application not yet decided.
  2. Everything overdue. Only applications past the processing standard.
  3. Overdue on the agency. Overdue cases that wait on staff, not on the resident.

Who relies on it

Staffing and overtime decisions, and what the legislature hears about the program.

Kept in

the case management system, its pending-documents status, the standard in policy

Decision this changes

Which cases are stalled and why, and whether the fix is staff, policy or a resident’s missing documents.

Example with numbers

1,200 applications are open and 500 are past the 30-day standard, 300 of them waiting on the applicant for documents. The backlog is 1,200, 500 or 200, depending on the rule.

Decide before it is published

Open, overdue, or overdue and waiting on staff?

First-contact resolution

Can be counted as

  1. By the call log. Resolved if the call ended without a transfer.
  2. By repeat contact. Resolved only if the resident did not contact the agency again about the same issue within seven days.
  3. Which channels. Phone only, or phone, chat, email and the counter together.

Who relies on it

Contact-center staffing, and whether self-service is working.

Kept in

the phone system, the 311 or CRM case records, chat and email logs

Decision this changes

Whether the answers and the hand-offs work, or residents are being sent around.

Example with numbers

Of 1,000 calls, 800 end without a transfer, but 150 of those callers call back within seven days about the same issue. First-contact resolution is 80% by the call log and 65% by repeat contact.

Decide before it is published

By the log or by repeat contact, and across which channels?

Call abandonment

Can be counted as

  1. Every hang-up. Every caller who hangs up before reaching an agent.
  2. Short abandons. Calls dropped within the first few seconds left out of both counts.
  3. Callbacks. Callers who accepted a callback counted as abandoned, or as served.

Who relies on it

Hold-time targets, staffing by hour, and the case for a callback option.

Kept in

the phone system’s queue reports

Decision this changes

Whether there are enough staff at the right hours, and whether residents can reach the agency at all.

Example with numbers

10,000 calls enter the queue and 1,200 hang up, 400 of them within ten seconds. Abandonment is 12% counting every hang-up and 8.3% when the short abandons leave both counts.

Decide before it is published

Do short abandons and accepted callbacks count?

Online share of transactions

Can be counted as

  1. Started or finished. Transactions that start online, or only those completed online.
  2. Per transaction or per resident. Each renewal counted, or each person who used the service.
  3. Assisted digital. Staff entering an application online for a resident counted as online, or not.

Who relies on it

Where to invest in the website, and how many counter and phone staff a service needs.

Kept in

web analytics, the transaction system, the counter log

Decision this changes

Whether the online service is usable, and how much help to keep at the counter and on the phone.

A higher share is not better if it shuts out residents who need a person.

Example with numbers

Of 5,000 renewals, 3,500 start online and 2,800 finish online; the other 700 move to the phone or the counter. The online share is 70% by starts and 56% by completions.

Decide before it is published

Started or finished, per transaction or per resident, and is assisted digital online?

Improper payment rate

Can be counted as

  1. Confirmed overpayments only. Payments shown to be too high.
  2. Underpayments as well. Payments too low count as improper too.
  3. Unknown payments. Payments without the documentation to tell either way, counted in or reported apart.

Who relies on it

Corrective action plans, federal reviews, and how the program is described in public.

Kept in

quality control samples, the eligibility system, payment records

Decision this changes

Where verification and program controls need work, and what goes in the corrective action plan.

An improper payment is not the same as fraud; many come from missing documentation or administrative error.

Example with numbers

A sample worth 10 million dollars has 300,000 dollars confirmed overpaid, 100,000 underpaid and 200,000 without the documentation to tell. The rate is 3%, 4% or 6%, depending on what is counted.

Decide before it is published

Overpayments only, underpayments too, and where do unknown payments go?

Cost per transaction

Can be counted as

  1. Staff time only. The cost of the people who handle the transaction.
  2. Fully loaded. Overhead, systems and space shared in as well.
  3. Which channel. One cost for all channels, or a cost for each.

Who relies on it

Budget requests, fee setting, and which channel to move residents toward.

Kept in

the financial system, timekeeping, the permitting or licensing system

Decision this changes

Whether modernization lowers what the service costs in total, or only moves the cost to another channel.

Example with numbers

A permit counter costs 900,000 dollars a year in staff and 300,000 in overhead, and handles 20,000 permits. That is 45 dollars a permit on staff cost alone and 60 dollars fully loaded.

Decide before it is published

Staff time or fully loaded, and one cost or one per channel?

Which of these gets asked for in every budget hearing? Bring it to a 30-minute call, and we will show you what it takes to count it once and report it the same way to every audience.

Settle one number before the next hearing

The work, set out by service line

Four sections, the way an agency organizes its services rather than its systems. Each item says what it is measured by; where a case study exists, it is linked.

Section AThe front door: 311, the contact center and the website

  • An assistant inside 311 and the contact center

    Plain-language answers from a knowledge base staff keep current without code, tailored to where the resident lives, and a hand-off to an agent with the context when the question needs a person.

    Measured by Repeat calls and first-contact resolution

    Case study: An AI assistant in a city’s 311 service

  • One answer source for every channel

    The web, the phone line, the counter and any assistant drawing on the same approved answers, each with an owner and a review date, written to the accessibility standard.

    Measured by Answers out of date, and calls about what the website already says

    First step: AI & Agent Pilot

Section BCasework: benefits, licensing and permits

  • Applications read on arrival and routed

    Documents classified and the key fields read as they arrive from mail, email, the web and the counter; missing items flagged to the resident at once; cases routed to the right worker, who makes every decision.

    Measured by Processing time and backlog

    First step: AI & Agent Pilot

  • Status residents can see

    Case status from the case system made available to the website, the phone line and staff, in plain language, so a resident can see where an application stands without calling.

    Measured by Status calls, and cases chased by hand

    First step: Data Platform Accelerator

Section CAcross agencies: records and data sharing

  • Residents matched across programs, under written agreements

    Records matched across programs with the data-sharing agreements written first, so each agency uses only what the law and the agreement allow, and every use is logged.

    Measured by Duplicate verification, and improper payments found late

    First step: Data Platform Accelerator

  • Verification without handing over the record

    Proof that a record is genuine, checked by another agency, a business or a resident, while the record itself stays with the agency that issued it.

    Measured by Manual checks, and time to verify

    Case study: Tamper-evident verification for a state

Section DOversight: reporting, budget and accessibility

  • Reports to the legislature from governed data

    The numbers the legislature, the council and federal partners ask for, defined once, with the lineage from every published figure back to its source.

    Measured by Days to produce a report, and figures that differ by agency

    First step: Data Platform Accelerator

  • Dashboards and documents everyone can use

    Public dashboards, PDFs and data portals rebuilt or checked to WCAG 2.1 Level AA, with text alternatives for every chart and a way to download the data behind it.

    Measured by Accessibility defects in public dashboards and documents

    First step: AI & Data Assessment Sprint

Section EBuilt for the public servants who will use it

A new tool changes who answers residents, which cases need professional judgment, and how supervisors spot a failure. Caseworkers need to understand a recommendation, recognize the exception, decide for themselves and record why. So every build includes:

  • Training by role, on the agency’s own cases, calls and policies, for caseworkers, agents and supervisors.

  • The workflow redesigned with the people in it, including who reviews what the tool suggests and what happens when it is wrong.

  • A way to report problems, from staff and residents, with a named owner who acts on them.

  • Accessibility for staff as well as residents, in the tools caseworkers use every day.

  • Use measured after launch, so the agency can see whether the tool is used, trusted and still working a year on.

First step: Adoption Review. Practice that leads it: AI & Data Literacy.

Across every program

  • An AI policy, an inventory and an approval path

    Every AI tool and AI feature in use listed with an owner, a purpose and a review level, and a written path for new ones, so a records request or an auditor’s question about AI gets a same-week answer.

    First step: AI Governance Quickstart. Practice that leads it: Strategy & Governance.

  • Security for criminal justice, tax and health data

    Where the agency holds criminal justice, federal tax or health data, the controls CJIS, Publication 1075 or HIPAA ask for, built into the data platform and every tool that reads from it, rather than checked at the end.

    First step: Data Platform Accelerator. Practice that leads it: AI Security & Cyber Risk.

Your agency’s longest queue is somewhere else? Bring the one residents complain about most to a 30-minute call.

Talk through your longest queue

Which requirements to evaluate for your agency’s data

Every public entity should check the first two rows. The rest depend on the data you hold, the programs you run and your funding. Select what describes your agency to mark the requirements worth evaluating; each links to its source. It is a screening aid for planning, not a determination of what applies.

Select what describes your agency

Requirements to evaluate are labeled; the others fade but stay on the page.

Requirements that may apply to a public agency’s data: who each covers, its status in October 2026, and what it means for the data
RequirementCoversStatus, October 2026What it means for the data
ADA Title II rule on web and mobile accessibility (opens in a new tab)Every agency
  • Every state and local government
Equal access is already required under Title II. The interim final rule (Apr 20, 2026) set the dates for the technical standard: Apr 26, 2027 for governments of 50,000 people or more; Apr 26, 2028 for smaller ones and special districts.Web content and mobile apps, including public dashboards, documents, forms and chatbots, meet WCAG 2.1 Level AA.
State public records and retention lawsEvery agency
  • Every state and local government
  • Varies by state
In force, under each state’s own law and retention schedules.Prompts, outputs and chat logs can be public records, depending on your state’s law and what they were used for. Where they are, they are kept as long as their retention schedule requires and produced on request.
FBI CJIS Security Policy, version 6.1 (opens in a new tab)Worth evaluating
  • Agencies with access to criminal justice information
  • Contractors who handle it
Version 6.1 released in June 2026. Priority 1 controls have been sanctionable since Oct 1, 2024; the rest are audited without sanctions until Sep 30, 2027, and your state’s CJIS Systems Agency sets its own audit timeline.Only where the agency or its contractors access criminal justice information: multi-factor sign-in, encryption of at least 256-bit strength in transit and at rest outside secure locations, monthly vulnerability scans, and the same controls on any analytics or AI tool that reads that data. Each control is checked against the policy text your state audits to.
IRS Publication 1075 (opens in a new tab)Worth evaluating
  • Agencies that receive federal tax information
  • Their contractors
In force; November 2021 revision.Federal tax information stays in approved environments, every access is logged, and the IRS Office of Safeguards is told in advance before contractors or a cloud service touch it (45 days ahead for contractors).
Federal benefit program confidentiality rules (opens in a new tab)Worth evaluating
  • SNAP (7 CFR 272.1)
  • Medicaid (42 CFR 431, subpart F)
  • Unemployment insurance (20 CFR 603)
In force.Applicant and recipient information is used and disclosed only for the purposes each program allows, which decides what can be matched, shared across agencies, or sent to an AI tool.
HIPAA, for agencies that are hybrid entities (opens in a new tab)Worth evaluating
  • Health departments, public hospitals and public health plans
In force.An agency with health care functions can designate those components as a hybrid entity; the Privacy Rule then applies to them, and otherwise to the whole agency.
Texas Responsible Artificial Intelligence Governance Act (HB 149) (opens in a new tab)Worth evaluating
  • Texas state agencies and local governments
  • Not hospital districts or higher education
In force since Jan 1, 2026; enforced by the attorney general, with 60 days to cure.An AI system meant to interact with residents tells them so before or at the time of the interaction, in plain language; the duty covers resident-facing systems, not every staff tool. AI is not used to score residents by their behavior or traits.

For orientation only, not legal advice: other federal, state and local rules may apply, and what applies depends on the data, the program, the contracts and the funding. Your counsel, security officer and records officer decide. Checked October 14, 2026.

Systems we connect, and the rules that usually come with them

Systems we connect, and the rules that usually come with them
Source systemHoldsRules that usually come with it
Case management and eligibilityApplications, cases, decisions and notices, often on a mainframe or an older platform.Program confidentiality; federal tax information where income is verified
311 and resident CRMRequests, calls, chats and service tickets, with locations.Public records; accessibility for anything residents see
Permitting and licensingApplications, inspections, plans and fees.Public records; accessibility
Financial and budget systemsSpending, revenue, grants and payroll.Federal grant reporting; public records
GISParcels, service areas, assets and boundaries.Public records; open data policies
Public safety records and dispatchIncidents, calls for service and reports.CJIS Security Policy
Health and clinical systemsClinic visits, immunizations and disease reports.HIPAA where the agency is covered; public health law
Documents and emailLetters, notices, policies and correspondence.Public records and retention schedules

Review an AI use case before residents meet it

Pick a use case to see what it does, its public impact and the data it touches, the constraints to examine, the safeguards to have in place, the offices that usually take part, and the first record to file.

Front door

An assistant answering residents on 311 or the website

What it does

Answers residents’ questions and points them to the right service or office.

Public impact and data: check all that apply

  • Decides or recommends about a residentNo
  • Touches criminal justice, tax or health dataNo
  • Faces the public directlyYes
  • Produces public recordsYes
  • Uses federal program dataLikely

Data and legal constraints to examine

  • ADA Title II: the assistant meets WCAG 2.1 Level AA like the rest of the website.
  • State public records law: conversations kept under the right retention schedule.
  • In Texas, residents are told they are talking to AI before or at the start.

Safeguards and evidence before launch

  • Answers only from content the agency has approved, each item with an owner.
  • Says it is an AI assistant, and offers a person at any point.
  • Anything about eligibility or a resident’s own case goes to staff.

Suggested review participants

Led by: Program review311 or contact-center directorRecords officerAccessibility coordinatorInformation security

First record to file

An inventory entry for the assistant, and a retention decision for its conversation logs.

First step: AI & Agent Pilot. Case study: An AI assistant in a city’s 311 service

Casework

Eligibility recommendations for benefits

What it does

Suggests whether an applicant qualifies, and for how much.

Public impact and data: check all that apply

  • Decides or recommends about a residentYes
  • Touches criminal justice, tax or health dataLikely
  • Faces the public directlyNo
  • Produces public recordsYes
  • Uses federal program dataYes

Data and legal constraints to examine

  • Program confidentiality rules for SNAP, Medicaid or unemployment insurance.
  • Notice and fair hearing rights: residents can see the basis for a decision and appeal it.
  • Publication 1075, where federal tax information is used to verify income.

Safeguards and evidence before launch

  • An eligibility worker makes every decision; the tool only suggests.
  • Every suggestion shows the facts and rules it rests on.
  • Outcomes checked by group before launch and every quarter after.

Suggested review participants

Led by: Executive reviewProgram directorCounselPrivacy officerInformation security

First record to file

A use-case assessment, with the plan for testing outcomes by group before launch.

First step: AI & Data Assessment Sprint

Casework

Reading documents as applications arrive

What it does

Classifies incoming documents, reads the key fields, and flags what is missing.

Public impact and data: check all that apply

  • Decides or recommends about a residentLikely
  • Touches criminal justice, tax or health dataLikely
  • Faces the public directlyNo
  • Produces public recordsYes
  • Uses federal program dataLikely

Data and legal constraints to examine

  • Program confidentiality rules for the data the documents contain.
  • Public records law: the original documents are kept, not just the extracted fields.

Safeguards and evidence before launch

  • Low-confidence fields sent to staff to confirm.
  • Original documents kept with the case.
  • Accuracy measured on the agency’s own documents before go-live.

Suggested review participants

Led by: Program reviewProgram directorRecords officerPrivacy officerInformation security

First record to file

An accuracy test on the agency’s own documents, with the result kept on file.

First step: AI & Agent Pilot

Program integrity

Flags for possible fraud or improper payments

What it does

Picks out cases for an integrity review.

Public impact and data: check all that apply

  • Decides or recommends about a residentYes
  • Touches criminal justice, tax or health dataLikely
  • Faces the public directlyNo
  • Produces public recordsYes
  • Uses federal program dataYes

Data and legal constraints to examine

  • Program rules on investigations and overpayment recovery.
  • Notice and fair hearing rights before any benefit is reduced or stopped.

Safeguards and evidence before launch

  • A flag opens a review; no benefit stops on a score.
  • Flag rates compared across groups and regions before launch.
  • Every flag recorded with its reasons and the reviewer’s decision.

Suggested review participants

Led by: Executive reviewProgram integrity directorCounselPrivacy officerFair hearings office

First record to file

A human-review procedure for every flag, written before the first one is raised.

First step: AI & Data Assessment Sprint

Front door

Translating notices and resident letters

What it does

Drafts translations of notices, letters and web pages.

Public impact and data: check all that apply

  • Decides or recommends about a residentNo
  • Touches criminal justice, tax or health dataLikely
  • Faces the public directlyYes
  • Produces public recordsYes
  • Uses federal program dataLikely

Data and legal constraints to examine

  • Language-access obligations under state law and the terms of federal funding.
  • Accessibility rules for anything published online.

Safeguards and evidence before launch

  • A qualified translator reviews vital documents before they go out.
  • Each translated notice marked with its review date.
  • Residents can ask for help in their language from a person.

Suggested review participants

Led by: Program reviewProgram directorLanguage access coordinatorAccessibility coordinator

First record to file

A review procedure naming which documents are vital and who checks them.

First step: AI & Agent Pilot

Records

Searching and redacting for public records requests

What it does

Finds responsive records and proposes redactions.

Public impact and data: check all that apply

  • Decides or recommends about a residentNo
  • Touches criminal justice, tax or health dataLikely
  • Faces the public directlyLikely
  • Produces public recordsYes
  • Uses federal program dataNo

Data and legal constraints to examine

  • State public records law: deadlines, exemptions, and what must be released.
  • CJIS, health and tax rules for the records it searches.

Safeguards and evidence before launch

  • A records officer approves every release and every redaction.
  • Searches logged, so the search can be shown if the response is challenged.
  • Protected data is never sent to tools outside the agency’s environment.

Suggested review participants

Led by: Program reviewRecords officerCounselInformation security

First record to file

A search log and a redaction review procedure.

First step: AI & Agent Pilot

Licensing and permits

Plan and permit review assistant

What it does

Checks applications and plans against the code and lists the issues for a reviewer.

Public impact and data: check all that apply

  • Decides or recommends about a residentYes
  • Touches criminal justice, tax or health dataNo
  • Faces the public directlyLikely
  • Produces public recordsYes
  • Uses federal program dataNo

Data and legal constraints to examine

  • The building, zoning and fire codes the jurisdiction has adopted.
  • State licensing rules for the reviewers who sign off.

Safeguards and evidence before launch

  • A licensed reviewer decides every application.
  • Each issue cites the code section it comes from.
  • Applicants see the reviewer’s decision, not the tool’s list.

Suggested review participants

Led by: Program reviewBuilding official or permitting directorCounselInformation security

First record to file

A test of the tool’s issue lists against applications already decided.

First step: AI & Agent Pilot

Transportation

Traffic and transit analytics from cameras and sensors

What it does

Counts vehicles, riders and incidents to plan service and signal timing.

Public impact and data: check all that apply

  • Decides or recommends about a residentNo
  • Touches criminal justice, tax or health dataLikely
  • Faces the public directlyNo
  • Produces public recordsYes
  • Uses federal program dataLikely

Data and legal constraints to examine

  • State and local rules on cameras, license plate readers and retention.
  • CJIS, if footage or plate data reaches law enforcement.

Safeguards and evidence before launch

  • Counts kept; faces and plates not stored unless the law requires it.
  • A written retention period for footage.
  • Access to raw footage logged and limited.

Suggested review participants

Led by: Program reviewTransportation directorPrivacy officerRecords officerInformation security

First record to file

A retention decision for footage, plate data and counts.

First step: AI & Data Assessment Sprint

Records

Drafting minutes and summaries of public meetings

What it does

Drafts minutes and plain-language summaries from recordings.

Public impact and data: check all that apply

  • Decides or recommends about a residentNo
  • Touches criminal justice, tax or health dataNo
  • Faces the public directlyYes
  • Produces public recordsYes
  • Uses federal program dataNo

Data and legal constraints to examine

  • Open meetings law: the official minutes are the ones the body approves.
  • Accessibility rules for the published minutes and summaries.

Safeguards and evidence before launch

  • The clerk edits and the body approves the official minutes.
  • The recording kept under its retention schedule.
  • Summaries labeled as summaries, not as the minutes.

Suggested review participants

Led by: System owner sign-offClerkRecords officerAccessibility coordinator

First record to file

A retention decision for recordings and drafts.

First step: AI & Agent Pilot

Internal

A policy and procedures assistant for staff

What it does

Answers staff questions from the agency’s own policies and procedures.

Public impact and data: check all that apply

  • Decides or recommends about a residentNo
  • Touches criminal justice, tax or health dataNo
  • Faces the public directlyNo
  • Produces public recordsLikely
  • Uses federal program dataNo

Data and legal constraints to examine

  • Little beyond the agency’s own policies, provided resident data is kept out of it by design.

Safeguards and evidence before launch

  • Only current, approved policies, each with an owner.
  • Every answer names the policy and its date.
  • Questions about a resident’s case referred to a supervisor.

Suggested review participants

Led by: System owner sign-offPolicy officeInformation security

First record to file

An inventory entry naming the owner and the policies it answers from.

First step: AI & Agent Pilot

Illustrative governance considerations, not a legal determination or a preapproved workflow. What applies depends on the program, the data, the funding and how the tool is used; your counsel, privacy officer and records officer decide. The requirements in full: Part V.

Who leads the review. Separate approvals may still be needed: program authorization, security, legal and privacy review, procurement for any new tool or contract, and operational readiness. Each use case lists the offices that usually take part; the level says who leads.

  • Executive review: the agency head or deputy and the CIO, before anything that decides about residents.
  • Program review: the program director or the office that owns the service.
  • System owner sign-off: a named owner, with access controls and a periodic check.

A use case not on the list, or one already in use without these conditions? Thirty minutes is usually enough to decide what review it needs.

Talk through one use case before it launches

Check one: where to start

Three first steps for three different readers. The first two are our assessment and governance offers, scoped to one program or one policy, and none commits you to the next.

  1. Box A

    One program’s data and AI readiness

    For a CIO, chief data officer or program director with one service to fix first

    Two weeks, one program · Runs as the AI & Data Assessment Sprint

    One program’s case, call and finance data traced to where the delays and repeat contacts start; the gaps on one list with their cost; and a plan in order, ready for a budget request.

    Ask about a two-week sprint

  2. Box B

    An AI policy, inventory and approval path

    For a state or city AI lead, or counsel, with AI already in use

    One use case, taken to a decision · Runs as the AI Governance Quickstart

    One AI use case taken through your rules, your records obligations and your security plan to a decision, with the inventory and approval path the rest of the agency’s AI can follow.

    Ask about the Governance Quickstart

  3. Box C

    Data Maturity Assessment

    For an IT director who wants a quick read before anything is bought

    Free, six minutes or so

    Fifteen questions on lineage, quality, ownership, access and platform. The result is instant and needs no email.

    Score your data foundation

After either of the first two: a first build on one service, measured against a number the program already reports, such as processing time, repeat calls or the online share, then a decision to scale, change or stop.

AI & Agent Pilot · Data Platform Accelerator

For the budget request

What you can take to the budget office

  1. A readiness score for the program’s data and AI.
  2. The gaps on one list, each with what it costs the program.
  3. An order-of-magnitude cost to build, and to run each year.
  4. The next steps in order, with the measure each should move.
  5. After the first build, a recommendation to fund, scale, revise or stop.

Lines 1 to 4 come from the two-week sprint in Box A. Line 5 comes from the first build that follows, measured against a number the program already reports.

Asked by agency leaders, IT and counsel

Can you work within our procurement rules?

We set out a first engagement with a defined scope, deliverables and price, so your procurement office can weigh it against your purchasing rules. Before any work begins, we work with procurement on the competitive process, the contracting route, the vendor documentation and the approvals those rules require. Your rules decide the route, not the size of the engagement.

What will our procurement office need from you?

What we set out in a proposal: the scope, deliverables and price; who is responsible for security and data handling; our accessibility commitments; any subcontractors; and how the work is handed over at the end. If your process asks for something else, tell us at the start and we will say plainly whether we can meet it.

Will resident data leave our environment?

It does not have to. We start from your data classification, your approved environments and your security plan, and where they require it, covered data stays in agency-approved infrastructure with access, processing, retention and vendor controls set out. Before anything is deployed, every data flow is written down, logs and backups included, with the approvals each needs. Training a model on resident data, using it to answer a question, and keeping prompts and outputs are three separate decisions, each yours. Where the agency holds criminal justice, federal tax or health data, CJIS, Publication 1075 or HIPAA apply to everyone who touches it, us included.

Our case system is decades old. Does it have to be replaced first?

Usually not. We first check its interfaces, its data quality and the security restrictions on it, and if reading from it will not work, we say so before anything is built. Where it will, we read from what you run, including mainframe extracts, model the data once, and leave the system of record where it is, so replacement can follow on its own schedule and budget.

Do we have to tell residents when they are dealing with AI?

In Texas, for AI systems meant to interact with residents, yes, before or at the start of the interaction. Elsewhere it depends on state law, the use and agency policy, and most agencies choose to say so anyway. We design the disclosure, and the way to reach a person, into the tool.

Are AI prompts and outputs public records?

Often, yes, though it turns on your state’s law and what they were used for. In many states, material made or received in the course of public business is a record whatever tool produced it. We set what is kept, and for how long, with your records officer before anything goes live.

How do we meet the accessibility deadline for dashboards and documents?

Start with what residents use most: the top pages, forms, PDFs and dashboards. Equal access is already required; the rule sets the technical standard from April 26, 2027 for governments of 50,000 people or more and April 26, 2028 for smaller ones. We rebuild dashboards to WCAG 2.1 Level AA, test each one with a keyboard and a screen reader, in a sensible focus order, through the whole journey a resident takes, add text alternatives for charts, and give the data behind each chart a download.

Do you work with cities and counties as well as states?

Yes. One of the two public engagements on this page is a major city’s 311 service, the other a state government.

Can a small team keep it running after you leave?

That is the plan from the first week: documentation, training by role, and a handover to named owners, with the tools built on what your team already supports.

Bring the request that waits longest in your queue.

The question 311 hears all day, the application that sits for weeks, the report the legislature keeps asking for, or the AI tool staff already use. Thirty minutes with the person who would lead the work, on what to change first.

Book 30 minutes for your agency

A queue that keeps growing? Thirty minutes on the call line, the caseload or the AI policy.

Book a call for your agency