AI use cases in the public sector: 12 systems agencies build in 2026

The AI use cases government agencies build first are resident contact center agents, benefits application intake, public records redaction, permit intake and code modernization for aging benefit and tax systems. Under OMB Memorandum M-25-21 for federal agencies, and similar policies in many states, any of these that affects rights, benefits or safety counts as high-impact and needs testing, monitoring and human oversight before and after launch.

This guide walks through twelve systems for federal, state and local agencies and public authorities, grouped into resident services, casework and records, and operations. Each entry covers the records used, the systems touched, the risks to manage and an illustrative staffing plan. Ryz serves enterprises and public organizations across industries. Our AI pod teams build in your cloud tenancy and repos, working on your team with agency IT, security and program staff.

Reading the numbers. Staffing plans and durations are illustrative planning assumptions, not quotes or bids. Costs rely on Ryz's public rates alone: mid-level $7,000 to $10,000, senior $10,000 to $15,000 and tech lead $15,000 and up per engineer per month, quoted per team, with the lead at its $15,000 floor.

The policy frame for government AI

• OMB M-25-21. Issued April 3, 2025, M-25-21 replaced M-24-10. Federal agencies must name a Chief AI Officer, publish an annual AI use case inventory and apply minimum risk management practices to high-impact AI, including pre-deployment testing, impact assessments, ongoing monitoring and human oversight, or stop using that functionality.
• NIST AI RMF. Many state and local policies point to the NIST AI Risk Management Framework for how to govern, map, measure and manage AI risk.
• Accessibility. The Department of Justice's ADA Title II web rule requires state and local government web content and mobile apps to meet WCAG 2.1 Level AA. After an April 2026 interim final rule, compliance dates are April 26, 2027 for entities serving 50,000 or more people and April 26, 2028 for the rest and special districts. Chat and voice agents are part of that surface.
• Hosting and data. Federal systems generally need FedRAMP-authorized cloud services at the right impact level, and criminal justice data falls under the FBI's CJIS Security Policy. Check the authorization of each model endpoint, not just the cloud.

Resident services

1. 311 and benefits contact center agent

A chat and voice agent answers the questions that fill call queues: office hours, application status, required documents, how to report a pothole or missed trash pickup. It opens service requests directly and hands complex or upset callers to staff with the conversation attached.

• Records used: knowledge articles, service request history, application status data, call transcripts.
• Systems touched: CRM or 311 platform (Salesforce Public Sector, ServiceNow, Microsoft Dynamics 365), telephony, case systems.
• Risks to manage: wrong eligibility guidance, language access for residents with limited English, and accessibility for screen reader and TTY users. See AI voice agent development.
• Illustrative staffing: tech lead plus three senior engineers, five to seven months: $45,000 to $60,000 a month, $225,000 to $420,000 in total.

2. Benefits application intake and document checks

Applicants upload pay stubs, leases, IDs and medical forms. The system classifies each document, extracts the fields eligibility workers need, flags missing or expired items and pre-fills the case, so workers spend their time on decisions rather than data entry.

• Records used: uploaded documents, program rules, labeled examples from past cases.
• Systems touched: integrated eligibility system, document imaging, applicant portal.
• Risks to manage: this sits next to a benefits decision, so it is high-impact territory. The worker makes the eligibility determination, applicants keep their notice and appeal rights, and error rates are tested across languages and document types. See intelligent document processing.
• Illustrative staffing: tech lead plus three senior engineers, five to seven months: $45,000 to $60,000 a month, $225,000 to $420,000 in total.

3. Plain-language and translated notices

Agency notices are often written at a level few residents read comfortably. The system rewrites notices in plain language and drafts translations into the agency's required languages, with legal and program staff approving every template before it goes out.

• Records used: notice templates, statutory language that must stay verbatim, approved glossaries per language.
• Systems touched: correspondence generation, content management, case systems.
• Risks to manage: legally required wording that cannot be paraphrased, and translation quality in less common languages, which needs native-speaker review.
• Illustrative staffing: tech lead plus one senior and one mid-level engineer, three to four months: $32,000 to $40,000 a month, $96,000 to $160,000 in total.

Casework and records

4. Public records request processing and redaction

For FOIA and state public records requests, the system searches email and document stores for responsive records, removes duplicates, proposes redactions for personal data and exempt content with the exemption cited, and builds the production set for a records officer to approve.

• Records used: email archives, document repositories, exemption rules, past redaction decisions.
• Systems touched: records request platforms (GovQA, NextRequest), Microsoft 365 or Google Workspace, eDiscovery tools.
• Risks to manage: under-redaction is a privacy incident and over-redaction invites appeals, so the officer reviews each proposed redaction. See PII redaction.
• Illustrative staffing: tech lead plus two senior engineers, four to five months: $35,000 to $45,000 a month, $140,000 to $225,000 in total.

5. Social services case file summaries

Caseworkers inherit files with years of notes. The system produces a summary of household members, history, services, court dates and open safety concerns, each linked to the source note, so a new worker gets oriented in minutes rather than days.

• Records used: case notes, assessments, service plans, court documents.
• Systems touched: child welfare or case management systems, document management.
• Risks to manage: highly sensitive data with strict confidentiality rules, and summaries that must surface conflicts and uncertainty rather than smooth them away. The model never recommends removal or placement.
• Illustrative staffing: tech lead plus two senior engineers, three to five months: $35,000 to $45,000 a month, $105,000 to $225,000 in total.

6. Permit application intake and plan pre-check

The system checks building and land use applications for completeness, extracts project data, compares submitted plans against a checklist of zoning and code items and drafts the correction letter, so plan reviewers start from a clean package.

• Records used: applications, plan sets, zoning code, past correction letters.
• Systems touched: permitting platforms (Accela, Tyler EnerGov), GIS parcel data, document review tools.
• Risks to manage: plan drawings are hard for models to read reliably, and code interpretation belongs to the reviewer. Start with completeness checks, which are lower risk and still save time.
• Illustrative staffing: tech lead plus three senior engineers, four to six months: $45,000 to $60,000 a month, $180,000 to $360,000 in total.

7. Solicitation drafting and proposal evaluation support

Procurement staff get help assembling solicitations from approved clauses, and evaluators get a structured comparison of proposals against the stated evaluation criteria, with page references, while scoring stays with the evaluation panel.

• Records used: clause libraries, past solicitations, proposals, evaluation criteria.
• Systems touched: e-procurement and contract writing systems, document management.
• Risks to manage: procurement integrity and bid protests. Every comparison must trace to the proposal text, and nothing outside the stated criteria can enter. See AI agents for procurement.
• Illustrative staffing: tech lead plus two senior engineers, three to four months: $35,000 to $45,000 a month, $105,000 to $180,000 in total.

Operations and modernization

8. COBOL and mainframe modernization

Unemployment insurance, tax and motor vehicle systems often run on COBOL, CICS and DB2. AI-assisted analysis maps the business rules buried in the code, generates documentation and tests, and helps engineers rebuild services as APIs in Python or Java while the old system keeps running.

• Records used: source code, copybooks, JCL, batch schedules, production data samples.
• Systems touched: mainframe environment, new cloud services, CI/CD.
• Risks to manage: undocumented rules that only show up in month-end batches, and parallel runs that must match to the cent before cutover. See mainframe modernization.
• Illustrative staffing: tech lead plus five senior engineers, nine to twelve months: $65,000 to $90,000 a month, $585,000 to $1,080,000 in total.

9. Improper payment and vendor fraud detection

Models score benefit claims and vendor invoices for patterns such as shared payment accounts, impossible addresses, duplicate invoices and identity anomalies, and send investigators a referral with the evidence. A Ryz AI pod built fraud detection for a global fleet company that scored 244K+ invoices in under 30 seconds each and surfaced $5.94M in fraud confirmed by the client's fraud team (case study).

• Records used: claims and payment data, vendor master data, confirmed fraud outcomes.
• Systems touched: financial system (Tyler Munis, Oracle, SAP), program integrity case management.
• Risks to manage: a score must not stop a payment or benefit without a human decision and due process, and disparate error rates across groups need testing.
• Illustrative staffing: tech lead plus three senior engineers, five to seven months: $45,000 to $60,000 a month, $225,000 to $420,000 in total.

10. Road and asset condition from vehicle imagery

Cameras on garbage trucks or fleet vehicles capture street imagery; a vision model detects potholes, faded markings, damaged signs and blocked storm drains, geotags them and creates work orders after a supervisor confirms.

• Records used: geotagged images, asset inventory, work order history.
• Systems touched: GIS, work and asset management (Cityworks, Maximo), fleet telematics.
• Risks to manage: faces and license plates in public imagery need blurring and a retention policy before anything else.
• Illustrative staffing: tech lead plus two senior engineers and one mid-level engineer, four to six months: $42,000 to $55,000 a month, $168,000 to $330,000 in total.

11. Call transcription and quality review for dispatch centers

Recorded calls in emergency communication centers are transcribed and checked against protocol for quality assurance reviews, and non-emergency lines get faster routing. The model never handles a live emergency call on its own.

• Records used: call recordings, protocol cards, QA scoring forms.
• Systems touched: call recording, CAD, QA tools.
• Risks to manage: CJIS requirements for any criminal justice information, and accuracy with noisy audio and many languages. See speech-to-text.
• Illustrative staffing: tech lead plus two senior engineers, three to five months: $35,000 to $45,000 a month, $105,000 to $225,000 in total.

12. Policy and legislative research assistant

Staff ask how a bill would change an existing statute or which regulations cite a section. The assistant searches the code, regulations, bill text and agency guidance, and shows a redline with citations for analysts to check.

• Records used: statutes, administrative code, bill text, fiscal notes, internal guidance.
• Systems touched: legislative information systems, document management.
• Risks to manage: version control across sessions and amendments. Analysts own the conclusions.
• Illustrative staffing: tech lead plus one senior engineer, two to three months: $25,000 to $30,000 a month, $50,000 to $90,000 in total.

How to pick your first use case

• Start below the high-impact line. Records search, notice rewriting, research assistants and modernization help staff without deciding anything about a resident, which keeps the first project out of the heaviest review.
• Write the inventory entry first. Purpose, data, impact determination and oversight plan take a page, and you will need them for the use case inventory anyway.
• Measure staff time and backlog. Public sector wins are usually measured in backlog days and wait times. Baseline them before the pilot.
• Plan for accessibility and language access. Build WCAG 2.1 AA and multilingual support into the first release instead of retrofitting.
• Score readiness honestly. Our AI readiness scorecard covers data access, ownership and governance.

What it costs

Multiply the number of engineers by the months and by each engineer's monthly rate. Ryz charges $7,000 to $10,000 for mid-level, $10,000 to $15,000 for senior and $15,000 and up for tech leads, quoted per team.

• Research or notice pilot (tech lead plus one senior engineer) for three months: $25,000 to $30,000 a month, $75,000 to $90,000 in total.
• Contact center pod (tech lead plus three senior engineers) for six months: $45,000 to $60,000 a month, $270,000 to $360,000.
• Full AI pod of about seven senior engineers including a tech lead and an ML engineer: $75,000 to $105,000 a month, the usual size for modernization programs.

Cloud, model usage and telephony run in the agency's own accounts and are not included. See AI development cost for a wider budget view. You meet the named engineers before work starts, and a team can usually start in under two weeks once you say go. If your procurement requires a prime integrator with specific clearances or set-aside status, use one; Ryz teams fit agencies and authorities that can bring engineers onto their own projects.

Related

• AI legacy modernization
• Hire mainframe developers
• AI chatbot development
• AI governance
• AI use cases in education
• AI use cases in energy and utilities

FAQ

What are the best AI use cases for government?

The most common production systems are contact center agents, benefits document intake, public records search and redaction, permit completeness checks, case summaries, legacy code modernization, improper payment detection and research assistants for policy staff.

What does OMB M-25-21 require for high-impact AI?

Federal agencies must apply minimum risk management practices to high-impact uses, including pre-deployment testing, an AI impact assessment, ongoing monitoring, adequate human oversight and training, and remedies for affected people. If a high-impact use cannot meet them, the agency must stop using it.

Can government agencies use commercial large language models?

Yes, through cloud services authorized for the data involved, such as FedRAMP-authorized offerings at the right impact level for federal work, with data handling terms that prevent training on agency data. State and local rules vary, so check your own policy.

Do AI chat and voice agents have to meet ADA requirements?

For state and local governments, web and mobile content, including chat interfaces, falls under the ADA Title II rule and WCAG 2.1 AA. Voice agents also need a clear route to a person and support for TTY and relay services.

How do Ryz engineers work with an agency?

They join the agency's project team, work in the agency's repos and cloud tenancy under its security controls, and ship in its environment. The agency controls access throughout and owns all code and IP.

Questions we didn't answer? Email info@ryzlabs.com.