Writing AI Use Cases for Your Agency

The Use Case Inventory Requirement

OMB M-25-21 requires every federal agency to maintain a public inventory of AI use cases. This is not optional, and it is not just a paperwork exercise. The inventory serves three purposes: transparency to the public, internal governance oversight, and accountability for how AI is actually being used in government operations.

The inventory must be submitted to OMB and published publicly. Most agencies post theirs on their website under an "AI Inventory" or "AI Use Cases" section. If you want to see what a completed inventory looks like, the GSA, HHS, and DOD all have published examples you can reference.

Who contributes to the inventory? Program offices, not just IT. If your office is using or planning to use an AI tool to support your mission, you are responsible for providing the use case entry. Your CAIO or CIO shop will compile and submit the inventory, but the content has to come from people who actually know the programs.

This is why you need to know how to write one.

What a Use Case Entry Must Contain

OMB specifies the minimum required fields for a use case inventory entry. Here is what you need to document:

The Use Case Template

Use this template to draft your entries. You can adapt the language but keep all required fields present:

USE CASE NAME: [Plain English description of what the AI does]
  Example: "Meeting Summary Generation for Policy Staff"
  NOT: "ChatGPT" or "AI Assistant"

SUMMARY DESCRIPTION:
  [Agency/Office] uses [AI tool/platform] to [specific function].
  The system takes [input data type] and produces [output type].
  This capability supports [mission function] by [specific benefit].

BUSINESS OWNER: [Office name, GS-level of responsible official]

AI TECHNOLOGY: [Tool name + version/model]
  FedRAMP Status: [Authorized / In Process / Not Applicable]
  Authorization Level: [Moderate / High / IL4 / etc.]

DATA USED:
  Data Type: [Description of data inputs]
  Classification: [Unclassified / CUI / IL4 / etc.]
  Contains PII: [Yes/No — if yes, specify type]
  Data Source: [Internal system / External / Both]

STAGE: [Initiated / In Development / Deployed / Retired]

IMPACT ASSESSMENT: [High / Medium / Low]
  Rights or Safety Impact: [Yes/No — explain if Yes]
  Significant Government Decision: [Yes/No — explain if Yes]

HUMAN OVERSIGHT:
  Review Mechanism: [How humans review AI outputs before use]
  Override Capability: [How humans can override AI recommendation]
  Responsible Role: [Title of person responsible for oversight]

BENEFITS:
  [Specific, quantified where possible: "Reduces meeting summary
  drafting time from 45 minutes to 5 minutes per meeting,
  freeing analyst time for substantive policy work."]

KNOWN RISKS AND MITIGATIONS:
  Risk: [Specific risk]
  Mitigation: [Specific control or procedure]

Risk Assessment Framework for Government AI

Not all AI use cases carry the same risk. OMB requires that you assess the impact level of your use case — specifically whether it affects individual rights, public safety, or significant government decisions. Here is a practical framework for making that determination:

High Impact Use Cases (require enhanced review)

Any AI system that directly influences decisions about: benefits eligibility, law enforcement actions, immigration status, credit or housing, employment, education, healthcare, criminal justice, or other rights-affecting determinations. These require mandatory human review before any action, detailed documentation, and regular audits.

Medium Impact Use Cases

AI systems that support internal government operations, process internal data, or assist with drafting and analysis — but where a human makes all final decisions and the AI output has no direct external effect. Most document processing, internal communications, and knowledge management tools fall here.

Low Impact Use Cases

Administrative tools, productivity assistance, internal search and retrieval, meeting notes, scheduling assistance. These have limited governance requirements but still need a human oversight description in the inventory.

Human Oversight: The Non-Negotiable Requirement

Every use case in the federal AI inventory must describe how humans maintain oversight of the AI system. "A human reviews the output" is not sufficient. You need to specify:

• Who reviews the output (by title/role, not by name)
• What they review it for — what checks are they making?
• When in the workflow the review occurs — before the output is acted on?
• How they can override or reject the AI output if it is wrong

Key Takeaways from Day 3

• Every federal agency must maintain a public AI use case inventory per OMB M-25-21. Program offices provide the content — this is your responsibility, not just IT's.
• A complete use case entry includes: name, description, business owner, technology, data, stage, impact level, human oversight mechanism, benefits, and risks.
• Impact assessment determines governance requirements. High-impact use cases require enhanced review and mandatory human oversight before any action.
• Human oversight must be specific: who reviews, for what, when, and how they can override the AI. Vague statements are not acceptable.
• You now have 3 draft use case entries. These are ready to share with your CAIO or CIO's office.