AI RMF Playbook: Map
OUR PICK FORFraming a new use case before selecting a solution
Suggested actions for documenting intended purpose, context, stakeholders and risk at the start of an AI use case.
Visit originalCapture the purpose, owner and likely impact before work begins.
Sources reviewed through
View 4 resourcesRECOMMENDED RESOURCES
Compare type, format and framework. Every link opens the original publisher.
OUR PICK FORFraming a new use case before selecting a solution
Suggested actions for documenting intended purpose, context, stakeholders and risk at the start of an AI use case.
Visit originalOUR PICK FORDeeper screening for automated decisions
A questionnaire covering business process, data, system design, algorithm and decision context.
Visit originalOUR PICK FORAdding ethical review prompts to intake
Principles and activities for responsible development, procurement and use, including early documentation of decisions.
Visit originalOUR PICK FORA lightweight form to adapt
An editable open repository form for an initial AI use-case gate and risk register entry.
Visit originalNo resources match these filters. Clear filters to see the full collection.
This is our suggested workflow for an organization reviewing a proposed use, not an online submission service or a mandatory NIST sequence. Adapt responsibilities to your policy and existing approval routes. Use the NIST AI RMF Playbook: Map for context on intended purpose, affected stakeholders and evaluation; the steps below turn those questions into a practical handoff.
1. Submit the proposed AI use case → 2. Screen the request and select a review route → 3. Assess risks and test the proposed controls → 4. Approve, restrict, return or reject the request → 5. Register the decision and hand over ownership
Owner: the business sponsor or service owner submits the request, with help from the people who will operate the system. Describe the task and the intended benefit before choosing a product. “Use a chatbot” is a solution preference; “help support agents draft replies for human review” is a scope a reviewer can examine. Include the current manual approach and why a change is worth evaluating.
Documents: submit a short intake form, a process sketch and links to the proposed product or internal design. Identify users, affected people, data categories, integrations and intended autonomy. Use synthetic examples if real records would expose confidential information. State unknowns openly, assign someone to resolve them and give the request a stable ID that can follow it through review.
Common blocker: nobody accepts ownership, or the request combines several unrelated uses. Return it with the specific missing information and a named contact; split materially different purposes into separate records. The output is a complete enough proposal for screening, not permission to upload data or begin a trial. A free account or existing software contract does not remove this gate in the suggested process.
Back to page sections ↑Owner: the intake coordinator, governance lead or designated reviewer checks completeness and routes the request. First confirm whether the organization already has an approved use covering the same purpose, data and configuration. Reuse a relevant decision by reference, while documenting any difference; do not assume that approval of a vendor covers every use of its products.
Documents: add a screening note identifying the proposed risk route, required specialists and questions to resolve. Consider sensitive information, decisions affecting people, external communications, security access and the ability to take actions. These signals guide review depth; they are not a universal legal classification. Record why a simplified route is sufficient or why more assessment is necessary.
Common blocker: the requester describes a low-risk drafting tool while the integration can send messages or alter records automatically. Resolve the actual permissions before routing. If a supplier is involved, start AI vendor assessment early enough to request product-specific evidence. If purchasing or a contract change is needed, notify the AI procurement owner so review and commercial work share the same scope and ID.
Back to page sections ↑Owner: the service owner coordinates the assessment; privacy, security, technical and other relevant specialists make findings within their responsibilities. Describe potential failures and who could be affected. For a drafting assistant, assess incorrect statements, confidential information in prompts, inappropriate replies and whether staff can recognize and correct an error before sending. Keep likelihood estimates and unresolved uncertainty visible.
Documents: prepare a risk and evidence log, data-flow description, evaluation plan and proposed oversight arrangements. Link evidence to the actual product, configuration and intended use. Use AI vendor assessment to record what a supplier claims, what evidence supports it and what remains unverified. Test important failure cases against agreed criteria; a generic demonstration cannot establish fitness for your workflow.
Common blocker: a critical claim has no evidence, or a safeguard exists only as an intention. Assign the gap to an owner and define the evidence needed to close it. Reduce the pilot scope if that makes the remaining uncertainty acceptable to the responsible authority; otherwise hold the request. The output is a recommendation with conditions and open issues, not a score that automatically grants approval.
Back to page sections ↑Owner: the authority named in your policy makes the decision using the proposal and review findings. The requester should not silently approve their own unresolved risks. Escalate questions outside the approver’s authority to the relevant decision owner, using an AI governance committee where an existing cross-functional decision route is appropriate.
Documents: retain a dated decision stating the permitted purpose, users, data, environment and autonomy. Distinguish a limited pilot from operational deployment. Name the approver, conditions, expiry or next review, monitoring owner and stop authority. Link the evidence considered and explain accepted gaps. For purchased tools, pass conditions into AI procurement so the contract, access configuration and acceptance criteria reflect the same decision.
Common blocker: “approved with conditions” leaves nobody responsible for meeting those conditions. State which actions must close before use and which may remain under monitoring, with dates and owners. A decision returned for changes should identify exactly what must be resubmitted. Approval is complete only when its scope and conditions are understandable to the people configuring and operating the system.
Back to page sections ↑Owner: the inventory custodian creates or updates the AI system inventory record, and the service owner accepts ongoing responsibility. Link the intake ID, system ID, approval and evidence instead of maintaining disconnected copies. Keep proposed, pilot, approved, rejected and retired statuses distinct so a recorded request cannot be mistaken for a live authorization.
Documents: retain the decision reference, product and configuration, permitted use, accountable owner, conditions, review date and material-change triggers. Store sensitive supporting records in the appropriate restricted system; the register can contain references rather than confidential documents. Record where staff report incidents and who can suspend the use. Set a practical review trigger for changed data, purpose, supplier behavior or autonomy.
Common blocker: the project closes but nobody maintains the record or monitors conditions. Make the operational handoff explicit before closing intake. For example, in a fictional support-assistant pilot, the support lead accepts monthly checks and IT owns access removal; those assignments still need confirmation in a real project. Reopen the relevant review when scope changes rather than treating the original approval as permanent.
Back to page sections ↑Copy these suggested fields into your existing form or document; record unknowns and owners rather than inventing answers.
| Field | What to enter | Why it is needed |
|---|---|---|
| Request ID and date | A stable identifier and submission date. | Connect reviews and prevent duplicate requests. |
| Business owner | Named accountable person and team. | Give reviewers a decision contact and ongoing owner. |
| Purpose and outcome | The task, expected benefit and success measure. | Evaluate the need and define a useful test. |
| Current alternative | How the task works today, including a non-AI option. | Compare benefit, effort and failure consequences. |
| Users and affected people | Operators and people affected by outputs. | Identify who needs consultation or safeguards. |
| Product and configuration | Provider, model or version, deployment and settings. | Bind evidence to the system actually proposed. |
| Data categories | Inputs, outputs and sensitive or confidential categories. | Route data handling and access review. |
| Data flow and retention | Where information goes, who accesses it and retention. | Expose dependencies and unresolved data questions. |
| Autonomy and integrations | Read, draft, send, write or payment permissions. | Separate assistance from consequential actions. |
| Expected failures | Plausible errors and their impact on people or operations. | Prioritize evaluation and risk treatment. |
| Evidence and test plan | Supplier references, test cases and acceptance criteria. | Distinguish claims from supported findings. |
| Human oversight | Reviewer, intervention point and stop mechanism. | Make control responsibilities testable. |
| Reviewers and approval | Required specialists, decision owner and status. | Route work without implying premature permission. |
| Conditions and review trigger | Restrictions, action owners, dates and change events. | Support follow-up and reassessment after approval. |
| Inventory reference | System ID or a pending registration action. | Connect intake with the maintained register. |
A short, repeatable request for a proposed AI use case, its context, data and expected outcome.
Business owners, product teams and governance reviewers.
Before a pilot, purchase or deployment.
OFFICIAL GUIDANCE
These sources explain the underlying expectations; the practical templates help teams apply them.
COMPARE RESOURCES
Start with a short project form. Add an impact assessment when the intended decision or affected people raise the stakes.