Copy the blank fields into your existing document or work system, then complete them with the people responsible for the workflow. The examples below are fictional. This page does not save your answers.
Purpose and limit
Use this worksheet to choose one bounded AI workflow candidate, identify a required redesign, or record a no-AI decision. It does not produce a maturity score.
This worksheet provides general operational education. It is not legal advice or a compliance determination. No attorney-client relationship exists, and none of its protections apply.
Who should complete it
- Workflow owner
- Qualified reviewer
- System or data owner when sensitive information or connected systems are involved
Inputs
- Current workflow steps and owner
- Sample inputs and outputs that may be used for testing
- Approved tool, account, and data-handling facts
- Current baseline for quality, time, rework, and exceptions
Part 1: Define the workflow
| Field | Response |
|---|---|
| Workflow name | |
| Trigger | |
| Current inputs | |
| Current output and destination | |
| Process owner | |
| Person affected by an error | |
| Current baseline | |
| Proposed AI role | |
| Actions the AI must never take |
Part 2: Run the selection screen
For each row, record evidence or an open question. Do not convert the answers into a numeric score.
| Question | Evidence or open question | Route if unresolved |
|---|---|---|
| Is the outcome specific and observable? | Redesign the workflow definition. | |
| Is the work recurring and testable with representative examples? | Keep manual or gather examples. | |
| Are the AI inputs, outputs, and permissions bounded? | Narrow the role. | |
| Can a qualified person verify the result from authoritative sources? | Do not rely on the output. | |
| Are the tool, account, settings, and data use approved? | Stop before data entry. | |
| Can a bad result be contained before an external or irreversible action? | Add a gate or choose no AI. | |
| Is a named owner accountable for the workflow? | Assign ownership before a pilot. | |
| Is the expected value greater than review, correction, and operating cost? | Keep the current method. |
Part 3: Place the review checkpoints
| Checkpoint | Reviewer | Evidence shown | Decision | If unavailable |
|---|---|---|---|---|
| Before sensitive access | allow, narrow, or stop | |||
| Before reliance or external action | approve, correct, reject, or escalate | |||
| After operation through sampling | continue, change, pause, or retire |
Part 4: Choose the route
Select one route and state why.
- [ ] Bounded pilot: task, tool, data, reviewer, test, and containment plan are clear.
- [ ] Redesign first: a promising use remains, but a named boundary is unresolved.
- [ ] No AI for this decision: the proposed role would replace judgment or create an unacceptable consequence.
Decision and rationale:
Named owner:
Next action and due date:
Synthetic completed example: candidate requiring confirmation
This example is fictional.
| Field | Completed response |
|---|---|
| Workflow name | Contract-intake routing |
| Trigger | A contract request arrives through the approved intake form. |
| Current inputs | Request form, document type, business unit, requested date, and attached draft. |
| Current output and destination | Draft intake record in the legal queue. |
| Process owner | Legal operations manager. |
| Person affected by an error | Requesting business owner and assigned legal reviewer. |
| Current baseline | Measure routing corrections, missing fields, elapsed time, and reviewer effort for two weeks. |
| Proposed AI role | Extract form fields and suggest a queue from an approved list. |
| Actions the AI must never take | Give legal advice, set priority alone, send external messages, or accept terms. |
| Selection question | Evidence or open question | Disposition |
|---|---|---|
| Specific outcome | A draft queue record with linked source fields. | Pass for pilot design. |
| Recurring and testable | Synthetic cases represent expected routine and exception paths. | Pass after test-set review. |
| Bounded role | Suggestion only, with no external action. | Pass. |
| Verifiable | Intake coordinator checks every field against the form. | Pass. |
| Approved data and tool | Use only the approved environment and approved test data. | Must be confirmed before use. |
| Containment | Nothing posts until the coordinator confirms it. | Pass. |
Route: Redesign first until the approved environment and test set are confirmed. Then run a bounded pilot.
Next action: The legal operations manager confirms the approved environment and prepares representative synthetic cases.
Synthetic completed example: no-AI candidate
This example is fictional.
| Field | Completed response |
|---|---|
| Workflow name | Final settlement acceptance decision |
| Proposed AI role | Choose whether to accept a settlement and communicate acceptance. |
| Verification limit | The decision depends on client objectives, legal judgment, and current facts. |
| Consequence | Acceptance is externally consequential and may be difficult to reverse. |
| Containment | A post-action review would occur after the material decision. |
Route: No AI for this decision.
Next action: Keep the decision with the responsible lawyer and client. Evaluate separate support tasks only under their own bounded worksheets.
Review and maintenance
Reopen this worksheet when the task, tool, model, data, permissions, reviewer, destination, or failure pattern changes. Preserve the prior decision and date.
Before the first worksheet, the free AI readiness assessment returns your priority gaps, a reasonable place to look for the first candidate.
Source basis
- NIST AI RMF 1.0, accessed September 18, 2026.
- NIST Generative AI Profile, accessed September 18, 2026.
- ABA Formal Opinion 512, accessed September 19, 2026.
