AI Pilot Planning Canvas
Start with a real workflow. Keep human judgment visible. Measure the whole task.
Your copy of the canvas.
Enter your name and email. We will send a verification link to unlock the download. Two A4 pages. Twelve practical prompts.
Start with the task. Then choose the tool.
- Choose one recurring task and identify who reviews the result.
- Complete page one before testing, including decisions and data boundaries.
- Use page two to review the evidence and decide what comes next.
Print and write in the spaces. This PDF has no interactive form fields.
What is inside?
Task and accountable owner
Name one recurring task, its users and the person accountable for the pilot. Who reviews the result?
Current workflow
Map the input, key steps, review and handoff. Where are information, time or context lost?
Scope and decision boundary
What may the tool assist with? What stays with a person? Write one explicit out-of-scope use.
Permitted data
List approved sources and information that must not be shared. Name who checks access and retention requirements.
Baseline before the tool
Record current total effort, including review and rework. Note sample size, errors and incomplete handoffs.
Human review and escalation
What must be checked? Who handles uncertainty or a harmful result? Who can pause the pilot?
Trial plan
Choose a small set of cases, reviewers and a trial period. Record any approvals needed before starting.
Evidence log
For each case, record a non-personal reference, baseline effort, assisted effort including review, and observed issues. Use a separate sheet if needed.
Success criteria
Set a quality threshold and an acceptable total effort before testing. Which errors would outweigh a time saving?
Stop or pause conditions
Write the trigger, responsible person and immediate action. Include unreliable output, unsafe data use or a changed workflow.
Decision and its limits
Continue, adjust or stop? Link the decision to observed evidence. What does the small sample still leave uncertain?
Next review and learning
Name the next owner and review date. What changed in the workflow, and what did the team learn even if the pilot stopped?
Read the thinking behind the canvas
Companion to: The Best AI Pilot Begins Before the Model