AI pilot planning worksheet

Test one workflow and its AI agent action boundaries, human approvals, pause, revocation and recovery.

For
Process owners and teams moving from AI experiments to daily use.
You leave with
A bounded pilot plan with a baseline, review responsibilities and a scale, redesign or stop decision.

Choose your format

DOCX is editable in Word or a compatible editor. PDF is ready to print. No signup is required.

Inside this template

The full document follows these sections. The first prompts are shown below so you can check the fit before downloading.

Choose the work

Define the workflow problem before selecting the tool.

Problem and desired outcome

Describe the task, current friction and the decision the pilot should support. Who benefits if it works?

Current workflow and baseline

Map the main steps and record current time, quality, corrections and review effort. State how the baseline was measured.

Participants and ownership

Name the business owner, pilot users, reviewer and the person making the final decision.

Boundaries

Define what the pilot includes and excludes, its duration and which data and tools are approved.

Design a fair test

Measure a useful outcome, including the work needed to review and correct it.

Proposed workflow

Describe where AI contributes, what stays with a person and what happens if the tool is unavailable.

Quality and human review

Define what must be checked, which source material the reviewer sees and when to correct, reject or escalate.

Success measures

Choose measurable targets for quality, total time including review, risk and actual use. Record a comparison method and sample.

Stop conditions and support

List unacceptable failures, the person who can pause the pilot and how users get training or help.

Review the evidence and decide

Complete this page at the pilot review, using observed results rather than impressions alone.

Results against the baseline

Record the sample tested, results for each measure and limitations of the evidence.

User and reviewer feedback

What helped, what caused friction and what work moved elsewhere? Include corrections and near misses.

Decision

Choose: scale, redesign, pause or stop. Record the reasons, approver and decision date.

Next actions and review date

Name owners and dates for process changes, guidance, training and the next review. Keep a record of what changed.

Benefits evidence and portfolio decision

Link this review to a stable use-case ID. Estimated benefits are not measured outcomes.

Baseline and benefits

Compare total work plus review effort, quality, usefulness and costs with the manual baseline. Record method, period and evidence.

Data and dependencies

Review data-quality gaps, access, integrations, suppliers and capacity. Name owners and readiness conditions.

Portfolio decision

Choose pilot, continue, change, scale or retire. Record evidence, rationale, authorised approver and conditions.

Handoff and follow-up

Update the use-case register. Assign implementation, supplier, training and incident-learning actions, with owners and review dates.

Action boundaries and recovery tests

Run in the approved test environment. Record expected and observed results, evidence, owner and date for each test.

Allowed and prohibited actions

Test permitted actions and attempts outside scope. In the fictional support example, read approved guidance and draft; verify that sending and customer-record changes are denied.

Approval boundary

Test the required approval and its absence. Change the proposed action, recipient or record after approval and check that it cannot execute under the old approval. Name the authorised reviewer.

Pause and delegated-access revocation

Pause new work and revoke the agent’s delegated access. Check queued and in-flight actions and retained logs. Record actual effects and any delay before access stops.

Recovery and reauthorisation

Validate restored state and manual fallback. Check uncertain outcomes before retrying to avoid duplicate sends or updates. Name who authorises restart; preserve failures, corrections and retest evidence.

Need clearer footing for an AI decision?

Start with a focused conversation about a live AI use case, workflow bottleneck, training need, or governance gap.

Contact

Talk to Ada Studio

Have an AI adoption question? Send Adriana a short message.