Pilot Card Builder

---
name: pilot-card-builder
description: Turn an open-ended AI pilot idea into a one-page, time-boxed test with a real decision, a success signal, and a stop point. Use when a manager is about to start or approve an AI pilot on their team.
license: All-Rights-Reserved
metadata:
  author: Zain Haseeb
  version: 1.0.1
  changed: "2026-09-08"
  category: ai-leadership
  tags: [pilots, management, scoping, ai-adoption]
---

# Pilot Card Builder

Convert an enthusiastic but vague AI pilot proposal into a bounded, one-page test that has a real end date, a real measure, and a named person who can stop it.

## Critical behavior

1. Name the specific decision the pilot exists to inform, not the general hope behind it.
2. Set a fixed start and end date before the pilot begins. A pilot without an end date is a quiet permanent adoption.
3. Define one success signal and one failure signal, both observable before the window closes.
4. Name what stays out of scope, including systems and processes the pilot must not touch.
5. Name who can pause or cancel the pilot and under what condition.
6. Do not invent adoption numbers, sentiment, or savings the manager did not supply.

## When to use

Use this before starting, approving, or reporting on an AI pilot on a team. It is especially useful when a pilot idea arrived from a vendor demo, a conference talk, or a peer's success story rather than from a named problem on the team.

Do not use it to add process to a tool a person is simply trying out on their own work with no team, budget, or customer impact.

## Build the card

### 1. Find the decision it informs

State in one sentence what decision the organization will be able to make once the pilot ends that it cannot make now. Reject pilots framed only as "try the tool" with no attached decision.

### 2. Set the window

Pick a start date, an end date, and the review date immediately after. A pilot with no end date should be rejected or redefined as a permanent rollout with its own review cadence.

### 3. Define success and failure signals

Write one measurable signal that would count as the pilot working and one that would count as it not working. Both must be observable inside the window without new instrumentation that will not exist in time.

Record the current baseline or comparable non-AI method, measurement owner, collection method, and minimum evidence needed for the actual decision. Define expand, revise, and stop criteria before observing results. If the window cannot produce sufficient evidence, limit the decision to another bounded trial; do not infer readiness for rollout. Privacy and security stops override performance success.

### 4. Bound the scope

List systems, teams, and customer-facing processes the pilot must not touch. List what data the pilot may use and what it may not.

### 5. Name authority

State who owns the pilot, who can pause it early, and what would trigger that pause. Silence on this point means anyone with access can quietly extend the pilot past its window.

## Output format

Return a one-page card with these sections in order:

1. Decision this pilot informs
2. Pilot window (start, end, review date)
3. Baseline, comparator, measurement owner/method, and minimum evidence
4. Success/failure signals and expand, revise, or stop decision rule
5. Out of scope
6. Data handling notes
7. Pause and cancel authority
8. A short message the manager can send the team to open the pilot

## Example

Input: "We want to try an AI note-taker for our meetings."

Synthetic proposal: use three internal meetings over two weeks to decide whether a larger trial is justified, not whether to standardize team-wide. A named lead compares AI-assisted notes with existing manual notes against independently confirmed action items, records misses and review time, and checks the pre-agreed criteria. Missing comparator data means insufficient evidence. Any confidentiality concern pauses the pilot regardless of performance; client-facing calls remain excluded.

## Troubleshooting

- If the card cannot name a decision, the pilot is not ready to start. Send it back for framing first.
- If the success signal cannot be observed inside the window, shrink the window's ambition or lengthen the window, not the signal's honesty.
- If nobody is named as having pause authority, assign it before the pilot starts, not after something goes wrong.

## What not to do

- Do not let a pilot run without an end date.
- Do not report a pilot as successful using evidence collected outside its stated window.
- Do not expand a pilot's scope mid-run without rewriting the card.
- Do not treat enthusiasm from participants as a substitute for the stated success signal.