AI MARKETING PILOT BRIEF
Companion to “From AI experiments to a marketing system that works”
By Janaha Vivek, Founding Partner, AiDU

Pilot name:
Owner:
Review date:

1. PROBLEM
Which recurring activity needs to improve, and for whom?
How often does it occur? What is the consequence of getting it wrong?

2. CURRENT BASELINE
Record total time, cost, rework and quality for comparable recent work.
Include preparation, review, correction and exception handling.

3. INPUTS
Approved sources and locations:
Who may access them:
Source dates and owner responsible for keeping them current:
Personal or sensitive information to exclude:

4. OUTPUT
Specific deliverable and intended user:
Example of a good result, with reasons:

5. ACCEPTANCE CRITERIA
What must a reviewer verify?
What makes the result unacceptable?
How will source references and uncertainty be checked?

6. BOUNDARIES
Allowed actions:
Actions requiring human approval:
What happens if evidence conflicts, is missing or is inaccessible:
How to stop the process and recover from a mistake:

7. OWNER
Reviewer and backup:
Person handling failures:
Person maintaining sources, instructions and integrations:

8. DECISION RULES
Expand when:
Revise when:
Stop when:
Representative cases to test, including difficult ones:
Review date and decision-maker:

TRIAL LOG (repeat for each trial)
Date and case:
Input/source versions:
Preparation time:
Generation and checking time:
Review/correction time:
Total time:
Tool and other costs:
Accepted, revised or rejected:
Unsupported claims or missing evidence:
What changed before the next trial:

Illustrative example only:
Task: turn approved customer questions into a campaign brief.
Allowed: read the approved pack, group questions and draft a brief.
Human decisions: verify evidence, select direction, approve claims.
Excluded: publishing, contacting customers, changing campaign spend.