The Decision Document
A one-page document to fill out before committing to any AI build — the six build questions plus the decision, a revisit date, and a written trigger for changing your mind.
The one-page document to fill out before committing to any AI build — the six questions plus the decision itself, with a revisit date and a written trigger for changing your mind. Thirty to forty minutes. If any section takes more than a few sentences, that’s a signal in itself: either the build is more complicated than it should be this early, or the answer is still too vague to be useful.
If sections on workflow or readiness come back vague, that’s not a form problem — run Is It Worth Automating? for the count and Build, Buy, or Skip? for the verdict.
The Decision Document
DECISION DOCUMENT
Build proposal: ______________ Author: ______________
Date: ______________ Revisit date: ______________
1. THE PROBLEM
In two sentences, your own words, no vendor terminology:
____________________________________________
2. THE WORKFLOW
Instances per period: ________ Time per instance: ________
Total annual time cost: ________
3. THE CUT — classify each step:
AI (judgment, varies by case) / Automation (same input, same
output) / Human (high-stakes, low-volume, or accountability)
Steps: ____________________________________________
4. READINESS
The build depends on these capabilities: ______________
Classification: Reliably ready / Almost ready / Not yet
Evidence: ____________________________________________
5. HUMAN PLACEMENT
Humans are positioned at: ______________
At each point, what specifically does the human do?
____________________________________________
6. PROTOTYPE PLAN
Start: ________ End: ________
What I'll do manually during it: ______________
What notes I'll keep: ______________
7. THE DECISION
Build / Build smaller / Wait / Don't build
Budget gate (if building): ______________
What evidence would change this decision:
____________________________________________
Why it works It makes the decision visible — to you now, to everyone who touches the build, and to you at the revisit date, when the only question that matters is: given what I know today, would I start this build? The failures travel in packs: a vendor-shaped problem statement almost always comes with a wrong AI-versus-automation cut and an aspirational readiness claim, which is why all six sections live on one page where the pattern can't hide. And section 7's last line — what evidence would change this — is the sunk-cost antidote, written while you still have nothing to protect.
Then what Fill it before resources move, keep it with the project, and actually honor the revisit date — a decision document that never gets re-read wasn't being used. If sections 2 or 4 come back vague, that's not a form problem; run Is It Worth Automating? for the count and Build / Buy / Skip for the verdict. A fillable version that renders a clean PDF is coming; this one's built to be photographed.
These prompts put the framework from Before You Build in your hands. More about the book →
Frequently asked questions
What is an AI build decision document?
A one-page document filled out before committing to any AI build: the problem, the workflow, the AI/automation/human cut, readiness, human placement, a prototype plan, and the decision itself — with a revisit date and a written trigger for changing your mind.
How long should the decision document take?
Thirty to forty minutes. If any section takes more than a few sentences, that's a signal the build is more complicated than it should be this early, or the answer is still too vague to be useful.