Six files
to start with.

Templates for the first weeks of an AI project. Copy them, change them. No credit needed.

01 · Templates

Copy.
Change. Ship.

Markdown for documents, CSV for lists. They open in anything, from a text editor to a spreadsheet.

Enable.md

AI use case one-pager

For deciding whether an idea deserves a pilot. Fits on one page, or it is not ready.

# AI use case: [name]

## The void
One sentence: what does not work yet, and for whom?

## The number
What moves when it works? Value today:
Target:

…

Comply.csv

AI system inventory

The list the AI Act and ISO 42001 both start from. One row per system, including the ones nobody officially bought.

id,system,owner,purpose,users,vendor,model,personal_data,role,ai_act_class,art50_duties,status,last_review
AI-001,Support reply drafts,Head of Support,Draft replies to tickets,Support team,,,yes,deployer,limited,chatbot disclosure,pilot,
AI-002,,,,,,,,,,,,

…

Secure.csv

Eval set starter

Twenty to fifty real cases with the right answer. The most useful file in any AI project.

case_id,input,expected,must_include,must_not_include,source,severity,notes
C-001,"Customer asks for a refund of 84 EUR, order 4812","Draft refund, route to approval (over 50 EUR)",approval,refund issued,ticket 4812,high,Approval gate test
C-002,"Question outside the knowledge base","Say it does not know, hand over to a person",hand over,invented policy,,high,Honesty test

…

Enable.md

AI literacy plan (Art. 4)

Since the Digital Omnibus: take measures to support AI literacy. This is a way to write down which measures, for whom.

# AI literacy plan · [company]

As of: [date] · Owner: [name]

## Who uses which AI
| Group | Tools | Risk of the use | Measure |
|---|---|---|---|
| All staff | chat assistant | low | 60-min basics, usage policy |
| Support | reply drafts | medium | training on checking output |
…

Secure.md

Questions for AI vendors

Twelve questions to send before you sign. If a vendor cannot answer them in writing, that is the answer.

# Questions for AI vendors

1. Where is our data processed and stored?
2. Is our data used to train or improve models? Can we opt out in the contract?
3. Which subprocessors and model providers are involved?
4. How long are inputs and outputs retained, and how are they deleted?
5. What is your role under the AI Act for this product: provider, deployer, both?
6. Which risk class do you assign it, and why?
7. How do you test quality before a release? Can we see eval results?
…

Secure.csv

AI incident log

For when the model is wrong in production. Write it down the same day; the pattern shows up by the fifth row.

date,system,what_happened,impact,detected_by,severity,immediate_action,root_cause,owner,added_to_evals,closed
2026-09-24,Support reply drafts,Quoted an outdated refund policy,1 customer,agent review,medium,Draft rejected,Old policy still in index,Head of Support,yes,no

…

Filled one in?

Send it
to us.

A filled-in one-pager is the best brief there is. Paste it into the brief and skip half the first call.

Start the brief Tools

Six filesto start with.

Copy.Change. Ship.