World Model Readiness
Engraved commercial instrument

For your company

Money & Vendors

Module · The pitch is not the product

The AI Procurement Discipline Check

AI vendors sell a demo, and a demo is a magic trick: rehearsed data, cherry-picked prompts, no edge cases. Your normal procurement muscle was built for software that either works or does not, but AI fails softly and expensively in the gap between the demo and your data. This module checks the five habits that let you buy on evidence instead of vibes: written requirements before demos, a proof of value on your own data, security and compliance gates, an exit that actually works, and sign-off from the people who will live with it.

Question 1 of 5 · Requirements before demos

Do you write down what you need before you watch a single vendor demo?

A demo you walk into without written requirements sets the agenda for you: you end up evaluating what the vendor is good at, not what you need. Requirements fixed in advance are the only thing that keeps a slick presentation from becoming your specification.

Question 2 of 5 · Proof on your data

Do you test AI tools on your own data before you buy, not just the vendor demo?

A demo runs on data chosen to make the tool look good. Your data is messier, and that is exactly where AI tools break. A proof of value on a real sample of your own work is the difference between what the tool can do and what it will do for you.

Question 3 of 5 · Security gate exists

Does every AI purchase pass a security and compliance review before it goes live?

An AI tool touches your data, and often sends it somewhere else to be processed. Where it goes, who can see it, and whether that is lawful are questions a review answers before signing, not after a breach. A gate that fires on every purchase is cheaper than a single incident.

Question 4 of 5 · Exit clause works

If an AI vendor fails you, can you leave with your data and switch?

The moment you depend on a tool is the moment its price and its terms are no longer negotiable. An exit you wrote into the contract, covering your data, your history and a real notice period, is the leverage that keeps a vendor honest for the whole relationship, not just the sale.

Question 5 of 5 · The users sign off

Do the people who will use the tool have a real say before you buy it?

A tool bought over the heads of the people who must use it becomes shelfware the day after the contract signs. Sign-off from the actual users, and from security and legal, is not bureaucracy: it is the difference between a purchase and an adoption.

For the statistics · one click each

Three questions for the public picture

These do not affect your score. They feed the anonymised, aggregated statistics; groups under 8 respondents are never shown.

How do you currently decide which AI tools to buy?

Ad hoc, case by case
IT decides
A standard process
Formal gated process
We do not know

Do you run a proof of value on your own data before buying AI?

Never
For big purchases
Usually
Always
No AI purchases yet

Could you leave your main AI vendor with your data intact?

No, locked in
We are not sure
Partly
Yes, cleanly
No main vendor yet

Your context

Used to calibrate the report. Company size and sector remain in the anonymized dataset; your email does not.

What the five levels look like

Every dimension in this assessment is scored 1 to 5. This is what the levels mean, dimension by dimension. The graded report diagnoses where your own answers land and what to do about it.

Requirements before demos

  1. 1No requirements
  2. 2Vague wish list
  3. 3Requirements after demos
  4. 4Written first
  5. 5Written and weighted

At the low end: Buying AI with no written requirements means the best demo wins, not the best fit. Write down the problem, the must-haves and the deal-breakers before you book the next call. What good looks like: Weighted requirements defined up front are exactly how you stay in control of the evaluation. Keep them honest by scoring every vendor against the same sheet, not against each other.

Proof on your data

  1. 1Demo only
  2. 2Reference calls
  3. 3Trial, generic data
  4. 4Pilot on real data
  5. 5Scored pilot, real data

At the low end: A decision made on the vendor demo alone is a decision made on their best day. Before you sign, run the tool on a real sample of your own data and see what actually happens. What good looks like: A scored pilot on your real data is the single strongest signal you can get before buying. Keep the scoring criteria fixed in advance so the pilot proves something instead of flattering the tool.

Security gate exists

  1. 1No review
  2. 2Review if remembered
  3. 3Review for big deals
  4. 4Standard gate
  5. 5Gate plus data mapping

At the low end: Buying AI with no security review means you learn where your data went after it has already gone. Put a mandatory review in the path before any tool touches real data. What good looks like: A standard gate with data mapping tells you exactly what leaves the building and where it lands. Keep the map current as vendors change their subprocessors; the answer expires.

Exit clause works

  1. 1Locked in
  2. 2No exit terms
  3. 3Exit on paper
  4. 4Exit with data export
  5. 5Tested exit path

At the low end: A vendor you cannot leave sets your price at renewal, not the market. Negotiate a data export and a clean exit now, while you still have a signature they want. What good looks like: A tested exit path with real data portability keeps every renewal a negotiation instead of a ransom. Re-check it when the vendor changes their platform; portability quietly rots.

The users sign off

  1. 1One person decides
  2. 2Rubber-stamp only
  3. 3Users consulted late
  4. 4Users sign off
  5. 5Cross-functional sign-off

At the low end: A tool chosen by one person and handed down is a tool the users will route around. Bring the people who do the work into the evaluation before the shortlist, not after. What good looks like: Cross-functional sign-off means the tool arrives with the users, security and legal already behind it. Keep the group small enough to decide; a sign-off that never converges is its own failure.