Skip to content
World Model Readiness
Engraved team-functions instrument

For your team

Teams by Function

Module · Seeing through the AI pitch

The Procurement Team AI Check

AI vendors sell the demo, and the demo is always the best day the product will ever have. Your procurement team stands between a slick pitch and a contract that is hard to leave, expensive to run, and built on claims nobody tested. This module checks the five disciplines that let the buying team see through the deck: testing the technical claims on your own data, checking references the vendor did not hand-pick, modelling the true cost past the licence line, holding a standard on exit terms, and scoring the specific risks an AI supplier brings that a normal software vendor does not.

Question 1 of 5 · Claims tested on your data

Do you test an AI vendor's technical claims on your own data before buying?

A vendor demo runs on data the vendor curated to make the product shine. Your data is messier, and the gap between the demo and your reality is where the disappointment lives. A proof of concept on your own inputs, measured against your own bar, is the only claim that counts. Everything before that is marketing.

Question 2 of 5 · References you chose

Do you check references you sourced yourself, not just the ones the vendor hand-picked?

The references a vendor offers are its happiest customers, coached and cherry-picked. The useful conversation is with a customer you found yourself, ideally one who churned or struggled. In a young AI market where products change monthly, a candid reference about what broke is worth more than three glowing ones about what worked.

Question 3 of 5 · True cost is modelled

Can your team model the real total cost of an AI tool, past the licence line?

The licence fee is the visible tip. Underneath sit usage-based charges that scale with success, integration and data-prep effort, the internal time to run and monitor it, and prices that reset at renewal once you are dependent. A team that models only the sticker price is comparing vendors on the one number designed to be compared.

Question 4 of 5 · Exit is negotiated in

Do your AI contracts include real exit terms: your data out, in a usable form, on notice?

The moment to negotiate leaving is before you sign, while you still have leverage. AI vendors lock you in through your accumulated data, tuned configurations, and embedded workflows. A contract with no clean exit, no data export in a usable format, no reasonable notice, turns a tool you outgrew into a hostage situation at renewal.

Question 5 of 5 · AI-specific risk scored

Do you score the risks specific to AI suppliers, not just the standard vendor checklist?

An AI supplier carries risks a normal software vendor does not: where your data goes and whether it trains their model, which foundation model sits underneath and what happens if it changes, how the tool behaves under the EU AI Act, and whether a young vendor will still exist at renewal. A generic security questionnaire asks none of this.

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 many AI tools has your team bought or is evaluating?

What do you require before an AI purchase is approved?

How do your AI contracts handle leaving the vendor?

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.

Claims tested on your data

  1. 1Demo taken on faith
  2. 2Vendor benchmarks only
  3. 3Informal trial
  4. 4Tested on our data
  5. 5Measured against our bar

At the low end: Buying on the strength of a curated demo is buying the best case you will never see again. Require a proof of concept on your own data before any AI purchase clears. What good looks like: Testing claims on your own data against your own bar is what separates a real capability from a good pitch. Keep the acceptance criteria written down; a proof of concept with no pass mark just relocates the optimism.

References you chose

  1. 1No references
  2. 2Vendor list only
  3. 3Vendor list, probed
  4. 4Own references sourced
  5. 5Sourced, including churned

At the low end: Skipping references entirely on a fast-moving AI product means learning its failure modes in production. Talk to real customers before you sign, starting with any you can find who left. What good looks like: References you sourced yourself, including customers who churned, tell you how the product behaves after the honeymoon. Keep asking about recent experience; in this market a reference from a year ago describes a different product.

True cost is modelled

  1. 1Licence price only
  2. 2Plus obvious add-ons
  3. 3Rough TCO
  4. 4Full TCO modelled
  5. 5TCO plus exit costs

At the low end: Comparing AI vendors on licence price alone hides the usage charges and internal effort that dominate the real bill. Build a total-cost model that includes consumption, integration and the people to run it. What good looks like: A full TCO that carries renewal and exit costs lets you compare vendors on what they will actually cost, not on the sticker. Keep the model honest as usage scales; consumption pricing is designed to surprise the buyer who modelled a flat fee.

Exit is negotiated in

  1. 1No exit terms
  2. 2Standard terms accepted
  3. 3Exit raised, weak
  4. 4Exit terms negotiated
  5. 5Exit tested and portable

At the low end: A contract with no exit terms hands the vendor all the leverage at every future renewal. Negotiate data export, format and notice into the agreement before you sign, not after you are stuck. What good looks like: Negotiated, tested exit terms mean you can leave a tool that stops earning its place. Actually verify the export works before you are relying on it; an export clause you have never exercised is a promise, not a capability.

AI-specific risk scored

  1. 1Standard checklist only
  2. 2Security review only
  3. 3Some AI questions
  4. 4AI risk scored
  5. 5Scored and monitored

At the low end: A standard vendor checklist misses every risk that makes AI different, from data training to model dependency. Add AI-specific questions on data use, the underlying model, and regulatory class before you approve a supplier. What good looks like: A dedicated AI-supplier risk score, kept current, catches the exposures a security review never looks for. Keep monitoring after signing; the model underneath, the data terms and the vendor's survival can all change during the contract.