World Model Readiness
Engraved commercial instrument

For your company

Money & Vendors

Module · Buying the story or the system

The M&A AI Due Diligence Check

Every target now has an AI story, and the story is priced into the deal. Some of it is a real system with real data and a defensible edge; some of it is a wrapper around a public model, a founder who leaves at closing, and a revenue line that AI touched but did not earn. This module checks whether your diligence can tell the two apart: verifying the technical claims, tracing who owns the training data, testing the dependency on key people, confirming the model and IP actually transfer, and attributing revenue to AI honestly.

Question 1 of 5 · Claims are verified

Can your diligence team actually test the target's AI claims, not just read them?

A deck says the model is proprietary and the accuracy is high. Diligence that only reads the deck is buying the claim at face value. Someone who can look under the hood, at the model, the code, the evaluation, is the difference between verifying an asset and financing a story.

Question 2 of 5 · Training data is owned

Do you know whether the target owns the data its models were trained on?

A model is only as transferable as the data behind it. If that data was scraped, licensed on terms that do not survive a sale, or belongs to customers who can withdraw it, the asset you are buying can evaporate at closing or surface later as a lawsuit.

Question 3 of 5 · Not one person's head

If the key AI people leave at closing, does the capability survive?

In many targets the real AI asset is two or three people, and the model is a thing they know how to keep working. If the capability walks out the door when the earn-out vests, you bought a snapshot, not a system. Retention and documentation decide which one it is.

Question 4 of 5 · The model transfers

Will the models, IP and infrastructure actually transfer with the deal?

The AI may run on a founder's personal cloud account, depend on a licence that is not assignable, or sit on IP the company never properly assigned to itself. What runs today is not automatically what you own tomorrow. The transfer is a legal and technical question, and both answers matter.

Question 5 of 5 · Revenue attribution is honest

Can you tell how much of the target's revenue AI actually earns?

AI-driven revenue is the line that justifies the premium, and the easiest one to inflate. Revenue that AI touched is not revenue AI earned. Diligence has to separate the sales that genuinely depend on the AI from the ones that would have closed anyway with a spreadsheet.

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 active are you in acquiring companies with AI capabilities?

Not acquiring
Exploring targets
The occasional deal
Actively acquiring
Prefer not to say

Who verifies the AI claims in a target you are buying?

Nobody specific
The deal or finance team
Internal technical staff
Independent AI experts
No AI deals yet

Has an acquired AI capability ever underdelivered against the deal case?

No
Yes, a little
Yes, significantly
We have not measured
No such deals 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.

Claims are verified

  1. 1Take claims on trust
  2. 2Read the deck
  3. 3Ask the founders
  4. 4Independent tech review
  5. 5Hands-on verification

At the low end: Taking AI claims on trust prices a story as if it were a system. Bring someone who can inspect the model and the code into diligence before you commit to a number. What good looks like: Hands-on verification of the AI is exactly what protects the price you pay. Keep the reviewers independent of the deal team; a verifier who wants the deal to close finds what they hope to.

Training data is owned

  1. 1Never asked
  2. 2Assumed owned
  3. 3Partially traced
  4. 4Provenance documented
  5. 5Provenance and rights confirmed

At the low end: If nobody has traced the training data, you do not know if the model survives the sale. Make data provenance a diligence workstream, because rights that do not transfer take the model with them. What good looks like: Confirmed provenance and transferable rights are what make the model an asset you can actually keep. Get the reps and warranties to match what diligence found, so the risk sits with the seller.

Not one person's head

  1. 1One irreplaceable person
  2. 2Key people, no lock-in
  3. 3Some documentation
  4. 4Retention plus handover
  5. 5Institutionalised capability

At the low end: A capability that lives in one person's head is an asset that can resign. Before closing, tie the key people in and get the knowledge out of their heads and into documentation. What good looks like: An institutionalised capability, documented and staffed beyond the founders, is what you actually want to buy. Verify the documentation is real by having someone else run the system from it.

The model transfers

  1. 1Transfer never checked
  2. 2Assumed to transfer
  3. 3Legal review only
  4. 4Legal and technical checked
  5. 5Transfer plan tested

At the low end: If nobody has checked whether the AI transfers, you may be buying access, not ownership. Confirm the models, IP and infrastructure are assignable before the transaction, not during integration. What good looks like: A tested transfer plan means the AI you value is the AI you will own. Keep it tied to the integration timeline; a transfer that works in principle can still fail on a hard cutover date.

Revenue attribution is honest

  1. 1Take the number given
  2. 2AI-touched counted
  3. 3Partially separated
  4. 4Attribution tested
  5. 5Independently modelled

At the low end: Accepting the AI revenue number as presented is how you pay an AI premium for ordinary sales. Ask what revenue actually depends on the AI, and what would close without it. What good looks like: An independently modelled attribution tells you what the AI is really worth to the business. Stress-test it against a downside where the AI edge erodes; that is the case that decides the price.