AI technical due diligence
Independent assessments of the AI and technology in software, IT-services, and tech-enabled businesses: for acquirers testing a target before they sign, and for companies preparing to go to market. Typically businesses with about $5–100M in revenue; larger by arrangement. Fixed scope, stated evidence, and the limits of that evidence.
What gets tested
Most deal teams read the architecture paper and interview engineering. That is necessary, not sufficient. Four areas get a direct look, whichever side of the deal you are on.
What management says the system does, tested against documents, interviews, and where scope allows, direct testing in the research lab.
Whether the data, licences, and pipelines can support the claims, and what the model depends on that the deck does not mention.
The controls and architecture we can see, reviewed for risk. This is a posture review, not a penetration test or a certification.
What happens when this stack meets yours: integration cost, dependency on third-party models, and the assumptions the plan rests on.
For acquirers
Pick the band that matches where the AI sits in the business. Final scope for Multi-Workstream is agreed before engagement.
Single workstream
$7,500 fixed fee
Timing: 1–3 weeks
A single-workstream assessment of the AI and technology claims in a target, its data readiness, its security posture, and an integration-risk memo.
Multi-workstream
$15,000–20,000
Timing: 2–4 weeks
For targets where AI sits in the core product or the cost base. Adds agreed workstreams on top of Standard. Final scope is agreed before engagement.
Timing depends on how quickly you provide access, documents, and answers: the clock runs from an agreed start date and stops while we wait on the client. Multi-Workstream scope is confirmed before engagement. A security posture review is not a penetration test or a compliance certification.
Request scope and availabilitySee what an acquirer receives: Worked example, the buyer’s view
What you receive
Model accuracy holds on the customer’s own data
Partly supported
Benchmark run on the vendor’s sample set only. No customer-data run was provided.
Severity: MediumConfidence: High
Inference cost per request falls as usage scales
Not supported
No cost-per-request data in the data room. Pricing terms imply the opposite at current volumes.
Severity: HighConfidence: Medium
Training data is owned outright
Open question
Licence terms for two source datasets were not provided. Ownership cannot be confirmed.
Severity: HighConfidence: Low
Authored for illustration. Not a completed engagement, and it contains no client, employer, or target material.
Full worked example: buyer’s view · seller’s view
How it runs
Timing assumes an agreed start date and timely access. Confidential material moves through a secure channel after the NDA and conflict check, never through the contact form.
We confirm the decision, the target type, your timetable, and any conflict with the target or its advisers. Then we agree written terms.
Data room or management access under written terms. Access timing sets the start date, so we agree it before kickoff.
Document review, management interviews, and research-lab testing where the agreed scope includes it.
A written report and a walkthrough with your deal team, with every finding traced to its evidence.
For companies preparing to go to market
Buyers now press hard on AI and data. Find the answers before they do, fix what can be fixed, and tell the story with evidence. AI maturity increasingly shapes buyer perception and exit outcomes, and the gaps are better found by you than in a buyer’s data room.
The same claim-by-claim assessment a buyer’s diligence team would run, applied to your own business, so there are no surprises in their data room.
A prioritised list of what to fix before launch, what to disclose and how, and what to leave alone, ahead of going to market.
A check that what the marketing materials and management presentation say about your AI, data, and technology matches what you can demonstrate.
The technical questions a buyer is likely to ask, with evidence-backed answers and a prepared section of the data room.
We quantify the gaps and build a defensible improvement path. We do not promise an “AI multiple” on valuation. Scope is fixed and quoted before work starts, and timing depends on how far ahead of launch we begin.
I am not your banker. I do not run the sale process, value the business, or negotiate. The work sits alongside your adviser and your counsel.
See what a seller receives: Worked example, the seller’s view
Optional additions
Scoped public-record checks on the target and its principals: litigation, regulatory exposure, and reputational signals not visible in the data room. Available only inside a diligence engagement.
An optional, separately agreed engagement to work the AI roadmap identified in diligence. It is never automatic and never a condition of the diligence.
Not part of this work
No insurance, accreditation, or guaranteed-confidentiality claim is made here. Terms are set in the engagement letter.
Questions
No. The first conversation covers the target type, the decision, and the timetable. Testing starts once access is authorized under agreed terms.
No. Confidential material moves through a secure channel after the NDA and conflict check, never through the public contact form.
Neither. It is a posture review of the controls and architecture we can see. It is not a penetration test and not a certification.
I prepare companies to go to market by assessing their AI and technology readiness. I do not act as banker, run the sale process, source targets, value the business, or negotiate.
Only if you choose. The report is yours, and you decide what is shared, with whom, and when.
No. It is an optional module inside a diligence engagement and is not sold as a standalone service.
Then the report says so, with the evidence. Telling a buyer where the AI is not what it is claimed to be is the point of the work.