AI due diligence for acquirers and private equity
AI due diligence from Paul Okhrem starts at USD 80,000 for acquirers and private equity teams. The review tests a target's AI claims against its systems, data rights, costs and controls. Findings separate evidence from assumptions and missing access. The 12-area checklist below helps prepare a review before signing or funding a transaction.
Price: USD 1,000 per hour, 80-hour minimum, USD 80,000 floor. Delivered personally by Paul Okhrem. See the published consulting terms. Engineering, software and external specialists cost extra unless included in writing.
When an AI diligence review fits the deal
Use a scoped review when an AI claim could change the investment decision, operating cost or post-close plan. Paul connects technical evidence to the commercial question without deciding the investment for you.
- A target markets an AI feature but has little evidence of use or quality.
- Customer data, licenses or a model vendor may limit the planned growth.
- The deal team needs to understand AI costs and key-person risk before signing.
A general code audit, legal opinion or financial valuation answers different questions. Those specialists retain their decisions. Diligence does not prove every claim or remove all transaction risk.
AI due diligence checklist
Use this checklist to test a target company's AI claims before you sign. Ask for the evidence in the third column. Treat the fourth column as a stop-and-investigate signal, not an automatic reason to reject a deal.
Download the AI due diligence checklist as a CSV. It contains the same 12 areas shown below, with no email gate.
| Area | Check | Evidence to request | Red flag |
|---|---|---|---|
| AI inventory | List every AI system in the product and in operations | System register with owners | No register, or claims exceed the register |
| Data rights | Confirm rights to train on and process customer data | Contracts, data processing agreements, consent records | Data used outside contract terms |
| Model provenance | Identify models, vendors and licenses | Vendor contracts, model documentation, license terms | License limits on commercial use |
| Revenue attribution | Test how much revenue depends on AI features | Feature usage and revenue by cohort | "AI-powered" label with no usage data |
| Unit economics | Check model and vendor cost per customer at scale | Cost per transaction and 12-month trend | Costs grow faster than revenue |
| Quality evidence | Review evaluation results and error rates | Test sets, evaluation reports, incident log | No evaluation before releases |
| Security | Check prompt injection, data leakage and access controls | Penetration test results, access reviews | Agents with broad write access |
| Regulation | Map EU AI Act risk tiers and sector rules | Risk classification and compliance plan | High-risk use without controls |
| Concentration risk | Check dependence on one model vendor | Vendor share of AI spend, exit plan | Single vendor and no fallback |
| Team | Assess who builds and runs AI, and key-person risk | Organization chart, retention plans | One person holds critical knowledge |
| Governance | Check ownership, approval and monitoring | Policies, decision logs, monitoring dashboards | Policy exists but there is no evidence of use |
| Roadmap realism | Test the AI roadmap against budget and team | Roadmap, budget, hiring plan | Roadmap needs capabilities the team lacks |
For a real model-risk failure, read the Zillow AI pricing lesson. It is an external company example, not a client result from Paul's consulting.
Each finding needs context. A license restriction may be acceptable for the planned use; a missing file may need further access. Qualified counsel assesses rights and legal duties. This checklist is not a complete security, legal or financial audit.
Want Paul to review your draft? Send a private brief
Separate a demonstration from a defensible AI capability
A polished demo can hide manual work, a small test sample or an unsupported claim. Request read-only evidence, representative tests and a walkthrough of the production path. Record what cannot be inspected and why.
A missing document does not prove that a control is absent. A management statement does not prove that it works. For each material finding, record the source, access limits, confidence and next question.
How findings change the questions a deal team asks
Turn a warning sign into a specific question about the deal thesis. Its importance depends on the target, intended use and specialist review.
- Vendor-dependent AI sold as proprietary: where is the actual value that a competitor cannot easily copy?
- Manual review missing from the cost model: does the margin include the full cost of completing a customer task?
- Unclear production testing: how much quality or customer risk is unmeasured?
- One person holds critical knowledge: what retention and handover work is needed?
Keep technology findings separate from valuation decisions. An unresolved question can require more evidence, a contract condition or a post-close action.
What the AI diligence report delivers
The scoped report gives the deal team an evidence register, ranked findings, open questions and remediation options. It separates issues to resolve before signing from improvements that can wait until after close.
- Map each material AI claim to the evidence reviewed.
- State access limits and unresolved dependencies.
- Rank findings by business significance, not technical novelty.
- Name remedial work, owners and assumptions behind cost or timing estimates.
- Prepare a decision note with clear next questions for the relevant specialists.
Security testing, code audits, accounting reviews and legal opinions need separate scopes with qualified providers. A technology finding alone does not determine deal valuation.
Connect diligence findings to the first 100 days
After close, management needs owners and budgets for accepted risks. Transfer the evidence and unresolved questions into an operating plan. Do not turn a diligence hypothesis into a promised synergy.
For portfolio-wide work after the deal, see AI consulting for private equity firms. The pre-close review and post-close value plan remain separate scopes.
Keep the source of each claim visible when briefing a new management team. The AI evidence and measurement standard explains how to separate observed results, assumptions and public company examples.
Access, confidentiality and diligence fees
Paul charges USD 1,000 per hour, with an 80-hour minimum and USD 80,000 floor. Agree the target, deadline, data-room access, interviews and specialist work before committing. The minimum does not cover unlimited targets.
Check conflicts and confidentiality before sensitive access. Start with a non-confidential deal-stage summary, not a data-room password or target documents. Turnaround depends on access, scope and specialist availability; it is agreed in writing.
FAQ
How much does AI due diligence cost?
Paul Okhrem charges USD 1,000 per hour, with an 80-hour minimum and USD 80,000 floor. The written scope names the target, evidence access, interviews and outputs. Legal, accounting, penetration testing and code reviews are separate specialist scopes, not services silently included in the minimum.
Can AI due diligence verify every claim a target makes?
No. Conclusions depend on scope, access, time and the evidence available. Paul records unresolved claims and the limits of the review. The checklist helps structure requests; completing it does not certify the company or remove investment risk. The deal team decides how to handle remaining uncertainty.
Does the review include intellectual-property legal advice?
No. The review identifies data, model and licensing questions for qualified counsel. It can request contracts, consent records and model terms, then record missing evidence. Counsel determines the legal meaning and permitted use. Paul's technology review does not substitute for that opinion.
Can AI due diligence fit a transaction deadline?
Timing must be agreed against access, scope and specialist availability. Start with the decision date and the evidence already available. Paul can define a focused review, but does not promise a fixed turnaround before those dependencies are checked. Record material questions that cannot be resolved in time.
How is AI diligence different from a general code audit?
AI diligence connects the investment claim to data rights, models, tests, operating economics and dependencies. A code audit examines a narrower technical scope and may supply evidence to that review. Buyers can need both. The agreed scope should name which provider owns each part.
What happens to the findings after acquisition?
Management can turn accepted risks and opportunities into named actions, budgets and review gates. Keep unresolved evidence requests visible during the handover. Post-close implementation or portfolio advice is a separate scope. A diligence finding is not a promised saving, synergy or timetable for a production release.
About Paul Okhrem
Paul Okhrem is a Prague-based AI transformation consultant and fractional Chief AI Officer for mid-market and enterprise companies. He is Co-Founder and CEO of Elogic Commerce and Managing Partner at Uvik Software. He has built B2B and enterprise software since 2009.
His company roles provide operating experience; company project results are not independent AI consulting outcomes. Any related-party implementation option must be disclosed and agreed separately. Read the evidence and measurement method.
Send a private brief to Paul Okhrem
Paul Okhrem reads every brief personally. Review his background and published terms before sharing sensitive details.