Before paying for AI upside in an acquisition, a buyer should see evidence that the gains are repeatable and owned by the business being bought. That means a measured baseline, costs that survive model price changes, clear rights to the data and models, and people who stay. Anything less is a forecast, and should be priced as one.
This question is now being asked openly by AI investors and technology leaders, as private equity takes a growing role in enterprise AI. The topic is the prompt. The guidance below is AIPEX’s own, and we make no comment on any company involved.
Why is AI now in the deal case?
Buyers increasingly underwrite acquisitions on the assumption that AI will lift margins after completion: fewer people per unit of output, faster service, better pricing. Sellers, in turn, present AI as part of the equity story. Both can be legitimate. The problem is that AI upside is easy to describe and hard to evidence, and deal timetables reward description.
For a board approving an acquisition, or preparing a business for sale, the question is not whether AI could create value in the target. It almost always could. The question is how much of that value is already demonstrated, how much depends on things the target does not control, and who pays if the plan slips.
What separates evidenced upside from a forecast?
Three tests do most of the work.
- Is there a baseline? A claimed efficiency gain needs a before and after, measured the same way, over enough time to rule out a one-off. If the target cannot show what the process cost before AI, the gain cannot be verified.
- Is it in production? A pilot that worked in one team is not the same as a change running across the business with support, monitoring and adoption. Many AI gains stall between the two, a pattern we describe in AI Pilot Purgatory.
- Is it in the numbers? A time saving only becomes value when it shows up as lower cost, higher capacity that is actually used, or revenue. Ask finance, not the AI team, to trace it.
Which dependencies can erode the value after completion?
AI value often rests on things outside the target’s control. Due diligence should test each one:
| Claim in the deal case | Evidence to ask for | If the evidence is weak |
|---|---|---|
| “AI has cut our cost to serve” | Baseline, current unit cost, and model and compute costs per transaction | Treat the saving as a forecast; adjust price or add an earn-out |
| “Our models are proprietary” | Ownership of models, fine-tunes, prompts and training data rights | Specific warranties and indemnities; a price adjustment if rights are unclear |
| “We can scale this across the group” | Production evidence in more than one team; data readiness in the rest | Fund integration separately; do not pay for it twice |
| “Our supplier terms are secure” | Contract length, price-change terms, change-of-control and exit clauses | Model a price rise or a forced switch in the downside case |
| “The capability is embedded” | Who built it, who runs it, and retention arrangements | Key-person retention as a condition, or a lower value |
Supplier dependency deserves particular care. Many AI-enabled businesses rely on one model provider. Pricing, usage limits and model behaviour can all change during the life of a contract, and the target may have little negotiating power. A deal case that assumes today’s cost of AI for five years is carrying a risk it has not priced.
What about AI risk that the deal inherits?
Upside is only half of the picture. A buyer also inherits the target’s AI liabilities: data used without clear rights, systems that fall into a regulated category, customer commitments about AI that cannot be met, and incidents that were never reported. Two questions are worth asking directly. Does the target’s insurance respond to losses caused by its AI systems, or are they excluded or untested? And does the target use AI in areas such as recruitment, creditworthiness or life and health insurance pricing, which are listed as high-risk in Annex III of the EU AI Act (Regulation (EU) 2024/1689, 13 June 2024) if the business touches the EU market? These are governance questions for the board, not legal advice; the deal’s lawyers will need to confirm the position. Our article on EU AI Act fines and enforcement explains what regulators can already do.
What should the board ask before approving the deal?
- How much of the value in the deal case depends on AI, and how much of that is already in production?
- Has anyone independent of the target checked the baseline behind the claimed gains?
- What happens to the case if the main model provider raises prices or changes terms?
- Do we own, or have secure rights to, the data, models and outputs we are paying for?
- Which people carry the AI capability, and are they staying?
- Who owns delivery of the AI plan in the first 100 days after completion?
If the board cannot get clear answers, that is not necessarily a reason to walk away. It is a reason to move AI value from the price into an earn-out, a warranty or the integration budget, where it belongs until it is proven.
Where to start
Our briefing on AI due diligence for mergers and acquisitions covers the eight risk areas buyers should check, the red flags that should reprice a deal and a 100-day integration plan. For a seller preparing for a process, or a buyer that has just completed, the step on its four-step path that fits is step 2, Assess: the AI Readiness Assessment (£999 + VAT) scores 30 areas across six domains with risk flags, a prioritised roadmap and a 60-minute debrief. The AI Due Diligence Checklist for M&A below gives deal teams the buyer and seller questions.
Frequently asked questions
How should a buyer value AI upside in an acquisition?
Pay for AI gains that are evidenced: a measured baseline, the change running in production and the saving visible in the numbers. Treat the rest as a forecast, and move it into an earn-out, a warranty or the integration budget until it is proven.
What dependencies can erode AI value after a deal?
Reliance on one model provider whose prices or terms can change, unclear rights to data and models, capability held by a few people, and pilots that never reached production. Each should be tested in due diligence and reflected in the price or the deal terms.
What AI risks does a buyer inherit?
Data used without clear rights, systems in regulated categories, AI commitments to customers that cannot be met, unreported incidents and insurance that may not respond to AI losses. See AI Due Diligence for M&A →
Correct as of 28 September 2026. This content reflects the law, official guidance and deadlines as published on that date. The EU AI Act and its guidance, standards and deadlines are still being amended and clarified. If you are reading this after 28 September 2026, check for later amendments, European Commission guidance and harmonised standards before relying on it. It is general information, not legal advice, and AIPEX accepts no liability for decisions taken on it without specific professional advice.