Ask what runs in production, who maintains it, what breaks if the founder leaves, and where the data lives. Real AI has running systems, named owners and an audit trail. Claimed AI has a demo and a roadmap. Post-deal, ask which jobs eat the week and what the smallest build that pays back looks like.
Why does AI now appear in every deal?
Because both sides of the table have decided it moves valuations. In Deloitte's 2025 study of 1,000 senior corporate and private equity leaders, 86% said their organisations had integrated generative AI into their M&A workflows, and 35% were using it for target screening and due diligence itself. AI is now simultaneously a tool used in the deal and a claim made inside the deal.
This guide is about the second part: the AI story a target tells, and the AI plan a new owner writes afterwards. Sellers know the multiple a credible AI story can attract, so every data room now contains one. Your job is to work out whether the story describes a working asset, a science project, or a slide. The good news is that a handful of plain questions will tell you, and none of them requires you to read code.
Which questions expose real AI versus claimed AI?
The ones about production, people and data, because demos can be staged and those three can't. Ask them in the room, and ask for evidence rather than descriptions:
- What actually runs in production today? Not what's planned, piloted or "in beta". Which systems ran on real data, for real users, last month? Ask to see usage logs, not a walkthrough. A demo built for the process is not an asset.
- Who maintains it? A named person or team with time allocated, a supplier on a current contract, or nobody? "The founder looks after it at weekends" is an honest answer you will hear more often than you'd expect, and it prices very differently from a maintained system.
- What breaks if the founder leaves? If the honest answer is "most of it", the AI is a key-person risk wearing a technology costume. Documentation, handover and a second person who can run it are worth more than any model choice.
- Where is the data, and who owns the workings? Is the capability built on the company's own data in its own systems, or is it a thin layer over a third-party tool that could reprice, change terms or cut access? If the vendor relationship ends, what is left?
A company with real AI answers these in minutes, with logs and names. A company with claimed AI answers with a roadmap. Both answers are useful; they just belong in different valuation conversations.
Which questions matter for post-deal plans?
The ones about the week, not the technology. Post-acquisition AI value creation fails when it starts from "what could AI do?" and succeeds when it starts from "where does this team's time actually go?"
Two questions do most of the work. First: which jobs eat the week? In every operating company there are a handful of repetitive, rule-heavy jobs, reporting, checking, matching, chasing, that consume days of skilled time every month. Those jobs are the value creation plan; find them by sitting with the people doing the work, not by surveying the leadership team. Second: what is the smallest build that pays back? Not the platform, not the transformation: the one system that hands back the most days for the least build. Prove that, then decide the next one with evidence in hand.
A plan written this way has a testable first quarter and honest arithmetic behind every line. A plan written from the technology down produces pilots, and pilots are where value creation goes to stall.
What are the red flags?
No owner, no audit trail, and lock-in. Any one of them turns an AI asset into an AI liability, whichever side of the deal you're on.
- No owner. Nobody named is responsible for the system's output, its errors or its costs. Unowned systems decay quietly until they fail loudly, usually in the first year of your hold.
- No audit trail. The system makes or shapes decisions, and nobody can show what it decided, when, or why. That is a compliance exposure and a sale-day problem: what can't be evidenced can't be diligenced by your eventual buyer either.
- Vendor lock-in. The "proprietary AI capability" turns out to be rented: it lives in a third party's cloud, on their terms, and walks out the door if the subscription ends. Rented capability can still be useful; it just isn't an asset you should pay an asset price for. We've written more on this in custom AI: build or buy?
The base rates justify the scepticism. In a 2025 S&P Global Market Intelligence survey, 42% of companies had abandoned most of their AI initiatives, up from 17% a year earlier, and the average organisation scrapped 46% of AI proof-of-concepts before production. Most AI claims you will meet in a process sit somewhere in that wreckage, and the questions above are how you find the exceptions.
How do we help?
With CTO-grade judgement before a line of code is written. Jim has spent 25 years building and running technology through three exits, which is the lens diligence actually needs: not "is this model clever?" but "is this a maintained, owned, evidenced system that survives the founder's departure?"
For diligence, that means plain-English answers to the questions above, backed by looking at the real systems. For post-deal work, it means a plan before a build: we map where the operating company's week goes, do the arithmetic on each candidate job, and put a straight recommendation in front of the board, including "don't build anything yet" when that's the truth. The first conversation is a free Impact Call, and it costs you nothing but an hour.
Sources
Figures and claims in this guide draw on our own delivery work and the sources below. We only publish numbers we can stand behind.
- Deloitte, 2025 M&A Generative AI Study (survey of 1,000 senior corporate and private equity leaders, first half of 2025): 86% have integrated generative AI into M&A workflows; 35% use it for target screening and due diligence. deloitte.com. Published 2025, accessed 8 July 2026.
- S&P Global Market Intelligence, 451 Research "Voice of the Enterprise: AI & Machine Learning, Use Cases 2025" (n=1,006 IT and business leaders): 42% abandoned most of their AI initiatives, up from 17%; 46% of proof-of-concepts scrapped before production. spglobal.com. Published 2025, accessed 8 July 2026.
- AI Nativ.es delivery experience, 2026: client work is described without names until we have written permission to use them.