Stop Asking Whether the Target Uses AI

Almost every technology diligence now contains an AI section. In most of them, it answers a version of the same question: is the target using AI, where, and how seriously?

It is a reasonable thing to want to know. It is also close to irrelevant to the decision being made.

Adoption is an operational fact

A company that has not adopted AI in its engineering, support or back office has a gap that a competent owner can close. It requires tooling, some training and a year of management attention. It is exactly the kind of improvement a private equity buyer expects to make, and it is priced accordingly — as an operational upside, not as a structural property of the asset.

The reverse is equally true. A target with impressive internal AI adoption has demonstrated competence. It has not demonstrated that its product is still going to be worth buying in four years, and the two are frequently confused because they are discussed in the same meeting.

Exposure is a valuation fact

The question that belongs in the investment case is not what the target does with AI. It is what AI does to the market the target sells into.

• How much of what customers pay for becomes reproducible at lower cost?
• Does cheaper software strengthen this company or strengthen entrants more?
• Does the category remain a purchase, or does it become a feature inside something larger?
• Which parts of the current price are protected by something other than the cost of building the product?
• What happens to pricing when three credible alternatives exist instead of one?

None of these can be answered by looking at the target alone, which is the practical reason they are so often skipped. They require a view of the market under a different cost structure, and that view is a judgement rather than a finding.

Why the adoption question persists

Because it is answerable. Adoption is observable, comparable across targets, and can be evidenced with documents. Exposure requires someone to take a position on the future and be wrong in public if it does not happen.

Diligence processes reward the first kind of question and punish the second. That is a structural bias, not an oversight, and it is worth naming explicitly when scoping the work — otherwise the report will be full of well-evidenced answers to questions nobody needed.

What replacing it looks like

In practice the substitution is not difficult. It changes what gets asked, not how much work is done.

1. Start from the thesis, not the technology

   Write down what the buyer is underwriting and the assumptions underneath it. AI exposure is only meaningful in relation to a specific claim about future cash flow.

2. Separate the product from the position

   Establish which part of the revenue is protected by the product itself, and which by distribution, data, contracts, workflow or regulation. Only the first part is exposed to reproduction cost.

3. Reconstruct rather than review

   Ask what it would take to rebuild the customer-visible product today. Reviewing what exists tells you about quality. Attempting to reconstruct it tells you about defensibility.

4. Underwrite the scenario, not the snapshot

   Three-to-five year cases under different assumptions about compression, platform behaviour and competitive response — including the case in which the thesis is right but early.

The AI adoption question can stay in the report. It simply belongs in the operational section, next to the ERP migration and the sales-ops headcount, rather than in the part that decides whether the deal is worth doing.

How this is done in a live process is described under AI Disruption Due Diligence. The margin half of the argument — why productivity gains do not reliably become profit — is in AI EBITDA is not free money.