Two different questions
When AI comes up in a deal, it usually means one of two things. Either the target says AI is part of its value, or you believe AI will help you create value once you own it. Both claims are common now, and both are easy to make in a management presentation. Diligence is where you find out which parts are real.
These are different questions and they need different work. The first is about what exists today. The second is about what it would take to build. Mixing them is how buyers end up paying for a capability that turns out to be a plan.
What the target really has
Start with specifics. Ask management to show you, not describe, where AI is used in the business today. For each use, the questions are simple.
- What decision or process does it change, and who relies on it day to day?
- What results has it produced, measured how, and over what period?
- Is it built in house, bought from a vendor, or a thin layer over a third-party model?
- Who built it, and are they staying after close?
- What does it cost to run at today's volume, and at twice that?
The answers usually sort capabilities into three groups: those that are core to how the business runs, those that are useful but easy to copy, and those that exist mainly in the deck. Each deserves a different weight in your valuation.
Be wary of a capability that only one or two people understand. If the model, the data pipeline and the know-how all leave with a founder or a small team, you are buying a dependency, not an asset. Retention terms and documentation belong in the deal for anything you are paying for.
Look hard at the data
In tech-enabled services, the most valuable AI asset is often not a model. It is proprietary data about clients, transactions, assets or operations that competitors do not have. Diligence should establish what data the target holds, how clean it is, and whether it has the rights to use it the way your value plan assumes.
Ask about the risk side as well. How is sensitive data handled, which vendors see it, and what oversight exists for automated decisions that affect clients? Weak answers here are not always deal breakers, but they belong in the integration plan and sometimes in the price.
Test your own value plan
The second question is harder because the person making the claim is you. If your investment case assumes AI will lift margins or growth after close, put that assumption through the same test you would apply to the seller.
Where exactly
Which processes or offers change, and which number moves in each?
With what
Which data, systems and people does the plan depend on, and does the target have them?
By whom
Who owns delivery after close, on your team or theirs?
By when
What has to be true in the first hundred days for the timeline in the model to hold?
This is where many value plans thin out. The idea is sound, but the data lives in three systems that do not talk to each other, or the team that would do the work is the same team running the integration. Better to find that now and price it in.
Plan for the first hundred days
Whatever diligence finds, decide before close who will own AI in the combined company and what the first priorities are. Integration is a rare moment when processes, systems and teams are all open to change at once. Companies that use it to build AI into the new operating model move faster than those that wait for things to settle.
Diligence is where you find out which parts of the AI story are real, and what the rest would cost to build.
Bring someone into diligence who has run AI programs inside an operating business, not only evaluated them. The questions above are straightforward. Knowing which answers to trust comes from having been on the other side of them.
