The questions global capital now asks about your AI

Fifteen years inside global investing, and across that time more than three thousand companies have passed in front of me in diligence. I write this in a personal capacity. What I want to offer founders and fund managers in Hong Kong and Singapore who are about to raise is simple: you can build your AI story to the standard international capital now recognises, and you can do it before the money asks. The lens is knowable in advance, which means the work is buildable in advance.

Start with the instinct, because it is the thing to unlearn. When an investor asks about your AI, the reflex is to describe it: what the product does, how capable the model is, how much faster the workflow has become. That is the pitch. Diligence is a different exercise. A serious investor is not trying to be impressed. They are trying to find the line in your story that does not hold under evidence. The gap between a company that can describe its AI and a company that can evidence it is now a pricing gap, and it compounds quietly between the first meeting and the term sheet.

So here is the lens, and here is how a prepared team builds against each part of it. These are the questions a capable diligence team works through, often before they tell you they are.

Provenance and intellectual property. Where did the model and its training data come from, and what do you actually own? An investor wants to see the chain of title on the thing that creates your advantage. Open weights fine-tuned on someone else's licence, data gathered without clear rights, a core capability that lives inside a third party's API: each is a question about whether the value is yours to sell. The buildable answer is a clean, documented chain of title you can put on the table.

Human oversight. Where does a person sit in the loop, and what happens when the model is wrong? Prepared teams can describe the review step, the escalation path, and the cases the system is not permitted to decide on its own. The absence of that answer reads as unmanaged risk, not as confidence.

Failure modes. Every model fails. The question is whether you know how yours fails, how often, and what it costs when it does. An investor is reassured by a team that can name its error surface and show the controls around it, and unsettled by a team that treats the model as if it has none. You build this by measuring it before they ask.

Unit economics of the AI claim. What does inference actually cost per transaction, and does that cost fall or rise as you scale? If your margin depends on token prices staying low, that is a dependency, not a moat. The diligence team will model it whether or not you do, so model it yourself first.

Vendor dependency. If the capability rests on one external provider, your roadmap and your cost base are exposed to that provider's pricing, terms, and continuity. Concentration here is a real risk, and a credible answer addresses substitutability rather than waving it away.

Attribution of uplift. When you claim AI improved a metric, can you separate the model's contribution from everything else that changed? A clean, modest, defensible number beats a large one you cannot stand behind when a new sales team and a pricing change landed in the same quarter. Clean attribution is something you can build into how you measure from the start.

None of this is exotic. It is the standard international capital now applies, and the uncomfortable part is that most teams, asked these questions plainly, can evidence fewer than half of them. That is not a failing of founders. The questions arrived faster than the playbooks.

The discipline is to see your own AI story the way a diligence room will see it, on evidence, and to do that quietly while you still control the words. The lens is knowable in advance, so the gaps are findable in advance, so they can be closed before the meeting rather than discovered inside it. It is far cheaper to find the weak answer yourself than to have it found for you when the money is in the room.

The companies that win the next round will not be the ones with the boldest AI claim. They will be the ones whose claim survives the questions.

Noel Lam, Founder and Managing Director, VeriQ Standards. If your next raise is on the horizon, the free Diligence Probe Library sets out the real questions, and a sample read shows the shape of the work.