Why Most Finance AI Pilots Die at the First Round of Internal Audit

At our recent CFO roundtables in Perth and Melbourne, one theme surfaced faster than any other: governance isn't a nice-to-have for finance AI projects, it's the reason most of them stall.

Finance teams don't operate under the same rules as the rest of the business. Data handling in finance carries obligations most other functions never have to think about — audit trails, regulatory reporting, materiality, segregation of duties, and a level of scrutiny that assumes every number will eventually be checked by someone whose job is to find the problem. That's a very different starting point to a marketing or ops team piloting a new tool.

The pattern we keep seeing

A finance team gets excited about an AI use case. They run a proof of concept. It works. And then it goes to internal audit — and stops. Not because the model was wrong, but because nobody could clearly answer the questions audit is required to ask: where did this data come from, who can access it, how is the output validated, and what happens when it's wrong.

Most POCs aren't built with those questions in mind, because they're built to prove the technology works, not to prove the technology is governable. By the time governance gets asked to look at it, it's too late to bolt on — and the project quietly dies.

Governance has to be a day-one input, not a day-ninety gate

Every finance leader in our sessions agreed on this once we said it out loud: governance can't be the thing that happens after the pilot. It has to shape the pilot from the start — the same way it shapes everything else finance does.

In practice that means being explicit, before a single model is touched, about: what data the use case actually needs (and what it doesn't), how outputs will be validated against existing controls, who owns the decision the AI is supporting, and what evidence audit will expect to see. None of this is exotic. It's the same discipline finance already applies everywhere else — it just hasn't been applied to AI yet.

The upside: finance is actually well positioned for this

Here's the part that surprised a few people in the room: finance teams' comfort with controls, audit, and evidence is an advantage, not an obstacle, when it's brought in early. Functions with less rigour around data handling often move faster on AI pilots and then hit exactly the same wall finance would have hit — just later, and more expensively. Finance just tends to hit it first, because the scrutiny arrives sooner.

The teams making real progress are the ones treating governance as part of use case design, not as a hurdle at the end. That's also exactly where the ROI conversation starts — which is the subject of our next post: the cost of frontier models, and why getting from pilot to enterprise value is proving so hard for finance to control.

Join the conversation. We're taking this discussion to Sydney next. If your finance team is trying to build AI governance in from the start rather than bolting it on at the end, come join us. Register here to save