Filip Pesek is the founder and CEO of DonnaPro
Finance teams are using more AI, but still don’t have more time. Why? Because AI has changed the essence of finance work from producing to checking, and nobody has staffed the checking.
Which means that as useful as automation is, it’s not actually solving any problems or giving finance users any time. While an AI model can manage multiple functions, doing everything from drafting board commentary to flagging exceptions in a reconciliation, it can’t perform the checks. Not reliably. You still need someone with the knowledge and expertise to verify the figures against the source, check the wording, and send it to the right person for approval. And you still need people to chase when sign-off is slow, and to take ownership when things go wrong. The work is still there; it’s just different.
Automation hides the work
This is something I see particularly clearly in fintech. There are so many different use cases where automation hides what really happens behind the scenes. You have the Series A payments businesses that openly extol the virtues of using AI to prepare parts of their board pack, ignoring the fact that the finance lead has to check every figure against the underlying management accounts before use. And the lending platforms that can automate the first pass of an audit request, but still need to use people to confirm practically everything. Even investor updates still have to be manually checked before they can be sent out, even if they can be drafted automatically. And that all takes work.
That is why a finance team can be ostensibly automated, but no tasks can be completed instantly. The senior team still has a huge amount of work to do, and none of it is quick. It looks like inefficiency, but it’s actually professionalism.
The big finance checking problem
The thing is, it’s not just fintech, although it’s here that AI has been most rapidly adopted. In their 2024 survey, the Bank of England and Financial Conduct Authority (FCA) found that 75% of UK financial services firms were already using AI, and another 10% were planning to adopt it within three years. But only 2% of AI use cases were fully autonomous. All of the others still involved people somewhere in the process. And that’s not really changed. Particularly in finance.
While a marketing team can often review a draft and decide whether it works, in finance it’s all about the numbers. Everything has to be checked and verified, because everything matters. From numbers matching sources, exceptions, audit trails, and approvals. Working with materiality thresholds alone takes time.
The Bank of England and FCA found that 62% of AI use cases in financial services were considered low materiality, compared with 16% considered high materiality. So, the more AI output a finance team produces, the greater the need for verification and the time you save in the early stages through AI use almost inevitably creates more work later on.
What being AI-trained should actually mean
This is also where the idea of being “AI-trained” needs more thought. AI training should never just be about knowing how to write better prompts. It should mean knowing which finance tasks are suitable for automation in the first place. This is relevant to every sector, but in finance, there is so much at risk. And anything involving client money, regulated communications, or sensitive personal data should always be looked at twice. The question needs to be asked whether AI has a place handling that information. And when it is used, who should be held accountable when AI gets it wrong?
Something else the Bank of England research has shown is that productivity gains from AI are strongest where skilled employees can validate and refine the generated output. And that’s important, because we haven’t reached the point where people are no longer needed. And we may not ever do. People provide an essential control point.
So, where AI is used, you still need human input to check and validate. And leaders can no longer be seen to be doing their jobs if the only tech question they’re asking is how much time AI is saving. The real question is checking. Who is checking what, and when? How are results being validated? Who is giving final approval?
In most finance teams, the answer is that it all leads back to one person: the CFO, the founder, or the finance lead. AI hasn’t removed the bottleneck, it’s moved it, and it’s made it worse, because now there’s more output flowing toward the same single point of approval. That’s not efficiency. That’s concentration of risk.
This is where the checking layer matters, and where the right support changes the picture entirely. An executive assistant who understands the business, and who is trained to know which tasks are safe to automate and which need a second pair of eyes, becomes the first line of validation. They chase the sign-offs, flag the exceptions before they reach the finance lead, verify the routine figures against source, and make sure nothing regulated or sensitive goes out unchecked. Not replacing professional judgment on the high-materiality calls, but making sure the finance lead’s attention is spent only where it genuinely has to be. The output still gets checked. It just stops all funnelling through one exhausted person.
That’s the shift AI actually demands. Not fewer people, but the right people in the right place, so validation is a system, not a single point of failure.
Filip Pesek is the founder and CEO of DonnaPro, a European executive assistant agency for CEOs and founders, and a bestselling author.
