Future CFO talks series: Most finance teams are using AI to do the wrong things faster

Notes from the third Future CFO Talks in Barcelona, where the conversation moved past "are we using AI?" to a harder question: are we using it on anything that matters?

A room full of CFOs spent a morning in Barcelona comparing notes on AI, and the most useful thing said all day was a warning about doing it badly.

Daniel Ruiz, a partner at the audit and advisory firm Crowe, repeated a line he treats almost as a religion: automate a broken process and all you get is a faster broken process. He was talking about the invoice-entry screen, the form where someone keys in supplier invoices by hand. Most teams look at that screen and ask how to make it quicker. The better question is why it still exists.

"In a few years," he said, "we'll look at that screen the way we look at a phone with a dial and a curly cord."

That captures the trap most finance functions are walking into. They are buying AI to speed up work that should not be there in the first place.

The input problem: you're automating the wrong thing

Ruiz's fix was to stop building everything around the ERP and put the data in the middle instead. For twenty years the ERP was the center of gravity. Every process had to route through it, which gave you control but made everything slow to connect. Take the ERP out of the center, he argued, and the manual screens start to disappear, because a person no longer has to be the one carrying data from one place to another.

This is the unglamorous version of "AI strategy." Before you automate a task, decide whether the task deserves to survive. Finding out is cheap now. Ruiz reckoned an AI pilot costs less than five percent of what it used to cost to replace an ERP, so the price of being wrong is low. His own firm ran one with its forensics team, pointing a private, contained language model at the mountains of documents an investigator has to read before giving an opinion. Early estimate: seventy percent of the reading time gone, with every answer traceable back to the exact paragraph it came from.

Which leads to the second trap.

The output problem: faster, and then what?

Pep Martorell, who spent a decade as associate director of the Barcelona Supercomputing Center, put it to the room plainly.

You finance people, he said, never show me a line in the accounts called "time saved."

Freed-up time only reaches the P&L if you turn it into lower cost or higher revenue. Automate a process, free up forty percent of someone's week, and nothing has actually happened yet. The value shows up only when you decide where that time goes.

That decision is the whole game, and most companies make the lazy version of it. They free the time and book it once, as a headcount cut. It works, but you can only do it once, and then you have stopped.

Iván Herrero, CFO of Cooltra, the Barcelona-based scooter-rental and shared-mobility company, showed the other version. His team had four people typing invoices into the system. After building a model to read the invoices instead, that became one person. The other three did not leave the company. They moved to balance analysis, treasury forecasting, and tax, the kind of work Cooltra never had the hands for before. Same headcount, pointed at things the business actually values. He did the same in customer support, where more than sixty percent of their hundreds of thousands of chats are now handled by AI. Rather than shrink the team, they redirected it to phone calls they had never been able to answer.

Herrero pushed back, politely, on Martorell's framing. For him, time absolutely does land in the P&L. But they were making the same point from opposite ends. Saved time only becomes money when a human decides what the freed capacity is for.

Two questions, not one

A year ago the question in most finance teams was "how do I automate this?" The room in Barcelona had moved on to two better ones.

Should this work exist at all? And once it doesn't, what do I want the freed time to do instead?

The first question kills the busywork rather than accelerating it. The second decides whether saved time becomes a one-off cut or a permanent gain. Neither is a technology question. Both are judgment calls, which is why they land on the CFO's desk rather than the IT department's.

None of this needs more software. It needs the thinking done before the software arrives, and that is the part most teams skip. At Future CFO Talks, the finance leaders who had something to show for their AI spending were the ones who had asked both questions first. Everyone else was still working out how to make the old screen run faster.