Twelve CFOs sat around a table in April. Chatham House Rules, no agenda beyond one honest question: where do you actually start with AI in your finance team?
Before they walked in, we’d surveyed nineteen finance leaders from the same group. The results were sobering in their sameness.
Fourteen had either not started with AI at all, or had team members quietly experimenting on their own with Claude or ChatGPT. Only two had it embedded in the day-to-day.
When we asked what was stopping them, one answer dwarfed everything else. Time.
The paradox nobody wants to own up to
The session opened with an observation that got a rueful laugh, then a long silence.
Every person in the room was too stretched to adopt the thing that would make them less stretched. Too busy with close, board, reforecast, close, board, reforecast to carve out the afternoon that would break the cycle.
It’s a funny sort of prison. The job fills the week. The week fills the month. The month-end fills the quarter. And every week of deferral widens the gap between where your team is and where the market is moving.
One CFO in the education sector was refreshingly honest…
‘Unfortunately I don’t have any answers. I’m here with more questions.’
In a room of twelve, that wasn’t an admission of failure. It was the honest starting position for most of the group.
Deloitte’s 2026 survey of 3,235 senior leaders put ‘insufficient worker skills’ as the biggest barrier to AI adoption. Grant Thornton’s research found that training is the single most underfunded piece of AI investment, and a third of finance leaders said so outright.
That isn’t a technology problem. It’s structural. Until somebody creates protected time, the paradox keeps feeding itself.
The bottleneck nobody expects
Peter Beard joined us as the external voice in the room. He sits on the board of the ICAEW and has walked hundreds of finance teams through AI implementation. His reframe landed in one sentence.
‘The challenge is coming up with the idea of what it can do. I don’t know what it can do, so I don’t know what use cases it can apply to. That’s what we’re seeing a lot of.’
Teams don’t need smarter tools. They need a clearer view of what to ask those tools to do for the specific work in front of them.
He described a one-day exercise he runs with finance teams. Everyone writes fifteen to twenty-five tasks they do in a typical month. Each one gets ranked on three factors. Effort to set it up. Reward if it works. Risk if it goes wrong. Risk runs from ‘a note to myself’ at the low end through to ‘the VAT return’ or ‘the annual report’ at the top.
A recent exercise of this kind produced one hundred and thirty-five mapped use cases in a single day. By the end, the CFO knew the top ten to action the next morning. Not in six months. Tomorrow.
The ROI he’s seeing across those implementations is enough to make anyone pause. Three to eight hours saved for every hour invested. Teams using general-purpose AI for ten-plus hours a week once they’re set up. That’s before you count the long tail of small wins.
What’s actually working
This is the part of the session where people stopped crossing their arms and started writing things down.
One CFO had trained a custom ChatGPT on every member of his board. Public writing, previous roles, speaking appearances. Before sending a paper in, he runs it through ‘Board GPT’ and asks how each director is likely to react, individually and collectively. He reckoned it had saved him from a couple of nasty surprises in recent months.
Another takes the finished board pack and prompts Claude to behave as a PE investor. Pick holes. Challenge the assumptions. Find the awkward question you don’t want to meet cold across the table. Cost: minutes. Value: one fewer reason to lose sleep the night before.
A PE-backed software CFO had built a Claude project for monthly commission calculations. Drop the Salesforce export in, get formatted commission statements back in a standard template. Hours of Excel work now runs itself.
The same CFO described a scheduled task that reads his meeting transcripts each week, compares them with the week before, and feeds back his own communication style. His verdict: ‘It’s not pleasant. But it is very helpful.’ The best line of the day.
Four people. Four different use cases. All built in general-purpose tools. None of them a stroke of genius. All of them started with someone finally giving it thirty minutes.
The barrier isn’t as thick as it looks
The line that kept coming back to me… phrased differently by different people: the barrier isn’t thick. It looks thick because you’re standing outside it.
The preparation work that used to fill a finance team’s week is being quietly taken over. The question is no longer whether AI will reshape finance. It already is. It’s whether you lead that adoption, or whether someone on your team starts doing it on the side, in tools you haven’t approved, with data you can’t see.
The CFOs in the room who were furthest along weren’t cleverer, didn’t have bigger teams, weren’t on fatter budgets. They’d simply stopped treating AI as a strategic project and started treating it as a thirty-minute experiment on a quiet Tuesday afternoon.
Where to start this week
If you only do one thing, do this.
Get every person in your finance team to write down fifteen to twenty-five tasks they do in a typical month. Rank each one by effort, reward, and risk. Pick the top three where the reward is high and the risk is low. Start there.
Don’t wait for a platform decision. Don’t wait for a budget line. Don’t wait for the week after month-end, because there’s always another month-end.
The teams that crack this never did it by finding a whole week free. They did it in thirty minutes between one thing and the next.
The capacity will come back. But only once you spend some of the capacity you already haven’t got.
The next PE CFO Roundtable is already taking registrations. Chatham House Rules, no pitch, just twelve finance leaders in a room working through what’s actually on their plate. If you’d like a seat at the table, register here.
