Generative AI makes the least experienced people in a finance business noticeably better and the most experienced only a little. That should change who you hire, how you train and what you check.
I run a UK accountancy practice, and AI is now part of how everyone in it works. When people ask what it has changed, the most honest answer I have is one sentence: “It doesn’t raise the ceiling but it raises the floor.”
Everyone becomes good with it. It cannot replace a person yet, but it makes every member of the team more efficient and more effective. What it does not do is turn a competent accountant into an exceptional one, and in a finance business most of the value sits in that gap.
The research points the same way
The best evidence we have says the gains go mostly to people at the bottom of the curve. A study of 5,179 customer support agents by Erik Brynjolfsson, Danielle Li and Lindsey Raymond, published as NBER Working Paper 31161, found an AI assistant raised productivity by 14 per cent on average. For novice and low-skilled workers the gain was 34 per cent, with minimal impact on the experienced and highly skilled. The authors suggest the tool spreads the habits of the best workers to the newer ones.
An experiment with 758 Boston Consulting Group consultants, run with Harvard researchers, found the same shape. Ethan Mollick, one of the authors, wrote that the consultants who scored worst at the start had the biggest jump in performance, 43 per cent, while the top consultants gained less.
The same study carries a warning that finance leaders should read twice. One task was chosen so that the AI would give a wrong but convincing answer. Consultants working without AI got it right 84 per cent of the time, and those using it got it right 60 to 70 per cent of the time. The floor rose, and so did the number of people who were confidently wrong.
Use case: the junior’s first question
The clearest change in my own firm is where questions go. Juniors used to bring their queries to me. Now everyone has an assistant, they ask it first, fewer questions reach me, and they feel better educated for it.
I see the same in myself. AI is a very good assistant and a very good educator, and I have learned more about marketing in the last 12 months than ever before. That is the floor rising.
What it has not given me is more time. I fill the hours it saves with more projects, and the business is moving faster than it ever has.
Where the ceiling sits
I learned what the ceiling is worth before AI arrived. In the early years I hired people who were not accountants, or who had done accounting jobs without the qualification. It was very, very tough: there were huge mistakes, and I wasted years. They could do 80 per cent of the job, and I always did the other 20 per cent.
I kept asking why I could not step back from the business, and the answer was that I never had the right people. That changed when I hired a fully qualified, experienced accountant. For the first time, work got finished without coming back to me.
AI looks a lot like those early hires: very useful, but it does not do a complete job. The temptation for any finance leader is to conclude that AI plus a cheaper hire now adds up to a qualified one. In my experience the 20 per cent is the job, and it is where a confident wrong answer costs the most.
Larger institutions face a version of the same problem. In the Bank of England and FCA survey of AI in UK financial services, published in November 2024, 75 per cent of the 118 responding firms were already using AI. Yet 46 per cent said they had only a partial understanding of the AI technologies they use. A floor you do not understand is a hard one to stand on.
The hiring problem nobody budgeted for
If AI raises the floor for staff, it raises it for candidates too. Interviewing has become harder, because CVs and even interview answers are now often written with AI. The polish that used to hint at ability now tells you very little about it.
Recruiters see it at scale. In Robert Half’s UK surveys of 500 hiring managers, in finance and accounting among other functions, 48 per cent said AI-generated CVs create challenges and 67 per cent reported a longer time to hire. Matt Weston, Robert Half’s senior managing director for the UK and Ireland, put it plainly: “Generative AI has made it considerably harder to evaluate CVs at face value.”
The tell is not always subtle. More than one new starter has replied to me with the prompt still in the message. It says something serious: the work passed through them without being read.
A framework: floor work and ceiling work
My policy for the team fits on one line. AI is encouraged, but the work must read as if you wrote it and understood it. I do not mind whether a tool was involved. I mind whether the person can stand behind every sentence.
For other leaders, I would sort the work in three steps.
- Label the task. Floor work is where competence is enough: a first draft, a summary of a rule, a routine explanation. Ceiling work is where being right matters most: a judgement call, an exception, anything with your name on it. Give AI the first and a qualified person the second.
- Hire for the ceiling. AI now gives everyone the floor for very little, so stop paying a premium for it. Pay for qualification and experience, because that is the part no tool supplies.
- Test understanding, not authorship. Trying to catch every use of AI is a losing game. Ask instead whether the person could explain the work with the screen switched off.
What this means for finance businesses
AI is changing finance businesses, but not in the way the replacement headlines suggest. It is levelling up the bottom of every team, quickly and cheaply. That makes the people above the floor, and the ability to tell the two apart, worth more than they were.
The firms that do well will not be the ones using the most AI. They will be the ones that know which of their work is floor and which is ceiling, and staff it accordingly.
Harvey Dhillon is Founder and CEO of Zmartly, a UK accountancy firm for small businesses.
