This is the question every CFO is asked and almost nobody answers straight. Ask a vendor and you get a reassuring line about augmentation, not replacement, delivered with the confidence of someone who would very much like you to buy something. Ask the internet and you get the opposite, a headline about the end of accounting. Both answers are worthless, because both are trying to make you feel something rather than tell you what is true.

The truth is more interesting and considerably more useful, and it takes some patience to see, because the evidence looks contradictory at first. Two of the most credible sources on this question appear to flatly disagree with each other. Understanding why they do not is the whole answer.

The Case That It Shrinks

Start by taking the pessimistic case seriously, because it is not stupid and it is not just hype.

Goldman Sachs, analyzing which occupations are most exposed to AI, rated accountants and auditors at the highest risk of displacement. Not high. Highest. And the logic is sound: accounting runs on structured information with repeatable patterns, and AI performs best exactly where a task has abundant historical examples, consistent formats, and verifiable right answers. Ledger entries, reconciliations, invoice matching, transaction coding. That is not a coincidence, it is a description of the work AI is best suited to take.

The evidence is not just theoretical. The Big Four have been quietly acting on it. Graduate intake fell sharply in 2025, with KPMG cutting its intake by 29 percent and Deloitte by 18 percent, while simultaneously committing enormous sums to AI infrastructure. Those are the firms whose entire business is understanding where accounting work is going, and they are hiring fewer of the people who do the entry-level version of it. That should get a CFO's attention.

So the shrinking case is real. If your finance team's work is mostly transaction processing, the machine is coming for a large share of it, and the people best positioned to know are already staffing accordingly.

The Case That It Doesn't

Now take the other side just as seriously, because the data there is at least as strong.

The Bureau of Labor Statistics projects that employment of accountants and auditors will grow 5 percent through 2034, faster than the average across all occupations. Gartner assesses AI's net impact on finance jobs as neutral through 2026, with augmentation dominating replacement. A Duke University study with the Federal Reserve Banks of Atlanta and Richmond, surveying more than 700 executives, found a negligible impact from AI on headcount in 2025 and close to zero expected effect in 2026. A JPMorgan survey of 1,500 middle-market CFOs found 60 percent expected no headcount impact at all.

And in the actual labor market, finance hiring is not collapsing, it is frantic. Unemployment for accountants and auditors sat at 1 percent in May 2026. Nearly three-quarters of finance and accounting leaders planned to increase permanent headcount in the second half of 2026, and 61 percent said skilled professionals were harder to find than a year ago. Three-quarters reported that skills shortages had already caused project delays.

That is not a profession being automated out of existence. That is a profession that cannot find enough people.

Why Both Are Right

Here is the seam, and once you see it the contradiction dissolves.

Goldman Sachs measured task exposure: what share of the work inside a role could theoretically be done by AI. BLS measured employment: how many people will actually be employed in that role. Those are different questions, and a role can score high on the first and still grow on the second, as long as the human work that remains becomes more valuable as the automatable work is taken away.

And that is precisely what the BLS numbers show when you look one line further down. Over the same decade in which accountant and auditor employment is projected to grow 5 percent, bookkeeping and accounting clerk positions are projected to decline 6 percent. Same profession, same technology, opposite directions. The dividing line is not the job title. It is the nature of the work inside it.

The work that is disappearing is high-volume, rules-based, and pattern-driven: data entry, transaction coding, matching, basic reconciliation, templated reporting. The work that is growing is interpretive: judgment, exception handling, investigation, explaining what the numbers mean to people who need to act on them. AI is not eliminating the finance function. It is eating the bottom of it and pushing everyone upward, whether or not they were ready to go.

So the honest answer to the question in the title is: neither, exactly. Your finance team probably will not shrink. But the composition of it will change, and the change is already underway.

How It Actually Shows Up on Your Payroll

This is where CFOs are most often surprised, because the effect is real but almost invisible in the way people expect.

AI does not usually arrive as a layoff. It arrives as a hire you never make. The pattern in the current data is that teams which would have added two people to handle growing volume are adding one instead, with the machine absorbing the increment. It surfaces as slower headcount growth, as an unfilled role that quietly stops being urgent, as a hiring plan that gets trimmed rather than a team that gets cut. One survey found the share of finance leaders using AI and automation specifically to reduce the need to fill open roles more than doubled in a single year, from 23 percent to 63 percent.

That distinction matters enormously for how a CFO should think about this. If you are waiting for AI's impact to announce itself as a restructuring, you will conclude it never came, while the effect accumulates silently in your hiring plan. The efficiency shows up as capacity you did not have to buy, not as people you had to let go. Which is, incidentally, the far more humane version of the same economics.

The Risk Nobody Puts in the Business Case

There is a second-order effect here that I think is genuinely underrated, and it deserves a CFO's attention precisely because it will not hurt this year.

Entry-level finance work has always been built on repetition, and that repetition was never only about getting the work done. It was how judgment got built. A junior accountant learned to sense when something was wrong by coding thousands of transactions, preparing reconciliations, chasing support, and working exceptions until the patterns became instinct. Nobody designed it as a training program, but that is what it was.

AI removes exactly that work. The new accountant sees fewer raw transactions and fewer messy reconciliations, which means fewer of the reps that used to build the intuition. And so a firm can automate its junior work, gain real efficiency, and quietly stop manufacturing the senior people it will need in five years. The bill for that does not arrive this quarter. It arrives when your bench is thin, when too few people can handle ambiguity without escalating, and when the person who could have spotted the anomaly is not there because they never had the years of pattern exposure that would have taught them to.

The controllers who are thinking clearly about this are already asking a simple diagnostic question of their own teams: how often do junior staff actually touch an exception, versus merely reviewing what the AI proposed? If the honest answer is almost never, you have an efficiency gain today and a capability gap being built underneath it.

What This Means for a Property CFO

Property finance sits unusually squarely in the path of all of this, because so much of its work is exactly the kind AI does well: high-volume recurring transactions, rent postings, invoice matching, reconciliation across entities, templated reporting for lenders and investors. If your team's days are consumed by that work, the efficiency available to you is genuinely large.

But the same fact contains the warning. The efficiency only arrives if the data is clean enough for the AI to work on, which for most property companies is precisely what is not true. A finance team drowning in reconciliation because its systems disagree cannot hand that reconciliation to a machine, because the machine cannot resolve which version is right either. The teams that will capture the headcount efficiency are the ones whose data was already coherent, and the teams that most need the relief are often the least able to receive it.

So the practical answer to "will AI shrink my finance team" is, for most property CFOs: not yet, and not the way you think. What it will do first is change who you should be hiring. The evidence is already clear that firms are struggling most to fill mid-career and specialized roles, and that the skills leaders now value are analytical and interpretive rather than task-based. The team you need in three years has fewer people whose job is to make the numbers agree and more people whose job is to understand what they mean.

Hire for that team now, while you still have time to build it, rather than discovering later that you automated the work that used to grow it.

Frequently Asked Questions

Q1. Will AI reduce the size of my finance team?
Probably not through layoffs. The evidence shows AI's near-term effect on finance headcount is close to neutral, with a Duke and Federal Reserve study of over 700 executives finding negligible impact, and 60 percent of middle-market CFOs expecting no headcount change. What it changes is hiring: teams that would have added two people add one instead, so the effect shows up as slower growth rather than reductions.

Q2. Why do Goldman Sachs and the BLS seem to disagree about this?
Because they measure different things. Goldman Sachs rated accountants highest for AI displacement risk based on task exposure, meaning how much of the work could theoretically be automated. BLS projects actual employment and expects 5 percent growth in accountants and auditors through 2034. A role can be highly exposed to automation and still grow, if the remaining human work becomes more valuable.

Q3. Which finance roles are actually shrinking?
The clerical ones. Over the same decade in which accountant and auditor employment is projected to grow 5 percent, bookkeeping and accounting clerk roles are projected to decline 6 percent. The dividing line is the nature of the work: high-volume, rules-based processing is being absorbed, while judgment, interpretation, and exception handling are growing.

Q4. If AI is so capable, why is finance hiring still so competitive?
Because the profession has a genuine talent shortage. Unemployment for accountants and auditors was around 1 percent in mid-2026, roughly three-quarters of finance leaders planned to increase headcount, and most report that skilled professionals are harder to find than a year ago. AI is changing who firms need, not reducing how many.

Q5. What is the hidden long-term risk of automating junior finance work?
That you stop producing senior people. Entry-level repetition, coding transactions, preparing reconciliations, working exceptions, is how judgment was traditionally built. Removing that work gains efficiency now but can thin your bench in three to five years. A useful diagnostic is how often junior staff actually handle exceptions rather than just reviewing AI output.

Q6. Are the Big Four evidence that accounting jobs are disappearing?
They are evidence that entry-level intake is shrinking. KPMG cut graduate intake by 29 percent and Deloitte by 18 percent in 2025 while investing heavily in AI. That is a real signal about the bottom of the pyramid, but it sits alongside continued growth in demand for experienced, judgment-heavy roles, which is the same pattern seen across the profession.

Q7. What should a property CFO do differently because of this?
Change what you hire for before you change how many you hire. The roles hardest to fill are mid-career and specialized, and the skills leaders now value are analytical and interpretive rather than task-based. Build the team you will need, one weighted toward judgment and interpretation, rather than waiting for automation to hollow out the team you have.

Q8. Can we capture these efficiencies if our data is messy?
Largely no, and this is the trap. AI can absorb reconciliation work only if the underlying data is coherent enough to reason over. A team stuck reconciling systems that disagree cannot hand that problem to a machine, because the machine cannot determine which version is correct either. The teams best positioned to gain the efficiency are the ones that least needed it.