“The exoskeleton, not the robot”: what AI’s impact is on jobs in finance

There is a scene that plays out somewhere in the world every few seconds. In fact, the chances are that you’ve probably experienced it yourself. You open LinkedIn, and it’s there: Accountants are finished, apparently. The CFO is next on the list. Reading on, the whole profession seems to have a shelf life measured in software release cycles.

This doomsday scenario is seemingly ubiquitous, and somewhere along the way, it’s easy for the reader to stop seeing a possibility and start seeing a prophecy. But it’s important to remember that social media being social media, this is only one notion of what AI might do, reposted and reworded until it’s everywhere.

The grim verdict is what feeds the algorithm of course, with no grounding other than ‘what if?’. So it was a relief, during a recent ApprovalMax webinar on the future of the profession, to hear someone posit a better picture. Dan Schonfeld, ApprovalMax’s CFO, was asked the question that hangs over every one of these conversations: If AI ends up running the reconciliations, the invoice processing, the anomaly detection and the audit trails, then what is the human being actually there for? His answer turns the question on its head.

“AI is like an exoskeleton, it’s not a robot. It doesn’t replace the human being. It’s an augmentation. You basically, as a human being, you step into it and then you’re suddenly faster, more powerful, you can do more stuff. There’s still the human being inside.”

It sounds like a small swap of metaphor, but it carries a great deal of weight. A robot is the thing that takes your place. An exoskeleton is the thing you climb into so you can lift more than you could before. One is a story about being made redundant. The other is a story about reach, and based on the evidence so far, the second story is the one that better matches what is happening.

AI is reshaping finance jobs by automating routine work rather than replacing professionals — 58% of finance functions already used AI in 2024, up 21 percentage points in a year (Gartner), mostly for process automation, anomaly detection, and analytics. The roles most exposed are those dominated by quality-assurance checks. What stays human is judgment, ethics, client relationships, and accountability - the discretion and liability no software can absorb. The practical rule: keep a human in the loop on every final decision.

The job was already changing long before AI showed up

Here is the part most of the doom coverage misses. AI did not start this transformation at all. It walked into a profession that had been quietly reinventing itself for years, and handed it a faster engine.

For a long time now, what we expect from finance people has been drifting away from recording what already happened and towards helping shape what happens next. Schonfeld described it as a move “from archaeology to astronomy, so from looking backward to looking forwards, from analyzing the past and providing a reflection of what happened to trying to help understand what’s going to happen and help figure out the future.” And he was careful to add that none of this is new. “This has been happening before AI,” he said. “That expectation has been growing... but that has been in place for years now, and this is not news.”

The rest of the profession tells the same story. A 2026 report covered by CPA Practice Advisor described a decisive shift away from backward-looking reporting and towards proactive value creation, with finance people increasingly embedded inside business units so they can influence decisions at the point they get made. The Swiss business school IMD put it about as plainly as anyone has, noting that the modern finance function “was already forward-looking, strategic, and deeply embedded in decision-making. What AI changes is the speed, scale, and depth at which this role can be exercised.”

That is the exoskeleton again, only dressed in business-school language. The road was mapped out long ago. AI is simply what lets a finance team of exactly the same size cover a great deal more of it.

What it actually looks like inside a finance team

Strategy is easy enough to talk about in the abstract. Where it gets real is in the question of time, and specifically how much of it you have and where it goes.

Senior people in any business tend to share one chronic complaint. “The one thing that they’re always short of is time,” Schonfeld said, and the consequence is that “leadership teams find themselves extremely poor on time to think long term, to be strategic, to really take a step back, look at the environment... and think long term.” There is never quite enough left over for the work that supposedly matters most.

So this is the bit where the augmentation earns its keep, and notably it does so without anybody losing their job. What shifts is the balance of how the hours get spent. For a team of the same size, Schonfeld explained, AI “gives you a better mix of time between the, let’s say, tactical operational day-to-day stuff and longer-term strategic work.”

In his own team that takes a fairly tangible shape. They run automated agents that keep an eye on the competitive landscape, pulling together what the significant players have published or announced and feeding it straight into senior leadership discussions. But the interesting part is not the clever automation. It is that the time it buys back gets pushed upwards, into the kind of judgment-heavy thinking that no machine can do for you. The whole thing, as he put it, “allows the entire business to be more strategic.”

And the hard numbers point the same way. Gartner found that 58% of finance functions were already using AI in 2024, a leap of twenty-one percentage points in a single year, and the most common uses were process automation, anomaly and error detection, and analytics. Which is to say, the repetitive, pattern-spotting grind that has always eaten up far too much of a finance professional’s week.

The quiet squeeze in the middle

If there is a genuinely awkward part of this story, it is sitting neither at the top of the org chart nor at the bottom. It is somewhere in the middle, and to their credit the panel did not dodge it.

Alastair Barlow, who has spent twenty-five years as a finance practitioner and is now building an AI-native accounting company of his own, made the case that runs against the obvious assumption. The senior people, he argued, “exercise judgment, experience, relationships, so that’s less at risk.” The juniors, meanwhile, are inexpensive to employ and can be amplified enormously, given what he called “a superpower by using some of this gen-AI where they can maybe punch way above the weight than what they could have done before.” Which leaves, in his words, “the bit in the middle.”

The temptation is to read that as the wholesale disappearance of mid-level roles, and Schonfeld stopped that reading in its tracks. “You seem to be thinking about this in binary terms, as in we either have mid-level roles or we don’t,” he said. “And it’s not binary, it’s just less.” His image was not of a flattening pyramid but of one that changes its angle. Future partners will still come up through middle management the way they always have, he reckoned, “however, there’s going to be fewer of them.”

So what is really exposed here? Not a particular layer of people so much as a particular kind of work. “A lot of what they do is QA,” Schonfeld pointed out, “and QA is very much at risk, because AI is exceptionally good at QA. It ultimately is a pattern recognition machine.” If your week is mostly checking other people’s work for errors, you should be paying attention. As he put it, “if somebody’s job is like 90% QA, I’d be worried.” The role itself tends to survive. What has to change is the substance of what fills it, which needs to climb a few rungs up the value chain.

Where the human refuses to be automated

This was the part of the conversation that mattered most, and it happens to be the part the headlines almost never bother with. If AI really is this capable, the honest question is where it ought to stop.

The panel had a clear answer, and it came down to four things that resist being turned into a formula. Judgment. Ethics. Relationships. Trust. These are not the consolation prizes you hand a human because there is nothing more useful for them to do. They are the genuinely irreducible parts of the work, the bits a spreadsheet has no way of touching.

Schonfeld built his case around discretion. “If everything was mechanistic and everything was a formula you could trust software to do it perfectly, but it isn’t,” he said. “There are intangibles... that go into decision-making that are currently at least impossible, in my view, to quantify.” He was thinking about the soft, human reading of a situation. Knowing your audience. Getting a feel for the character of the people across the table. Sensing the philosophy that sits underneath a board’s position. “These are all very, very soft elements that currently software can’t capture,” he said. “You can do a lot of the analysis through software, yes... But there’s that additional layer of judgment, of discretion at the end, which is really where the value happens.”

Barlow took the same instinct somewhere even more fundamental, which was liability. Accountants, he argued, “inject trust into the entire system,” and trust turns out to be tangled up with accountability in a way no software vendor is ever going to sign up for. “Are we ever going to see a vendor have professional indemnity insurance?” he asked, and the question rather answers itself. The big firms carry billions of pounds of lawsuits between them, and the notion of a software provider volunteering to absorb that is, frankly, fanciful. When something goes wrong there has to be, in his words, “someone to blame, someone to... be there to pick up the liability.” That someone is always going to be a person.

And none of this is wishful thinking dressed up as reassurance. It has become something close to the mainstream professional view. The consultancy McCracken warns that organisations risk leaning so heavily on automated insight that they lose proper human oversight, precisely because AI systems are brilliant at spotting patterns yet have none of the business context, industry knowledge or strategic judgment that an experienced finance professional carries around in their head. Follow the machine blindly, the firm cautions, and you end up making strategic mistakes you would never have made yourself.

Drawing the line: a practical set of dos and don’ts

If the real question is where the human stops and the machine starts, it helps to get specific about it. And there is a principle worth borrowing from how ApprovalMax has built its own product, which is that a human always stays in the loop. As Schonfeld put it, the AI features in the product "are always designed with the basic assumption that there's always a human in the loop. We never build something that takes away discretion from a human being." That is not a marketing line bolted on afterwards. It is the assumption the approval workflows are built around, and it turns out to be a remarkably good rule for deciding where AI belongs in any finance team. So here, roughly, is where the line sits.

Do hand over the high-volume, rules-based grind. Reconciliations, invoice processing, transaction-level anomaly detection, the first pass at variance analysis. This is the territory where the software beats a human working by hand on both speed and accuracy, and it is exactly where Gartner's figures show the adoption has already gathered. None of it asks the machine to make a judgment call, which is precisely why it is safe to let go of.

Don't hand over the final call. This is the line that matters, and it is the one ApprovalMax draws inside its own product. The machine can surface ideas and flag the patterns, but a person makes the decision and a person owns whatever follows from it. The moment AI starts quietly deciding rather than suggesting, you have handed away the very thing a finance professional is there to provide.

Do use it to widen your field of view. Baselining a client against its peer group, digesting more data than you could ever read alone, clawing back the hours that real strategic thinking demands. But you should not treat whatever it produces as the verdict. It is a place to start and nothing more, something to bring to a human who then decides what to do with it. Barlow had the nicest way of putting it, saying that AI "may not bake the cake itself, but it'll give you a lot of the ingredients."

Don't be careless about what you feed it. Think hard about the data you hand over in the first place. As Barlow noted, you "need to be comfortable with the data you're putting in, either anonymized or you need to be in a secure environment." The trust has to run in both directions, which is rather the point. A human stays in the loop not only on the decisions coming out, but on what goes in to begin with.

Breathing new life into the role

So which is it, then? Is AI lifting finance into something richer, or quietly hollowing it out? On the strength of this conversation, it’s the former, and it manages it by clearing away the parts of the job that were never really the point in the first place.

The reconciliations and the audit trails were always the price of admission rather than the thing of value. What clients and boards are actually paying for is judgment, a sense of where things are heading, the trust that lets a number be acted on with confidence, and a human being prepared to stand behind it when it counts. AI cannot manufacture any of that. What it can do is sweep enough of the routine off the desk that finance people finally have the room to deliver it.

Which is the exoskeleton doing its work. The accountant, the controller and the CFO are not being written out of the story at all. They are being asked, after years of being too busy for it, to do the part of the job that was only ever theirs to do. Not replaced by the machine, but standing inside it, reaching further than they ever could unaided.

This piece draws on ApprovalMax’s webinar “If AI runs the numbers, what do you run?”, featuring Dan Schonfeld (CFO & COO, ApprovalMax) and Alastair Barlow (finance practitioner and AI-native accounting founder), in conversation with Dan Cockerton of the Digital Accountancy Show.