AI in Professional Services: 5 Things We're Noticing

Over the last few years at Acuru, we've had hundreds of conversations with people across professional services, from juniors and managers through to partners and global L&D heads.

Nothing has quite matched the last few months in terms of what's happening with AI. There's a heady mix of excitement about the possibilities (more efficient work, higher margins, less dependence on external software) alongside underlying concerns that seem squarely existential to the business model.

There's been plenty written about the big AI trends, but what interests me is what's happening at a cultural and operational level. Changes in how people work, what they're valued for, and how firms respond to the pressures coming from clients.

The sort of stuff discussed at the watercooler but not necessarily making it into the Financial Times.

Here are five things we've been noticing.

01The Apprenticeship Model in Flux

The other day I was chatting to a friend who is now a partner in a large professional services firm. We were laughing about our relative levels of cluelessness when we first started out in our careers, and reflecting thankfully on how forgiving the apprenticeship model had been at the time.

We learned through drafting the first version of the memo, the working papers, the research note. Someone senior covered it in red ink. Over time the junior person absorbed context, built pattern recognition and slowly developed their own intuitions for what excellent work looked like.

It was slow and not especially efficient. But it worked, and it has underpinned the professional services model for generations.

AI is now doing much of that early-stage drafting. Which is great for the timesheet, and a real problem for the apprenticeship.

The pressure on firms is to make juniors useful much sooner. Clients are understandably less keen to pay for work a model can produce in minutes. And yet the things that clients are willing to pay for - spotting the issue that matters, knowing when something doesn't smell right, having a view, were precisely what those drafting years were building.

Nobody seems quite sure what the apprenticeship of the future looks like. Firms need judgement sooner while AI threatens the work that built it. Questions are being asked about how to compress a learning curve that traditionally took years, without losing what made it valuable in the first place.

02The Wildly Uneven Distribution of Gains

Two years ago, we thought there were three things that would determine how valuable someone would be in professional services: their technical knowledge, higher-order skills like judgement, synthesis and problem-solving, and their facility with AI models.

As the models have improved, the first and third categories have become much less of a differentiator.

On technical knowledge, anyone with access to a good model can now draw on an extraordinary breadth of information, tailored to their particular circumstances.

As for AI proficiency, an unbelievable amount of routine knowledge work can now be done through a simple chat interface. The major models can work directly with documents, Excel spreadsheets and PowerPoint presentations, without requiring much technical know-how.

In other words, professionals can now access a substantial share of AI's efficiency gains without having any particular technical expertise, or even much interest in the technology itself.

But while access to AI has been democratised, the gains are anything but evenly distributed.

Much of a firm's real value lies in its internal precedents, house views and accumulated experience, often sitting in the heads of directors and partners and difficult to fully codify. AI may actually widen the gap between those who possess that knowledge and judgement and those who don't.

Take advising a client on setting up a foreign corporate structure. There's a big difference between producing technically sound advice and getting it implemented. Knowing how the tax authorities are likely to approach it, who to speak to in the relevant jurisdiction, and having the contacts and initiative to make things happen.

We've heard plenty about the 10x engineer in technology. I wonder if we're going to see something similar in accounting, law and consulting, where the billable-hours model has traditionally maintained a fairly linear relationship between time spent and revenue generated. AI could start to break that relationship, with those who can use it to amplify their existing judgement, reputation and ability to get things done pulling much further ahead.

Which brings us back to the apprenticeship. If judgement is one of the things that separates the 10x professional from everyone else, firms can't afford to leave its development to chance.

03The Review Bottleneck

One of the more immediate consequences of AI is the sheer amount of material people can now produce.

Most of us can get something to a reasonably convincing 80% in a fraction of the time it would previously have taken. Reports, presentations, research, analysis. AI is a ceaseless machine, happy to keep producing as long as you keep asking.

The problem is that someone still has to review all this stuff.

And there's something about AI-generated writing itself that makes that job particularly difficult. Its probabilistic tendency towards the generic, the abstract and the bland. That almost alabaster smoothness to the sentences, the uniformity of tone and structure that makes your eyes slide over the words without quite taking them in.

Human writing, especially from a variety of people, has more variation, specificity and occasional flaws or awkwardness. Things that catch your attention and force you to think about what you're reading.

AI can be "bad" in new and perplexing ways, producing errors that are difficult to see or anticipate, all beneath a convincing veneer of confidence and correctness.

So the reviewer has three problems: more work coming their way, more judgement required to distinguish what's correct and valuable to the client, and writing whose very fluency can make errors harder to spot.

And if directors and partners are spending increasing amounts of time reviewing and correcting material that would previously have been developed and refined several levels below them, what happens to the margins?

04'Slop' Enters the Chat

On that note, the term "AI slop" seems to be making its way into the historically polite and diplomatic discourse of professional services, an industry where even an egregious error, perhaps three zeros out, might traditionally have been met with a gentle suggestion that "you might want to take another look at that."

Until recently, producing a polished report or presentation required a fair amount of effort. It wasn't necessarily an indication of great thinking, but at least there was usually some work and thought behind it. Firms are now grappling with the need for a new language and new feedback frameworks for work that looks good but is actually bad.

I've written before about how AI slop has become a real tax on doing business, and I've tried to adopt a zero-tolerance approach to it, both as an originator and a recipient. But in a knowledge economy where the written word is the coin of the realm, how do we reap the productivity gains of AI without paying an ever-growing slop tax?

As polish becomes increasingly easy to produce, the real value of professional advice comes into sharper focus: a clear point of view, a well-judged question and a recommendation that someone has the expertise and conviction to put their name to.

I think there will also be value in the unmistakable tells of a living, breathing human behind the output. Could the passive-aggressive work email, neutralised into oblivion by AI models, be poised for a comeback?

05The Asymmetric Downside Lurking in Efficiency Gains

There's a reason the old saying "Nobody ever got fired for buying IBM" became so well known. A large part of what clients pay for in professional services is confidence. Confidence that the difficult questions have been asked, the risks properly considered, and that a firm with the necessary expertise and reputation is prepared to put its name to the advice.

For a board making a major decision, or a company facing regulatory scrutiny, that reassurance is worth paying a premium for.

AI offers tantalising efficiency gains, but the downside risk hasn't gone anywhere. If a transaction unravels because of an overlooked tax issue, or a regulator challenges a conclusion that should never have been signed off, nobody will care how quickly the report was produced. One serious error can undo the savings from hundreds of successful engagements, with reputational consequences that go well beyond the financial cost.

And this brings several of the earlier threads together. Firms are under pressure to review work more efficiently, just as AI makes errors harder to spot and removes some of the routine drafting through which professionals traditionally developed their judgement.

The very capability that clients are ultimately paying for is becoming both harder to develop and harder to exercise at scale.

What Does This All Mean?

What's interesting about these trends is that they're not all pointing in the same direction. Some work is getting dramatically easier, while other parts, particularly reviewing, challenging and taking responsibility for it, are getting harder.

But the hardest problem may be the one firms have taken for granted longest: how people actually become good at this job.

For generations, expertise was built almost by accident, as a by-product of years spent doing routine work under the eye of someone more experienced. If AI removes much of that routine work, expertise will have to be built on purpose.

That's the work we're doing at Acuru. We're giving professionals structured practice in identifying issues, applying judgement, questioning plausible-looking outputs and making recommendations they'd be prepared to stand behind. It won't replace years of professional experience, but it can help firms build those capabilities much more deliberately, rather than leaving their development to chance.

The firms that get this right won't just have more efficient juniors. They'll still have partners in fifteen years' time.

Next in Acuru InsightsThe unreasonable difficulty of creating a firmwide AI adoption policy in professional services.