Digital Economy Dispatch #303 -- What Am I Worth?

Like many people, I wear multiple hats: academic, commentator and, increasingly, investigator and paid adviser. I enjoy the advisory work and learn a great deal from it about how complex, ambiguous organisations really operate. But each time, I ask myself the same question: what value am I bringing? The wide availability of AI tools makes that harder to answer. Can’t the client just ask Claude or ChatGPT? So, in recent discussions with potential clients, one of the conversations I fear most is before I even start: agreeing a price. And it seems I am not alone.

Last year KPMG, a firm that earns its living selling professional hours, told its own auditor that the audit should cost less because AI would make it quicker. According to Financial Times reporting in February, summarised by Going Concern, KPMG International threatened to take its business elsewhere unless Grant Thornton UK passed on the savings. The fee duly fell from $416,000 for 2024 to $357,000 for 2025, a cut of about 14%.

I find it hard to think of a better illustration of where we are, and why this issue is so contentious. A Big Four firm has made, on its own behalf, exactly the argument its clients are now making to it: if the hours shrink and AI does most of the groundwork, why should the bill stay the same? It is a fair question. However, it exposes a much older problem. In professional services we were never very sure what we were paying for in the first place, and AI has made that uncertainty impossible to ignore.

By professional services I mean any work in which people are paid for expertise and judgement, and where the bill is usually set by time. Educators, lawyers and business consultants are the examples I focus on here. However, the list is much longer: auditors and accountants, architects, engineers, financial advisers, marketing agencies, software developers and many more. All of them sell knowledge by the hour, the day or the term, and all of them now face the same question. What is their work worth?

The meter was always a poor proxy

For decades, professional services have priced inputs: the billable hour, the day rate, the contact hour. There is a good reason for this. The client usually can’t judge the quality of advice, an audit or an educational experience at the point of purchase, and often not for years afterwards. Time became the stand-in for value because time was the one thing both sides could count.

That form of accounting was never very reliable. And yet, Thomson Reuters estimates that 90% of US legal spending still flows through hourly billing, a structure that rewards time consumed more than problems solved. In technology consulting, research by Source Global cited by the Financial Times found that only one in three clients considered an adviser-led IT transformation wholly successful. After more than thirty years in and around enterprise software, I am not surprised. Much of what clients bought was never the document or the deliverable. It was reassurance, a credential, or someone to hold accountable when things went wrong.

AI has not broken this model so much as exposed it. When a first draft takes minutes, hours stop measuring anything useful, and buyer and seller are forced to discuss what the work is worth. Neither side has had much practice.

Faster is not the same as better

The obvious response is to celebrate the speed. I think that’s a mistake, for two reasons.

First, in these professions time was often part of the product. A student struggling with a complex idea, a lawyer weighing a judgement, a consultant walking a recommendation through a sceptical organisation: the slowness was doing some of the work. Compressing it does not deliver the same value sooner.

Second, cheap production shifts the cost of verification to the reader. Research from Stanford and BetterUp Labs, published in Harvard Business Review, found that 40% of US desk workers had received "workslop" in the previous month: polished AI-generated material that does nothing to advance the task. Recipients spent almost two hours trying to make sense of each instance. Deloitte Australia learned the lesson in public last year when it agreed to a partial refund on an A$440,000 government report containing a fabricated court quotation and references to papers that do not exist. The courts have been blunter still. In June 2025 the High Court in England warned lawyers of severe sanctions after one filing cited 45 authorities, 18 of which were invented.

In each case, the work arrived quickly. The checking, and the liability, landed on someone else.

The consequences arrive at different speeds

What strikes me most is the breadth of the effects. They reach well beyond one sector, and they are not arriving together. I see three speeds.

The fastest is fee negotiation, where the impacts are already here. A recent Financial Times investigation found large companies demanding lower fees or taking work in-house. Greg Meyers, chief digital and technology officer at Bristol Myers Squibb, has pressed advisers to move from hourly billing to fixed-price or performance-related contracts. UniCredit cut its spending on consultants by 24% in the first half of 2026, and the share of clients expecting to use the Big Four more has fallen from 80% to 55% in a year. McKinsey says around a quarter of its fees are now tied to outcomes.

Yet the picture is not one of simple decline. Source Global still expects technology consulting spend to grow 8% this year, and one US accounting firm has begun charging premium rates for AI-assisted work. The same technology is producing both discounts and surcharges.

The technology vendors, also moving at this first speed, are no more certain. Salesforce tried charging per conversation for its AI agents and is now edging back towards per-seat licences after customers pushed for more flexibility. Meanwhile J.P. Morgan reports that token prices have risen by more than 60% since December 2025, leaving many companies with AI costs that behave like a utility bill. If the firms that build AI cannot settle on how to charge for it, we should not be surprised that those selling services built on it are struggling too.

The second speed is slower and more contested. Many fear that firms are dismantling the apprenticeship on which their future partners depend. PwC cut its UK entry-level intake from 1,500 to 1,300 last year, although its UK head pointed to a weak economy, more than AI, as the main cause. Among more than 100 leading UK law firms, trainee numbers are down just 2%, a fall Legal Cheek attributes largely to the rise of solicitor apprenticeships. The concern is reasonable. The evidence that AI is responsible is thinner than the headlines suggest.

The slowest is education. I expected the evidence to show students questioning their fees. It does not. The share of UK undergraduates rating their course good value rose from 37% to 45% this year, the highest in more than a decade, even though 94% now use generative AI in assessed work. The doubt comes from outside the university itself: Ipsos finds that only 28% of Britons think a degree is good value, the lowest of 31 countries surveyed. Universities have time, but not immunity.

So, what are they buying?

My expectation is that the fee and vendor pricing questions will come to a head very soon, because the negotiations are already under way. The others will take longer. For senior decision makers, the practical response is the same in each case: stop buying time and decide what you are buying instead. Three questions are worth asking before your next contract or renewal.

1. What are we paying for: the output, the judgement behind it, or someone accountable if it is wrong?

2. Who checks the work, and who carries the cost when the checking fails?

3. If our supplier's own AI costs double, does the contract say who pays?

None of these questions is new. Professional services have avoided them for decades by pointing at the clock. This reckoning was always coming. AI has brought it forward.

Which brings me back to my own uncomfortable conversation. When a client asks what they are paying me for, the honest answer was never my hours. Nor is it the reports I write, because Claude or ChatGPT can produce a passable first draft of those in an afternoon. What they are buying, if they are buying anything worthwhile, is judgement shaped by experience, a willingness to say what others in the room will not, and someone who puts their name to the advice and stands behind it when it is tested. That’s harder to price than a day rate, and harder to defend. But it may be the only answer that survives the arrival of AI.

So, the next time a client asks what I am worth, I intend to answer with a question of my own: what are you trying to achieve, and what would it be worth to you if we got it right…or you got it wrong?