BILL’s latest US survey of accounting firm leaders found that 85% expect AI to improve their business model, but only 17% are planning genuine transformation. As those figures circulate through UK trade commentary and LinkedIn debate, they’re being read almost entirely as a pricing problem. We think that reading misses the point, and skips past several questions nobody in the discussion is asking.
Daniel Lawrence | Founder & CEO, bots for that
BILL’s fourth ‘NewtonX-run Accounting Firm AI Ambition Survey’ has been circulating widely this week, and the headline numbers are stark: 85% of US firm leaders expect AI to improve their business model over the next five years, yet only 17% foresee genuine structural transformation. The remaining 68% report that they’re treating AI as an efficiency upgrade rather than a reason to rebuild anything.
The way this has been picked up and discussed across UK accounting commentary has settled into a familiar shape: firms are stuck on legacy pricing models, hourly billing collides with automation, and the fix is pricing reform, followed swiftly by a shift into Client Advisory Services (CAS). [I can almost hear UK firms collectively groaning and rolling their eyes at that insight.]
I don’t believe this shape is necessarily wrong, but I do feel it’s possibly incomplete in a way that’s important, and in places I think it gets the causality backwards (just my view, from my experience across this and other sectors). In this short article, I outline where I think the prevailing UK interpretation might fall short, followed by my own set of observations that I don’t feel the discussion has touched yet, some of which I’ve addressed at greater length in my own work on AI adoption in the profession.
Part One: Where the UK Read May Fall Short
1. Pricing lag is a symptom, not the disease
The accusation that appears to be taking hold is that firms are slow to reprice because of partnership-model inertia and an attachment to the long-standing billable hour. But I’m finding that any pricing reforms in relation to the impact of AI, generally depend on one clear thing that has to exist first: structural visibility into exactly where AI is saving time, on which tasks or outcomes, for which clients, and at what confidence level. Most firms don’t seem to have that, any side of the pond. They have deployed tools or agents for sure, but without the governance layer needed to measure what those tools or agents actually changed and what they cost.
You can’t reprice what you cannot see, at least not in a defendable way. The advice doing the rounds seems to be to ‘audit your fee structures’, is only possible once that measurement layer exists. And anyone who has truly tried to get to grips with the ‘tokenomics’ of AI at scale will know this all too well.
2. A US survey wearing a UK verdict
The underlying data comes from BILL’s fourth NewtonX-run survey wave, conducted in late 2025 among more than 200 accounting firm leaders. Nothing in BILL’s own methodology note claims a UK sample, yet much of the commentary circulating in UK trade press and on LinkedIn treats the findings as a verdict on UK managing partners. UK and US practices sit inside different pricing cultures, audit regulation and private equity dynamics, and none of the survey’s headline percentages have been demonstrated to hold on this side of the Atlantic. That distinction deserves to be made explicit, not folded in silently and left for the reader to assume.
3. The billable-hour ‘collision’ is not a new phenomenon
This tension is not new and is not an AI-specific point of contention. Cloud accounting, practice management software and automated bank feeds have all compressed billable hours long before AI arrived. What is genuinely new with AI is the height of exposure, not the issue itself. I don’t think the constraint has ever been the technology, as much as it has capacity to absorb change at the pace that technology and people allow, which is a pattern the accounting profession has experienced a few times in the past twenty years.
CPA.com’s Kimberly Blascoe, quoted in an earlier volume of the same BILL survey series, makes the same point independently: the movement toward CAS has been debated for years, and what AI adds is the ability to capture and interpret the data CAS depends on, not the underlying shift itself.
4. The talent section is missing a framework, not just nuance
The survey reports that 52% of leaders expect headcount reductions overall, rising to 68% among large firms, while 66% expect material skill shifts. When you read that first, you’d be forgiven for thinking those numbers seem contradictory. They aren’t, once you separate the work or tasks themselves into what is genuinely routine and automatable, what is tacit and only ever approximated by an AI model, and what is accountable and irreducibly human. Headcount can fall in the first category. Skill shifts tend to happen in the second. The third category is where the profession’s durable value really sits, and this survey (or any other) never seems to address which category is actually shrinking (if at all).
5. The case studies are carrying more weight than they can bear
Several pieces circulating alongside the survey seem to lean on anonymous illustrative case studies, a mid-tier firm in the North West, a three-partner firm in London, that are unlinked and impossible to verify. That doesn’t make the underlying survey data wrong, but it’s worth separating BILL’s verified statistics from the vendor-style anecdotes layered on top by secondary commentary. The distinction matters more in a profession that is, rightly, sceptical of unverified claims dressed up as case evidence. I know I am.
6. The most important number in the whole report is getting the least attention
Only 9% of leaders report substantial results scaling capacity without adding headcount. That’s the real headline I think, and it’s barely featuring in the commentary building up around the survey. It says most firms can’t yet convert AI-driven capacity into anything, not primarily because of which pricing model they chose, but because they lack the operational infrastructure to redeploy saved hours at all. Everything else in the discussion is downstream of that one figure. I think it needs to be turned around, it’s a leading indicator.
BILL’s own data sharpens this further. Scaling capacity without adding headcount is described as one of the most cited AI goals among surveyed leaders, yet it carries just a 9% result rate. Meanwhile only 29% of firms set a goal to build new AI-powered revenue lines at all, and of that minority, one in four reported substantial returns, the highest success rate of any goal measured in the survey. Taken together, those two figures might be saying something more visceral than the headline most of the commentary settled on, that firms are concentrating effort on the goal least likely to pay off, and neglecting the one most likely to.
7. ‘Appoint a process owner’ is gesture governance
A recurring recommendation also doing the rounds is to appoint a named process owner. I advocate for this too, but not without any substance. A named process owner without any actual real data controls or structural guardrails underneath them is merely policy intent with a job title. It sounds like progress on a slide or board deck, but it changes next to bog all in practice. Real governance is behavioural and structural: it lives in how systems are configured, how data flows are controlled, how people behave and act, not in who chairs the AI steering meeting.
Part Two: What the Discussion Isn’t Asking
The points above are about where I think the prevailing UK reading of this survey falls short. The points below are things I think matter more than anything in that discussion, and that nobody seems to be asking.
You don’t know what you don’t know
Very few firm leaders seem to genuinely understand how AI works, and understanding is the precondition for knowing where and how to use it well. This is what’s called the ‘productivity paradox’ playing out in real time: a transformative technology gets treated as an incremental one, because the people evaluating it are missing the frame required to see the transformation. I don’t think the 17% versus 83% split in the survey is a courage gap. I think it’s a comprehension gap. You can’t restructure the future of a firm around a technology you have already filed under ‘slightly faster version of what we already do’.
Drowning in manual, repetitive work? Tell us the task and we’ll show you what to automate.
Reimagining a firm is genuinely hard, and uncertainty is genuinely scary
Even leaders who suspect AI is more than an efficiency tool often don’t know how to picture what their firm would look like rebuilt around a different model, and many would rather not try. That is not laziness. People make decisions from what they already know, and there is no template on the shelf for ‘accounting firm rebuilt around AI-native workflows’ the way there was for ‘accounting firm that adopted the cloud’. It’s much easier to debate pricing tweaks than to sit with the discomfort of not knowing what the destination might actually look like.
BILL’s own advisors make a version of this point their survey results, even if most of the surrounding commentary doesn’t carry it through. Ariege Misherghi is quoted in the underlying research, warning that if AI has only sped up a broken workflow, firms have missed the point entirely, the real bar is whether work was actually deleted and capacity freed for judgement, not simply accelerated. That is the comprehension gap stated plainly, from inside the source material itself.
Pricing is dynamic, and most of the commentary treats it as a one-way ratchet
Several assumptions running through the current discussion are deserving of further analysis:
- Cost does not disappear when AI enters a workflow, it moves. Data governance, model oversight, human-in-the-loop review and security all become new cost centres. Today’s AI pricing is artificially low relative to what running it properly actually costs, and that gap will close (ignoring all the talk and consequence of the bubble bursting for the moment).
- If time (to complete tasks / deliver outcomes) falls but cost does not fall in step, this doesn’t automatically point to a lower price for it. Value-based pricing means price follows value, not time, and value can move in either direction.
- Nobody asks their broadband provider for a discount because fibre is faster than copper. Speed and accuracy are increasingly the product itself, not a saving to be passed back by default.
- As AI platforms move directly into accounting services, the real pricing pressure on firms is substitution risk from AI-native entrants, not internal efficiency gains. The market’s perceived value of doing it yourself changes even when the actual cost of doing it well does not fall as fast as people assume.
| “When you hire a master practitioner, you’re not paying for 10 minutes of their time, you’re paying for the 20 years it took them to make a 10-minute solution look effortless.” |
Most of the discussion only examines the supply side of this equation and the firm’s own pricing choices. It rarely asks what happens to a firm’s pricing power once clients believe they can get comparable value without paying a firm at all.
CAS is a destination, but not the only one
Much of the commentary treats advisory as the inevitable home for displaced compliance revenue. Speaking as a client of accounting services rather than a provider of them: I have never had an accountant deliver everything right, first time, on time. A faster, error-free, real-time service is valuable on its own terms. I would happily pay the same fee, or more, for accounts that are simply correct and current, without needing an advisory call bolted on to justify the price. Reliability and speed are a product. CAS is one route to capturing displaced value. It is not the only one, and treating it as the default might very well risk overlooking the simpler win sitting in front of most firms.
BILL’s own data backs this up more directly than most of the surrounding commentary lets on. Tax planning and preparation, not CAS, is the service firms most often say AI is pushing them to expand, at 50% versus 39% for CAS. A separate wave of the same survey found 82% of firm leaders now believe AI has raised client expectations, and speed of service tops that list at 79%, ahead of advisory at 67%. With MTD running and stalling here in the UK, I’d say Tax planning and preparation has the biggest opportunity for growth, as once the data is “out there”, it’s too late.
Clients are telling the profession, through the same research this discussion draws on, that they want things done faster and right, not necessarily discussed more. I’d go as far as to suggest, there are many businesses coming through, in the next generation especially, that aren’t even looking for the accountant relationship and regular engagement, they just want real-time accurate accounts. They’ll ask AI the rest (right or wrong).
Building your own AI infrastructure is usually the wrong call
| “The single largest constraint on any AI or automation program is data. No fix for that exists inside the technology itself.” |
Most firms do not need to configure AI infrastructure from scratch, and for the majority the cost of trying to get that right does not go down over time. Core software platforms will only take a firm part of the way, and leaning on them exclusively increases dependency without delivering the benefit firms are hoping for. A hybrid approach, part platform, part purpose-configured automation, consistently outperforms either extreme.
Underneath all of it sits the constraint that no capital raise or vendor relationship fixes: data. If the underlying data is not governed, structured and trustworthy, no amount of AI layered on top will work. The second constraint is understanding, of the technology, its risks, how to build it to reduce and correct when it goes wrong. That still requires a human in the loop. There is no version of this where that responsibility can be handed off to a platform or a vendor.
The headcount debate is premature
Debating whether AI reduces headcount before a firm has decided what it wants to become is arguing about an outcome nobody has earned the right to discuss yet. The 52% and 68% figures in the survey implicitly assume the current operating model, minus some hours, projected forward. A firm that genuinely reimagines its service line, more advisory, more real-time client contact, more oversight roles per the accountable and irreducible categories above, may well need more people, doing meaningfully different work. Until a firm has done the destination-and-journey thinking, the headcount conversation is noise not a conclusion.
Worth noting alongside the 52% and 68% headcount figures: 45% of leaders expect AI to make work easier for existing staff, against only 6% who expect it to get harder. Sentiment on day-to-day work quality is broadly positive even where headcount forecasts are negative, which is itself a sign that the headcount question and the ‘will this be good for our people’ question are being confused as one debate when they really aren’t.
The Real Question
BILL’s own researchers are right that capacity gains do not automatically translate into revenue. Where I think the prevailing UK reading is falling short is treating that as a pricing problem to be solved with a new fee structure and a named process owner. It’s a question of clear understanding, comprehension and governance, and pricing reform is simply one more obvious place the gap becomes audible.
Firms that get this right will not be the ones that reprice fastest. They will be the ones that build the structural, data-level visibility to know what AI is actually doing inside their practice, before they touch a fee schedule at all. That is the same foundational work Bots For That has written about in our own thinking on AI adoption in the profession, which goes further into the governance and readiness gaps.
Further reading: our‘The Self-Improving Firm’ nine-part series https://youtube.com/playlist?list=PLEtOBa16yEvOV-DLgV4KN1iGN6aY_lAoL&si=oSFvGu_3ODUZIzXi which addresses the governance-versus-gesture distinction raised here in more depth.
Sources: BILL, Accounting Firm AI Ambition Survey, Vol. 3 [https://www.bill.com/blog/accounting-firm-ai-survey-client-expectations-rise], produced with NewtonX.
