AI for accountants is the use of artificial intelligence to automate routine administrative tasks so the human focuses on advice, judgement and client relationship. The question is not whether the accountant's role disappears, but which part of the daily work keeps adding value. Routine bookkeeping and statutory compilation get done by software, and what stays is precisely what made you an accountant in the first place. The part clients call you for. The advice, the judgement, the professional liability, the relationship. According to Exact's KMO Barometer, 71 percent of Dutch accounting firms used AI in some form by April 2026. At the same time many of those firms wrestle with implementation and shifting client expectations. This essay explains why that tension is not about your role; it is about your billing model.
The difference sits in your billing model, not in your profession
In the conversations we have, we keep seeing the same pattern. Accountants billing on an hourly rate are often reluctant to start with a tool like Nance or any comparable AI platform. Not because they do not understand the technology, but because they reason that if the software does their work, their invoice gets shorter. Their clients pay for hours; removing hours looks like removing revenue. That reasoning is valid inside the hourly model and breaks outside it.
Accountants on a fixed monthly fee or retainer think the opposite, almost automatically. Their client pays for outcome, not for hours. If AI handles the routine processing faster, the outcome stays the same and the firm becomes more profitable. More to the point: they can serve more clients with the same team, or serve the same clients deeper with the same team. For them AI is a margin multiplier and a capacity multiplier.
Which model you run determines how AI feels to you. That explains why inside the same firm a partner can say "we should be doing this" while a senior associate says "this eats my work". They see the same pattern; they draw a different conclusion because they sit in a different economics of their own labour. The conversations we end up having rarely stop at "should we use AI"; they stop at "what do we do with our pricing model".
What AI does well today, including the parts nobody talks about
The obvious list is everywhere: invoice coding, VAT validation, bank reconciliation, period-end checks, first-pass statutory compilation, draft notes to the financial statements. On those tasks AI sits at useful-accuracy levels in 2026 (the exact percentages vary by tool and dataset, and you must measure them on your own client mix before you ship anything AI-prepared to a client).
What rarely appears in those lists are the tasks that today do not happen at all because nobody has the time. Three examples we see often.
The VAT-code cross-check between invoice, ledger account and supplier. On paper simple: does the VAT code on this invoice match this ledger account, and does it line up with how we have booked this supplier historically? In practice almost nobody does this periodically. An AI layer running it weekly surfaces deviations that otherwise only appear at year-end or in a sample audit.
Supplier data hygiene. Bank account still correct, email still live, address details still current, tax identity still valid. Nobody dedicates time to this until something goes wrong. AI can cross-check this monthly against external sources and alert on what has shifted. The result is a cleaner accounts payable file and fewer payment-day surprises.
ERP data enrichment. Most ERPs have fields that are structurally underfilled: cost centre per supplier, project link per purchase order, segment per customer. Not critical, but valuable when populated. AI can fill these slowly based on patterns in the existing data, always with a four-eyes step.
The shared characteristic: these are tasks nobody was waiting for because they cost no measurable hours; they simply did not happen. AI turns that absence into a presence, and that is an underrated lever.
What AI cannot do today, and why that will not change quickly
The limit is not in the models. The limit is in what happens when the models are wrong. An accountant who gives conceptually wrong advice carries that into their own professional liability. An AI tool giving the same advice pushes the liability onto the human who passes it on. That is a transfer the human cannot refuse and should not want. The value of an accountant sits precisely in that judgement moment.
There is also a behavioural limit we see more often. When AI is woven through your whole workflow, the temptation is to skip the thinking step and just forward AI output to client or colleague. That is not efficiency; it is shifting work to the recipient. The client or colleague now gets analysis that nobody understood before sending and has to figure out what is correct themselves. In practice this creates more work, not less. Not outsourcing your thinking to AI is not an ethical principle; it is a productivity rule.
Finally: non-routine decisions. "My client is considering a €100k loan because they foresee cashflow trouble in six months. What do I advise?" A capable AI system can pull the data, run the scenarios, plot the chart. It cannot determine what this entrepreneur can carry, the state of their banking relationship, their risk tolerance on a variable rate, what they decided in their last two crises and why. That sits in a conversation between the accountant and the client, not in a prompt.
Three roles that emerge in an AI-first practice
What shifts is not the accountant's identity but the daily task mix. In an AI-first practice three roles emerge that did not exist or were marginal a decade ago.
The AI supervisor. Someone inside the firm who watches model performance, reads audit trails, sets thresholds for when human-in-the-loop is required, and periodically samples whether AI output stays aligned with what an experienced associate would have produced. Salary band comparable to a senior associate because the substantive quality is the same; the difference sits in where the time goes (reviewing and calibrating rather than booking yourself).
The client strategist. The human who has forward-looking conversations with clients: cashflow scenarios, M&A preparation, fundraising, pricing strategy, segment strategy. Work that ten years ago lived only inside the advisory arm of large firms now sits within reach of an SME practice because time freed from bookings can be reallocated. Salary band higher; you are not selling hours, you are selling outcome.
The tooling integrator. The human who designs workflows, decides where AI sits in the process, integrates with clients' ERPs and maintains the firm-level platform. In small firms this is a partner's part-time role; in larger firms a specialised one. Background typically finance plus IT, sometimes an experienced controller who has made the step into systems thinking.
None of these three roles replaces the existing senior associate. They emerge alongside and on top, and they bend the career steps that historically sat between junior and partner in a different direction.
What this does to your pricing model
This is where it gets uncomfortable. When routine work gets cheaper to deliver, you can do one of two things: let the margin expand inside your hourly model (short-lived; the market corrects), or change the pricing model.
Three models we see working in 2026, side by side.
| Model | How you bill | Strengths | Weaknesses | Fit |
|---|---|---|---|---|
| Fixed monthly fee per client segment | Monthly price per client based on segment (revenue, complexity, advisory frequency) plus a set number of advisory moments per quarter | Predictable for client and firm, scalable, margin grows with efficiency | Requires sharp segmentation and discipline on scope creep | SME firms with a broad client base |
| Value-based pricing on advisory | Admin on fixed monthly fee or hourly, advisory on outcome: percentage of interest saved, fixed price per negotiated supplier deal, advisory hour 10 to 15 times the admin hour | High margin on advisory, makes expertise visible in the invoice | Requires measurable outcomes and clients in a phase where value is demonstrable | Practices serving scale-ups and growth SMEs, or with strong treasury / M&A practice |
| Hybrid | Fixed monthly fee for admin, value-based for advisory, both on one invoice | Low transition risk, gives you time to learn what clients want in the advisory part | Client sees two components and can negotiate both | Most practices starting their transition in 2026 |
In practice 80 percent of firms that make this choice somewhere in 2026 land on hybrid, because the transition risk is small and it gives you time to learn what clients want to buy in the advisory part.
For anyone still purely hourly and a few years from retirement, staying the course is a legitimate strategy. For anyone with 10 or 20 years left in the practice, revisiting the pricing model is not an optional action; it is an agenda item for the next strategy meeting.
What you can do this month without buying anything
Three experiments that need no tooling and that you can finish in a month.
Map three processes you currently run and would want to speed up. Not the processes AI logically fits, but the processes that cost the most time in your practice. We frequently see firms discover at the mapping stage that their ERP already automates two or three of those processes without AI. That is a near-term win and it makes visible where AI actually adds value later.
Run one client experiment. Pick one client, one workflow (for example draft notes to financial statements, or VAT coding on international invoices), and run a month with AI support. At the end: how much time did it really take, what mistakes did AI make, what mistakes did the human catch, what was the client's reaction when they were told AI was in production?
Open the pricing conversation with your partners. Not to decide today, but to make the option space visible. Which clients could move to a monthly fee, which to value-based? What would you have to measure differently in your admin to support that? What conversation do you have with clients when they ask "why is my invoice changing?" tomorrow?
Plus: invest half a day in the change-management side of your client relationships. The biggest delay in a typical administrative workflow does not sit with you; it sits with the client who submits supplier invoices on the seventh, forgets the bank documents, or never tells you a supplier changed addresses. How do you automate that on the client side? What standard can you impose? What do you give clients as a reason to comply? The savings for an average SME practice here are bigger than in any AI tool.
Frequently asked questions
Which accountancy tasks does AI automate in 2026?
In 2026 AI sits at useful-accuracy levels for invoice coding, VAT validation, bank reconciliation, period-end checks, first-pass statutory compilation and draft notes to financial statements. Underrated categories that today often happen manually or not at all: VAT-code cross-checks between invoice and ledger account, supplier data hygiene, and ERP field enrichment. For every task: the exact accuracy varies by tool and client mix; always measure on your own data before sending AI output to a client.
What remains the accountant's work?
Judgement under uncertainty, client relationship, non-routine decisions, and professional liability. AI can pull data and rerun scenarios, but cannot determine what a specific entrepreneur can carry, the state of their banking relationship, their risk tolerance, and what they decided in previous crises. That sits in a conversation between accountant and client, not in a prompt.
How does AI change the hourly-rate structure?
The hourly-rate model becomes untenable as soon as a large share of the work gets cheaper to deliver. Three models we see working in 2026 practices: a fixed monthly fee per client segment, value-based pricing on advisory, or a hybrid of the two. In practice 80 percent of firms starting the transition in 2026 land on hybrid, because the transition risk is small and it gives you time to learn what clients want to buy in the advisory part.
What new roles emerge in an AI-first accounting practice?
Three roles emerge alongside the existing ones: the AI supervisor (model performance, audit trails, thresholds for human-in-the-loop), the client strategist (forward-looking advisory on cashflow, M&A, fundraising), and the tooling integrator (workflows, ERP integrations, firm platform maintenance). None of the three replaces the senior associate; they emerge alongside and bend the career steps between junior and partner in a different direction.
How do you start with AI as a firm without big risks?
Three low-risk actions without buying a tool. (1) Map the three processes that cost you the most time and check whether your own ERP can already automate two or three of them. (2) Run one client experiment: one client, one workflow, one month of AI support, with a clear definition of success up front. (3) Open the pricing conversation with your partners. Only then move to tooling choices. For those who want to move faster: invest half a day in the change-management side of your client relationships (delays often sit with the client, not with you).
Closing
The accountant does not disappear. The admin work moves into the background and what your job actually is moves into the foreground. For those who wanted that for years but had no time, this is an opportunity. For those who built the practice on billing that admin, this is a renovation. Which of the two you are, you know best. Start this month with the three processes that cost you the most. Then think about the tool.
Get in touch if you are an accountant and want to think through how AI fits in your practice without having to revisit your pricing model tomorrow.
Further reading: AI in Exact Online: what you can actually automate today (the integration side of the story), AI for your books: when chat AI is enough, and when you need an AI colleague (the strategic context), everything Nance automates across the finance function, or Solutions for accounting firms for the full offering.
