AI automation for accounting firms: what actually works

AI automation in an accounting firm is nearly always about the same things: booking invoices automatically, categorising bank statements, preparing VAT returns. Understandable, because that is where the visible manual work sits. But ask a firm owner where the crunch actually comes from and you hear something else. Not the bookkeeping. The waiting.

In the accountancy sector report of the Exact MKB Barometer 2026, 52 percent of Dutch firms name chasing and waiting for missing information as a cause of peak workload. 50 percent name late delivery by clients. All four of the biggest causes sit outside the firm, with the client. That is an uncomfortable finding, because virtually all AI automation sold to accounting firms targets the other half: processing documents that have already arrived.

This article is about that split. Where the time really leaks away, what your own software vendor is already building for you, and which question you need to answer before you automate anything at all, because in this profession, automating can cost you revenue instead of earning it.

Where time really leaks away in an accounting firm

You know the rhythm. Filing deadline in sight, and for part of your client base plenty is still missing. Someone in the office sends an email. No reply. A week later, another email. Then a phone call. Then a message to the one person who does pick up. Meanwhile the work shifts into the final two weeks, where your team works overtime on files that could have been on the desk six weeks earlier.

That chasing appears in nobody's time sheet as a process. It is scattered across ten people and three hundred separate moments. Which is exactly why it does not feel like something you can automate, and exactly why it is the biggest silent cost in the firm.

The same research contains another number worth knowing. Asked how often they use AI, 17 percent of firms say daily and 12 percent weekly. So fewer than one in three use AI weekly or more often. The majority has tried it, finds it interesting, and uses it a few times a month to draft a paragraph. If you think you are behind: probably not. If you think you are ahead because something AI-related happens now and then: probably also not.

Your software vendor is already building one half, often for free

Before you pay an agency to read invoices and propose bookings, check what is already in your subscription or arriving in it this year. The vendors of bookkeeping and reporting software are building that layer themselves, and they are deliberately pricing it at the floor. Visionplanner currently offers its AI assistant VAIA free to all users. The word the vendor itself attaches to that is: for now.

That is not charity, it is strategy. A software vendor earns on the number of ledgers you run with them, not on the AI. So the AI that fills, checks and summarises your files becomes part of the package. Anyone buying a custom project for that today is buying something that will sit inside the subscription they already pay for twelve months from now.

What structurally will not arrive in the package is everything that runs outside your software. The client calling with a status question. The bank or the estate agent asking for a figure. Chasing documents that have not arrived anywhere yet, because by definition you cannot pull those out of a system they are not in. That part stays yours, and it is therefore the part where automating pays off over time.

First this question: do you bill by the hour or by subscription?

This is the question almost nobody asks, and it decides everything. Automate client work in a firm that bills by the hour and you shrink your own invoice. You prepare the VAT return in thirty minutes instead of three hours, and then you write down thirty minutes. Your productivity goes up, your revenue goes down. That is not a theoretical risk, that is simple arithmetic.

Work with fixed subscriptions and that same time saving lands straight in your margin. The Benchmark Kantoorcijfers by NOAB and Fiscount, with 439 participating firms, shows that 44.1 percent now work with fixed subscriptions. The hourly rate is still dominant, but no longer a given.

If you bill by the hour, the conclusion is not that automating is pointless. The conclusion is that you should aim at something else: at work you already do but never invoice, and at capacity you simply do not have. More on that below.

What you can automate without touching your own revenue

The intake chain. Everything between the moment you know what you need and the moment it lands. That is work you already do, that you charge nothing for, and that your client will never miss if it happens more politely and more consistently than a rushed employee manages on a Thursday afternoon.

• Track per client and per period what is still missing, and request it automatically in plain language, not as a template email everyone recognises and ignores
• Send reminders on a rhythm you set, and only escalate to a person when it genuinely drags on
• Confirm what arrives straight away: thank you, this is in, this is still missing
• Answer client status questions without anyone having to open a file
• A weekly overview for you: which files are behind, whose court the ball is in, and what will make the deadline
• Acknowledge and log the receipts an owner sends through an app or WhatsApp, so nobody in the office spends time taking stock

An important detail: none of this requires substantive access to files. The system does not need to know what is in a set of annual accounts. It needs to know that three bank statements for the second quarter are still missing for client X, and that they were requested seven working days ago. That is status, not content.

The money sits in write-offs and capacity, not in cost savings

Accounting firms are one of the few groups of business owners who can simply pass their costs on. ING Economisch Bureau expects fees for accountancy services in the Netherlands to rise by at least 4 percent in 2026, the fifth consecutive year, while billed hours grow by roughly 2 percent. Staff shortages and heavier regulation push the price up and the client pays it.

Which means a cost-saving pitch does not land here, and rightly so. The real problem is not that the work is too expensive, it is that you cannot get all of it done. Large firms buy that capacity in delivery centres in India, South Africa or Eastern Europe. For a firm of five to fifty people that is not an option. So for you, automating is not the cheapest route to more capacity, it is the only one.

And there is a second pot of money almost nobody talks about: work you do but never invoice. In the same Benchmark Kantoorcijfers, the write-off percentage at accounting firms up to 20 FTE sits at 4.0 percent. On an average firm revenue of 795,200 euro, that is roughly 31,800 euro a year in hours worked and then struck through.

A solid share of that write-off is rework: figuring it out again, requesting it again, starting again because the file was incomplete when someone picked it up. That is money you win back without a single client paying less. Which is exactly what separates it from efficiency gains on billable work.

What does AI automation cost an accounting firm?

Put it next to what you already spend. From the Benchmark Kantoorcijfers: IT costs per FTE came to 8,300 euro in 2024, up 11.6 percent on the year before, roughly 7 percent of revenue. So a firm of 6.5 FTE already spends around 54,000 euro a year on software and systems. At the same time, only 12.7 percent of firms rate their IT supplier as excellent. The budget grows, the satisfaction does not.

I start with a free introductory conversation. We discuss what takes time, which systems you use and where automation could help. You then receive a proposal with automation opportunities, integrations, a schedule and costs.

Convert it to your own denominator. With 200 clients in your portfolio, that retainer is four to eight euro per client per month. That is less than you pay per ledger for reporting software. It is also the test: if this does not close a substantial part of that 31,800 euro in write-offs or free up half an FTE of capacity, do not do it. We would rather say that upfront.

Where AI does not belong in an accounting firm

Professional judgement. Not because it could not be done, but because you sign and you carry the liability. A system that prepares, collects and flags is fine. A system that decides what goes into the annual accounts is a risk no amount of time saved is worth.

The conversation where the news is bad. The client hearing that the assessment is higher than expected, or that you will not get the file finished before the deadline. You make that call yourself. One badly handled moment costs you more than a hundred smooth reminders earn.

And since 2 August 2026 there is a hard rule on top. Under the European AI Act, a person communicating with an AI system has to be able to know it. If you let clients chat with something that answers automatically, that needs a clear notice, not buried in the terms, but visible in the moment. That is no reason to skip it. It is a reason to have it built properly.

How to start

• Answer the pricing question: hourly or fixed subscriptions? That decides whether you steer on margin or on capacity
• Call your software vendor and ask what AI is coming to your package this year, cross that off your list
• Pick one filing period and one client group, for example the quarterly returns of your twenty slowest deliverers
• Measure two things beforehand: how many days before the deadline a file is complete on average, and how many hours you write off on that group
• Build only the chasing. Nothing inside the files, nothing on the professional work. If that runs measurably better after a quarter, expand

It is not a spectacular project. No robot arrives to write your annual accounts. What arrives is something that politely stays on your clients every day, so your team can start in week six on work that currently sits in a pile in week eleven. That is dull, and it is exactly where the crunch comes from.