Experiences and developments

What are the experiences with AI workflow automation at accounting firms?

Firms use AI widely but shallowly, mainly for text and less for audits. Reliability and review time hold them back more than money.

Piet Baudoin··13 min read

Public accounting-firm experiences in 2026 show broad but shallow AI use. Over three quarters use AI, mainly for text and communication. Only a quarter use it for compliance and auditing. Reliability, at 44 percent, is a greater obstacle than costs, at 33 percent.

This page compares Dutch firms, trade publications, professional bodies, and regulators’ public statements in 2025 and 2026. It identifies sample sizes and commercial interests, recalculates time-saving claims, examines failures, and sets out our expectation for the end of 2027.

The short answer

  • Exact’s June 2026 SME Barometer reports 78 percent of Dutch accounting firms using AI: 37 percent for text and communication, 24 percent for compliance and auditing. Exact makes Dutch accounting software.
  • The same study finds 49 percent unsure how to get the most from AI.
  • Silverfin’s September 2026 study of 100 Dutch respondents finds 44 percent citing accuracy and reliability as the largest obstacle, versus 33 percent for costs.
  • Dutch respondents still combining Excel with manual checks total 57 percent, versus Belgium’s 41, Luxembourg’s 44, Denmark’s 48, and the UK’s 31.
  • AFM estimates AI components are used in about 12 percent of statutory audits at the largest six firms. Most had fragmented or absent policies.
  • The only measured effect comes from Stanford’s US field study of 79 businesses: 8,5 percent of time shifts away from routine work, and monthly closing is 7,5 days faster.
  • No public independent Dutch measurement covers an entire AI workflow including review. Claims of 8 to 9 hours weekly come from vendors.

How many Dutch accounting firms use AI?

Exact’s SME Barometer puts use at 78 percent (Accountant.nl, June 2026). A qualification often missing: it surveyed over 1.700 SME entrepreneurs, rather than 1.700 accounting firms. Exact sells accounting software with AI.

Growth is real and rapid. Accountancy Vanmorgen, September 3, 2026 reports SME AI use rising from 6 percent in 2024 to 70 percent in 2026. Firms’ 78 percent is higher still.

Silverfin measures expectations: among accountants anticipating professional change, 62 percent identify AI and automation as the main driver (September 15, 2026). The US Karbon report reports 98 percent without specifying its sample.

Use means everything here. Rewriting one customer email counts just as much as AI preparing the whole tax-return cycle. Adoption percentages say little about workflows. Examine tasks instead.

What do firms use AI for, and what do they not yet use it for?

Text dominates the auditing work central to the profession. Exact figures in Accountant magazine 3-2026, September 9, 2026 show 37 percent text and communication, 32 percent analysis and advice, and 24 percent compliance and auditing. Text leads roughly one and a half to one.

Karbon includes workflow automation only in its more expensive Business plan, $ 89 per user monthly with annual billing. Text assistants can be bought separately; workflows require setup.

The US Firm of the Future study, June 23, 2026, 725 professionals finds firms averaging 10 apps. Fragmented systems are reported by 48 percent, integrated ones by 41 percent. Employees spend about 5 weekly hours transferring data between systems. AI is routine for 30 percent and occasional for 54 percent. High-stakes decisions remain almost entirely human at 64 percent.

An Acuity partner article, July 22, 2026 describes the Dutch equivalent: duplicate entry, approvals in separate emails, and AI unable to identify patterns across disconnected systems. It is partner content without numbers, but recognizable.

See which processes to automate first.

What do accounting firms themselves report?

Two firms have publicly described traceable experiences. Both begin with working methods rather than tools.

EasyMKB, with 1.300 to 1.500 clients, reports 75 to 80 percent of accounting actions fully automated in a Blinqx partner article, April 1, 2026. Its first lesson, translated: “Do not start with the tool; start with your working method.” All clients follow one fixed approach, avoiding exceptions that stall automation. Another statement, translated: “Technology does not replace people. But technology does replace people who do not want to use it.” This is partner content and does not define an action.

Vermetten gives the most candid account. The SRA AI dossier calls AI, translated, “now an established part of audit practice.” Its entire audit team, partner to intern, has tools (SRA, July 17, 2026). It started with document searches, meeting transcription, and technical questions. It also told Accountantweek, July 17, 2026 that financial-statement document review and consistency checking fall short. Ideas are plentiful; everyone is busy. Its greatest risk is people stopping thinking and trusting outputs too much.

Both first established the work. Neither supplies measured savings. Nart Wielaard, September 9, 2026 argues firms buy separate tools while gains lie in redesigned workflows.

See broader administrative AI agent experiences.

What holds firms back from AI workflow automation?

Reliability matters more than money. Dutch respondents in Silverfin’s study, September 15, 2026 report:

Obstacle Netherlands
Accuracy and reliability 44 percent
Setup 38 percent
Costs 33 percent
Connecting systems 25 percent

Source: Silverfin/Censuswide, 100 Dutch respondents, published September 15, 2026. Silverfin sells AI accounting software.

US Karbon research finds 83 percent concerned about data breaches, up 7 points from a year earlier.

Consider margins. With 100 respondents and a 57 percent result, the 95-percent margin is about 9,7 percentage points: our calculation, 1,96 times the square root of 0,57 times 0,43 divided by 100. The Netherlands’ 57 percent using Excel and manual checks differs from Belgium’s 41 by more than that margin. Reliability’s 44 versus setup’s 38 does not. Dutch firms relatively often check manually; both reliability and setup are top concerns.

Nextens, September 17, 2026, which sells tax AI, points out the sample of 100 was commissioned by a software provider.

See errors and review.

How much time does AI really save an accounting firm?

No public measurement answers this for a Dutch firm. Circulating figures are expectations or vendor claims. The only actual measurement is US research and is lower.

Source Claim Weekly equivalent Method Commercial interest
Silverfin/Censuswide, 15-09-2026 1,48 hours daily recoverable About 7,4 hours Respondents’ estimates Sells AI accounting software
Accountee partner article, 25-09-2026 8 to 9 hours per employee weekly 8 to 9 hours No measurement method Sells product
aiagency.nl, 20-03-2026 Anonymous case: 180 to 65 hours monthly Not per employee No source Sells implementation
Choi and Xie, Stanford, 07-05-2025 8,5 percent of time moves away from routine work About 3,4 hours at 40 hours Field data from 79 small and medium-sized businesses Academic, data from software partner

Conversions are our calculations: 5 working days, 40 hours weekly.

Comparable employee figures range from 3,4 to 9 hours. Only the lowest was measured. Stanford also found closing 7,5 days faster and 55 percent more clients weekly. These are US businesses using one vendor’s software, rather than a Dutch SME firm.

Accountee’s € 120.000 to € 150.000 annual firm claim is internally consistent: 8 to 9 hours over roughly 46 working weeks means 368 to 414 annual hours per employee. Ten employees give around 4.000 hours, over two FTE at 1.800 productive hours. But the starting 8 to 9 hours was not measured.

Conversely, Firm of the Future identifies 5 weekly hours lost moving data between systems. That is identifiable loss rather than a promised gain.

What AI errors do accountants make, and why?

Errors are real and arise especially under time pressure. Accountant.nl, August 10, 2026 reports invented content at PwC, EY, and KPMG. Speakers cite workload, insufficient AI knowledge, and checks abandoned near deadlines. CMS’s Bert Vries says, translated: “Employees often work under tight deadlines, creating a risk that they adopt AI outputs too quickly.” Julia van Leeuwen notes convincing answers can still be wrong.

Vermetten’s warning about excessive trust and Silverfin’s reliability obstacle point to the same cause from three angles.

The busiest weeks are precisely when AI should save the most time and review time is scarcest. Tax-return or financial-statement errors create recovery work and potentially worse consequences.

What do AFM, NBA, and SRA say?

The regulator and professional bodies agree: foundations before speed.

Date Event Source
11-12-2025 AFM: about 12 percent of statutory audits at largest six use AI components; twelve building blocks for safe use Accountancy Vanmorgen
01-04-2026 EasyMKB: 75 to 80 percent of accounting actions automated Accountancy Vanmorgen/Blinqx
April 2026 EY deploys agentic AI in EY Canvas, targeting about 160.000 audit engagements within several years Accountant.nl, 08-06-2026
04-06-2026 NBA/NOREA publish Guideline 2: Applied AI, dozens of examples including SME accountants Accountant.nl, 05-06-2026
20-05 to 24-07-2026 SRA publishes four white papers; AI becomes continuing education theme for 2026-2028 SRA AI dossier
17-07-2026 Vermetten: AI established in audit practice SRA
10-08-2026 Deadline pressure increases AI error risk Accountant.nl
03-09-2026 Exact launches financial analysis agent Accountancy Vanmorgen
15-09-2026 Silverfin: 44 percent cite reliability obstacle Accountancy Vanmorgen

AFM’s strongest statement, translated: “At the majority, policy is fragmented or even entirely absent.” It studied large audit firms rather than SMEs. The lesson still applies: define responsibility, data quality, and review ownership. NBA/NOREA say validation and human judgment remain the basis of public trust. VEB asks firms to disclose AI use in audits.

Why do firms save time only when they include review?

Our position: firms struggle with review time rather than finding tools. AI moves work from creation to checking, requiring exactly the time firms lack.

Silverfin identifies reliability, Vermetten busyness and overtrust, Accountant.nl deadline failures, and Exact only 24 percent using AI for compliance and audits. Together they point to approval effort as the brake.

Take a twenty-person firm. Silverfin research via Accountant.nl, September 15, 2026, 500 respondents in five countries, finds 84 percent expecting at least an hour daily. At 220 working days, that is 220 hours per person, 4.400 hours total, about 2,4 FTE. If a quarter goes back into checking and recovery, 1,8 FTE remains; if half does, 1,2 FTE. Nobody has publicly measured that share. It determines returns.

Measure remaining human time per completed case, including review and recovery. Evidence beside a proposal makes approval different from reopening an entire case. Designing approval is the real work. See more work with the same people and becoming an AI-native service business.

What does AI mean for hourly billing and juniors?

Saving time while billing hourly loses revenue. Vermetten asks whether hourly billing remains viable in five years (Accountantweek). MKB Accountants’ Geert Dreschler says hourly billing offers no AI incentive, whereas fixed pricing with a value proposition does (Accountant.nl, October 20, 2025).

Van Garderen, January 15, 2026 describes the junior-assistant base gradually disappearing. Routine work that taught juniors goes to AI first. Future seniors learn less by doing.

Firms using savings to serve more clients benefit from fixed prices. Teach retained juniors checking rather than entry. Reviewing supported work develops the judgment the profession needs.

Where is this heading?

Growth is broad but shallow. Exact reports SME use rising from 6 percent in 2024 to 70 percent in 2026; accounting firms already reach 78. Further adoption growth is limited. Depth comes next: workflow preparation versus manual work. With 24 percent compliance/audit use and 57 percent manual Excel checking, there is room.

Our expectation: by the end of 2027, at least one Dutch firm, SRA, or NBA publishes measured net savings for one workflow including review. It will be below today’s expected 1,48 daily hours.

Professional bodies already build frameworks: Guideline 2, four SRA papers, continuing education through 2028. Measured practice is next. Vermetten, AFM, and August incidents show review and policy time missing from expectations. Stanford’s measured 3,4 weekly hours is already under half the expectation.

An independent Dutch measurement with higher net savings, including review, would break this expectation. Faster approval would be good news.

Bombos records each proposal, approval, and correction so workflow measurements show prepared work and review time.

What can accounting workflow automation not yet do?

  • Evidence is thin. No public independent Dutch SME account measures an entire AI workflow. Hourly claims are expectations or sales claims, including ours.
  • Audit document review is insufficient. Vermetten reports document review and consistency checking falling short.
  • AI sounds certain when wrong. Invented report content at three large firms makes unreviewed use indefensible.
  • Exceptions remain human work. EasyMKB’s high rate relies on one method for every client. Customized firms must change methods first.
  • AI cannot find unwritten knowledge. Record customer tax arrangements, for example through interviews.
  • The accountant remains responsible. NBA/NOREA retain human judgment as the foundation.

How does Bombos approach this?

Bombos helps firms grow without adding people: more work at higher quality with the same team. It starts with one workflow and human approval.

Chef distributes incoming work. Wegwijzer guides you. Specialists fit your tasks, such as email into the right case or requesting missing attachments. Bombos reads email, finds the client and case in packages such as Exact Online, and prepares a supported proposal. You approve, edit, or reject in Bombos. Only then does the result enter the package or reach the client. Corrections become rules for coworkers too.

Reviewers see evidence beside proposals rather than reopening cases. Every approval is logged. Customer messages await your approval by default.

Your team then teaches the next task without technical skills. We guide you; afterward you can do it yourself. See AI automation for Dutch accounting firms, agent costs, best agent platform, and which AI solution fits.

Sources

Each source was opened on September 30, 2026. Original excerpts appear verbatim; translations are identified above.

Free, no obligation

More work done, at a higher quality, with the same team.

That is what Bombos is for: companies that grow fast and want to keep the same team. We start with one task that keeps piling up and guide you until your team can handle it. Then your team teaches Bombos the next task. Leave your number and we will call you back to talk about your situation.

We read what you write. Within one working day you hear from the one of us who knows your kind of work best.

More about experiences and developments

All questions in the knowledge base →