Administrative processes increasingly automated with AI involve incoming text that needs processing: sorting and answering email, reading and summarizing documents, processing purchase invoices, taking meeting minutes, scheduling appointments, and updating customer files. Growth is measured across process groups. In the Netherlands, 27 percent of small businesses now use AI, but only a small share lets it handle an entire process.
This page compares the best available measurements: CBS, Eurostat, the SME Barometer by Exact, a Dutch accounting and business software provider, McKinsey, the Microsoft Work Trend Index, and the Anthropic Economic Index. For each process, it explains what was actually observed, how strong the evidence for growth is, and where it is heading. To choose your first process, see which administrative processes to automate first. Broader technology developments appear in AI agent trends for businesses.
The short answer
- Among small Dutch businesses, using the CBS classification, 11 percent used AI in 2023, 19 percent in 2024, and 27 percent in 2025, a provisional figure. CBS, December 12, 2025
- Of Dutch businesses using AI, 32 percent use it for business administration or management tasks, and 17 percent for accounting, auditing, or financial management. CBS, December 12, 2025
- Across the EU, 19.95 percent of businesses with at least ten people employed used AI in 2025. Of those users, 31.05 percent use it for administration and management. Eurostat, December 2025
- Administrative work grew from about 10 to 13 percent of Anthropic’s business API traffic between August and November 2025. That is relative growth of about 30 percent in three months. Anthropic, January 15, 2026
- Active agents in Microsoft 365 increased fifteenfold between March 2025 and March 2026. Microsoft, May 5, 2026
- Among large enterprises, the share with scaled agents rose from 27 to 40 percent. Among smaller organizations, it remained at 22 percent. McKinsey, August 25, 2026
- Our calculation using CBS figures: the share of all Dutch microbusinesses using AI for administration rose from 1.46 percent in 2023 to 3.57 percent in 2025. CBS, March 16, 2026
Which administrative processes most often use AI today?
The administrative processes most often done with AI involve a person reading text and acting on it. The Anthropic Economic Index is the only source showing this at task level. It classifies actual usage traffic instead of asking business owners what they do. In November 2025, its business API traffic included these clusters: analyzing email and preparing replies, 0.28 percent of all traffic; classifying email, 0.23 percent; calendars and appointments, 0.16 percent; and building and maintaining invoice-processing systems, 0.24 percent. Anthropic, January 15, 2026
Those numbers look small, but traffic covers thousands of tasks, from programming to translation. Administrative work together accounted for 13 percent of the API traffic studied.
In its June edition, Anthropic counted documents and reports as 15 percent of conversations in chat and Cowork. Anthropic, June 26, 2026 Work and personal use are combined, and the classification differs from January. You therefore cannot join that 15 percent to the 13 percent.
In the Dutch SME survey by Exact, a Dutch accounting and business software provider, AI users mainly mention meeting minutes and translations, 37 percent, and insights into business performance, 36 percent. Exact SME Barometer 2026 The pattern is consistent: AI first appears where the work involves reading, summarizing, and drafting a proposal.
How many Dutch businesses use AI for administration?
About one in three Dutch businesses using AI applies it to administration. Across all businesses, that is still a small minority. CBS measures that 32 percent of AI-using businesses with at least two people employed use AI for business administration or management tasks, and 17 percent for accounting, auditing, or financial management. CBS, December 12, 2025 Marketing and sales lead; administration follows.
The denominator is where many articles go wrong. That 32 percent applies to businesses already using AI, rather than all businesses. In the CBS small-business category, 27 percent used AI in 2025. Rough multiplication, 27 percent times about a third, suggests around 9 percent of those businesses use AI somewhere in administration. This is an estimate: CBS does not publish purposes by size category in that same release.
For the broader group with at least ten people employed, the older CBS edition allows a precise calculation. Across all businesses, about 4.17 percent used AI for administration and management in 2023, 14 percent times 29.8 percent, and 6.88 percent in 2024, 22.7 percent times 30.3 percent. CBS AI Monitor 2024, February 27, 2025 The share among users barely changed. Growth came from additional users.
“Using AI for administration” in these surveys also does not mean administration runs itself. It could equally mean one employee having an email rewritten.
Which processes are growing, and how strong is the evidence?
No individual administrative process has a Dutch time series for small businesses. Anyone promising a top five of fastest-growing processes is inventing it. What we can show is what was actually observed for each process, the strength of evidence for “increasingly,” and its direction. The direction column is our own expectation through the end of 2027.
| Process | What was observed | Evidence for growth | Direction through late 2027, our expectation | What remains a human decision |
|---|---|---|---|---|
| Email sorting and draft replies | Separate clusters of 0.28 and 0.23 percent of API traffic (Anthropic) | Moderate: growth measured for the entire administrative group, from 10 to 13 percent | From text proposals to handling within the correct file | What you promise and whether it is sent |
| Reading and summarizing documents | Documents and reports account for 15 percent of chat and Cowork conversations (Anthropic) | Moderate: a different measurement unit from the earlier series | From summaries to checked fields with the source document | Missing pages, linking to the wrong customer |
| Purchase invoices, receipts, and expense claims | Recognition and AI analysis are in standard software from AFAS, a Dutch business software maker (AFAS); Oracle offers an invoice agent in an early adoption program (Oracle 26A) | Weak adoption evidence: product availability, no usage figure | Less retyping, more attention to exceptions and coding | Duplicate invoices, posting logic, payment approval |
| Meeting minutes and translations | Mentioned by 37 percent of SME users (Exact) | Moderate: one measurement, covering Dutch SMEs | From minutes to actions with an owner | A decision attributed to the wrong person |
| Appointments and scheduling | Separate usage cluster of 0.16 percent (Anthropic) | Weak: no separate time series | More coordination of availability and exceptions | Promises to customers |
| Customer and file maintenance | Anthropic includes customer management in administrative work | Weak: no separate growth figure | Emails and conversations become proposed file changes | Wrong customer, outdated information |
| Reporting and financial planning | Performance insights mentioned by 36 percent of SME users (Exact); 76 percent of large companies use AI in financial planning (KPMG) | Moderate for large companies, weak for SMEs | Faster preparation of explanations and scenarios | Incomplete figures, the final decision |
| HR administration and onboarding | Microsoft uses agents internally for onboarding (Microsoft, September 17, 2026) | Weak: one vendor case | Coordinating documents, tasks, and missing information | Granting access, decisions about people |
We deliberately left two processes out. Payment reminders have long been automated with ordinary rules. AI growth there is unproven. Reconciliation and period closing feature in international agent research (Avalara, August 11, 2026), but there is no Dutch growth measurement for each closing process. Familiar examples of these tasks, grouped into five streams, appear in our task register.
Is AI use in accounting rising or falling?
AI use in accounting rises slightly across all small businesses, while its share among AI users falls. That sounds contradictory, but it is arithmetic. In March 2026, CBS published a separate series for microbusinesses with two to nine people employed. Definitions remained consistent across three years. We multiplied AI adoption by the share for each purpose.
| Year | Any AI use | Administration and management, among users | Accounting and financial management, among users | Administration, all microbusinesses, our calculation | Accounting, all microbusinesses, our calculation |
|---|---|---|---|---|---|
| 2023 | 6.8% | 21.4% | 20.9% | 1.46% | 1.42% |
| 2024 | 10.6% | 23.0% | 14.9% | 2.44% | 1.58% |
| 2025 | 13.8% | 25.9% | 12.9% | 3.57% | 1.78% |
Source: CBS, Use of AI technology by Dutch microbusinesses, March 16, 2026, tables 2.1 and 4.1. The final two columns are our estimates from rounded percentages, rather than CBS indicators.
AI for administration and management grows in two ways: more users and a larger share per user. In two years, it reached almost two and a half times its earlier level. Accounting grows only through additional users: 0.36 percentage points in two years. New AI users therefore start with email, text, and organization, much less with the books. Our explanation, which has not been measured: accounting packages already automate posting through fixed rules, e-invoices, and document recognition. Business owners therefore less often add their own AI tool there. See AI within AFAS or Exact, or an AI agent alongside it.
Why do AI adoption figures differ so much?
Figures differ because they count different things: businesses, respondents, conversations, or agents. The Exact SME Barometer reports that 70 percent of businesses with 2 to 250 employees use AI in 2026, compared with 33 percent in 2025 and 6 percent in 2024. Exact SME Barometer 2026 CBS reports 27 percent for small businesses in 2025. Both can be correct. CBS asks about seven specific AI technologies. Exact asks 1,719 business owners online whether they use AI. Someone using only a chat assistant is more likely to answer yes to Exact. Exact itself sells administrative software.
| Measurement | What it counts | Start | End | What you can conclude |
|---|---|---|---|---|
| CBS, small businesses | Businesses with any AI use | 2023: 11% | 2025: 27% | Adoption within the relevant business size |
| CBS AI Monitor, at least 10 people employed | Administration as a purpose, among users | 2023: 29.8% | 2024: 30.3% | The purpose grows alongside adoption, no faster |
| Anthropic API traffic | Administrative work’s share of transcripts | Aug. 2025: about 10% | Nov. 2025: 13% | A different traffic mix at one provider |
| McKinsey, smaller organizations | Organizations with scaled agents | 2025: 22% | 2026: 22% | Not every group accelerates |
| Microsoft 365 | Active agents | March 2025: 1x | March 2026: 15x | More agents by Microsoft’s own threshold |
These are four different units. Never plot them as one line. Microsoft counts an agent as active after at least one day of use or one autonomous run per 28 days. Fifteen times more agents therefore does not mean fifteen times more completed work. Eurostat marks a break in the Netherlands’ 2025 series. Do not splice Eurostat figures onto CBS figures. KPMG itself warns that questions differed between survey rounds, so the finance jump from 30 to 75 percent is not a clean growth measurement. KPMG AI in Finance 2026
Do small businesses automate entire administrative processes with AI agents?
Small businesses mainly automate individual steps with AI, much less entire processes with agents. McKinsey sees the share of large enterprises, above $1 billion in revenue, with scaled agents rise from 27 to 40 percent in a year. Smaller organizations remain at 22 percent. McKinsey, August 25, 2026 The study surveyed 1,719 people in 97 countries. Its smaller-organization group is much broader than Dutch SMEs. Still, the signal is clear: the shift from tool to workflow is accelerating at large companies and is not at small ones. For the actual position of Dutch SMEs, see AI workflow automation for Dutch SMEs.
The gap in finance is equally large. KPMG reports that 76 percent of 1,013 surveyed leaders at businesses with revenue of at least $250 million actively use AI for financial planning. KPMG, 2026 PEX reports in a broader group that 31 percent use AI in finance, but only 9 percent deploy it widely across the team. PEX, 2026 The distance between “someone uses it” and “the team works with it” is larger than the first figure suggests.
Dutch accountants show the same pattern. Silverfin research among one hundred Dutch respondents finds that 57 percent still reconcile and review manually in Excel. Meanwhile, 44 percent cite AI reliability as a barrier. Accountancy Vanmorgen, September 15, 2026 AI also leads their expected changes. Enthusiasm and manual work coexist.
Why is the next growth area between intake and decision?
Our position: the next growth in administrative AI lies in the handoff between intake and decision: turning incoming information into a complete, reviewable proposal that enters the right process.
Three facts support this. First, measured use is almost entirely preparation: classifying email, preparing replies, summarizing documents, and taking minutes. Second, smaller organizations stall at scaled workflows, 22 percent unchanged for two years, while Microsoft’s agent count grew fifteenfold. Third, people who must finish the work, such as the 57 percent of accountants still reconciling in Excel, remain stuck at the end of the process.
Together, this means a twenty-person business already has an AI assistant that can summarize an email. The next connection is missing. Which customer does the email concern? What agreement is in the file? What is the proposal, who approves it, and where is the outcome recorded? While an employee must find and retype that information, the assistant produces better text but does not complete a file. The blog about how a service business becomes AI-native develops this distinction between using AI and completing files.
There is a second-order effect. Faster preparation moves the queue to the reviewer. Receiving more prepared proposals without making review easier saves little turnaround time. Ordinary cases disappear, leaving difficult exceptions for people. What counts is how many tasks finish correctly and how many human minutes they need, including review and recovery. See AI agents automating work between software systems for keeping work moving between systems.
Where is this heading?
Measured adoption is fast, but deeper use is slow. AI use at small Dutch businesses rose 16 percentage points in two years, from 11 to 27 percent, relative growth of about 145 percent. Administration’s share of Anthropic traffic grew 3 percentage points in three months. Scaled agents at smaller organizations did not grow.
Our expectation: by the end of 2027, a larger share of Dutch small businesses using AI will process incoming email and documents in a connected workflow than on September 30, 2026. Preparation with human review will grow faster than fully autonomous sending, payments, or contractual actions.
Here, a connected workflow means recognizing incoming work, finding the customer and file, preparing a proposal, and recording the approved outcome in a business system. Pasting standalone text into a chatbot does not count. We give no growth figure because no baseline exists.
The reasoning: adoption is already there, packages such as AFAS and Oracle are adding AI, and the buying barrier for individual steps has disappeared. Remaining gains lie in connecting them. Suppose a future CBS measurement shows 20 percent AI adoption among microbusinesses, with 30 percent of those using it for administration. That gives 6 percent of all microbusinesses, compared with 3.57 now. This is a calculation with chosen assumptions, rather than a forecast.
The expectation fails if connected workflows’ share stays flat or falls, or if fully autonomous handling grows faster instead. The economics fail if connection, review, and recovery consume the time saved. Faster models alone do not prove those conditions disappear.
By then, Bombos aims to handle preparation for you: reading email, finding the file, preparing a supported proposal, and recording the outcome in your own package after your approval.
What can this overview not yet prove?
This overview cannot prove which individual administrative process grows fastest, because nobody measures that. No representative Dutch time series separately measures email, invoices, files, onboarding, and scheduling at small businesses.
- CBS purposes are broad. “Business administration or management tasks” does not distinguish invoices, leave requests, or an email. Before 2023, CBS counted HR separately. It has since merged that category, so old and new series do not align.
- Anthropic measures one provider. A few intensive users can shift traffic shares. The same January edition also mentions both 13 and 15 percent administrative API traffic, in two chapters with different analytical bases.
- Microsoft and Exact sell what they measure. Their figures are real, but thresholds are low: one usage day per 28 days, or an online question about using AI.
- Product pages are not adoption. Oracle says its invoice agent is “not available to be enabled outside of the early adoption program.” An automatically posted invoice in AFAS may result from fixed rules or e-invoices, without generative AI.
- Use is not a result. McKinsey reports that 37 percent of organizations see some positive impact on profit, almost unchanged from 2025. More AI does not automatically mean less administrative time. Cheaper reports may even prompt more reporting.
- Our calculations are estimates from rounded percentages, without confidence intervals.
No AI system can retrieve knowledge that exists only in someone’s head, such as an agreement never written down. This limit applies to every process in the table. Bombos addresses it by interviewing your team and recording that knowledge.
How does Bombos approach this?
Bombos starts at the handoff where measurements reveal a gap. It reads incoming information across connected email addresses and documents, finds the customer and file in your systems, and prepares a proposal. Each proposal shows its basis. You approve, change, or reject in Bombos. Only then does the result appear in your package. Bombos reads and writes in systems including Exact, AFAS, Visma, a business software group, Twinfield, cloud accounting software, Moneybird, online accounting software, SnelStart, Dutch bookkeeping software, Nmbrs, payroll and HR software, and Microsoft 365.
Payments, customer messages, and contracts wait for your approval by default. That boundary is enabled by default and technically enforced. A correction becomes a rule, including for coworkers. If something exists only in someone’s head, Bombos interviews you to record it, making it available next time.
You start with two regular workers: Chef distributes incoming work, and Wegwijzer helps you find your way and suggests what Bombos can take over. Specialists around them are built around your tasks. Examples by work type appear under use cases. For combining email, documents, and Excel, see automating documents, Excel files, and email with AI.
The goal is more work at a higher quality with the same team. The first task is the beginning. Your own team then teaches Bombos the next task, without needing technical skills. To try a small start, see can I start with an AI automation pilot?.
Sources
Each source was opened on September 30, 2026, and each original excerpt appears verbatim in it. Sources selling something related to the claim are identified. Our calculations are labeled in the text. Dutch excerpts below are identified as translations.
- CBS: Businesses most often use AI for marketing or sales. December 12, 2025, figures 2023-2025, 2025 provisional. Translated excerpt: “32 percent use AI for business administration or management tasks.” Statistics agency.
- CBS: Use of AI technology by Dutch microbusinesses. March 16, 2026, tables 2.1 and 4.1. Statistics agency.
- CBS: AI Monitor 2024, chapter 2. February 27, 2025, tables 2.1.1 and 2.3.1, partly provisional at the time. Statistics agency.
- Eurostat: Use of artificial intelligence in enterprises. Data December 2025. Quotes: “In 2025, 19.95% of EU enterprises used AI technologies.” and “2025 break in time series for Netherlands (NL).” Statistics agency.
- Exact: SME Barometer 2026. Fieldwork first quarter 2026, 1,719 respondents. Exact sells administrative software.
- McKinsey: The state of AI in 2026. August 25, 2026, fieldwork May-June 2026. McKinsey sells consulting and AI services.
- Anthropic: Economic Index report, Economic primitives. January 15, 2026, November 2025 sample. Quote: “Build and maintain invoice processing systems.” Anthropic sells Claude.
- Anthropic: Economic Index, Uneven AI adoption. September 15, 2025, August 2025 measurement. Anthropic sells Claude.
- Anthropic: Economic Index report, Cadences. June 26, 2026. Anthropic sells Claude.
- Microsoft: 2026 Work Trend Index. May 5, 2026, telemetry March 2025-March 2026, 20,000 AI-using knowledge workers in ten countries. Quote: “growth in active agents in Microsoft 365 year over year.” Microsoft sells Copilot.
- Microsoft: What we’ve learned from Microsoft’s own AI transformation. September 17, 2026. The vendor and studied organization are the same party.
- KPMG: AI in Finance Report 2026. Fieldwork March 2026, 1,013 leaders in twenty countries. Quote: “question scope differed between waves.” KPMG sells consulting and AI services.
- PEX: The state of finance. 2026 edition, 687 respondents. PEX sells expense management.
- Avalara: How the race to deploy AI agents is outrunning finance governance. August 11, 2026, Censuswide survey of over 1,500 finance leaders. Avalara sells tax automation.
- Accountancy Vanmorgen: AI leads expected changes in accountancy. September 15, 2026. Coverage of Silverfin research; Silverfin sells accounting practice software.
- Silverfin: Firms’ advisory gap in 2026. 2026 edition, fieldwork July 2026, 500 respondents in five countries. Silverfin sells accounting practice software.
- AFAS: From invoice to dashboard, with automation and AI. Undated product page. Translated excerpt: “And let Jonas AI prepare an analysis or fill fields in the workflow.” AFAS sells this software.
- Oracle: Payables Agent for Invoice Ingestion, Compliance, and Control. Release 26A. Quote: “The feature is not available to be enabled outside of the early adoption program.” Oracle sells ERP software.