AI workflow automation for a Dutch small or midsize business means work moves from incoming email into the right software itself, with AI only where judgment is needed. Almost every business now uses AI, but the only measurement explicitly asking about routine-task automation finds just 15 percent doing it. That gap offers the opportunity.
This page compares Dutch figures from CBS, Eurostat, KVK, Exact, ABN Amro, and STRATO, explaining what each measures. Then it covers suitable processes, what agents add to classic Zapier or Power Automate flows, vendor prices, and why experience is the barrier rather than money. Platform choice is in the best AI automation platform for Dutch businesses. Definitions are in AI agents versus workflow automation.
The short answer
- In 2025, 27 percent in the Dutch 10-49-worker statistical category used at least one AI technology, compared with 11 percent in 2023 (CBS, 12-12-2025).
- Seventy percent of small and midsize business owners say they use AI, mainly standalone tools: 56 percent for text and images (Exact, 03-08-2026).
- Only 15 percent automate routine tasks with AI, the lowest of nine countries studied (Ondernemen & Internet on STRATO/YouGov, 30-06-2026).
- Small and midsize businesses use AI most for administration or management, ahead of marketing (CBS, 16-03-2026). Administration is the Netherlands’ most cited purpose at 39.3 percent (Eurostat, 2026).
- Those considering but rejecting AI cite lack of experience (73 percent), privacy (49), and legal consequences (42) (CBS, 12-12-2025). Only 20.3 percent of microbusinesses cite excessive costs (CBS, 16-03-2026).
- Across 375 mid-market and enterprise businesses, AI accounted for only 18 percent of workflow steps; the rest involved rules and moving data (Zapier, July 2026 update).
- Our position: the Netherlands has a work-chain problem. Most figures measure chat AI; nobody separately measures the email-to-software chain, where work stalls.
How many Dutch small and midsize businesses use AI in 2026?
In the Dutch 10-49-worker category, 27 percent used at least one AI technology in 2025, fewer than larger businesses. CBS annually asks whether a company uses technologies from a fixed list, such as text mining, text generation, speech recognition, or machine learning. Its series by statistical size class:
| Statistical workforce category | 2023 | 2024 | 2025 |
|---|---|---|---|
| 2-10 | 7% | 11% | 14% |
| 10-49 | 11% | 19% | 27% |
| 50-250 | 20% | 31% | 45% |
| 250 or more | 42% | 53% | 66% |
| All, 2 or more | 8% | 13% | 17% |
| All, 10 or more | 14% | 23% | 33% |
Sources: CBS, 12-12-2025 and, for the 10-plus row, CBS Digitalization and knowledge economy 2025, 10-02-2026. The 2025 figures are provisional. Eurostat marks a Dutch time-series break in 2025.
Three findings stand out. The 10-49 category grew fastest proportionally: 11 to 27 percent is 2.45 times in two years, versus 2.25 for 50-250 and 1.57 for 250-plus. The gap remains: a thirty-person company is 18 percentage points below a hundred-person company. Sector differences exceed size differences. Among businesses with at least 10 workers, AI use is 70 percent in ICT, 59 percent in financial services, and 55 percent in specialist business services. Hospitality, transport, and construction show 16 to 21 percent (CBS, 10-02-2026).
The Netherlands ranks highly in Europe. Eurostat lists 33.2 percent AI use in its Dutch 10-plus category against 20.0 percent EU-wide. For the 10-49 category, it is 28.8 percent versus 17.0 percent (Eurostat, 2026). A twenty-person Dutch company doing nothing with AI therefore trails its own sector rather than Europe.
Why do the figures say 15, 27, 70, and 78 percent simultaneously?
They measure different things. An open ChatGPT tab, a specific AI technology, and routine-work automation are different measures. Figures from 15 to 78 percent can all be correct for different questions.
| Source | Measurement | Figure | Sample | Does the commissioner sell something? |
|---|---|---|---|---|
| ABN Amro, 24-06-2026 | AI “in some form” (translated) | 78% | 777 organizations | Bank |
| Exact, 03-08-2026 | Owner uses technology (2-250 FTE survey category, Q1 2026) | 70% | Over 1,700 | Yes, accounting software |
| KVK, 11-09-2025 | Uses ChatGPT or Microsoft Copilot (June 2025) | 30% | 651 | No |
| CBS, 12-12-2025 | At least one listed AI technology, 10-49 category | 27% | Official statistics | No |
| STRATO/YouGov, 30-06-2026 | AI for routine tasks, such as organizing email or scheduling tasks | 15% | 219 decision-makers | Yes, web hosting |
| CBS, 12-12-2025 | Robotic process automation, all businesses with at least 2 workers | 3% | Official statistics | No |
Only the bottom two concern workflows. The top two mainly indicate almost everyone has opened a chat window. Exact says use is “mainly as a standalone tool and for accessible tasks” (translated). Only 36 percent use AI for business performance insight. Forty-seven percent lack knowledge of how to use it (Exact, 03-08-2026).
Two cautions: dividing 15 by 70 is invalid because samples differ. “One in five AI users automates” is no valid calculation. STRATO’s 219 Dutch decision-makers form a small, hosting-vendor-funded sample. Nevertheless, it is the only Dutch measurement we found explicitly asking about routine automation. It agrees directionally with CBS process automation, unchanged between 2024 and 2025.
Do Dutch small and midsize businesses have an AI problem or a work-chain problem?
A work-chain problem: AI is already present, but work does not yet move between places automatically. Combining figures produces a picture none of the sources draws itself.
The digital foundation exists. Text mining (22 percent) and text generation (16 percent) lead AI technologies among Dutch 10-plus businesses (CBS, 10-02-2026). These are the technologies needed to read incoming email and write replies. The AI required for a chain is already available.
The automation layer remains small and static. Robotic process automation, closest to “work moves by itself,” is 3 percent and did not grow (CBS, 12-12-2025). Only 15 percent automate routine tasks with AI (STRATO/YouGov, 30-06-2026).
For your company, “we use AI” will cease to distinguish you within a year. If 70 percent of competitors have ChatGPT open, everyone writes faster. The difference appears when employees stop retyping: an emailed invoice reaches Exact with the right cost center, or a report becomes an order for the right technician. A chat window does not see incoming work or write into your software. See why ChatGPT and Copilot are not enough.
Cheaper chat tools therefore do not close the gap. Access to language models is already widespread. The missing element is the connection among inbox, file, and software, with someone deciding which steps require judgment and approval.
Which processes use AI workflow automation in small and midsize businesses?
Administration leads, ahead of marketing. This difference from the smallest businesses is substantial and rarely noted.
AI-using microbusinesses in the 2-9-worker category most often use it for marketing and sales (32.7 percent), then administration or management (25.9 percent). Small, midsize, and large businesses instead prioritize administration or management (CBS, 16-03-2026). Eurostat confirms the Dutch pattern: 39.3 percent of AI users cite administrative organization or management, versus 37.2 percent marketing and 22.7 percent accounting and financial management. The Netherlands is one of seven EU countries where administration leads (Eurostat, 2026).
| Purpose among AI-using businesses | All Dutch businesses, CBS 2025 | Dutch 10-plus, Eurostat 2025 |
|---|---|---|
| Marketing or sales | 35% | 37.2% |
| Administration or management | 32% | 39.3% |
| Research and development | 25% | 25.6% |
| Production or service processes | 18% | 20.8% |
| Accounting, auditing, or financial management | 17% | 22.7% |
| ICT security | 11% | 19.9% |
| Logistics | 4% | 6.3% |
Sources: CBS, 12-12-2025; Eurostat, 2026. CBS starts at 2 workers, Eurostat at 10, explaining differences.
In practice, this means work arriving by email that must end in software: purchase invoices, orders, reports and faults, hours and weekly timesheets, planning, status questions, and requests for missing data. See tasks for AI, first administrative processes to automate, and processes increasingly automated with AI.
What does an AI agent add to a Zapier or Power Automate flow?
An agent handles judgment, such as reading messy email and identifying tasks. Ordinary rules handle the rest. Classic Zapier, Make, n8n, or Power Automate flows use if-then logic: email from this address means put the attachment in that folder. Structured input works. One email containing an invoice question, address change, and complaint defeats a simple rule.
Zapier defines AI workflow automation as “the use of AI models inside automated workflows to handle steps that require judgment”. Across 375 mid-market and enterprise businesses, AI was “just 18% of workflow steps. The rest ran on conventional automation”. Flows using AI only for reasoning “cost 71% less to run than workflows that route everything through a model” (Zapier, July 2026 update). Zapier sells this, but the direction applies to smaller businesses: agents fill gaps in flows.
| Chain step | Classic flow | AI contribution | What stays fixed |
|---|---|---|---|
| Email arrives | Sender or subject rule | Reads and identifies tasks, including three in one email | Trigger |
| Retrieve data | Fixed field mapping | Finds customer and file from vague or incomplete input | Software connection |
| Decide | If-then | Judgment under uncertainty or request missing data | Approval for payments, customer messages, contracts |
| Write to software | Fixed action | Supported proposal | Write action after approval |
Makers move the same way. Zapier says it “is migrating Agents (agents.zapier.com) to AI by Zapier, a more powerful AI experience built directly into the Zap editor” (Zapier Help, 07-09-2026). The standalone agent returns to the workflow. Make has beta AI Agents within its scenarios (Make, accessed 30-09-2026). The question becomes which step requires judgment and which waits for a person. See AI agents versus traditional workflow automation.
What do Zapier, Make, n8n, Power Automate, and Copilot Studio cost?
Software is cheap: paid Zapier starts at $19,99 monthly and Power Automate Premium costs $15 per user monthly. Setup, integration, and maintenance are expensive. Prices were read September 30, 2026. Zapier’s pricing page had no readable amount; its article supplies the starting price.
| Product | Vendor price | Billing unit | Source |
|---|---|---|---|
| Zapier | Paid from $19,99 monthly | Tasks; each AI step counts | Zapier |
| Make | Free $0, Core $9, Pro $16, Teams $29 monthly; Enterprise on request | Credits, from 10,000 monthly | Make |
| n8n | Starter €20, Pro €50, Business €667 monthly with annual billing; self-hosted Community Edition free | Full workflow execution | n8n |
| Power Automate Premium | $15,00 | Per user monthly, annual contract | Microsoft |
| Power Automate Process | $150,00 | Per bot monthly, annual contract | Microsoft |
| Copilot Studio | $200,00 per 25,000-credit package | Monthly, organization-wide | Microsoft |
| Microsoft 365 Copilot | $30,00 | Per user monthly, annual contract | Microsoft |
Microsoft’s example agent takes four actions per new order: check inventory, find delivery time, approve order, email customer. Each costs 5 credits, or 20 per order (Microsoft Learn, updated 14-09-2026). At $0,008 per package credit, that is about $0,16 per order; $200 covers 1,250 monthly orders. A wholesaler handling fifty daily orders must consider the limit: Microsoft disables the agent above 125 percent of capacity.
n8n charges per complete execution, “regardless of complexity”, and gives businesses with fewer than 20 employees 50 percent off Business (n8n, accessed 30-09-2026). Self-hosting has no license cost, but needs a server and maintainer.
None includes the most time-consuming work: connecting Exact, AFAS, or Syntess, finding exceptions, and updating flows when suppliers change invoices. Total costs are in administrative automation with AI pricing.
Why do Dutch AI workflows stall at Exact, AFAS, and Outlook?
Dutch work runs through Dutch software, absent from almost all English-language workflow guides. Zapier claims over 9,000 apps (Zapier, July 2026 update); Make over 3,000 (Make, accessed 30-09-2026). But the English comparisons we read omit Exact Online, AFAS, Twinfield, Moneybird, SnelStart, Nmbrs, and Syntess.
An installation company, owners’ association manager, or twenty-person accounting firm works differently. Reports arrive in shared Outlook mail; customers, planning, and invoices live in separate packages. None sees the work between them. Exact and AFAS build assistants within their own walls. Five packages create five assistants unable to see each other. See AI inside AFAS or Exact, or an AI agent alongside.
If your business runs entirely in Microsoft 365, Power Automate is the obvious first layer because licensing is often already present. If industry software drives work, first check existing connections and who maintains them after updates. See workflow software and existing business systems, Exact Online, AFAS, and Microsoft 365.
What stops Dutch businesses from automating workflows with AI?
Lack of experience, privacy concerns, and unclear liability. Money ranks low. Among businesses considering but rejecting AI, CBS finds 73 percent citing experience, 49 percent privacy, and 42 percent legal consequences “such as liability for damage” (translated). Only 15 percent find AI unhelpful (CBS, 12-12-2025). Microbusiness figures are 71.6 percent for experience and 20.3 percent for excessive cost (CBS, 16-03-2026). EU-wide: “lack of relevant expertise (70.89%), followed by lack of clarity about the legal consequences (52.52%)” (Eurostat, 2026).
Users have barely addressed risks. Seventy-eight percent use AI in some form, but 9 percent have formal external-AI policies, versus a third of large businesses. Almost a third worry about employees sharing confidential information (Computable on ABN Amro, 24-06-2026). Forty-eight percent in Exact’s survey worry about employees entering business data (Exact, 03-08-2026). In June 2025, KVK found only 2 to 3 percent actively preparing for the European AI regulation; half did not know it (KVK, 11-09-2025).
Cheaper tools do not solve the main barrier. Experience and clear permissions do. A flow built in one afternoon moves the experience problem to when it stalls inexplicably. A sound start records decision sources and steps requiring a person. See AI without an IT department.
Where is AI workflow automation heading for small and midsize businesses?
AI use grows quickly while chain automation barely grows. The gap will widen before narrowing. In the 10-49 category, use rose 11 to 19 to 27 percent, 8 percentage points yearly (CBS, 12-12-2025). Straight-line continuation reaches 43 percent in late 2027; relative growth at 1.45 annually reaches 57 percent. RPA stayed at 3 percent. Thirty-two percent intend to use AI for routine tasks within six months; 24 percent see no potential (STRATO/YouGov, 30-06-2026). Intentions are not behavior.
Our expectation: by late 2027, 43 to 57 percent in the Dutch 10-49 statistical category use an AI technology, but fewer than a quarter run an email-to-software chain without retyping. The 25 percent is an expectation, rather than extrapolation: nobody measures it yet.
Reasoning: makers move agents into workflows (Zapier, September 2026), packages build assistants, and CBS and Eurostat identify an experience shortage. Cheaper components do not replace deciding which steps require judgment or approval. That takes scarce experience in a twenty-person company.
The hypothesis fails if Exact, AFAS, or Microsoft supply standard cross-package agents that write after approval, increasing adoption faster. It also fails if a new CBS or Eurostat administrative-chain indicator shows higher adoption. Until then, Bombos arranges the first chain with your team, which learns to create the next itself.
What can AI workflow automation not yet do?
A workflow depends on what it can see and the rules you give it. Everything beyond remains human work. Five limits survive better models:
- Knowledge only in someone’s head. A customer always wanting to pay by mail, or a technician avoiding ladders, may be undocumented. No agent finds it. Someone must ask and record it.
- Messy sources. Conflicting customer data in three packages can make agents choose wrongly. Connections do not fix this; cleaned sources do.
- Liability. The AI regulation and GDPR place responsibility on the business. Forty-two percent of those considering AI rejected it partly over legal consequences (CBS, 12-12-2025). Payments, customer messages, and contracts should await approval.
- Maintenance. Updates or changed supplier invoices can break working flows. Someone must notice.
- Measurement. We found no public Dutch measurement of time saved by small-business AI workflows. Advertising figures usually rely on self-estimates without a measurement design.
How does Bombos approach this?
Bombos builds AI agents for growing Dutch businesses, enabling more work, at a higher quality, with the same team. It provides the layer between inbox and software.
Chef checks connected email addresses, identifies tasks, and assigns specialists. Wegwijzer explains, helps set boundaries, and proposes work to take over. Specialists are built around your tasks, systems, and rules. Bombos reads and writes in Exact, Dutch accounting software, AFAS, Dutch business software, Visma, business and accounting software, Twinfield, Moneybird, and SnelStart, accounting packages, Nmbrs, payroll software, Syntess, installation-industry software, Microsoft 365, and Google Workspace.
Each proposal shows its basis. You approve, change, or reject in Bombos; the software then holds only the result. Corrections become rules for coworkers too. Payments, customer messages, and contracts wait for approval by default. Bombos interviews you to record knowledge only in people’s heads.
This addresses CBS’s leading barrier: experience. The first task starts the process. Then your team teaches Bombos the next without technical skills or needing us. See where to start with AI in your business.
Sources
Each source was opened on September 30, 2026, and each original quotation appears verbatim in it. Dutch quotations above are labeled as translations.
- CBS, Businesses use AI most often for marketing or sales, 12-12-2025: https://www.cbs.nl/nl-nl/nieuws/2025/50/bedrijven-gebruiken-ai-vaakst-voor-marketing-of-verkoop
- CBS, Digitalization and knowledge economy 2025, chapter 3: ICT use by businesses, 10-02-2026: https://www.cbs.nl/nl-nl/longread/rapportages/2026/digitalisering-en-kenniseconomie-2025/3-ict-gebruik-bij-bedrijven
- CBS, AI technology use by Dutch microbusinesses, 16-03-2026: https://www.cbs.nl/nl-nl/longread/rapportages/2026/gebruik-van-ai-technologie-door-nederlandse-microbedrijven?onepage=true
- Eurostat, Use of artificial intelligence in enterprises, Statistics Explained, 2026: https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises
- Exact, Businesses embrace AI but strategic use lags, 03-08-2026 (sells accounting software): https://www.exact.com/nl/nieuws/mkb-omarmt-ai-massaal-maar-strategische-inzet-blijft-achter
- Ondernemen & Internet, Only 15% use AI for routine tasks (STRATO/YouGov; STRATO sells hosting), 30-06-2026: https://www.ondernemeneninternet.nl/maar-15-van-mkb-zet-ai-in-voor-routinetaken/
- Computable, ABN Amro: businesses use AI but lack relevant risk policies, 24-06-2026: https://www.computable.nl/2026/06/24/abn-amro-mkb-gebruikt-ai-maar-ontbeert-beleid-voor-relevante-risicos/
- KVK, Entrepreneurs barely prepare for European AI legislation, 11-09-2025: https://www.kvk.nl/pers/kvk-onderzoek-ondernemers-zetten-nauwelijks-stappen-rond-europese-ai-wetgeving/
- Zapier, AI workflow automation, updated July 2026 (sells workflow automation): https://zapier.com/blog/ai-workflows/
- Zapier Help, Migrating from Agents to AI by Zapier, updated 07-09-2026: https://help.zapier.com/hc/en-us/articles/47402591569805-Migrating-from-Agents-to-AI-by-Zapier
- Microsoft, Power Automate pricing, accessed 30-09-2026: https://www.microsoft.com/en-us/power-platform/products/power-automate/pricing
- Microsoft, Copilot Studio plans and pricing, accessed 30-09-2026: https://www.microsoft.com/en-us/microsoft-365-copilot/pricing/copilot-studio
- Microsoft Learn, Billing rates and management (Copilot Studio), updated 14-09-2026: https://learn.microsoft.com/en-us/microsoft-copilot-studio/requirements-messages-management
- n8n, Pricing, accessed 30-09-2026: https://n8n.io/pricing/
- Make, Pricing, accessed 30-09-2026: https://www.make.com/en/pricing