A staff member uses a chatbot to draft a customer reply. It saves ten minutes and sounds polished. Nobody checks one small promise hidden in the final paragraph. Two days later, the customer expects something the company never intended to offer.
That ordinary scene matters more to a small employer than another grand prediction about artificial intelligence. The firmer Dutch signal comes from CBS and UWV: AI use is increasing, but its immediate effect is on everyday tasks, supervision and responsibility rather than wholesale replacement of staff.
CBS found that 13.8 percent of Dutch microbusinesses with two to nine workers used at least one measured AI technology in 2025. That was up from 10.6 percent in 2024 and 6.8 percent in 2023. Among users, marketing and sales were the most common applications, followed by administrative and management work.
The job changes quietly
This is how workplace change usually arrives in a small company. There is no formal launch. Someone finds a useful tool, another colleague copies the habit, and soon part of the company’s work is being done differently.
The first visible benefit is speed. Drafts appear faster. Notes become summaries. Recruitment text takes minutes. Information can be sorted without asking a senior employee to spend an afternoon on it. Yet the remaining human work often becomes harder. Someone must recognise an exception, test a claim, protect personal information and decide whether the answer suits the person receiving it.
UWV reports that about half of employers expect AI to alter training needs and required skills. It also found that around 60 percent of employers using AI guide their staff in that use. A substantial group is therefore using AI without reported staff guidance.
I read this less as a technology gap than as a management gap. Employees cannot respect boundaries that nobody has discussed. Managers cannot review work properly if they do not know where software entered the process.
Faster work still needs an owner
In a small business, responsibility has nowhere to hide. The founder may still approve the quotation, explain the invoice, answer the rejected applicant or repair the customer relationship. Software can prepare the words, but it does not carry the commercial consequence.
This becomes particularly sensitive when AI touches recruitment, appraisal, scheduling or work allocation. The European AI Regulation follows a risk-based approach. The Dutch government has highlighted data quality, risk management, transparency and human oversight among the requirements for high-risk systems. The treatment of a particular HR tool depends on what the system does and how the employer uses it.
For an employer, the practical question comes earlier than the legal classification. Does the company know what information enters the tool, what influence its output has, who checks it and who makes the final decision? A recruitment assistant that improves wording presents a different issue from software that ranks applicants or influences someone’s prospects.
The same discipline belongs in customer work and administration. Names, CVs, prices, internal notes, payroll information and commercial plans can move outside the familiar company process through one casual prompt. A policy may help, but a document alone cannot supervise the work. People need to understand the boundary during an ordinary Tuesday afternoon.
Do not remove the junior steps
There is another risk that receives less attention. Many routine tasks are also training tasks. A junior employee learns the company by drafting the first reply, checking a standard document or sorting information before a manager reviews it.
If software takes that first step, the company may gain time while losing part of its learning route. The junior worker sees the finished answer without developing the judgment behind it. Later, that same person may need to detect an error or handle an exception with too little practical experience.
This does not make AI unsuitable for junior work. It means the employer must consider how judgment will be taught when repetition disappears. Review can become a learning moment rather than a final signature. Staff can explain why an output is reliable, what they checked and where they disagree with it.
That matters in the present Dutch labour market. CBS counted 375,000 open vacancies at the end of the second quarter of 2026, with 95 vacancies for every 100 unemployed people. Unemployment stood at 3.9 percent. The market has cooled from its tightest point, but capable workers remain difficult to find in many businesses. UWV says employers increasingly point to missing skills, occupational knowledge and experience.
Time saved is not cash earned
The tempting payroll story is that faster software must mean fewer paid hours. Small-company economics rarely work so neatly. Ten minutes saved creates value only when the company uses the capacity well. It may support faster invoicing, more completed customer work, fewer corrections or lower external costs. Otherwise, it is simply spare time beside another monthly software subscription.
I would separate the tool cost, the staff time affected and the verified business result. That makes the conversation calmer. It also prevents speed from becoming the only performance measure. When people are rewarded merely for producing more, weak checking can look like productivity until the correction arrives.
Return to that customer email. The useful question is not whether the chatbot wrote it. The question is whether the employee knew what could be promised, checked the details and remained responsible for what was sent. That is work design, not technological enthusiasm.
AI may eventually alter staffing levels in particular companies. For most small Dutch employers, the earlier decision is more immediate: which tasks can move faster without making judgment, learning and accountability disappear? The business that answers that carefully may gain time. More importantly, it will still know who owns the final answer.
Need practical rules for AI use, staff guidance and review in your business? We can help you set them
The data, sourcing, and analysis behind this article were conducted by Linda Pavan. AI was not used to identify sources, build the factual basis, or produce the analytical judgment contained here. AI was used only as a drafting aid. The final English text was personally reviewed, edited, and approved by Linda Pavan before publication.
References
- AI Jobs Barometer | PwC
- CBS - AI adoption by microbusinesses
- CBS - Sector and size differences in AI use
- UWV - Employer preparation, training and staff guidance
- UWV - Labour-market pressure and the skills mismatch
- CBS - Latest actual labour-market position
- Rijksoverheid - AI governance and human oversight
- CBS - Worker perception and the need for credible change management
