Silvar
← Blog

More AI, less control: how to turn loose experiments into a workable AI policy

The AI Trend Study 2026 shows that organizations are using AI more often ad hoc and less often according to policy, even though oversight of AI literacy is possible. Here's how an SME can lay the foundation in an afternoon.

Bureau met losse notities aan de ene kant en een geordende stapel kaarten aan de andere kant

At many companies it goes like this: one colleague discovers that ChatGPT can get a quote on paper faster. Another has Copilot summarize meeting minutes. Yet another pastes a customer email into a free tool to have a reply written. It all works, and nobody has ever made any agreements about it. New Dutch research shows that this pattern is more widespread than you might think. AI use is growing, but the control over it is decreasing.

What the AI Trend Study 2026 shows

Berenschot and Nyenrode presented the results of the AI Trend Study 2026 on 23 September 2026 in Breukelen, under the telling theme 'Strategy first, then AI'. The figures appeared in the trade press in mid-October, and they reveal a striking contrast.

  • Enthusiasm is growing: three quarters of respondents now consider AI use desirable, compared with 65% a year earlier.
  • Ad hoc use is rising: from 32% to 38%.
  • Policy-based use is falling: from 27% to 19%.
  • A vision on paper is growing: 83% say they have a guiding vision on AI, up from 74% in 2025.
  • Training is lagging behind: 37% do not offer employees any AI training.

There are also differences between sectors. In services, nearly 40% say they already deploy AI structurally; in industry that is 20%. In government, use has actually declined compared with 2025. AI is mainly used for operational support and rarely plays a role in strategic decisions. Responsibility is increasingly landing with the IT department.

One caveat applies. Just over 200 respondents took part in the study, and a large share comes from the public sector and the built environment. It is therefore not a representative picture of the SME sector. Still, the core will sound familiar to many entrepreneurs: more use, less control.

Why this matters now: AI literacy

The timing makes the topic even more pressing. Article 4 of the European AI Act asks organizations to work on the AI literacy of their people. That obligation has applied since February 2025, and since 2 August 2026 supervisory authorities may check compliance. According to various sources, this also applies if you only use ready-made tools such as ChatGPT or Microsoft Copilot.

The sources do not agree on the precise interpretation. Kennisdomein reports that Article 4 was replaced on 27 July 2026 by a lighter version. According to that version, it concerns measures that support the development of AI literacy, without a specific level being guaranteed. The reports diverge on sanctions. One source mentions fines of up to 15 million euros or 3% of global annual turnover. Another source states that the Dutch draft implementing act mainly opts for an order subject to penalty, and that this draft has not yet entered into force. The same source reports that the rules for high-risk AI have been postponed to 2 December 2027. Our advice: check the current position with official sources, such as the Dutch Data Protection Authority (Autoriteit Persoonsgegevens). On one point, all sources do agree: being able to demonstrate that you are working on AI literacy is wise.

An AI policy on one A4

In its Strategy Trend Study 2026, Berenschot offers a clear piece of advice. Determine who is ultimately responsible, build in safety checks, set a clear framework for AI use by employees, and regularly test whether it works. For an SME, this does not have to be a bulky document. One A4 with five components is already a significant step.

Blank A4 sheet with five coloured tabs as a symbol of a concise AI policy
  1. Owner: who is ultimately responsible for AI use? Assign this to one person. It does not necessarily have to be someone from IT, as long as that person keeps an overview.
  2. Approved tools: which tools may be used, and with what type of account? A business licence often comes with different data agreements than a free version.
  3. Red lines for data: what must never go in? Think of customers' personal data, medical information, passwords and confidential contracts.
  4. Short training per role: a salesperson needs different knowledge than someone in administration. For each role, cover what AI does well, where it goes wrong and how to check outcomes.
  5. Record and evaluate: note who has completed which training and check every quarter whether the agreements still match practice.

Start small, but start

The research also shows that a vision on paper is something different from policy in practice. An A4 that disappears into a drawer changes nothing. So begin with what is already happening. Ask your team which tools they use and for what. You will usually discover two things. There are applications that work so well that you want to make them structural, for example in document processing or answering customer questions. And there are habits that you would rather agree on today than tomorrow.

Moving from ad hoc to structural therefore does not happen by buying more tools. It comes down to deliberate choices: where does AI genuinely deliver value, and under which conditions?

At Silvar we work with AI every day, and we help SMEs turn loose experiments into applications that reliably fit their existing processes. Would you like to brainstorm about where AI makes sense in your business, and how to arrange it properly? Ruben is happy to think along with you.

Sources

Made with Startspot — build your own site in minutes