According to Exact's MKB Barometer, 70 percent of Dutch SMEs now use AI, up from 33 percent last year. But mostly for text and images. Below is why that is no coincidence, and what sits between a loose tool and AI on your own business data.
In August 2026, Exact had research agency Markteffect survey more than 1,700 Dutch SME owners. Seven in ten use AI, compared with a third in 2025. That is a doubling in a year. The survey was commissioned by a software vendor, though, so read it as a snapshot of usage and not as a measure of return.
What is interesting is the how. More than half use AI for writing text and creating images. A little over a third use it to gain insight into business performance.
Up from 33 percent in 2025. Usage has doubled in a year.
The most common use: writing and creating pictures.
The share that points AI at its own business performance.
To let employees work with business data through AI.
A text or an image does not need business data. You type an instruction, you get something back and you judge it yourself. If it is not good, you throw it away. The risk is small and the payoff is immediate.
Insight works the other way round. A question like "which customers earn us the most margin" can only be answered with your own data: invoices, hours, quotes, the CRM. That data rarely lives in one place, and what a customer or a margin is differs per system. An AI tool that cannot reach it gives a generic answer. A tool that can half reach it gives a confident answer that is wrong, which is worse. What that fragmentation looks like and how to spot it in five minutes is covered in what is a data layer.
In the same survey, 48 percent of owners say it feels daunting to let employees work with business data through AI, and 47 percent are not sure how to use the technology effectively. Those are two different problems, and the first is more concrete than it looks.
Anyone who pastes business data into a chat window hands that data over without anything being defined about who may see what. That unease is justified. The answer is not to paste more carefully, but to have a place where permissions are already settled, so the AI only sees what the person asking the question is allowed to see.
The model knows what someone happened to copy, and nothing more.
The same models, but with a fixed source and fixed rules around them.
Between a loose tool and AI on your own data there is no big programme, just three agreements.
One definition. Settle what a customer, an open quote and a margin mean in your company. Without that, every system and every model gives its own answer to the same question.
Permissions per role. Decide who may see which data and have the AI follow the same limits. Then the answer to whether employees may work with it is a yes with conditions you can point to, instead of a feeling.
A question you can measure. Do not start with "we want to do something with AI", but with a question where you already know how much time it takes to answer. Then after a few weeks you can see whether it delivered anything. How to set up that baseline is covered in automating repetitive tasks.
Pick one question, bring the data it needs into an open, in-house database and let the existing package keep running next to it. Connect the AI using the permissions you just defined. If it works, the next question follows. If it does not, you have lost weeks instead of a year.
How we put that into practice, with an open in-house data layer that dashboards, AI chat and agents connect to, is described on the page about the AI Native Data Layer. What AI can do on top of that in your processes is covered in automating business processes with AI.
If you use AI to write texts, rephrase quotes or summarise notes, you do not need a data layer and a subscription to a tool is exactly right. The step to your own numbers only matters when you want answers that only your data can give. That is then not a matter of more AI, but of better data.
Usage has doubled in a year. What it delivers is a different question, and this survey does not answer it. If you already use AI, look at how much of it touches your own data. If that is hardly any, the gain is not in a better tool but in where your data lives.
According to Exact's MKB Barometer, conducted by Markteffect among more than 1,700 Dutch SME owners, 70 percent of SMEs use AI, compared with 33 percent in 2025. More than half (56 percent) use it for text and images and 36 percent for insight into business performance. The survey was commissioned by a software vendor and measures usage, not return.
A text or an image needs no business data, the risk is small and you judge the result yourself. Insight into your own numbers requires access to invoices, hours, quotes and the CRM, which are often scattered and defined differently per system. That makes the step to your own data bigger than the step to a writing assistant.
That depends on how. Pasting business data into a chat window gives you little control over who shared what. Safer is a setup where the AI reads from your own data layer, with permissions per role, so the model sees no more than the person asking the question. Almost half of owners (48 percent) find this daunting, which shows the concern is widespread.
Three agreements: a fixed definition of terms like customer, quote and margin, permissions per role that also apply to the AI, and a concrete question where you know beforehand how much time it takes today. A data layer is where those agreements come together, but you start with a question and not with a year-long programme.
In the whitepaper on the AI Native Data Layer you can read how to make AI work on your own business data, with permissions per role, in-house and without turning everything upside down at once.
Download the whitepaper