AI automation in the Netherlands doubled in two years, reaching 17% of businesses in 2025, but the industrial base sits at the bottom of that curve. Manufacturing runs at 16%, transport and storage at 7%. Our view at Lleverage is that the gap now sits between front-office text work and back-office execution.
Read the national statistics and the Netherlands looks like a country that already made the leap. Read the sector splits underneath them and a different picture appears. The companies that make, move and sell physical products are the least automated of all. That is not a story about Dutch caution. It is a story about which kind of AI got bought first.
We are usually called in after a year of promising pilots has left a company with four people still typing orders into an ERP. That is the vantage point this article is written from. We build agents for that work, so our interest is plain. What we hold ourselves to here is published statistics and named Dutch customers. If you would rather see the mechanics than read about them, the plan and produce page walks through one of these processes end to end. You can also book a demo and have it run against your own documents.
How many Dutch businesses actually use AI?
In 2025, 17% of Dutch businesses with two or more employees used at least one AI technology, according to Statistics Netherlands. That is double the 8% recorded in 2023, and up from 13% in 2024. Adoption climbs steeply with size, from 14% among the smallest firms to 66% among those with 250 employees or more.
The sharpest movement is in the middle of the market. Companies with 50 to 250 employees went from 20% in 2023 to 45% in 2025, more than doubling in two years. Firms with 10 to 50 employees reached 27%. That middle band is where most Dutch manufacturers, wholesalers and distributors sit. They are large enough to feel the administrative load, and small enough that nobody has a spare team to fix it.
The European comparison runs on a different basis, covering firms with ten or more employees, and on that measure the Netherlands ranked sixth in the EU in 2024 at 23% against an EU-27 average near 13%. Both readings point the same way. On any national scorecard, Dutch business is not the laggard.
Why does the Netherlands lead Europe on AI adoption?
Three structural advantages put the Netherlands near the top of the European table. Dense digital infrastructure, an unusually international and English-fluent workforce, and a business culture comfortable with flat decision-making. Together they shorten the distance between someone spotting a problem and someone being allowed to fix it.
Infrastructure comes first. Household broadband access sits near 98%, among the highest in the European Union, and Amsterdam hosts AMS-IX, one of the largest internet exchanges in the world. The practical barriers that slow adoption elsewhere barely register here. Digital public services set a baseline expectation too. People file taxes and manage healthcare digitally without friction. Against that, a manual paper process at work looks like a choice rather than a fact of life.
Culture does more work than people credit. Dutch organisations tend to be flat. An operations controller who can demonstrate a better way of processing orders will usually get a hearing without a six-month steering committee. That is a genuine advantage over more hierarchical markets. It shows up in how quickly a pilot moves from one desk to a department.
Regulation is the third factor, and it cuts both ways. Dutch firms have had strong data-protection practice since well before the EU AI Act. Governance questions therefore tend to be answered early rather than discovered late. That slows the first month of a project and speeds up everything after it.
Where is Dutch AI adoption actually happening?
Adoption is concentrated in the sectors that were already digital and in the tasks closest to language. Information and communication led at 54% in 2025, followed by specialised business services at 31% and financial services at 28%. Manufacturing sat at 16%, trade at 12%, and transport and storage at 7%, the lowest of any major sector.
| Sector | AI adoption, 2025 |
|---|---|
| Information and communication | 54% |
| Specialised business services | 31% |
| Financial services | 28% |
| Energy, water and waste | 18% |
| Manufacturing | 16% |
| Trade | 12% |
| Transport and storage | 7% |
The task split tells the same story from another angle. Among businesses using AI, 35% apply it to marketing or sales and 32% to administrative or management tasks. Accounting and finance accounts for 17%, production and service processes 18%, and logistics just 4%. The most common technologies were text mining at 12% and natural language generation at 8%. In other words, the Netherlands has adopted AI that writes and reads text, mostly in the front office.
Our reading of these two splits together is straightforward. The national lead is real, but it has been earned largely by sectors whose product is already information. The industrial base, where the country's actual economic weight sits, is running several years behind on the same clock. A wholesaler at 12% adoption and a logistics operator at 7% are not resisting AI. They have simply not been sold anything that touches the work that costs them money. Logistics at 4% of AI use is the clearest single number in the whole dataset.
What does back-office AI look like in a Dutch company?
Back-office AI reads the documents a business already receives and extracts what matters. It applies the company's own rules, then posts the result into the ERP without a person retyping it. It handles order intake, invoice matching, quotes and technical queries. The measure of success is not text generated but transactions completed correctly.
Take order intake, the single most common starting point among Dutch product businesses. Topa Bathroom Products is a Dutch importer and wholesaler serving around 700 customers. Four and a half people once entered every incoming order into Microsoft Dynamics 365 Business Central by hand. Orders arrived as plain emails, PDFs and spreadsheets, each customer with their own format.
"We had four and a half people, about 3.8 full-time equivalents, sitting there all day long, manually entering every single order into our Business Central ERP." Bryan van Ingen, Operations Director, Topa Bathroom Products
Today more than 90% of Topa's incoming orders are processed straight into Business Central without manual input. Customers receive confirmation within 30 seconds. The people who did that entry now plan service mechanics and handle after-sales work instead. Nothing about that outcome required a new ERP or a change to how customers place orders. That is the shape of an autonomous back office. The process runs, and people handle only what genuinely needs judgement.
The pattern repeats across other Dutch operations. At Koninklijke Dekker, a timber business trading for more than 140 years, an agent reads PDFs, spreadsheets and emails directly into the same ERP. At the flower wholesaler Xpol, roughly 20 minutes is saved on each large order across about 150 orders a week. That absorbed both a retiring specialist's workload and volume growth, with no new hires. At J. Kisch and Zonen, a 130-year-old Amsterdam furniture supplier, the co-owner's observation was that 80% of daily customer questions were the same question.
"We're constantly looking for ways to ensure we can last another 100 years and further improve our service and quality." Mart, Continuous Improvement Team, Koninklijke Dekker
None of these are technology showcases. They are ordinary Dutch companies removing a specific, repetitive, expensive task. That is what the adoption statistics miss when they count whether a business "uses AI" without asking what the AI is doing.
What is holding Dutch manufacturers and wholesalers back?
Statistics Netherlands put the question to the non-adopters directly in 2025. Among businesses not using AI, 73% cited a lack of experience or knowledge, 49% named privacy concerns and 42% pointed to legal uncertainty about liability. Not one of those is a statement about the technology being unproven. The real constraint is not talent, budget or appetite. It is that the AI most companies met first was a chat assistant, and a chat assistant cannot post an order. When the first thing a business tries produces drafts rather than completed transactions, the natural conclusion follows. AI is interesting, but not yet operational.
The second constraint is the ERP itself. A Dutch SME running Business Central, Exact, AFAS or SAP carries years of accumulated rules and customer-specific quirks. Most of it is encoded in the heads of two or three long-serving staff. Any automation that cannot work inside that system, and cannot absorb those rules, ends up creating a second place where work happens. The ERP integration question is therefore the whole question, not a technical footnote to it.
Third is the pilot trap. A pilot that never touches the production ERP teaches an organisation nothing. It says nothing about whether the work can actually be automated. It also burns the credibility needed for the next attempt. Lleverage's position is that a first project should be chosen for how measurable it is, not for how impressive it looks. One document type, one process, one system, with a number attached before you start.
Finally, there is a quiet demographic pressure the statistics do not capture. Dutch industrial firms are losing experienced staff to retirement faster than they can replace them. Much of what those people know was never written down. Xpol's agent codifies 25 separate customer rulesets that previously lived in one specialist's memory. That is a knowledge-retention project as much as an efficiency one.
How should a Dutch SME start with AI automation?
Start with one process that is high volume, rule-bound and currently manual, and attach a number to it before any work begins. Order intake, supplier invoice processing and repetitive customer queries are the three most reliable entry points for Dutch product businesses. All three are measurable within weeks rather than quarters.
Pick the process by counting, not by intuition. Ask how many documents of one type arrive each week. Ask how many minutes each absorbs from arrival to posting, and how often a person has to look something up to finish it. If a process clears roughly 100 items a week at more than five minutes each, the arithmetic settles the question on its own.
Then insist the pilot writes into the production system. A proof of concept that ends in a spreadsheet has proved nothing about the part that is hard. The hard part is posting a correct record into an ERP under real rules and real exceptions. This is where we land after enough of these projects. If it does not touch the ERP, it is a demonstration, not a pilot.
Keep the people who own the exceptions close to it. The controllers and order-desk staff who know why a particular customer's units convert differently are the source of the rules the agent needs. They are also the ones who will spot a wrong answer on day one. Automating order processing works when their knowledge goes into the system rather than out of the door with them.
Frequently Asked Questions
How many companies in the Netherlands use AI?
Statistics Netherlands recorded 17% of businesses with two or more employees using at least one AI technology in 2025, double the 8% of 2023 and up from 13% in 2024. Adoption reaches 45% among companies with 50 to 250 employees and 66% among those with 250 or more. On the EU-comparable measure the Netherlands ranked sixth in 2024.
Which Dutch sectors use AI the least?
Transport and storage recorded the lowest adoption of any major sector at 7% in 2025, level with construction. Trade followed at 12% and manufacturing at 16%, all below the national average of 17%. These are precisely the sectors handling the highest volumes of structured documents. That is why the back office holds the largest untapped gains in the Dutch economy.
Does AI automation require replacing our ERP?
No. Systems worth considering run inside the ERP you already have, whether that is Business Central, Exact, AFAS or SAP. They read incoming documents and post completed records into it. Koninklijke Dekker and Topa Bathroom Products both automated order intake directly into Dynamics 365 Business Central without changing their underlying system.
How long does a first automation project take?
For a single well-chosen process, Dutch SMEs typically see production results in weeks rather than quarters. That holds provided the pilot writes into the live ERP from the start. The timeline stretches when a project covers several document types at once. It also stretches when the business rules have never been written down and must be gathered first.
Is the EU AI Act a barrier for Dutch companies?
Not for this kind of work. Back-office document processing sits well outside the high-risk categories the AI Act concentrates on. Dutch firms generally arrive with mature data-protection practice already in place. The governance work is real but front-loaded, and most of it concerns access and audit trails rather than the model itself.
Where Dutch companies go from here
The Netherlands earned its position near the top of the European adoption tables, and the front-office half of the job is largely done. The half that is left decides something concrete. It decides whether a Dutch manufacturer still needs four people typing documents into an ERP in 2027. That is our reading of the numbers. It is also the work we do every week with companies whose order desks look much like yours. Book a demo and we will run an agent against a sample of your own documents, in your own system, before you commit.