AI Barriers in Danish SMEs: What Is Holding Businesses Back?
In a Danish SME survey from autumn 2025, 44% had worked with AI, while 10% had implemented it systematically. Read about five themes and possible first steps.
Table of contents
AI barriers for Danish SMEs can be technical, organisational and legal. This article covers five themes: skills and time, scattered data, GDPR and compliance, the business case, and dependence on individuals. Agent360 can build a unified workflow for your use of AI. We clarify which systems and accounts it should use, who has access, and which rights to data and to the solution the agreement must secure.
The short answer: Look at both the technology and the framework around it. According to the Danish Business Promotion Board's Business Panel (autumn 2025, 4,128 SMEs), 44% had worked with artificial intelligence, while 10% had systematically implemented solutions in business procedures, processes or value chains. The survey describes uptake and barriers; it doesn't on its own explain why a particular business has gone from trying things out to live operation. The article goes through five themes and possible first steps that you can assess in your own business.
AI barriers are the practical, organisational and legal obstacles: a lack of skills and time, scattered data, compliance uncertainty, an unclear business case and dependence on individuals, that prevent a business from going from scattered AI experiments to artificial intelligence as a reliable part of day-to-day operations.
If you're thinking "we should be doing more with AI, but we don't know where to start", you can use the themes here to delimit the task. First look into which specific circumstances stand in your way in your business, and then choose a first step.
Status: how far have Danish SMEs got with AI?
Let's start with the picture the figures actually paint, because it's both more encouraging and more sober than the debate often suggests.
The figures from Statistics Denmark
Statistics Denmark put AI use among private, non-financial urban industries (byerhverv) with at least 10 full-time employees at 15% in 2023 and 28% in 2024. The most widespread uses in 2024 were, according to the same release, producing written text or spoken language (18%) and text analysis (17%). These are historical figures for this defined population.
The figures show how widespread the uses mentioned are. They don't establish which technologies or organisational circumstances caused the growth. A browser-based text tool can be easy to try out, while live operation still requires an assessment of the task, the data and the workflow.
Experiment vs. operation: the big gap
The Business Promotion Board's survey from autumn 2025 zooms in on SMEs specifically, and here the picture gets sharper: 60% of SMEs find artificial intelligence relevant, 44% have worked with it, but only 10% have implemented it systematically in business procedures, processes or value chains, and just 2% have developed products or tools themselves that are actively used in the business. Of the SMEs that use AI, three out of four according to the report have only experimented with common tools such as ChatGPT or Copilot.
That's the gap this article is about. Danish SMEs aren't behind when it comes to trying AI; they're well ahead. But there's a funnel from "finds it relevant" (60%) through "has tried it" (44%) to "has it in operation" (10%), and each step down the funnel has its own barriers. The same report shows that 84% of the SMEs that see a large or very large AI potential in at least one business area also experience barriers to using it. The potential and the obstacles, in other words, live in the same businesses.
The potential question was put to the 2,464 SMEs in the sample that found AI relevant to their business. Where did this group see the greatest potential? According to the survey, sales and marketing top the list, with 38% judging that AI has a large or very large potential, followed by product and business development (29%), administration and finance (26%), customer service and support (23%) and production and operations (22%). That's worth noting for any SME manager: the areas where your peers see the most value are the customer-facing and administrative processes, not exotic specialist uses.
Barrier 1: Skills and time
In the defined barrier survey, lack of knowledge, skills and experience came up most among the answer options.
What the figures show
Across five business areas in the Business Promotion Board's autumn 2025 survey, "lacks relevant knowledge, skills and experience" was the most widespread barrier: 40-48%, with administration and finance at the top (48%). The figures apply to the SMEs that saw a large or very large AI potential in the business area in question; the barrier analysis covers 1,372 SMEs in total. In the same delimitation, 22-30% cited costs as a barrier.
The question about support offers was put to the 1,146 SMEs that had stated at least one barrier. They could choose up to three offers. When asked in the same survey which offers would help them move on, 58% answered advice on the use of artificial intelligence, 33% upskilling of employees and 29% upskilling of managers and the board. What's most in demand isn't subsidies or technology; it's someone who can translate AI into their business.
Why time is the hidden half of the barrier
The skills barrier has a twin that rarely appears in the statistics but that every SME manager knows: time. In a business with 15 employees there's no digitalisation department. The person who was supposed to "look at AI" is the same person who closes the accounts, handles the customer complaint and covers for a colleague on sick leave. Skills can be bought or learned, but the hours to try out, evaluate and embed a new tool have to be taken from operations.
That has an important consequence for how an SME should approach AI: solutions that require months of project work before the first gain lose out to day-to-day demands. Solutions that deliver a noticeable time saving within the first few weeks on one specific process win, because they fund the next hour you invest. So don't start with an "AI strategy for the whole business". Start with the process that annoys the most employees the most.
Barrier 2: Data scattered across systems that don't talk to each other
The second big barrier is one most people only discover once they're under way: AI is only as good as the data it can reach.
You probably recognise the symptoms
Customer data may sit in one system, orders in another, and agreements in emails or spreadsheets. In the Business Promotion Board's autumn 2025 survey, 16-24% cited problems with the availability or quality of necessary data as a barrier. The figures apply to SMEs that saw a large or very large AI potential in the business area in question. The share was highest in product and business development and in production and operations (both 24%). Among the same area-specific groups, up to 14% stated that existing equipment, software or IT systems didn't support the use.
The paradox is that many Danish SMEs actually have good data. The report itself highlights that the high degree of digitalisation means that many SMEs already hold structured business data, order, stock and sales figures, and customer data from CRM systems. The problem is rarely that data doesn't exist. The problem is that it lives in silos.
Why AI amplifies the data problem
An employee can compensate for messy data: you just ring Lars and ask what was actually agreed. An AI assistant can't. Put AI on top of scattered and inconsistent data and you get scattered and inconsistent answers, only faster. That's why many SMEs find that their first serious attempt at AI runs aground on a data clean-up project they hadn't counted on.
It isn't a reason to give up; it's a reason to reverse the order. Clean up where the AI is going to work first, not across the whole business. If the first use is sales support, it's the CRM data that needs to be accurate: duplicates merged, standard fields filled in, dead leads archived. We've written a practical walkthrough of exactly that in our guide to CRM data hygiene, and if the CRM concepts are new to you, the introduction to CRM automation is a good place to start. The point is the same in both: data quality isn't a prerequisite for all AI, only for the process you start with. That turns the mountain into a hill.
Barrier 3: Uncertainty about GDPR and compliance
The third barrier is less about what the rules say and more about the fact that nobody in the business is sure what they say.
What the uncertainty is actually about
In the Business Promotion Board's autumn 2025 survey, 11-28% cited worry about breaking the law as a barrier, highest in administration and finance (28%). The figures apply to SMEs that saw a large or very large AI potential in the business area in question. In these area-specific groups, 14-23% cited concerns about digital security or dependence. In the survey's question on support offers, 23% pointed to advice on legal matters and data protection rules.
Three practical supplier questions can help the clarification, but they don't replace an assessment of the legal basis for processing, the purpose, the data needed, informing the data subjects and the data subjects' rights:
- Where is data processed? Clarify the supplier's role, a data processing agreement if the supplier processes on your behalf, and any basis for transfers outside the EU/EEA.
- Is training done on our data? Does the supplier use your input to train models that serve other customers, and can that be switched off?
- Who can see what? Does every employee suddenly have access to summaries of data they weren't allowed to see in the source system?
None of the three questions requires a law-trained AI expert to ask. They require someone to have been given the job of asking them.
The EU AI Act on top, but no need to panic
The EU AI Act can impose further requirements, which depend on the system, the use and the business's role. Text drafting, meeting summaries and customer service must therefore be assessed specifically; they don't automatically carry the same obligations. We've put together a walkthrough aimed at Danish businesses in our guide to the EU AI Act for sales organisations.
Start the compliance clarification with an overview of which AI tools you use, for which purposes, with which data and with which suppliers. The overview makes it possible to examine the relevant requirements and any risks. Internal guidelines and company accounts can support governance, but do not in themselves make processing lawful.
Barrier 4: An unclear business case
"We can see it's clever, but we can't work out what it's worth." That's the fourth barrier, and it's more rational than it's often made out to be.
Why the gain is hard to calculate
In the Business Promotion Board's autumn 2025 survey, 22-30% cited implementation costs as a barrier. That applies to SMEs that saw a large or very large AI potential in the business area in question. The report suggests as one possible explanation that the cost concern may also reflect uncertainty about the gains. So examine the licence price, employee time and the results you want to measure.
One possible challenge is that small time savings are spread across several employees and don't automatically turn into a financial gain. Measure your current time spent and quality before you assess a change. Also look into what any freed-up time is actually used for.
From "exciting" to measurable
The solution isn't a bigger spreadsheet; it's a smaller claim. Instead of building a business case for "AI in the business", you build one process by process: Today we spend an estimated N hours a week writing quotes / answering standard enquiries / transferring data between systems. We'll try AI on that process for four weeks and measure what the figure is then. A test that narrow costs little, reveals a lot and gives a figure that management can relate to, whether it argues for or against continuing.
The processes best suited to this exercise are the repetitive and measurable ones: sales follow-up, quote writing, meeting preparation, reporting. Our complete guide to automating sales processes goes through how to select and delimit that kind of process, and for businesses that want to put measurement on a systematic footing across sales, marketing and customer service, the revenue operations approach is a good framework. What both have in common: you decide before the experiment what success means in numbers. Then the business case becomes a result, not a belief.
Barrier 5: Dependence on individuals and enthusiasts
The fifth barrier is the most overlooked, perhaps because at first it looks like a strength.
The enthusiast pattern
Dependence on individuals can arise if one employee is responsible for prompts, tool choices and helping colleagues. Look into whether knowledge, access and workflows can be taken over by others in your business. The survey doesn't establish how widespread this specific situation is.
Three out of four AI-using SMEs had, according to the report, only experimented with common tools such as ChatGPT or Copilot. That figure doesn't in itself show whether knowledge and accounts sit with individual employees. So look into who has access in your own business, and whether prompts and workflows can be taken over by others.
The risk of person-borne AI
If knowledge and access sit with only one person, absence or a change of job can make it hard to continue the work. Documentation and shared access as needed can reduce this dependence; look into which parts are actually missing in your business.
Dependence on individuals isn't specific to AI: every SME knows it from the bookkeeper who's the only one who understands the accounting system. But the AI version is more insidious, because it arises without a decision: nobody chose for Mette from marketing to be in charge of AI. It just happened. The countermove is to make the informal formal while the enthusiast is still there: write the working prompts and workflows down in a shared place, move AI use from private accounts to company accounts, and make "this is how we use AI here" an item in onboarding for new employees. The enthusiast shouldn't be held back: they should be copied while that's still possible.
Ways of working you can try
The Business Promotion Board's survey describes uptake and barriers; it doesn't document that particular ways of working have produced the results. The following are suggestions for ways of working that you can try out and measure in your own business:
- Start in the process, not in the technology. The question isn't "what can AI do?" but "where do we waste the most time?". The choice of tool comes last, not first.
- Choose a few uses and finish them. Delimit a trial, follow it through, write down the workflow and train the colleagues who are going to use it.
- Involve management. Agree responsibility and resources for the trial, and look into the need for training. In the survey, 29% asked for upskilling of managers and the board; the figure doesn't in itself document the effect of management's involvement.
- Treat data as part of the project. Look into the data quality in the system the AI is going to work in, and limit the necessary clean-up to the chosen task.
- Make rules early and keep them simple. Describe briefly what employees may do, may not do and must ask about. Test whether the rules can be understood and used in the specific workflow.
- Measure before and after. Compare time spent, quality and results, and look into other changes in the period. A before/after difference shows a change, but doesn't on its own prove that AI was the cause.
The suggestions require you to set aside relevant resources and follow the results. The difference between AI trials and systematic implementation in the survey doesn't in itself document which measures work in your business.
How to lower the barriers one at a time
The barriers shouldn't be solved all at once; they should be lowered in sequence. The table here gives the overview, and the text below it gives the order.
The five AI barriers in Danish SMEs: symptom and first step
| Barrier | Typical symptom | First step |
|---|---|---|
| Skills and time | "We don't know where to start", AI work is put off for operations | Pick one process and one person responsible; set aside 2-3 hours a week for four weeks |
| Scattered data | AI answers are imprecise because the truth lives in three systems and an inbox | Clean up the one system that the first use will need |
| GDPR and compliance | Nobody dares use AI on customer data, or everybody does so uncontrollably on private accounts | Make a one-page overview: which tools, which data, which suppliers |
| Unclear business case | "It seems clever, but what is it worth?" Decisions get put off | Measure one process's current time spent; run a four-week limited test and measure again |
| Dependence on individuals | All AI knowledge sits with one enthusiast with private prompts | Write working prompts and workflows down in a shared place; move to company accounts |
A sensible order for most SMEs looks like this:
- Week 1: Choose a process and a person responsible. One repeated, time-consuming, measurable process: preferably within sales, administration or customer service, where SMEs themselves see the greatest potential. Measure what it costs in hours today.
- Weeks 1-2: Clarify the rules for that specific process. Look into personal data, purpose, legal basis for processing, supplier roles, relevant agreements and any transfers. Write down the open questions and get them clarified before the test.
- Weeks 2-3: Get the data ready for that specific process. Merge duplicates, fill in key fields, archive the dead records, only in the system the process uses.
- Weeks 3-6: Run the test and measure. Same process, same measure, now with AI support. Note time spent and quality.
- Week 6: Decide on the figures. If it worked: write the workflow down, train your colleagues and choose the next process. If it didn't: you've bought clarity cheaply.
A word on choice of tools, because this is where barriers 2, 3 and 5 meet: if you choose a patchwork of individual tools on private accounts, you recreate the silos, the compliance uncertainty and the dependence on individuals in a new form. We can build the solution in your own systems. Whatever you choose, choose something where the data, access and knowledge belong to the business, not to the individual employee's browser.
SME checklist: are you ready to go from experiment to operation?
Use the list to find open tasks in your own business. The number of ticks isn't a measured benchmark for where you stand relative to others.
- [ ] We've designated one specific process as our first (or next) AI use
- [ ] One named person is responsible and has been given hours for it
- [ ] We know what the process costs in hours today (measured, not guessed)
- [ ] We've defined in advance what success means in numbers
- [ ] Data in the system the process uses has been cleaned up (duplicates, key fields, dead records)
- [ ] We have a one-page overview of AI tools in use: purpose, data, supplier
- [ ] We know whether our suppliers train on our data, and where data is processed
- [ ] Employees know our "may / may not / ask first" rules for AI
- [ ] AI use takes place on company accounts, not private accounts
- [ ] Working prompts and workflows are written down in a shared place
- [ ] At least one person from management uses the tools themselves day to day
- [ ] We have a plan for what happens if our most AI-savvy employee leaves
If many ticks are missing in one particular cluster, you've also found your biggest barrier, and with it your starting point.
Frequently asked questions
Q: How many Danish SMEs actually use AI today?
In the Danish Business Promotion Board's Business Panel from autumn 2025 (4,128 SMEs surveyed), 44% had worked with artificial intelligence, 10% had implemented it systematically, and 2% had developed their own products or tools in use. Statistics Denmark put AI use among private, non-financial urban industries (byerhverv) with at least 10 full-time employees at 28% in 2024, against 15% in 2023. The surveys have different populations and dates, and aren't a measurement of the situation in October 2026.
Q: What is the biggest barrier to AI in Danish SMEs?
In the Business Promotion Board's autumn 2025 survey, lack of knowledge, skills and experience was the most widespread of the barriers examined: 40-48% across five business areas. The figures apply to SMEs that saw a large or very large AI potential in the business area in question. In these groups, 22-30% cited costs and 11-28% concerns about the law. Among the 1,146 SMEs that had stated at least one barrier and were asked about support offers, 58% asked for advice on AI use. The figures don't rule out technical barriers in individual businesses.
Q: Do we need to have all our data under control before we can use AI?
No: only the data your first use will need. It's a common misconception that AI requires a total data clean-up project first. In practice, it's enough to clean up the one system the first process works in: merging duplicates, filling in key fields and archiving dead records. Many Danish SMEs already have structured data in CRM, order and accounting systems; the challenge is the silos between them. Limit the clean-up to the first use, and expand as AI use grows.
Q: Is GDPR a real obstacle to AI in a small business?
AI can be used within GDPR, but it requires a specific assessment of purpose, legal basis for processing, necessary data, information and the data subjects' rights. Also look into the supplier's role, access, model training and any third-country transfers. A data processing agreement is relevant when the supplier processes personal data on your behalf. The three practical supplier questions above are a start on the clarification, not a guarantee of lawfulness.
Q: How do we make a business case for AI when we can't predict the gain?
By making the claim smaller instead of the spreadsheet bigger. Measure what one specific process costs in hours today, run a limited test with AI support for four weeks, and measure again. Then you have an actual figure instead of a guess, and a decision that can be made on evidence, whether the test argues for or against. Repetitive, measurable processes such as quote writing, sales follow-up and standard enquiries are best suited to the first test.
Q: Our AI use depends on one skilled employee: is that a problem?
It's a strength with an expiry date. The enthusiast is often the reason the business got started at all, but if prompts, tool knowledge and workflows exist only in one person's head and private accounts, it all disappears the day that person leaves or gets busy with something else. The countermove is to formalise while the enthusiast is still there: write down working prompts and workflows in a shared place, move use to company accounts, and make AI workflows part of onboarding. Copy the enthusiast; don't hold them back.
#AI in SMEs #AI barriers #artificial intelligence #digitalisation #AI strategy
Contents
What takes up your time?
Write down the task that takes up the most time. We reply whether it is worth building, and we say so if it is not.
Write what takes up your timeRelated articles

Own or rent software? A decision guide for Danish businesses
Own or rent software? Cost over time, data lock-in, GDPR and exit, self-host and fair source, plus a concrete decision framework for Danish businesses.

Claude Code in a business: what an AI developer agent can do, and what it requires
What Claude Code, Anthropic's agentic developer tool, can do in a business, and what it requires in security, governance and organisation.

AI strategy for businesses: a step-by-step guide
A practical guide to an AI strategy: map workflows, choose use cases, fix your data, run a pilot, bring staff along and know when not to automate.