When I give talks on artificial intelligence, the question almost always arrives in the same form: it sounds interesting – but what do you actually use it for? The question is fair. The conversation about AI is often conducted in big words and future scenarios, while everyday life in a Danish company is about newsletters that have to go out, minutes that have to be written, and decisions that have to be made on a solid basis.
This article is my answer. I have gathered eleven examples of how Danish companies and organisations use artificial intelligence today. Three of them are cases from my own work as an adviser and teacher – anonymised, but real. The rest are typical patterns I meet across Danish organisations. Each example follows the same structure: the situation, what the AI does, the result, and what can be learned from it. If you need a basic explanation first, I have written a calm walk-through of what AI actually is.
Communications departments are furthest ahead
It is no coincidence that most of my examples begin here. Generative AI works in language, and language is the communications department's raw material – so this is also where the technology first found a settled place in the workflows.
First drafts and quality assurance in public-sector communications departments. In the public-sector communications departments I teach and advise, AI has moved into both ends of the writing process. Texts begin with an AI first draft: press releases, newsletters and other texts are written up from a memo and a fixed prompt describing purpose, audience and the sender's use of language. And before the content goes out, it is quality-assured with AI assistants that hold the text up against the department's tone of voice. The time from memo to draft has fallen from hours to minutes, and every text is still checked and approved by a human. The lesson is that AI is most useful at both ends of the writing process – as the first draft and the final check – while the responsibility stays with the sender.
AI as a critical reader. A typical pattern in membership organisations and other organisations with many writers: many different employees write to the members, and the texts sound like it – some formal, some chummy, none quite alike. Instead of letting the AI write, it is used as a critical reader: every text is held up against the organisation's written tone of voice, and the model points out ambiguities, officialese and unsupported claims before the text is sent. The number of rewrite rounds falls, and the discussions about language become more concrete. The lesson is that AI as a reader is often a safer place to start than AI as a writer – the responsibility stays where it always was.
Translation and communication in several registers. Danish companies with English as their corporate language increasingly use AI to translate internal content and – more importantly – to turn the same content into several levels: the technical version for the specialists, the short version for the intranet, the customer-facing version without jargon. The result is that content which used to exist in only one form now reaches more recipients. The lesson is that the same content in several forms is one of the tasks language models solve best – if someone has defined the forms in advance.
Leadership and HR use AI to listen and prepare
AI training in the leadership of a large NGO. A large Danish NGO I work with chose to begin with AI training in the leadership group – before a strategy for AI in the organisation was laid down. The order is the whole point: the digital imagination of the people who will make the decisions has to be expanded first, so they understand what the choices and trade-offs will come to mean – and can imagine a new version of the organisation with new technology. The lesson is that an AI strategy laid down before the leadership has had its own hands on the technology becomes a strategy for the organisation you were – not the one you could become.
Free-text answers in employee surveys. At many public-sector workplaces, the employee survey's free-text answers end up in the same place every year: in a file no one has time to read systematically. A growing pattern is to let AI gather the many hundreds of anonymised answers into themes, with quotes as evidence, so leadership gets an overview in a day that used to take weeks – or never got done. The decisions about what to act on still sit with leadership. The lesson is that AI can support a decision by making the material readable – but it should not make it.
Job adverts and onboarding material in HR. In many HR teams, job adverts are written under time pressure, and old templates are recycled with last round's dates and clichés. A typical pattern is to let AI draft from the role profile and check the language for bias and insider references before the advert is published – and to use the same approach for welcome material for new colleagues. The adverts get finished faster and become more consistent in quality. The lesson is that the return follows the competencies: things only move when the team is trained to instruct the model properly, not when the licences are bought.
Marketing uses AI for breadth and insight
Product texts and marketing copy in an import company. A medium-sized Danish import company I have worked with uses AI today to generate product texts and marketing copy. But the decisive part happened before the first text was written: a thorough piece of strategic groundwork in which we defined the company's core narrative, the role the brand should play for its customers, a clear tone of voice and criteria for the different text types – because what a product text has to do and what a campaign text has to do are not the same.
Without that groundwork, the result would have been generic: AI texts that could have appeared at any competitor, because the model only has the average to fall back on. With the groundwork, AI instead became an amplifier of a clarified identity – every text draws on the core narrative and is measured against the criteria, and production has become both faster and more precise in its language. The lesson is that the strategic groundwork is not a delay to the AI work but the precondition for it: clarification comes first, amplification afterwards.
Variants of adverts and campaign copy. In many Danish marketing teams the bottleneck is not the idea but the volume: the same campaign has to live on five channels, in three lengths, for several audiences. The AI produces the variants from one carefully worked core message, and the team selects, adjusts and tests. The result is more systematic testing than there used to be time for. The lesson is a division of labour: the machine delivers the breadth, the human makes the choice – and without a clear brief, the variants are just repetitions of each other.
Customer insight from reviews and enquiries. Companies sit on text data they rarely use: reviews, NPS comments, customer service enquiries. The AI reads across the material and gathers it into patterns – what customers praise, where they meet friction, which words they themselves use about the product. That gives marketing and product development a prioritised list instead of a gut feeling. The lesson is that most companies do not lack data about their customers – they lack a way of reading it.
Operations is where the gains are easiest to measure
Knowledge look-ups in your own documents. In many knowledge-based firms, the company's knowledge lies scattered across proposals, memos and folders, each with its own logic – new employees find answers by interrupting the experienced ones. A typical pattern is an AI assistant that answers only from the firm's own documents and points to the source, so the answer can be verified. Time spent searching falls noticeably, and onboarding gets shorter. The lesson is that the gain only comes once the documents are tidied up – AI rewards order and amplifies mess.
Minutes and follow-up on meetings. In many leadership teams, meeting minutes are written late, reluctantly or not at all – and decisions evaporate between meetings. A widespread pattern is to let AI transcribe the meeting and draft a decision log with owners and deadlines, which the chair adjusts and approves before it is shared. Follow-up on decisions becomes measurably better, and no one misses the old minute-taking duty. The lesson is perhaps the most important in the whole article: the best AI uses rarely start with a strategy – they start with an annoyance.
Three patterns across the examples
Eleven examples are eleven different stories, but step back and three patterns recur – and they are worth more than any single example, because they hold when the technology moves on.
The first pattern: the value arises in the workflow, not in the tool. None of the examples is about buying something and waiting for an effect. They are about changing a concrete way of working – a shared prompt, a fixed check before sending, an agreed division of labour between human and machine. The import company's strategic groundwork is the clearest example: the identity was clarified before the machine wrote a word. The same tool without that change delivers almost nothing.
The second pattern: the best cases start with an annoyance. The minutes that never got written. The free-text answers no one read. The documents no one could find. Tasks that are recurring, well defined and irritating are the best place to begin – not the big visionary projects no one can measure.
The third pattern: competencies decide the return. In several of the examples the tools had been available long before they made a difference – the difference came when the people learned to use them. The NGO that trained its leadership group before the strategy was laid down drew the consequence of that pattern from the start. That is why I spend most of my time on AI training and teaching rather than on technology choices, and why my AI advisory is more about workflows and people than about systems.
If you want to put the examples to use tomorrow, start with three questions: which task in your organisation is recurring, well defined and irritating? How will you measure whether the AI actually makes it better – not just faster? And who needs to be able to do more before the tool turns into value? The examples in this article will quickly multiply, and some of them will look different in a year. The patterns behind them have proved far more stable – they are what is worth building on.
