##  [AI strategy and AI architecture](/services/consulting/ai-strategy-and-architecture)

This page is about two things we help businesses with when they want to start using AI. The first is deciding what AI should be used for. We call that strategy. The second is deciding how the solution should be built so it can be used day to day. We call that architecture. We take the two together because they depend on each other: what you decide determines what has to be built, and what can be built determines what makes sense to decide.

By AI we mainly mean language models such as ChatGPT, Copilot and Claude. They are programs that can read and write text, answer questions and summarise documents. Many businesses have tried them in a small pilot, but most of those pilots never become something that is used for real. A 2025 study from MIT found that 95% of companies' pilots had no measurable effect on the bottom line. The cause was rarely the technology. More often it was unclear what the pilot was meant to solve, or the solution could not work with the company's own systems and data.

Thinking about what AI could do for you? [**Write to us, and we'll take a 20-minute conversation →**](/contact)

## What is an AI strategy?

An AI strategy is a written plan for what your business will use AI for, in what order, and how you can tell whether it works. It has to be something management can read and defend, not just IT.

We start by looking at where you spend the most time and money. That might be tasks repeated every day, questions only the same few employees can answer, or documents someone has to read through again and again. From there we usually find three to five places where AI can make a noticeable difference, and we assess each one on how much it will deliver and how hard it is to carry out.

Then we look at the rules. GDPR governs how you may handle information about people, and the EU AI Act sets requirements for the use of AI, especially when it affects decisions about people. It is cheaper to deal with that from the start than to fix a finished solution afterwards.

Finally, we lay out a plan for the next 6 to 12 months. It describes what should happen first, who in the business needs to learn what, which rules employees must follow when they use AI, and which numbers you should watch to see whether it pays off.

Sometimes the review shows that AI is not the right answer yet, or that it will cost more than it delivers. Then we say so.

## What is AI architecture?

Architecture is the description of how a solution is put together: which parts it consists of, where the data lives, and how the parts talk to each other. In an AI solution, the language model is only a small part. The rest is ordinary software that decides what information the model gets to see, checks what it answers, controls who may see what, and keeps a log of what has happened. Most of what goes wrong in AI projects goes wrong in that part.

Architecture is also about making your business logic and the artificial intelligence fit together. Business logic is the rules your company works by: how a price is calculated, when an order needs approval, and which customers get which terms. Those rules are fixed and must give the same result every time. A language model works differently. It works out the most likely answer and can answer differently from one time to the next. Good architecture makes sure your rules always have the final say, and that the AI is used for what the rules cannot handle alone, such as understanding an email from a customer or finding the right clause in a contract. That way the AI cannot grant a discount or promise a delivery date your rules do not allow.

We help you with the choices that follow from this. One of them is how much the AI should do. Every answer from a language model costs money, so we let it do only what requires a language model and let ordinary code handle the rest. That makes the solution more predictable and cheaper to run.

Another choice is where your data should go. The language models from the big American companies run on their servers. There are also European models, and models that can run on your own machines. What fits depends on how sensitive your information is and what your customers are entitled to.

A third is whether you can see afterwards why the solution answered the way it did. That is a requirement in several industries, and it is what makes it possible for management to stand behind the solution.

Three concepts come up in most of the solutions we work with.

RAG means the solution first finds the relevant documents in your own files, such as handbooks, contracts or previous quotes, and gives them to the language model along with the question. It then answers from your knowledge rather than from what it learned on the internet, and it can show where the answer comes from.

An AI agent is a solution where the language model does not just write answers but also carries out actions itself, such as sending an email, updating a system or ordering an item. That requires firm limits on what the agent may do without asking a human first.

Prompt injection is an attack where someone hides instructions in text the AI reads, for example in an email or on a website, so that it does something other than you intended. This has to be accounted for from the moment the solution is built, especially if it can act on its own.

## Why both from the same adviser

A plan made without knowledge of the technology often promises something that cannot be built for the price. A solution built without a plan often solves a problem nobody in management has prioritised. We build and run AI solutions ourselves, and we sell neither software nor licences. So we have no interest in which product you end up choosing.

## How to get started

- [AI Starter Kit](/packages/ai-starter-kit), €10,000. A workshop and a review of your business that ends with a prioritised list of concrete AI projects. For those who do not yet know whether AI can be used in their business.
- [Consulting Package](/packages/consulting-package), €12,000. An analysis and a workshop with management that ends with a finished strategy and plan. For those who have decided to use AI but lack the plan.
- [AI Implementation package](/packages/ai-implementation-package), from €50,000. We design and build the solution on your own data, typically a RAG solution over the company's documents.
- [Ongoing AI support](/packages/ai-support-retainer), from €2,000 per month. A fixed number of hours each month that you can use for questions, assessing vendor proposals and help while the plan is carried out.

All prices exclude VAT.

## Who it is for

Owners and managing directors of small and medium-sized businesses with up to 250 employees that do not have their own AI specialists. Boards that need to take a position on AI without being able to assess the technology themselves. For them we also offer the course [AI: Essential for the board](/da/node/201) (in Danish). And IT managers who want an independent assessment of a solution they have been offered.

## Contact us

Tell us briefly what you are thinking of using AI for. We'll take a conversation of about 20 minutes and give you an honest assessment of whether it is realistic and what it will take.

[**Write to us**](/contact)