AI as a strength: using it is not the same as building on it

The advice sounds the same in most boardrooms in the autumn of 2026: we need to use more AI. It is the wrong goal. Using AI is not a strength. A strength is something your competitors cannot simply buy tomorrow, and a ChatGPT subscription for the whole department is exactly what everyone can buy tomorrow.

The numbers behind that claim are in our own whitepaper, AI in Danish business 2026: eight in ten Danish companies use AI, but only one in ten does it systematically. The seven in between have the tools. They do not have the strength. And the difference between the two groups is not how much technology they bought.

What does it mean for AI to be a strength?

AI is a strength in a company when it sits inside a workflow with an owner, a measured starting point, and a result that leadership can put a number on.

Everything else is use. Use is fine. Use is where everyone starts, and we never advise anyone to ban it. But use does not move from the individual employee to the company by itself. The salesperson who writes proposals with AI has become faster. The company has not become better at selling. It only does when the proposal process changes, when the template is shared, when the pricing logic is built in, and when the next salesperson inherits all of it on day one.

We see the same pattern in most of the reviews we run. Someone in the company is far ahead. Usually one or two people who figured it out on their own. The rest do what they always did. And leadership believes the company uses AI, because they have seen the invoice.

Three questions that reveal the difference

We ask three questions when we need to find out whether AI is use or strength somewhere. They can be asked at a management meeting in ten minutes.

• Who owns it? If the answer is a person and not a role, it is use. When the person resigns, it leaves with her.
• What was the starting point? If nobody measured how long the task took before AI came in, nobody can prove anything got better. Then it is a feeling, and feelings do not survive a budget meeting.
• What is the number? Not "we save time". A number. Hours per week, days of average response time, errors per hundred orders, cost per case.

Most companies can answer the first. Fewer the second. Almost none the third, and not because they are badly run. It is because nobody told them to measure before they began.

A wholesaler with 80 employees

An example, assembled from several engagements so that nobody recognises themselves.

A wholesaler in the Triangle Region of Denmark with 80 employees receives around 300 orders a day. A third arrive as email, in free text, with product numbers typed wrong and quantities in the wrong unit. Three employees spend their day keying them into the finance system. Two of them have, for about a year, used an AI chat to clean up the worst emails before they type. They are good at it. They are also the only ones who can.

Question one: who owns it? The two employees. Question two: the starting point? Nobody had measured anything. Question three: the number? Nobody had a clue.

So we measured. Two weeks with a stopwatch and a simple tally: 14 minutes per email order on average, and 6 percent of them came back from the warehouse with an error. That is the starting point. Only then was something built. A small solution that reads the order email, looks up the product numbers in the system, proposes a finished order line and places it with an employee for approval. Not automatic posting. Approval.

After eight weeks: 4 minutes per order, and the error rate down to 2 percent. The number leadership can say out loud is roughly 1,100 hours a year and a warehouse manager who no longer calls sales every morning. The three employees still key in orders. They just key in the ones that need a human, and they have taken over the purchasing follow up that used to sit with an external consultant.

That is the difference. Before, AI was something two capable people did. Now it is something the company does, and it would still be there tomorrow even if both of them resigned today.

Why the order of things decides it

The mistake we see most often is that the order is reversed. First the technology is bought, then a task is found for it, and at the end somebody asks what the employees think. It is the reliable route to a pilot that never reaches operations.

We work across three battlegrounds, and the order is not accidental. The business first: where in operations do time and errors pile up, and what is it worth to solve? The technology second: what is the smallest solution that solves exactly that, and can it live inside the systems you already have? And people all the way through, not at the end: who will work differently, what do they need to learn, and what do they get out of it themselves?

The wholesaler did not get a 40 page AI strategy. It got one place in operations where AI became a strength, and it got a measurement method it can use again at the next place. The next place was credit assessment of new customers. It took four weeks, because the method was already there.

An AI strategy that cannot point to a specific workflow with a measured starting point is a shopping list.

How to get from use to strength

If you want to know where your own company stands, there is a short route. Write the list of places where you use AI today. Ask the three questions of each item. Anything that fails question two is a candidate: that is where measuring has to happen before building does.

That is exactly what our method is built for. Target is an AI Potential Review, 30 minutes, free and online, where we ask about your operations and not about technology. You get a prioritised list of the places where AI can become a strength for you, and an honest view on whether to start now. The plan that follows puts a number on every item before money is spent. The pilot measures the starting point first. And the embedding makes sure it still works once the first enthusiasm has settled.

Eight in ten use AI. That is no longer interesting. What is interesting is whether you can answer the three questions, and in how many places in your operations the answer is yes.