Are we ready for AI?That is the wrong question

The readiness trap

We often hear the same sentence: we first need to get our house in order before we start with AI. The data has to be cleaned, the old system replaced, everyone needs a course. It sounds responsible. In practice it often means nothing happens for a year, while competitors learn by doing.

Readiness assessments and maturity scans can be useful, but they easily turn into a project of their own. You end up with a report that says you are not ready yet, and a list of things to fix before you are allowed to begin.

Why it is the wrong question

Nobody is ready for AI in general, because AI in general does not exist in your company. What exists is the way orders come in, how a quote gets made, how a complaint gets followed up. Each of those processes has its own data, its own exceptions and its own people. One of them might be ready tomorrow, another one in two years.

Asking whether the company is ready is like asking whether a house is ready for renovation. The answer depends on which room you start with.

The better question

Ask instead: which process costs our team the most time, and what would it take to make that one run better? That question has a concrete answer. You can look at how the work flows today, where it gets stuck and which data that process really needs.

Often the answer is smaller than expected. To let AI read incoming orders, you need your article list and your customer list to be usable. You do not need to clean up ten years of archive first.

Data: good enough for one flow

Data quality matters, but only for the data the flow touches. If a client exists three times under different names, someone has to decide which one is right, and the system can help by proposing merges. That is work you do while building, not a precondition before you may start.

The same goes for old software. Many companies run on systems that are fifteen years old. Sometimes they can be connected, sometimes a part needs to be rebuilt. That becomes clear when you map one flow, not when you assess the whole company.

People learn by using it

Training helps, and the AI Act asks companies to take measures for the AI literacy of their people. But the best training is working with a system that helps you in your own work, with clear rules on what AI may do and where a person decides. People who see AI take over the work they dislike become the strongest supporters of the next step.

What the numbers say

Large studies keep finding the same thing: AI projects rarely fail because of the technology. In 2025 MIT's NANDA initiative reported that about 95% of the generative AI pilots it studied had no measurable effect on profit and loss. The successful ones were narrow, built into the actual work and measured on results. That is not an argument to wait. It is an argument to start small and concrete.

Signs you are waiting too long

The same complaints come back in every management meeting, but the plan is still to fix the data first. A maturity scan was done a year ago and nothing has been built since. People already use free AI tools on their own, without guidelines, because the company has not offered anything better. Each of these means the cost of waiting is higher than the risk of starting.

None of this means you should rush. It means the first step should be small enough to start next month, and concrete enough that you can measure whether it worked.

What a good first project looks like

A good first project has a clear owner, happens often, costs a measurable amount of time and has a limited risk per mistake. Reading incoming orders, linking mails to the right file or following up quotes that stay unanswered are typical examples. Writing a company wide AI strategy is not.

Agree upfront how you will measure success: hours saved, mails handled without a person, mistakes avoided. After a few months you have real numbers, a team that knows what AI can and cannot do, and a foundation of connections and rules that the next process can reuse.

How to start instead

Pick one process. Map how it really runs, with the people who do the work. Decide per step what it needs: AI, plain software or nothing at all. Build the first step, measure it, and use what you learn for the next one. That way readiness is not something you wait for, but something you build, one flow at a time.