When everyone is talking, shouting and selling, the hardest question is: Where does it concretely begin for us? Six answers - beyond the hype.
The most common question we hear in 2026 is not "What can AI do?", but "Where should I begin?"
The two are connected. But the second question is the honest one. It means: I have understood that it is relevant. I just don't know where the right place is for me. This page is a guide for exactly this moment.
The most common mistake in the mid-market: "We have to do something with AI now." Starting there is like going to the hardware store with the brief "as long as it's cordless". You then buy some drill and afterwards wonder what you actually wanted to drill.
The right order is the other way around: where does work pile up? Which task repeats itself every week? What annoys your best people the most? That is your first AI spot - regardless of whether the solution ends up being called AI or not. AI is a tool. You choose a tool according to the task, not the other way around.
If you don't want to call a consultant, first answer these three questions. A sheet of paper is enough.
One: Which three tasks cost the most hours per week in your business? Write them down, with a rough number of hours.
Two: Which of these tasks consists more than half of reading, sorting, summarising, transferring or replying? Those are your AI candidates.
Three: Where would an improvement in response time be worth hard money for your customers? Enquiries that today sit for three days, because no one gets to them. That is your leverage point.
In 2026, AI is so good in many areas that a mid-sized business can use it productively. In some, though, not yet. The honest breakdown:
Very strong today: Summarising incoming texts (emails, enquiries, contracts). Writing draft replies. Turning unstructured data into tables. First-pass translations. Code help for small scripts. Image recognition in clearly defined areas.
Usable, with supervision: Transcribing and summarising phone calls. Classifying customer enquiries. First proposals for quotes or invoices from orders. Search in your own documents.
Not yet reliable: Autonomous decisions with legal consequences. Complex accounting logic. Advice on sensitive topics without human control. Fully unsupervised customer communication.
Not every business is at the same level. An honest self-check helps to find the right first step.
Level one - Curious. ChatGPT is installed, individual employees play around privately. In the business nothing systematic happens yet. First step: one person who is officially allowed to experiment. Time-limited, with a report to the team.
Level two - Sporadic. Individuals use AI for their tasks - but everyone differently, without coordination. First step: a shared rule on which data may go in, which may not. Plus one tool that everyone uses.
Level three - Integrated. AI is firmly built into a concrete process, with accountability. First step: identify the second process. The first one worked - the second is usually easier.
Level four - Systematic. AI is part of the strategy. There are your own models or bespoke applications. Here we are often at your side, when the standard solutions are not enough.
This will surprise many who have read this page with the plan to quickly buy AI. The honest truth: before AI can really help, your data has to be graspable somehow.
If your customer information sits in four different Excel lists, your enquiries in the mailbox of a single employee and your orders on paper - then the AI is not the problem, but your data flow. First the tools, then the AI. That is not only our principle, but almost always also the reality: three weeks of ordering data save three months of AI frustration.
Most AI consultancies sell you the AI layer on top - because it is expensive and sounds like the future. But it only works when cleanly written data lies beneath it. Half of the failed AI projects in the mid-market fail not because of the AI, but because of this layer below it.
Anyone who sells you AI without having seen your data builds on sand. The invisible work is the expensive one. But it is also the one without which nothing holds up.
With every client who comes to us with the wish "we want AI", the first meeting looks the same. We talk for two hours, without speaking of AI a single time.
We talk about your day. About your order situations. About the three tasks you put off on Friday afternoon because you have no energy left. About the employee who wants to leave, because he has been doing the same routine for two years. Once we have heard that, we know where AI helps. Before that, every recommendation is a guess - and guessed recommendations always cost more in the end than they bring.
Where you begin with AI is not a technology decision. It is a question you have to put to yourself and your people.
Most of our clients don't need the biggest model, but the smartest first step. And that is almost always smaller than you think - but it holds up. A well-placed small automation saves more over months than a large AI project that never really gets going.
Why we never begin with AI can be found under First the tools, then the AI. What AI can and cannot do today can be found under What AI makes possible and What AI cannot do. How we clarify the problem before every build can be found under Before we build, we work out what is broken.