You may know the situation: at association meetings, everyone is talking about AI. A competitor supposedly already has "something running". Your own kids do half their homework with a chatbot. And somewhere in the back of your mind sits the question: shouldn't we have done something by now too? At the same time, there is no concrete starting point. No project, no use case, just a diffuse feeling of pressure to act.
We are writing this article because we encounter exactly this starting position all the time, and because it is no cause for concern. If you don't yet know where AI helps in your own business, you haven't missed anything. It only becomes dangerous when the pressure produces the wrong sequence: first buy a tool, then look for a problem for it.
Don't start with the tool, start with the process
Most of the failed AI initiatives we hear about have one thing in common: they started with the technology. Someone saw a tool, bought a license, was shown an estimate. And then people wondered where to fit the whole thing into the business. That is like buying a machine and only afterwards asking what it is actually supposed to produce.
The better sequence begins with a sober look at your own workflows. Four questions help more here than any product demo:
Where does waiting time occur? Tasks that sit idle because a specific person has to review, summarize, or write something up. Where does duplicate work occur? The same information typed in, reconciled, or reformatted multiple times. Where does search effort occur? Knowledge that exists in the business, in folders, inboxes, heads, but has to be laboriously tracked down every time. Where do transfer errors occur? Points where data is moved by hand from one system to the next, and occasionally something slips.
Anyone who can answer these four questions for their own business already has a list of candidates, without a single AI buzzword.
Three honest questions before any AI initiative
Before a candidate becomes a project, we work through three questions together with our customers. They sound trivial, but they reliably filter out what would later fail expensively.
First: does the process actually run in a defined way today? AI can speed up a workflow, but it cannot invent a workflow that doesn't exist. If three employees do the same task in three different ways and nobody can say which one is correct, then the first project is not an AI project. It is a clarification exercise. That is no disgrace. It is the norm in businesses that have grown over time.
Second: do the data and documents for it exist? A system that is supposed to answer questions about your records needs those records: findable, reasonably current, in readable form. A system that is supposed to draft proposals needs examples of what good proposals look like at your company. Often the material exists, just scattered. Sometimes, though, it only exists in the heads of two long-serving employees. Knowing that before you start saves months.
Third: who would use it in everyday work? Not: who could. But: who would. A tool that nobody builds into their working day is worthless, no matter how good it is technically. If the answer to this question remains vague ("well, anyone could use it now and then"), that is a warning sign. The best first projects have a specific person waiting for the relief.
The most expensive briefing in the world
"We want to do something with AI." This sentence costs more money than any misguided investment in a single tool. Because it shifts the actual work, namely clarifying the problem, onto someone who doesn't know your business. A service provider who receives such a briefing will deliver what demos well, not necessarily what helps you in daily operations. The result is a pilot project that shines in the presentation and then quietly falls asleep.
The flip side is reassuring: you don't need to build AI expertise to be well prepared. You need to be able to describe your own business, and you can do that better than any consultant. The technology side is our part of the work, not yours.
The most important preparation for an AI initiative has nothing to do with AI: it is the honest answer to the question of where in your business time, knowledge, or care is being lost, and who is most affected by it.
Where SMEs typically start
There are categories of tasks that repeatedly prove to be sensible first candidates in small and medium-sized businesses, because they occur frequently, are clearly delineated, and their benefit can be verified quickly:
Recurring texts. Replies to similar inquiries, standard letters, meeting minutes, job postings; everything that is written from scratch every time, even though ninety percent of the content repeats.
Searching your own records. "How did we solve this back then?", "What does the contract say about this?", "Which requirements apply here?" Questions whose answers exist in the business, but finding them requires experience or patience.
Transferring between systems. Data from emails, PDFs, or forms that is entered by hand into an ERP, accounting, or planning system. Monotonous, error-prone, and popular with nobody.
First drafts. The blank screen is often the biggest hurdle. A solid first draft, for a proposal, a report, a piece of documentation, that a human then reviews and finishes saves time without shifting the responsibility.
What is striking is what is missing from this list: the grand transformation promises. No "fully automated sales", no "factory of the future". First projects are small. That is their strength.
Sometimes the right answer is: no AI project
We say this to customers proactively, and we consider it one of the most important sentences in our work: a proven workflow that runs reliably may stay as it is. Technology is not an end in itself. If a process runs fast enough, produces few errors, and torments nobody, then AI creates no meaningful advantage there, only effort, training needs, and one more dependency.
Anyone who wants to sell you an AI solution for every process in your company is not advising you; they are selling to you. A good partner distinguishes between places where technology helps measurably and places where the business works the way it does for good reasons.
What a structured start looks like
If you put the previous sections together, you get an approach that requires no prior knowledge and starts without big budgets:
Review your processes. Go through the four questions from the beginning: waiting time, duplicate work, search effort, transfer errors. And don't just ask the leadership level; ask the people who do the work every day. They know best where things get stuck.
Estimate effort and benefit honestly. How many hours per week go into a task? What happens if an error slips through there? Rough estimates are enough, but they have to be honest, even if the result is: not worth it.
Start small and measurable. One process, one clearly named benefit, one time period after which you evaluate soberly: did it deliver the promised relief? If yes, expand. If no, stop, without having lost much.
Involve employees early. Not just at rollout, but already during selection. Those who had a say in which problem gets tackled will actually use the solution later. And those who experience a tedious task becoming easier will bring the next idea to you on their own.
This kind of review is, incidentally, exactly what we at AI Værk offer as orientation: together with you, in your business, often in the form of a workshop. Not in order to sell you as much as possible afterwards, but to end up with an honest list: worth it here, maybe later here, everything stays as it is here. If that is where you stand, feel free to get in touch.
