Enej Gradišek has more than 500 workshops on artificial intelligence for companies and teams behind him. When we asked him what the biggest mistakes are, he answered with an image that is hard to forget: people imagine you go to the shop, buy a kilo of AI, tip it out on the table at home and say you are now an AI company.
In the third episode of the podcast we talked with him about what, in practice, separates companies that really adopt AI from those that stay at "we did try it, but it didn’t work".
With AI, a bad process becomes a bigger mess
Garbage in, garbage out. If you have bad data or disorganised processes and throw artificial intelligence at them, you do not get order, you get a bigger mess. AI is built on data, and if the data is scattered, there is nothing to get out of it.
That is why every implementation starts with a review. What the company actually needs, what its data looks like and where its processes stand. The hardest part of that conversation is the question for management: is it all right if we turn this process upside down? Because an existing process is not necessarily optimal just because it exists.
A small company has the advantage here
In small and medium-sized companies this often happens: they come wanting AI, we say let us look at the processes first, and they reply that they have wanted to sort that out for years. In the end they put other things in the business in order so that they can add artificial intelligence. That turns out to be very productive.
In a corporation you are moving a tanker. The ship turns over months, not days. Enej says that is not necessarily bad. Every new model comes with hype that lasts three days, and if 10,000 or 15,000 people start using it at once, without a policy and without oversight, that is chaos. Stability has its value. What he misses are teams inside large organisations that would test the latest tools in a closed sandbox and then roll them out gradually.
The biggest risk is the status quo
This is the most useful part of the conversation for any owner. Companies worry about what will happen to their data. But when you ask whether they use artificial intelligence, the answer is often: not really, except that everyone uses a bit of ChatGPT. And when you ask whether those are business accounts, it turns out everyone has their own.
That is the real risk. Personal accounts can be used to train models, and the data bypasses any oversight. Simply moving from personal accounts to a business team account on a server in Europe makes a huge difference. Yet companies see that move as a bigger risk than leaving things as they are.
Enej adds one more provocation. Most larger companies run on Microsoft. An email you send through Outlook goes to a server and back. A language model within the same ecosystem processes the data in the same place. So the question is not whether data leaves the company, but whether you know where it goes, who processes it and how it is encrypted in transit and at rest.
If I do the work in one hour instead of four, do I go home three hours early?
Enej asked this question, and it is a fair one. If a task used to take four hours and now takes one, what happens to the other three?
The answer from our own practice: we ask this question in every diagnostic review, directly to employees. The reality is that people are overloaded. They are not working at 100%, they are working at 120%. So the first win is relief, not extra work. Only after that comes the work there is never time for but that would move the company forward.
From owners we hear two things, and the first one often surprises. The first is that they want to take load off the team, because they do not want people at breaking point. The second is that they want the company to grow without having to hire on a large scale.
A prototype in twenty minutes, production in six months
A mistake that keeps repeating: a lot of hype around a prototype, and then the thing falls apart in real production. Today you can build a prototype in twenty minutes and it looks great in a screenshot. But making it actually work can take six months of demanding work.
The most obvious example is voice assistants. Anyone can build a voicebot for a demo. Real customer support that speaks good Slovenian, knows the numbers and pauses when it should is a completely different story.
Enej has a good comparison for demos. A demo is like parking a Ferrari in front of someone’s house and asking whether it is not a beautiful car. Everyone says wow, but nobody knows how to drive it. That is why, for him, a workshop works instead of a demo: you take people through their own process.
Adoption stands or falls with people
Leadership is a precondition, but on its own it is not enough. You need both directions, top-down and bottom-up. If only management is pushing, the team feels that someone has once again come up with something and dumped it on them.
This is classic change management and has been known since the 1960s. If you are part of the change and someone listens to you, you go through the process more easily, because you have a stake in it. It is quite different if you get a sheet of paper telling you that from tomorrow things will be different.
We came to the same conclusion, just from the other side. In an audit we always ask the end users what they want and what would actually help them. When you come to an employee with a question about how to make their working day easier, and not with a ready-made solution, their defences drop.
What to take away
Enej warns that those of us in this field in Slovenia live in a bubble. To us it all seems trivial, but in reality a large share of people still cannot write a prompt, and according to the latest survey Slovenia is below the European average. His advice for everyone working in this space is to play the long game. Slovenia is small, everyone knows each other, and whoever does not deliver gets pushed out by the market.
The goal of a workshop must not be a conversation about AI. If nothing is different on Monday morning, nothing has happened.