Most Management / Executive teams have long been aware that AI is relevant. That is not the problem. The problem is the next step – moving from ‘this is important’ to ‘we’ve got this sorted’. And this step is not a technological issue. It is a leadership issue.
The blind spot lies not in the content, but in the organisation
Management / Executive and HR rarely underestimate the substance of AI. Word has got round that it is relevant. What is often missing is the transition from general understanding to robust internal practice – and this fails not because of a lack of knowledge, but because of unresolved organisational issues:
- Who sets the standards?
- How are employees trained?
- How are changes communicated?
- How is knowledge kept up to date?
- How can we track who has already completed what?
No tool can answer any of these questions. They concern responsibility, process and communication. A company may have the best AI tool and the smartest guidelines – but if no one is in charge, no one provides training and no one knows who is at what stage, it remains a patchwork affair.
This is precisely where a technological issue becomes a leadership task.
A first, clear version is better than the perfect solution
The good news for any manager who senses the effort involved here: there’s no need for a fundamental debate about AI. What’s needed is a first, clear version. Not perfect – but clear.
A good start often consists of just a few, but solid, building blocks:
- Internal guidelines – what applies in our organisation, in simple terms.
- Clear communication – explained in such a way that the message gets through, not just filed away.
- Practical training – based on real-life situations, not on legal clauses.
- Consistent standards – so that not every department does things differently.
- Transparent documentation – so that what has happened can be verified later.
The mistake is waiting for the perfect solution. Because whilst the ideal AI strategy is maturing in workshops and coordination loops, staff have long since carried on working as best they can – each in their own way. Every month without a framework is a month in which haphazard habits become entrenched. A perfect solution in twelve months is ultimately worse than a solid one in four weeks
It’s a bit like house rules: you don’t wait until you have a legally watertight version running to thirty pages. You put up the clear ground rules – and refine them over time.
Structure beats perfection – especially at the start.
Guidance works better than bans
A common mistake in AI communication: too many bans, too little guidance. When companies focus solely on setting boundaries when it comes to AI, good practice rarely emerges. Instead, uncertainty tends to arise.
People work better with clear dos and don’ts than with abstract warnings. What employees need are answers to specific questions: What’s okay? What’s sensitive? What internal procedures are in place? And what should I do if in doubt? A list of prohibitions alone answers none of these – it merely prompts the question: “What am I actually allowed to do then?”
This is the point that is often overlooked: A ban without a clearly indicated safe path either paralyses people or is circumvented. Anyone who only hears ‘Caution: dangerous’ will either do nothing at all – thereby squandering potential productivity gains – or carry on secretly, beyond any control. Nobody wants either of these outcomes. Permission with clear boundaries, on the other hand, sets people free: it shows the way forward and clears the mind so they can get on with their work.
Good compliance communication takes the pressure off. Poor communication complicates matters.
Evidence needs to be factored in from the outset
When it comes to AI, evidence is often considered too late. Many companies start by discussing usage, tools and rules. That’s understandable – but it’s not enough. Just as important is the question of how internal standards can be verified later on:
- Who has been trained?
- To what standard?
- Based on what content?
- How are changes addressed?
The moment when it really matters is rarely a quiet day in the office. It’s during an audit, a dispute or a claim. When the question then arises as to whether staff have been trained, a vague memory is worthless. What counts is the verifiable status: who, when, and on what basis.
And with AI, there’s an additional factor: the field is evolving rapidly. Training based on the state of affairs from the day before yesterday is already half out of date today. Proof is only as good as its timeliness – anyone who documents ‘trained in 2024’ but has never updated the content has a record that amounts to nothing. Particularly with dynamic topics, timeliness is therefore not a secondary consideration, but part of organisational quality.
AI does not organise itself – it is organised
All of this – setting standards, daring to launch an initial version, providing guidance rather than imposing bans, and thinking through certification requirements – is not the IT department’s responsibility. It is a leadership task. AI within a company does not organise itself; it is organised. And that is the real good news: it is in the hands of leadership.
This does not require a major project, but rather the decision to get started – cleanly, clearly, and in the knowledge that the first version will not be the last. Three sentences to bear in mind: Structure beats perfection. Guidance beats prohibition. Verifiability beats memory.
This is exactly what the EU AI Act Academy is designed for: practical training, uniform standards and documentation that shows who has been trained, when and on which topics – and which remains up to date as the legal landscape changes.