Introducing AI in the company: Training, use cases and the first steps

Introducing AI in a company is no longer just a question of technology. It is a question of three things at the same time: specific use cases, empowered employees and a comprehensible framework. If you just buy a licence for a language model and distribute it to the workforce, you haven't introduced AI - you've distributed a tool.

This article is written for the role responsible for enablement in the company, not just for procurement. It organises what "introducing" means in concrete terms, what obligation has been behind it since the beginning of 2025, why use cases come before tools and which first steps are important and in what order.

The state of 2026: High usage, low enablement

Adoption is no longer the open question. According to the Bitkom study "Artificial Intelligence in Germany" (survey 2025, published 2026, basis: 604 companies with 20 or more employees), 41 per cent of companies are actively using AI and a further 48 per cent are planning or discussing its use. A year earlier, active use was only 17 per cent. The figure has therefore more than doubled.

Enabling, on the other hand, is lagging well behind. Only 8 per cent of companies offer AI training for all employees. 21 per cent train the majority of employees, 25 per cent train selected employees and 43 per cent offer no training at all. This is the real gap: The tools are in-house, but the expertise behind them is completely lacking in almost half of the companies.

This gap has a double cost. It is expensive because untrained users rarely go beyond trial and error and the promised productivity gains fail to materialise. And it is risky because AI expertise is now also a legal requirement.

.

What "introducing AI in the company" actually involves

The word "introducing" has four components that belong together. Naming them one after the other will show you where the work actually begins:

1. Use cases: defined tasks in which AI makes a measurable contribution. Without them, the introduction remains a solution without a problem.
2. Enablement: Employees who know when a tool is suitable, how to use it and where its limits lie. This is the part that is most often missing.
3. Tools and integration: the specific tools, their connection to existing systems and data.
4. Framework / governance: rules for handling data, approvals, documentation and responsibilities.

These are not optional stages, but parallel tracks. An introduction that only serves the third is the standard variant of failure: licences run, usage trickles away. In order of value creation, the work starts with points 1 and 2, not with purchasing.

The legal obligation: Article 4 EU AI Act

Article 4 of the AI Regulation (EU AI Act) has been in force since 2 February 2025. It obliges providers and operators of AI systems to ensure that their staff have a "sufficient level of AI competence". The provision is often inaccurately reproduced, so here are three points for clarification.

Firstly, the obligation applies regardless of risk class. It applies not only to high-risk systems, but to every operational use of AI, from language models in customer communication to internal data analysis. AI competence means the skills, knowledge and understanding to use AI competently and to be aware of its opportunities, risks and potential harm.

Secondly, the article does not prescribe a specific training format. The EU Commission has made it clear that certification is not mandatory. What is required is sufficient knowledge, not a specific certificate. However, it is recommended that the time and participants of training courses be documented; this is relevant in the event of liability issues.

Thirdly, to categorise the consequences without drama: there is no separate fine for a violation of Article 4 itself. The risk is indirect. If damage occurs due to a lack of AI competence, the lack of competence falls back on the company. Training is therefore not a "nice-to-have", but a documentable duty of care.

Use cases first, tools afterwards

The most common order in practice is the wrong one: first select a tool, then search for purposes. The reverse makes sense. A use case is suitable as an entry point if three conditions apply. The task is recurring, the effort involved is measurable and the consequences of an error are manageable or easily controllable.

Three examples that typically fulfil these criteria:

• Customer support: Pre-formulated draft responses to standard enquiries that an employee checks and approves. Recurring, measurable via the processing time per ticket, errors intercepted by the release.
• Sales and quotation: First drafts for quotation texts or summaries of long tenders. Saves preparation time, the decision remains with the human.
• Internal documentation: Research in internal knowledge databases via a chatbot instead of manual searches. The data connection is the critical factor here, not the model.

What is not suitable as a starting point is just as important to name: Tasks with a high potential for damage in the event of errors, legally sensitive decisions or processes without a clearly measurable result. If you start there, you risk an expensive lesson and burn out internal acceptance for the next round.

The first steps in the right order

A structured introduction follows a sequence because later steps build on earlier ones. The following five steps are intended as a sequence, not as a checklist to tick off at random:

1. Inventory. Where is AI already being used, even unofficially via private accounts for work purposes (shadow AI)? This is the real starting point, not the green field.
2. Prioritise use cases. Select two to three use cases according to the criteria mentioned above. It is better to have a few that work than a broad catalogue that does not bind anyone.
3. Define the framework Before rolling out widely: Which data is allowed in which tools? Who releases it? How is usage documented? This step also fulfils part of the Article 4 obligation.
4. Target group-specific training. Not everyone needs the same thing (see the next section). Training should be based on the prioritised use case, not on abstract AI theory.
5. Measure and adjust. A use case without a defined metric cannot be evaluated. Check after a set period of time: Is it bearing fruit, or is it being replaced?

The most common mistake in this sequence is to skip step 3. Rolling out without a framework produces the very uncontrolled use that article 4 addresses, and only postpones the governance problem.

What an AI training course must do

An effective AI training course is not a one-off lecture entitled "What is AI". It covers three levels of competence, and not every role needs all three in the same depth:

• Basic technical understanding: What a language model can and cannot do, why it produces errors (hallucinations) and what distinguishes a good prompt from a bad one. Everyone who uses AI needs this level.
• Legal and organisational framework: Which data may be processed, what the internal framework stipulates and when a human remains responsible. Everyone who uses AI in a work context needs this level.
• Depth of application in the use case: The specific operation in the prioritised use case. This level is role-specific and belongs to the introduction itself, not in a general webinar.

The third level almost always determines whether a training course is effective or just ticked off. General AI knowledge is quickly conveyed and just as quickly forgotten. Empowerment arises from the specific use case, with the data and tools that employees work with the next day.

Conclusion

Establishing AI in the company means bringing four things together: Use cases, enablement, tools and frameworks, in this order of priority and not the other way round. The status of 2026 is clear. The use is there, training is lacking in almost half of the companies, and since February 2025, AI expertise has been a documentable obligation under Article 4 of the EU AI Act.

The pragmatic approach is small and structured: Take stock, prioritise two to three use cases, set a framework, train specifically on the use case, measure. If you proceed in this order, you will avoid the most expensive option: tools without the necessary skills.

A concrete first step if you are just starting out: taking stock. If you are already further along and want to structure the use case selection, the process analysis and AI strategy provides the right framework for this. AI workshops and team enablement are the direct lever for empowering teams, and our AI consulting provides an overview of the procedure.