Integrating AI sounds simple at first. In day-to-day business, though, companies run into hurdles that a private user rarely has in mind: data protection, co-determination, training obligations and the concerns of employees. Whether a rollout succeeds therefore depends less on the choice of tool than on how a company deals with these hurdles.
How widespread is concern about AI at work?
The concern is measurable. In a Bitkom survey published on 14 September 2026 (in German) of 603 companies in Germany, 45 percent expect a slight and 12 percent a significant drop in headcount due to AI over the next two years. Among companies that already use AI, 18 percent expect headcount to grow.
The same survey shows that 57 percent of companies now use AI, up from 36 percent a year earlier. Among users, 42 percent name a lack of acceptance among staff as an obstacle. Attitudes within the team are therefore one of the major hurdles of a rollout.
At European level, Eurostat shows a similar picture from the organisation's side. Companies that had considered using AI cite a lack of expertise (around 71 percent), unclear legal consequences (around 53 percent) and data protection (around 49 percent) as the main barriers.
Which questions should a rollout answer?
Four questions affect employees directly. A rollout that answers them openly from the start takes pressure off.
- Jobs: how will tasks and roles change?
- Monitoring: will anyone analyse who uses the AI and how?
- Skills: who gets which training, and when?
- Responsibility: who is accountable when an AI result is wrong and goes out under an employee's name?
An honest answer can also be: this is still open and will be settled by a specific date. Promises that later turn out to be wrong damage trust more than an open question.
Which rules are needed before the start?
Informing means the decision has been made and the team is told about it. Involving means the team helps shape it before the details are fixed. A small pilot group of volunteers, deliberately including sceptical employees, uncovers weak spots before a tool is rolled out widely.
If an AI tool processes employee data, co-determination has to be settled in Austria. In companies with a works council, its consent is usually required under sections 96 and 96a of the Labour Constitution Act (ArbVG), typically as a works agreement. Involving the works council early therefore saves time.
The ground rules should be in place before day one: which data may go into which tool, which results someone must check, and who helps with questions. Ideally in writing, on one page and in plain language. Nobody reads a twenty-page policy in legalese, and then no rule applies in practice.
What does the AI Act require for training?
According to Bitkom, 70 percent of companies train their staff in using AI, while one in four offers no training at all. Article 4 of the AI Act requires measures to promote AI literacy. The Digital Omnibus softened this wording in summer 2026, but the obligation itself remains, according to an analysis on lawandtechnology.eu.
Effective training starts from real tasks. A tool demo in a large hall shows what is possible but does not build routine. Short sessions in small groups, where each person solves one of their own tasks with AI, are closer to everyday work. After that, teams need named contacts and time to practise during working hours.
Which use cases are suitable for the start?
The first use case shapes how the team thinks about AI. Three criteria help with the choice:
- It noticeably and immediately takes work off the person using it.
- An AI error is easy to spot and does little harm.
- A human checks the result before it leaves the company.
Good candidates are minutes drafted from meeting notes, first drafts of recurring replies, or searching internal handbooks and policies. Unsuitable for the start is anything that measures employee performance or makes automatic decisions about people: such cases confirm exactly the concern a rollout is meant to reduce.
After the first few weeks, a review with the pilot group helps: what saved time, what got in the way, what is still missing? The answers are the basis for the next round and the most credible material for internal communication.
What does a communication plan look like?
- Explain the why: which problem AI is meant to solve and what is still open.
- Say clearly what changes and what stays the same.
- Publish the ground rules on data, review and responsibility.
- Set up a feedback channel where questions may also come in anonymously, and answer them visibly for everyone.
- Let the pilot group report on its results, including what did not work. Reports from within the team are more credible than a presentation from the top.
Communication is not a side programme in an AI rollout. It is a large part of the work, and it starts before the first licence.
Training formats for teams are described on the courses page.