In September 2026, Bitkom reported that, for the first time, a majority of German companies were using artificial intelligence. Fifty-seven percent of the 603 companies surveyed with 20 or more employees use AI, up from 36 percent the previous year and 20 percent in 2024. At the same time, there is Not a single one of the companies that use AI to fully tap into its potential; 59 percent of them say they do not use it at all.
„Artificial intelligence is not just the domain of a few pioneers; it has become widespread across the economy,“ said Bitkom President Ralf Wintergerst during the presentation of the figures. There is a gap between widespread adoption and actual benefit that cannot be bridged by licenses alone. This gap arises when employees are given a tool but have yet to develop the methods for using it effectively.
This article describes what an AI workshop for businesses must achieve in order to narrow this gap, what research says about productivity, and what obligations the AI Regulation will impose starting in February 2025.
What Sets an AI Workshop for Businesses Apart from a Training Course
The term is used for three formats that differ in their purpose and outcome. Categorizing them helps with booking, because providers sell all three under the same heading.
Training
A training session provides knowledge about tools: what a language model is, how a prompt is structured, and what the limitations are. It is suitable for the entire workforce and can be conducted online in just a few hours. The outcome is understanding; the process change follows in the workshop.
Workshop
A workshop focuses on a specific case from the company, such as the approval of incoming invoices or the implementation of a new feature in software development. The team tests the tools in its own environment, using its own data and conventions. The result is a set of rules and guidelines that take effect the very next day.
Program
A program links several sessions over the course of weeks or months, with tasks in between. This ensures that what’s learned becomes part of everyday life and doesn’t fade away after two weeks. The difference from a workshop lies in the commitment it requires. Our article on how changes are received within a team describes Change Management Methods.
The Gap Between Usage and Competence
The use of AI is growing faster than the availability of training for it. The TÜV Association noted this in its 2026 Continuing Education Study measured. From January through March 2026, Forsa surveyed 500 companies with 20 or more employees on behalf of the association; the respondents included training managers, managing directors, and board members. At the beginning of 2026, 27 percent of these companies had trained their employees in the use of AI; in 2024, that figure was 12 percent.
At the same time, 45 percent saw no need for AI training. In 56 percent of companies, employees were already using ChatGPT, Gemini, or Copilot. „Artificial intelligence has long since become part of everyday work in many companies, but skills development is not keeping pace with its use,“ said Joachim Bühler, managing director of the TÜV Association, in April 2026.
Company size is also a factor: 49 percent of companies with 250 or more employees had provided training, as did 32 percent of medium-sized companies with 50 to 249 employees and 21 percent of small companies with 20 to 49 employees. Bitkom expects higher figures by the summer of 2026 because it surveys companies based on the training they offer: 70 percent of companies provide training to employees on how to use AI, of which 37 percent train only selected employees and 11 percent train all employees. One-quarter do not offer any training; last year, that figure was 43 percent.
From the employees' perspective, the picture is similar. Bitkom Research conducted a survey in the spring of 2025 1,005 people aged 16 and older, including 513 employed individuals. Of those employed, 20 percent had received AI training from their employer, while 70 percent were not offered any.
When the necessary resources are not available, employees turn to their own tools. In the summer of 2025, 42 percent of companies expected that employees Private AI Tools for work. Only 23 percent had guidelines for AI use (Bitkom Research, 604 companies, summer 2025).
What Research on Productivity Shows
A field study and an experiment quantify the impact of AI tools in the workplace, and both explain why a workshop amplifies that impact.
Customer Service: 14 percent more resolved inquiries
Erik Brynjolfsson (Stanford), Danielle Li, and Lindsey Raymond (both MIT) examined the phased rollout of an AI assistant among 5,179 customer service representatives at a U.S. software company. The study was published in 2023 as NBER Working Paper 31161 and in 2025 in the *Quarterly Journal of Economics*.
The number of resolved inquiries per hour was measured. On average, it increased by 14 percent; among new hires and lower-performing employees, it rose by 34 percent, while there was hardly any increase among experienced employees. The authors see evidence that the assistant passed on the high-performing employees’ methods to the new hires, thereby shortening the onboarding process.
Writing assignments: 40 percent less time
Shakked Noy and Whitney Zhang of MIT had 453 college graduates with professional experience complete work-related writing tasks, such as press releases and reports. Half of them were given access to ChatGPT. The study was published in Science, July 2023.
The time required decreased by 40 percent, while the quality—as assessed by experienced professionals in the same fields—increased by 18 percent. The gap between weaker and stronger participants narrowed. Two weeks after the experiment, participants with access to ChatGPT reported using the tool in their actual jobs twice as often as the control group.
One study measures a task during normal operations, while the other measures tasks in an experimental setting; neither provides a department’s annual performance report. Our conclusion is that these findings can be applied to an entire company. The benefits accrue to those who use the tool, and they are greatest where the workflow is not yet well-established. A workshop supports this transition because the initial application takes place within the company’s own processes.
Why Your Own Process Matters
Our own case study demonstrates what a tool can achieve within a company, given its existing systems, data protection requirements, and established practices. The rules and findings from the workshop are thus seamlessly integrated into day-to-day operations. Microsoft's Work Trend Index 2024 shows what happens without this integration. The survey polled 31,000 knowledge workers in 31 countries in February and March 2024. 75 percent of knowledge workers used generative AI, and 78 percent of them brought their own tools to work. Sixty percent of executives were concerned that their organization lacked a plan for this.
Example: a development team
Traditional process: A team receives a requirements document and a mockup for a new feature. One developer reviews the materials, estimates the effort required, and writes code, while a second developer reviews it. Everyone uses AI assistants on their own, without shared guidelines, and no one knows which suggestions were implemented.
AI-supported process following the workshop: The team has a set of project rules—a configuration file in the repository—that specifies conventions, tests, and restricted areas for the coding agent (an AI assistant that writes and modifies code on its own). Requirements and mockups are fed into the agent as context; the developer reviews every change, and tests run before anything is merged. Changes made by the agent can be traced in the log; this also addresses the data protection officer’s concerns.
Example: Document Workflows
The same pattern applies to back-office operations. In accounts payable, for example, a workshop can be structured around the 20 most frequent suppliers, using actual documents from the last quarter. The team learns where Intelligent Document Processing (IDP) extracts the data, where a human performs the review, and what approval thresholds apply. Our article describes how AI projects grow beyond their initial use case: Article on AI Implementation.
Article 4 of the AI Regulation: The Requirement for AI Competence
Effective February 2, 2025 Article 4 of the EU AI Regulation (Regulation (EU) 2024/1689). Under the original version, providers and operators were required to use their best efforts to ensure that their staff possessed a sufficient level of AI competence. As of July 27, 2026, the version amended by the Digital Omnibus Regulation (EU) 2026/1744 applies: providers and operators shall take measures to support the development of their staff’s AI competence. Technical knowledge, experience, training and continuing education, and the context in which the systems are used must still be taken into account.
An "operator" is any company that uses an AI system in a professional capacity under its own responsibility; this includes the use of ChatGPT for administrative tasks.
The regulation does not prescribe a specific level or an exam, and since the amendment, it explicitly states that no one is required to guarantee a specific level of competence. It requires action. A documented workshop—including a list of participants, content, and the resulting rules—is an obvious way to provide such evidence. In the same Bitkom survey from September 2026 (603 companies), 37 percent expect to be affected by the AI Regulation as users, up from 23 percent the previous year. Of the companies that consider themselves affected or are assessing the situation, 89 percent anticipate a high implementation burden.
How to Tell if an AI Workshop Is Good
When booking, it's worth considering six factors that determine the outcome:
- Own cases: The workshop focuses on a current project within the company, such as a feature from the backlog or the financial statements for a quarter.
- Practitioners as Trainers: The trainers use the tools themselves in their day-to-day work and are familiar with the common pitfalls from their own projects.
- Guardrails as a result: In the end, there are rules: which data can be entered into which tool, who reviews changes, and what is logged.
- Assignments between meetings: An exercise after each step ensures that the team applies what it has learned before the next session builds on that foundation.
- Privacy by Design: Data processing (the data protection agreement with the tool provider), the data storage location, and access rights are clarified before the first meeting so that the data protection officer is present at the meeting.
- Protecting the Team: Work is done in pairs, and there are no individual performance metrics. This lowers the barrier for skeptics on the team who will later be expected to use the tool on a daily basis.
Tip: A vendor that promises a productivity figure—such as twice the speed—should be able to explain how it measures that figure. Developer output and administrative performance can rarely be quantified precisely. A reliable offer specifies maturity levels (stages that describe how independently a team uses AI tools) and concrete results, because both can be verified after the program is implemented.
Conclusion
The majority of companies use AI, but only one in nine trains all employees in its use, and almost none of them fully tap its potential. Research shows that productivity gains accrue to those who apply the tool in their work and are greatest where routines have yet to be established. An AI workshop for companies begins with its own processes, its own team, and the rules that will apply going forward.
For development teams, Konfuzio offers this type of program through the Konfuzio Academy for Development: two days on-site working with your own code and processes, followed by a four-month program. Are you looking for a starting point that meets your team where they are in their day-to-day work? Contact us, to get advice from experts in the field.
EN 
DE 
ES 
FR