AI implementation for small and midsize companies in Germany

Why companies introduce AI with a plan

A plan decides which data goes into the text tools that staff already use. In the DIHK’s digitalization survey of nearly 5,000 companies, published on January 28, 2026, generative AI for texts, images or code was the most common use of AI, at 78 percent. Legal uncertainty remained the largest obstacle to using data (DIHK).

A text tool needs no IT department: a member of staff opens it in the browser and signs up with a private address. The company then has data in accounts it does not administer, and nobody knows which answers went to customers unchecked. The term for this is shadow AI. An introduction with a plan gives staff an allowed way to do the same work, with accounts the company controls.

What AI implementation covers

AI implementation replaces private experiments with approved accounts and rules that staff know. It has these parts:

• An inventory of the tools staff use today and what they use them for. I ask openly, and nobody gets blamed for an answer.
• The tool decision: Claude, ChatGPT or both, on a business plan with admin rights. Where data may not leave the company, an open-weight model on your own server.
• The usage policy, written with your data protection officer (the ten questions below).
• Accounts with an owner: someone creates them, and someone removes them when a person leaves.
• A pilot team first, then the other teams, each trained on its own tasks.
• A review after the rollout: which tools staff use, and where a rule is too tight or too loose.

The AI usage policy in ten questions

A usage policy answers these questions. You can draft your own answers before our call; the ones you cannot answer yet show where the work starts.

1. Which AI tools are approved, and on which plan?
2. Who administers the accounts, and who removes an account when someone leaves?
3. Which data may staff enter: public texts, internal documents, customer data, personnel data?
4. Which settings are fixed for all accounts, such as whether the vendor may use your chats to train its models?
5. Who checks an AI draft before it reaches a customer?
6. Do texts written with AI get a note for the reader? Your company decides.
7. Which tasks stay off limits, such as decisions about applicants or employees?
8. Where do staff report a mistake or data that went into the wrong tool?
9. Who answers staff questions about the policy?
10. When does the policy get its next review?

Who I am

I am Michael Wutzke, an AI engineer and consultant in Frankfurt with more than twenty years in IT and media. As CIO of the Frankfurt-based company Blocksize Capital I had disciplinary responsibility for teams and was responsible for budgets and for the management of information technology. I teach practical work with Claude at Claude Hacker House and hold a training authorization from IHK Frankfurt for two IT apprenticeships. In client work I help rebuild a platform with Claude and Codex as coding agents, and I am interested in open-source AI models that a company runs on its own servers. Details: Career stages.