Five AI trends German businesses should plan for in 2027

Predictions about AI age badly, so this list stays close to the ground. These are five developments that are already visible in 2026 and that we expect to shape the AI decisions of German companies in 2027. Each comes with something you can do about it now.

1. From chatbots to agents

Until now, most business AI has answered questions. The next step is AI that completes tasks: it books the appointment, creates the ticket, updates the customer record, prepares the order. These systems are called agents, because they act through the same tools your employees use.

This raises a question that a chatbot never raised: what may the system do without asking? A good agent has narrow permissions, asks for approval before anything that costs money or cannot be undone, and keeps a record of every action.

What to do: Look for tasks with clear rules and a clear end, and write down which steps need a human signature. That list is the specification for your first agent.

2. Data sovereignty becomes a buying criterion

German customers, works councils and data protection officers ask where data is processed. In 2027 we expect that question to be on every procurement checklist. The options have improved: the large model providers offer processing in EU regions, European providers have grown, and open-weight models can run on your own servers.

No single setup is right for everyone. A public website assistant and a system that reads personnel files have different requirements.

What to do: Sort your data into classes, for example public, internal, personal and confidential, and decide for each class where it may be processed. Then the choice of model and hosting follows from the data, not from the marketing.

3. Smaller, specialised models

The largest models get the headlines, but many business tasks do not need them. Sorting emails, extracting fields from a delivery note or checking a form are narrow tasks. Smaller models handle them faster and at a fraction of the cost, and some run on hardware you can put in your own server room.

We expect more systems that combine both: a small model for the routine bulk, a large one for the hard cases.

What to do: When you plan a project, ask what the cost per case will be at full volume, not only whether the demo works. Build so that the model can be exchanged later.

4. Compliance moves into the design

The obligations for high-risk AI under the EU AI Act were postponed to 2 December 2027, and the transparency rules have applied since August 2026. Germany now has a supervising authority, the Bundesnetzagentur. From 2027, a customer or auditor asking "how do you make sure this system does what it should?" will be normal.

Logging, human oversight and clear labelling are cheap when they are part of the first design, and expensive when they are added later. Our article on the AI Act for German SMEs has the dates and a checklist.

What to do: For every AI system, new or existing, be able to answer three questions: what does it decide, who can overrule it, and where can we see what it did?

5. E-invoicing opens the door to automation

Since 1 January 2025, every company in Germany must be able to receive electronic invoices in business-to-business trade. From 1 January 2027, companies with a prior-year turnover above €800,000 must also issue them, and from 2028 everyone must.

This is a tax rule, not an AI rule, but it changes what is possible. An invoice that arrives as structured data can be checked, matched to an order and prepared for payment automatically. Many companies will rebuild their invoice process for the deadline anyway. Doing it once, with automation in mind, is cheaper than doing it twice.

What to do: If you are changing your invoicing for 2027, map the whole path from incoming order to payment and mark the steps where people only copy data from one system to another.

What these have in common

In all five, the model is not the hard part. The work is in the process around it: clean data, clear permissions, a person who is responsible and a way to see what happened. Companies that get this right in 2026 can switch to better models as they arrive without rebuilding the rest.

Sources

• Bundesnetzagentur: key role in the implementation of the AI Act, 29 July 2026
• Orrick: Digital Omnibus finalizes 8 compliance changes
• E-invoicing dates: German Growth Opportunities Act (Wachstumschancengesetz) and Section 14 of the VAT Act (UStG)