AI for SMEs:from first trialsto a working routine

AI for SMEs is no longer a niche topic in 2026, but in many businesses it is still a collection of individual tools rather than an orderly process. Depending on the survey, between a fifth and just over half of German companies use AI, and medium-sized companies are not in the lead. The most frequently cited barrier is a lack of knowledge, not the technology. If you want to introduce AI in your company, a clear route works best: choose a few suitable tasks, test them in a pilot, train your team as required by Article 4 of the AI Act, set approval rules and measure the benefit.

Nikolai Schöbel and Jeremias Burger

Co-founders of Scalableloops GmbH. Nikolai Schöbel leads online marketing and AI strategy, Jeremias Burger the AI architecture. Both build AI systems and train teams on them in their own agency work.

What do these figures mean for your own company?

Two patterns run through every survey. First, adoption is rising fast: in the ifo survey from 40.9 percent in 2025 to 54.5 percent in May 2026, at Bitkom from 36 to 57 percent within one year. Second, medium-sized companies are not in the lead. In the ifo survey they use AI less often (47.2 percent) than large companies and even less often than small ones.

KfW's analysis from July 2026 contains a finding that matters more to managing directors than any single rate: whether a company uses AI depends less on its size than on its level of digitalisation, its existing digital know-how and its general capacity for innovation. Companies that already work digitally find it easier to introduce AI. Where there are gaps, close them as part of getting started.

Using AI is also not the same as making full use of it. In the Bitkom survey, not a single company that uses AI says it exploits the technology's potential fully. Many businesses have individual tools in place, but no structured way of bringing AI into their processes.

How do you introduce AI in your business?

A route in small, verifiable steps has proven itself. It starts with tasks, not with tools, and it is only finished once the benefit has been measured. The order is deliberate: if you start by picking a tool, you end up looking for tasks to justify it.

1. 01

   01

   Choose the tasks

   Collect activities that recur, take a lot of time and follow a recognisable pattern: answering enquiries, preparing quotes, drafting texts, summarising documents, moving data between programs. Pick two or three where a mistake can be corrected and the result is easy to check.

2. 02

   02

   Record the baseline

   Before you start, note how long the task takes today, how often it occurs and how many queries or corrections it causes. Without this reference, you will not be able to say later whether the effort paid off.

3. 03

   03

   Pilot with real work

   Test for a few weeks with a small group and real cases. Decide in advance which data may be entered into the tool and which may not, and check with your data protection officer where the data is processed.

4. 04

   04

   Train the team

   Under Article 4 of the AI Act, businesses that use AI must take measures that support AI literacy among their staff. Training on your own tasks serves this purpose and lowers the biggest barrier, the lack of knowledge. Keep a record of who was trained on which tool and when.

5. 05

   05

   Set approval rules

   Define which results a person checks before they leave the building: customer emails, quotes, publications, anything involving figures or legal effect. AI systems can produce errors that sound convincing, so the check belongs in the process rather than in each individual's judgement.

6. 06

   06

   Measure, then scale

   After the pilot, compare time spent, error rate and satisfaction with the baseline. Only what holds up moves into everyday operations and is extended to further tasks. Whatever does not is adjusted or dropped.

What does the AI Act require when you introduce AI?

The key obligation for getting started is AI literacy under Article 4 of the AI Act. It has applied since 2 February 2025 to businesses that use AI systems, regardless of sector or size, so it also applies if your team only works with a chatbot such as ChatGPT.

Since 27 July 2026, an amended version applies. You must take measures that support the development of AI literacy, but you no longer owe a specific level of competence for each individual. A certificate is not required, although both the European Commission and Germany's Federal Network Agency recommend documenting the measures internally.

What this involves and how to record it is explained in detail in our article on the AI literacy obligation under Article 4. If you later deploy a chatbot that talks to customers, disclosure duties are added, which we cover in the article on chatbot disclosure under Article 50.

Which mistakes slow down AI adoption in SMEs?

Starting with the tool instead of the task. A licence for everyone is quickly bought. Without selected tasks, everyone uses the tool differently and nobody can say what it delivers.

Starting without data rules. If nobody has defined which customer, personnel or contract data may go into an AI tool, each person decides alone. That is exactly the uncertainty the surveys list as a barrier.

Skipping the training. A team that starts without instruction has no measure to show for the Article 4 obligation and gets weaker results, because it can neither write good instructions nor reliably spot errors.

Accepting results unchecked. Language models can invent details. An approval rule for everything that goes to customers or the public keeps such errors from carrying your name.

Not measuring. Without a baseline, every rollout remains a matter of gut feeling. Noting time spent and corrections before and after costs little and decides what you expand.