Key Points at a Glance
- Only 7% of SMEs with 10–49 employees are using AI to address the shortage of skilled workers (Bitkom, June 2026).
- 64% see themselves as laggards; this is a result of hype without substance, not a failure.
- Four entry-level areas (writing texts, structuring data, service pre-qualification, and research) can be implemented in 4–6 weeks.
- Multi-vendor instead of single-vendor: the Anthropic block on June 15, 2026, shows the supply risk.
The latest figures are clear. According to a Bitkom survey (June 2026), only 7 percent of companies with 10 to 49 employees use AI to address the shortage of skilled workers. For companies with fewer than 10 employees, the figure is 2 percent. For companies with 50 to 249 employees, it is 12 percent. Only from 250 employees upward does the rate jump to 21 percent.
64 percent of all companies surveyed see themselves as laggards. 22 percent believe they have missed the boat. Only 10 percent consider themselves pioneers.
We see these figures confirmed daily in our consulting practice. Managing directors ask us: Where do we start? Which tool? What does it cost? And above all: What does it actually deliver?
This article answers those questions without the hype. With concrete figures, clear examples, and an honest assessment of what is realistic in 2026.
The numbers in detail
The Bitkom study is based on 852 companies with three or more employees surveyed by phone. It is representative. The survey ran from calendar weeks 38 to 43 of 2024 and was published in June 2026.
AI adoption to address the skilled labor shortage by company size:
| Company size | AI adoption rate |
|---|---|
| Fewer than 10 employees | 2 percent |
| 10–49 employees | 7 percent |
| 50–249 employees | 12 percent |
| 250 or more employees | 21 percent |
The gap between micro-enterprises and large corporations is a factor of 10. That is not a minor detail. That is the core of the problem.
35 percent of all companies expect AI to help ease the shortage of skilled workers. But only 5 percent have actively implemented it. There is a gap between expectation and reality.
Source: Bitkom press release dated June 19, 2026.
Why adoption is stalling
In our client conversations, we hear four reasons.
1. Legal uncertainty. The EU AI Act has been phased in since August 2025. Which obligations take effect when? Who is liable if an AI system makes a wrong decision? Bitkom itself names data protection, AI Act implementation, and labor law consequences as the top barriers.
2. An overabundance of tools. ChatGPT, Claude, Gemini, Mistral, Perplexity. Plus hundreds of specialized tools for accounting, marketing, and sales. Who has time to compare them all? No one in day-to-day operations.
3. Lack of use cases. “Using AI” is not a goal. Specific tasks are the goal. Categorizing invoices, preparing quotes, translating texts. That is exactly what the South Westphalia Chamber of Industry and Commerce (IHK) points to in its latest magazine (May–June 2026): tools like Gemini or Claude can formulate texts, structure data, and prepare quotes, with no programming knowledge required.
4. Concerns about data protection. In the same article, the IHK warns: do not enter any personal or confidential data as long as data protection questions remain unresolved. This warning is justified. But it paralyzes many business owners.
Source: Südwestfälische Wirtschaft May–June 2026 (IHK).
The US block as a new variable
On June 15, 2026, Anthropic blocked its most powerful AI model for foreign users on the orders of the US government. Bitkom’s stance on this is sharp: the block also hits German companies that work productively with Claude.
In concrete terms, this means an SME that builds its workflow on a US frontier model carries a supply risk. Today everything runs. Tomorrow an export control adjustment arrives. The day after, the process grinds to a halt.
Source: Bitkom on the US block for AI models.
Our recommendation: test several providers in parallel. One model as the primary tool, a second as backup. Ideally, keep at least one European model (Mistral, Aleph Alpha) in the mix. This is not a hype argument. It is business continuity.
A pragmatic start for SMEs
We see four sensible areas to start with. All four can be implemented in 4–6 weeks. No million-euro investment. No team of consultants.
Area 1: Writing and proofreading texts. Quotes, product descriptions, newsletters, job postings. Standard LLMs deliver solid results here. Effort: 1–2 days of onboarding per team. Tool costs: 20–30 euros per user per month.
Area 2: Structuring data. Turning PDF invoices into Excel tables. Creating tickets from customer emails. Extracting order lists from order confirmations. Here SMEs often save 3–5 hours of routine work per employee per week.
Area 3: Customer service pre-qualification. Inquiries are sorted and summarized before they are handled. The employee gets a prepared draft, not a raw inquiry. Handling time typically drops by 30 to 40 percent.
Area 4: Research and benchmarking. Competitive analyses, market data, searching for studies. What used to take half a day is done in 30 minutes. Important: always double-check the results.
These four areas cover the majority of the inquiries we receive. We recommend starting with one area. Test it for three months. Then expand.
What we notice with our clients
Successful SME projects have three things in common.
First: there is an internal point person. No external consultant for ongoing operations. One person from the team who owns the topic. 4–8 hours per week, firmly blocked out.
Second: there is a written AI policy. Which data may go in and which may not. Who is allowed to use what. Who reviews outputs before they are published. This policy is no longer than two pages. But it exists.
Third: there is honest measurement of success. Not “AI has made us more productive,” but “we save 12 hours per week in order intake.” Concrete numbers, collected monthly.
Anyone who lacks these three points is running AI theater. Not AI use.
Our assessment
The 64 percent laggard rate is not a failure of the Mittelstand. It is the result of a market that has communicated in buzzwords for three years. Without substance. Without an honest weighing of risks.
An SME that starts in 2026 is not too late. On the contrary: the tools are more mature, the best practices clearer, and the prices more stable than in 2023.
But: anyone still standing there without a plan in 2027 will fall noticeably behind in the market. Not because AI is magic. But because the competition is 3–5 hours more productive per employee per week.
The question is not whether, but how. And above all: with which first step.
Checklist for getting started
- Name one person to own AI. In writing. With an hourly budget.
- Choose one starting area. Writing texts is the easiest start.
- Test two tools in parallel. At least one non-US provider in the mix.
- Set up an AI policy. Two pages maximum. Data classes, approvals, review obligation.
- Define success measurement. Hours, error rates, handling times. Collect monthly.
- Three-month trial phase. Then evaluate honestly: does the tool stay, does it go, does a second area come in.
This process usually costs less than 1,500 euros per month for a 10-person team. Tools included, consulting fees excluded.
We support SMEs getting started with AI, without hype and without hocus-pocus. We analyze your use case, recommend suitable tools, build the policy, and measure the results. If you want to know which starting area fits your business, write to us: waterproof.agency/kontakt/.
Frequently Asked Questions
What happens if our main tool, like Claude, gets blocked?
That is exactly what happened on June 15, 2026. Anthropic blocked its top model for foreign users. Anyone with a backup tool in their workflow simply switches over. Anyone without one waits. For this reason, we generally recommend multi-vendor setups over single-vendor strategies.
Do we need an external consultant for ongoing operations?
No. Successful SME projects have an internal owner with a fixed hourly budget. External consulting only makes sense for setup, policy, and success measurement. Anyone permanently dependent on outside help does not have an AI project. They have a dependency.
What data are we allowed to enter into an AI tool?
As long as data protection questions remain unresolved, the IHK recommendation applies: no personal data, no confidential business information, no draft contracts with real names. For these cases, there are GDPR-compliant enterprise plans with separate data processing. They cost more, but they are necessary for legally sound operation.
What are the monthly costs for a 10-person team?
Realistically, the pure tool costs run 200–300 euros per month for a 10-person team. On top of that come internal hours for the AI owner (4–8 hours per week). Consulting fees for setup and policy are a one-time cost in the low four figures.
Which AI tool is best for SMEs in 2026?
There is no single tool. For texts and standard tasks, Claude, ChatGPT, and Gemini deliver comparable results. We recommend testing two providers in parallel and keeping at least one non-US provider (Mistral, Aleph Alpha) in the mix. That lowers the supply risk when export controls change, as with the Anthropic block on June 15, 2026.