Lots of AI, little plan: the real gap in the Mittelstand
57 percent of companies use AI, but hardly any get out of it what they could. Why an AI strategy in the Mittelstand is almost always a question of systems, and where you start.
Moritz Grellner · Consultant · 7 min read
For the first time, most companies in Germany use AI: 57 percent, reported the digital association Bitkom on September 14, 2026. Yet not a single company that uses AI says it taps its full potential. 59 percent don't tap it at all. The gap rarely lies in the tool, but in the plan behind it.
In this article we show you why an AI strategy in the Mittelstand is almost always a question of systems. And in which order you can proceed.
AI has arrived, but mostly next to the process
According to Bitkom, 20 percent of companies used AI two years ago, 36 percent a year ago, and 57 percent today. The survey covered 603 companies with 20 or more employees. Bitkom speaks of a pace not seen with any technology before.
In everyday work, using AI often means a chat window in the browser. Someone has it draft an email or summarize meeting notes. That saves minutes, but it doesn't change the process. The invoice is still retyped, the Excel list next to the ERP still maintained.
Companies' own assessment shows this too. Only 3 percent of AI users say they tap its potential to a large extent. 28 percent use it rather little, 59 percent not at all.
According to Bitkom, this share is almost the same across all sizes, from the 20-person business to the large corporation. So it isn't a question of company size.
At a glance
- 57%
- of companies use AI (2024: 20%)
- 59%
- of AI users don't tap its potential at all, 0% fully
- 21%
- name licenses as a major cost item, 50% data preparation
Source: Bitkom, press release of Sep 14, 2026; survey of 603 companies with 20 or more employees
Without a plan, chance decides where AI lands
A third of the companies that use AI do so mainly for fear of falling behind otherwise. That, too, is in the Bitkom survey. The impulse is understandable. But it often means the tool comes first and the question of what it's for comes second.
The consequences are easy to see in the Mittelstand. We come across three patterns again and again:
Costs run away.54 percent of AI users name hard-to-calculate costs as an obstacle. If you start without a goal, you can't weigh any benefit against them.
Shadow AI emerges.This means AI services that employees use on their own initiative, often with private accounts. Company data then ends up where nobody manages it. Data protection is the biggest obstacle for 66 percent of users, and 45 percent worry that company data could fall into the wrong hands.
Pilots stay pilots.A test works in the demo. Because it isn't connected to the ERP, it never makes it into day-to-day business.
None of these patterns is a problem with AI itself. All three arise because nobody decided beforehand where AI should work and with which data.
Licenses are the smallest item
Many assume the big costs lie in software subscriptions. The Bitkom figures paint a different picture.
51 percent of AI users name infrastructure such as cloud and computing power as a major cost item. 50 percent name data preparation, 41 percent integration into existing systems. Licenses are a major cost item for only 21 percent.
“If you only buy a license, you've bought AI, but you haven't changed anything yet.”
For you, this means: the language model is interchangeable. The real work lies in clean data and in the connections between your systems. That is exactly where it's decided whether AI changes a process or just stands next to it.
Almost every business knows what data preparation means in practice. The same supplier is in the system twice, spelled slightly differently. Item numbers in the shop differ from those in the ERP.
Prices are maintained in an Excel list that only one colleague really understands. An AI that accesses data like this delivers results nobody trusts.
The incoming invoice shows the problem in miniature
There's an example in almost every inbox. According to a study by IW Consult of November 4, 2025, only one in five incoming invoices in the Mittelstand arrives as an e-invoice. 26 percent are paper, 55 percent digital formats that can't be processed directly, such as PDF.
These invoices are read, retyped and checked. AI can take over the reading, and it increasingly does. According to Bitkom, AI use in controlling and accounting rose from 17 to 25 percent within a year.
But the extracted data has to move on: into the vendor account in the ERP, to be matched against the purchase order and goods receipt, into approval. If that connection is missing, the AI reads the invoice and a person still types the result into the ERP. Why the structured e-invoice sets the pace here anyway is covered in our article on mandatory e-invoicing from 2027.
In the Mittelstand, an AI strategy is a question of systems
Anyone deciding on AI in their own business should be able to answer three questions. Where is your data? Which systems talk to each other and which don't? And where is data retyped or copied today?
This becomes even more important once AI agents come into play. That means software that completes tasks on its own, such as comparing prices or requesting quotes. According to Bitkom, 11 percent of companies that use, plan or discuss AI already use such agents, and 29 percent plan to. Bitkom President Ralf Wintergerst names reliable data, connected systems and fixed rules for what an agent may do as prerequisites.
In the Mittelstand, these systems are rarely new. There's an ERP such as Business Central, a CRM, Microsoft 365, payroll at the tax advisor's office and a few Excel lists holding everything together. That's why an AI strategy doesn't start with choosing a model. It starts with an honest look at this landscape.
For companies that don't use AI at all yet, the hurdle lies there too. According to Bitkom, 85 percent of them cite a lack of technical know-how for introducing it. In our view, the model is rarely the hard part. The hard part is connecting it to your own data and processes.
First the map, then the speed
A sound plan follows a simple order. Three steps have proven themselves:
Map.First, a map of your systems and processes is created. It shows which systems exist, which data lives where and where handovers happen by hand. This is also called system mapping.
Prioritize.Use cases emerge from the map. Each one gets a rough estimate of effort, budget and benefit. That way you see which step pays off first.
Implement where the data already is.AI works best in the systems your team already uses. These are usually ERP, CRM and Microsoft 365. Switching systems is rarely necessary for this.
Implementation also needs clear rules: who checks AI results, who approves them, which data is off-limits. And the team has to come along. According to Bitkom, one in four companies still offers no AI training at all.
Our principle for this is simple: people think, systems work. AI takes over reading, sorting and transferring. The decision stays with your team.
What you can do now
You don't need a big project to get started. You can take these steps with what you already have:
Write down where AI is already used in your business today, including unofficially.
Take a process that costs time every day, such as incoming invoices. Trace which systems it runs through.
Count the points where someone retypes, copies or updates Excel by hand.
Decide who checks and approves AI results.
Only then decide on tools and licenses.
You can estimate in advance roughly what retyping and media breaks cost your business with our system map.
Our take
We rarely see a company lacking the right AI. What's usually missing is an overview of where it should connect. That's why we don't start with tools, but with your systems and processes.
That's what the System Audit is for. In a ½-day workshop, we map together which systems you have, how data flows and where things are transferred by hand. Internally we call this discovery & system mapping. The result is a roadmap with prioritized use cases, budget and ROI.
The System Audit costs €2,999. We connect your systems instead of replacing them. That's how lots of AI becomes a plan that makes it into day-to-day business.
Next step
First the map, then the AI
In the System Audit, we look at your systems and processes in a ½-day workshop. Afterwards you have a roadmap with budget and ROI.
