**Generative AI in Mid-Market Companies: Why the 78-Percent Figure Is Misleading**
7 min read
According to the DIHK digitalisation survey, 78 percent of companies are using generative AI for text, images or code. The figure sounds like a breakthrough. In reality, it’s a trap: usage is being confused with impact. Turning on a tool doesn’t change a single process. It’s at this very gap that it will be decided which mid-sized firms are 20 % more productive by 2027-and which simply own a fancier autocomplete.
Key Takeaways
- Usage is not value creation. The 78 % figure counts who opens an AI tool, not who rebuilds a workflow around it.
- The hurdle is legal and organisational. According to the DIHK, it’s uncertainty and missing integration-not budget-that are holding things back.
- The leverage lies in the handoff. AI only pays off once it’s clear who reviews the output, owns the result and feeds it back into the process.
Related:Three AI failures and the lessons they teach / Stanford AI Index: reliability as the bottleneck
What the 78 % really means
The DIHK surveyed nearly 5,000 companies across all sectors for its digitalisation poll. The headline result that ricocheted through the press: 78 % are using generative AI, mainly for drafting text, generating images and writing code. More than a third of users expect a strong impact on their own productivity.
That’s the good news. The uncomfortable part is buried in the fine print. “Using” in most cases means someone in marketing pastes a prompt into a chat window, or a developer asks for a function stub. That’s handy. It’s not a changed process, merely a faster tool at one desk. The figure measures adoption, not embedding.
Anyone who has ever shepherded a transformation knows the difference. Having a tool on the premises is easy. Integrating it so the output reliably feeds the next step is the real work. It’s at the handoff-where AI results are handed to the next human or system-that the difference between usage and impact is made.
Why the gap isn’t about money
The DIHK (Association of German Chambers of Commerce and Industry) clearly identifies the biggest hurdle: legal uncertainty. Lack of expertise, limited data access, and costs are cited less often than last year-a remarkable shift. Just two years ago, the standard answer to questions about digitalization bottlenecks was: too expensive, no staff. Today, the question is: what are we even allowed to do?
For SMEs, this is a different challenge than submitting a budget request. If a company doesn’t know whether it’s permitted to feed customer data into an AI tool, throwing more money at the problem won’t bring clarity. What’s needed is a clear framework: which data goes into which system, with what approval, and under whose responsibility. That’s governance work. It can’t be delegated to IT like buying a server-it belongs on the executive table.
Those who shy away from this groundwork get the worst possible outcome: employees use AI anyway, just unofficially and without rules. Shadow AI isn’t the result of too few tools; it’s the result of too little clarity. The unspoken ban collides with the unissued permission in the legal gray zone of daily operations.
Regulatory pressure is also mounting. The EU AI Act requires companies to assess and document the risks of their AI applications. If AI is used informally on individual workstations, that’s simply impossible. Scattered tool usage can’t be audited; a defined process can. The legal uncertainty the DIHK flags as a brake won’t be resolved by waiting-it will only be resolved by the same anchoring that also unlocks economic benefits. Here, compliance and productivity finally point in the same direction.
What breaks
- AI as a one-off workaround with no process integration
- No clear rules on which data is permitted
- No checks on whether the output is accurate
What works
- A process where AI has a defined role
- Clear data approvals with named owners
- A designated team or role to validate results
How usage turns into impact
The path is unspectacular. Perhaps that’s why it’s skipped. Instead of rolling out yet another tool, it’s worth focusing on a single process that eats up time today. Creating quotes, handling complaints, reporting. The question asked here is: at which point can AI make a concrete step faster? And who takes ownership of the result afterward?
An example from quote creation makes the difference tangible. A sales team lets AI suggest text modules. That saves ten minutes per quote-a clear win at the individual level. But impact only materializes when the draft automatically flows into the CRM, a second person double-checks the prices, and approval is documented. Without this chain, the time saved remains a private trick that leaves the company the moment the employee does.
This second question is the crucial one. An AI draft that no one reviews and no one owns is no progress, but a new risk. Only when the handoff is clarified-who accepts, corrects, and feeds it back-does a tool become a process component. That’s exactly what the 78-percent figure doesn’t measure. That’s exactly what separates the companies that will be measurably more productive in 2027 from those that are just typing more expensively.
Having an AI tool in the building isn’t an achievement. The achievement begins where the result is handed off and someone stands behind it.
Iterative doesn’t mean hesitant. It means starting with one process, setting a clear metric, and after four weeks honestly checking whether the number has moved. If it has, the next process follows. If it hasn’t, it was the wrong tool or the wrong spot. A cheap insight-as long as it comes early.
The reflex to roll out as broadly as possible is understandable, but it’s precisely what produces the 78 percent that proves nothing. Depth beats breadth. A process that has truly been rebuilt convinces the workforce more than ten licenses nobody brings into daily use. If you want to say next year that AI paid off, you don’t need a higher usage rate. You need a number from day-to-day business that has demonstrably shifted. And a name behind it that owns the result.
Frequently Asked Questions
Is 78 percent AI usage not a good sign?
As a signal of adoption, yes. The figure only indicates that a tool is being used, not that a process has been transformed to become more productive. It’s that second step that determines the economic benefit.
What does DIHK identify as the biggest hurdle?
Legal uncertainties. Costs, lack of know-how, and limited data access are cited less often than last year. The bottleneck is therefore governance, not money.
Where should a mid-sized company start?
With a single process that currently consumes time, plus a clear KPI and a designated owner for the AI outcome. Only when that metric moves should the next process be tackled.
What about shadow AI in the company?
It arises from a lack of clarity, not a lack of tools. When no one defines which data is permitted, employees continue to use AI unofficially. A clear approval rule is better protection than a silent ban.
Does this require a large budget?
No. Research shows the bottleneck is not money but organization and accountability. A well-anchored process beats an expensive tool without handover.
Image source: AI-generated (May 2026)
More from the MBF Media Network