German SMEs are losing too many hours to administrative work while many of the people who hold their expertise approach retirement. The evidence makes a practical case for putting AI to work before that knowledge leaves the business.
KfW Research found that German SMEs spend around 7% of their working time meeting statutory requirements. That is 32 hours per company every month, 1.5 billion hours across German SMEs each year, and an estimated annual cost of €61 billion. The representative panel included around 10,000 companies.
Not every statutory task is pointless or automatable. KfW identifies tax, record-keeping, documentation, and accounting requirements among the biggest sources of effort. Many contain structured, repeated, reviewable steps that make them sensible candidates for AI support, although the study does not assess automation potential.
One model puts the opportunity in the trillions
The Institute for Employment Research modelled what stronger AI development and integration could mean for Germany. In its AI scenario, annual economic growth averages 0.8 percentage points above the reference scenario, adding €4.5 trillion in value over 15 years.
That is a scenario, not a promise. It depends on investment and companies changing how work moves. Buying licences is not enough.
Total employment could remain close to the reference scenario even as around 1.6 million jobs move between parts of the economy. SMEs need to redesign work early enough to use scarce people where they add the most value.
German companies expect productivity gains
In the ifo Institute’s 2024 Business Survey , 70% of German companies expected AI to improve their productivity over the next five years. Based on their estimates, ifo calculated an average economy-wide gain of about 8%. These are expectations, but they come from businesses assessing their own processes and staff.
An Institut der deutschen Wirtschaft study found that nearly four in ten companies with established AI applications believed labour productivity had increased. Among employees who had used AI for some time, 45% reported better work performance. These effects are self-reported, but they add operating experience to the scenario estimates.
Workplace field evidence shows what good implementation can do
A peer-reviewed field study by Stanford and MIT researchers followed 5,172 customer-support agents during an AI-assistant rollout. Issues resolved per hour increased by 15% on average, with the largest gains among less experienced workers.
Small and medium-sized companies often hold much of their knowledge in a few experienced people. New colleagues learn it slowly through examples, corrections, and questions. The field study suggests a mechanism worth testing: a well-designed assistant can make approved answers and proven working methods available at the moment of need.
The expert remains accountable while the rest of the team benefits from proven knowledge.
The result cannot be applied to every occupation. AI performs best when the task is clear, context is available, and the output can be checked.
Capture the knowledge before people retire
German SMEs face two retirement problems at once. KfW Research reports that 57% of SME owners are at least 55 years old. Of those planning to step down by the end of 2029, 569,000 do not plan for their company to continue, equivalent to 114,000 closures a year.
The workforce is shrinking at the same time. Destatis calculates that 13.3 million economically active people will pass the statutory retirement age by 2040. That is 30% of the people currently available to the labour market, and younger age groups will not fully replace them.
For companies that will continue operating, AI offers a practical response to workforce contraction and knowledge loss. It can help them embed deep domain knowledge in AI-supported processes.
That knowledge has to be captured before the people who hold it retire. Experienced people need to help design and test these processes while they are still in the business, including how they judge exceptions, diagnose problems, and use context that was never written down. The goal is for a smaller group to take over more of the work without training one replacement for every retiree. Companies will have to prove that this works through repeated testing, correction, and use on real work. They need to start while the experts are still there.
Large companies report higher AI use
According to Destatis , 26% of German companies with at least ten employees used AI in 2025. The rate was 23% among companies with 10 to 49 employees, 36% among those with 50 to 249 employees, and 57% among large companies.
Higher adoption gives larger companies more opportunities to learn which data helps, which outputs need review, and where integrations fail. A smaller company can often implement a focused workflow faster, but only if it starts.
Waiting for models to become perfect misses the point. Models will keep changing. The durable advantage is knowing how to define the workflow, set its limits, and measure the result.
German SMEs should act now
The Federal Ministry for Economic Affairs and Energy reports that SMEs account for more than 99% of German companies and around 55% of net value added. A large share of the predicted AI productivity gains will therefore be won or lost across millions of SMEs. Germany cannot capture the full economic opportunity described in the studies above without them.
Germany’s past economic success grew from this broad base. Specialised, often family-owned SMEs turned close customer relationships, long-term investment, and fast decisions into export strength and global leadership in narrow markets. The ministry still points to short decision paths and flexibility as defining strengths of SMEs.
General-purpose AI models will be available to every company. SMEs own something harder to copy: detailed knowledge of machines, processes, and customer problems. TRUMPF CEO Nicola Leibinger-Kammüller put it plainly:
After all, only those who know how a machine works down to the last screw can make effective use of AI in production.
This is where German companies can turn a widely available technology into a specific competitive advantage.
Large companies have more capital, but their scale and layers can make them slower to redesign daily work. Companies that can identify a bottleneck, test a new workflow, and adapt quickly have an advantage. SMEs are where the real action will be.
Delay is easy to justify. Data protection, works council concerns, data sovereignty, and supplier questions all deserve proper answers. They should prompt an investigation, not end one. Our article AI and Data Protection: The Cost of Waiting explains how companies can separate everyday, controlled, and sensitive work, then mitigate specific risks instead of treating them as a general ban on AI.
Build the strategy through targeted experiments
SMEs need a systematic AI strategy. It does not have to begin with a large transformation programme. Start with a targeted experiment and ask why an operational bottleneck exists:
- Why does this bottleneck exist?
- Does the constraint still apply when AI can retrieve, draft, check, classify, or coordinate the work?
- Where could a small, low-risk change produce a measurable result quickly?
- Which KPI should move, and what is its current baseline?
- How will the team measure the result, review failures, and decide whether to stop, redesign, or scale?
Choose one repeated workflow, name its owner, and set a six-week target. Measure cycle time, staff hours, delays, error rates, or another business outcome. Keep sensitive decisions with a named person. If the experiment creates no meaningful improvement, stop or redesign it. If it works, embed the proven steps, exception handling, and review rules in the workflow, then apply what the team learned to the next bottleneck.
Each experiment builds operating experience. SMEs have the agility to lead this shift, but they need to start before experienced people retire and take expertise that has not yet been embedded in the business with them.
Start with the bottleneck costing your team the most time. If you want help turning it into a measured first experiment and a practical AI roadmap, book a Roadmap Agentic Review .