The “New German AI Wave” in SMEs

The first wave of AI in German SMEs was an assistant that made the people already on staff faster. The second wave will be agents that work in positions that would never have been filled anyway – and that quietly changes what a bank should finance.

The number is small enough to overlook. According to the Institute for Employment Research (IAB), the research institute of Germany’s Federal Employment Agency, 2026 is the first year in which the country’s potential labor force – everyone in work plus everyone who could be in work – actually shrinks. The projected decline is 35,000 people, a mere rounding error in a pool of roughly 46 million.

What matters is the direction. For two decades, Germany’s demographic arithmetic was a forecast. This year it has become a data point, and the projections behind it are anything but kind. Without immigration and without further gains in labor force participation, the IAB projects that the labor supply would shrink by about seven million people by 2035, or roughly 15 percent. Even in the baseline scenario, which assumes continued immigration and rising participation rates among women and older workers, the pool falls to about 43.9 million by 2035 and to 40.4 million by 2060. The institute has calculated that only net immigration of around 400,000 people a year would keep the labor supply constant – far above the long-term average and, by the IAB’s own assessment, hardly achievable for long, because the countries that used to send workers to Germany are now aging as well.

Against that backdrop, the good news of 2026 reads rather differently. The KfW-ifo Skilled Labor Barometer found that in April, 22 percent of small and medium-sized enterprises said a shortage of skilled workers was hampering their business; among large companies, the figure was 19.3 percent. In manufacturing the gap was even wider – 18.3 percent of mid-sized firms compared with 8 percent of large ones. KfW chief economist Dirk Schumacher was blunt about why the pressure has eased: because of the weak economy, not because of a solution – and he expects the shortage to tighten again as soon as the economy recovers.

That is the setting in which German SMEs will make their next decisions about artificial intelligence. And it is the reason why the next wave of AI adoption in small and mid-sized German companies will follow a different logic than the first one did. The first wave was about making the employees a company already has faster. The second will be about work for which no employee is going to come anymore.

The first wave was an assistant

In February, KfW Research published its most thorough study to date of AI in SMEs. Twenty percent of small and medium-sized enterprises used AI in the period from 2022 to 2024, compared with four percent in the years 2016 to 2018 – a fivefold increase, or just under 780,000 companies in absolute terms. Size matters: 36 percent of companies with more than 50 employees use AI, but even among micro-enterprises with fewer than five employees the figure is 19 percent. The pioneers are companies with their own research and development, along with knowledge-intensive services and R&D-intensive manufacturing.

But look at what they actually use AI for, and the picture is narrower than the headline suggests. The most widespread applications are broad and easy to deploy: natural language generation, used by 14 percent of companies, and text recognition, used by ten percent. The demanding, less generic uses – data analytics at three percent, AI for the autonomous movement of machines at one percent – barely register. Across all technologies, the KfW analysts found the same determining factors: the level of digitalization, the digital know-how inside the company and its general capacity for innovation decide whether AI is used at all.

The Bitkom survey of 604 companies with 20 or more employees, published in March, shows the same shape from a different angle. Active AI use in this group more than doubled within a single year, from 17 to 41 percent, and another 48 percent plan to start. But 77 percent name data protection requirements as an obstacle to digitalization, and 70 percent name the skilled labor shortage itself – a shortage of exactly the people who are supposed to introduce the technology that is supposed to ease the shortage. Most revealing of all: 61 percent say they do not make full use of the data they already have, and only 32 percent believe they are using their data’s potential.

The Institute for SME Research in Bonn (IfM), which published a study on AI and the demand for skilled workers in March, describes what this amounts to in practice. In small companies today, AI mainly relieves employees of time-consuming and resource-intensive routine work. Tax advisory firms automate their bookkeeping and the processing of receipts; painting contractors use robots to prime large surfaces; trade businesses use digital assistance systems for route planning, answering the telephone and workflow. This relief, according to the IfM, can already reduce the problems of filling vacant positions. But adoption depends above all on the owner’s own digital skills and affinity, and many management teams are uncertain which of the many applications available actually fit their own business.

This is the era of the assistant: a tool that sits next to a person and makes that person faster. It is real, it is spreading, and it is not the part that is interesting.

The second wave does the job

An agent differs from an assistant in kind. It does not draft the payment reminder that a clerk then sends out; it runs the entire collections process – it checks the claim, applies the payment terms, sends the reminders and escalates the exceptions to a person. It does not summarize the supplier’s quote; it requests the quotes, checks them against the specification and prepares the purchase order. The assistant multiplies a person. The agent fills a position.

In an economy with a growing labor supply, that difference would be politically explosive and commercially of minor interest. In an economy with a shrinking supply, it is the whole point. The IfM’s Future Panel found that four in ten companies have difficulty finding qualified staff despite the weak economy. When the bookkeeper of a machine builder with 40 employees in the Sauerland region of western Germany retires and the position stays open for eighteen months, the company has not chosen automation over people. It has chosen automation over nothing at all. Nobody came.

This reframing matters for how SMEs will decide. The IfM found that some owners already use AI deliberately to become more attractive to young talent – the technology as an argument for hiring, not as a replacement. Agents carry this logic one step further: the scarce, qualified people a company can still attract – the master craftsman, the engineer, the site manager – spend less of their week on administration that a machine can handle, and workforce planning stops pretending that the back office will ever be fully staffed again.

The demographic floor also explains why this wave will not be a voluntary one. A company can turn down a new productivity tool. It cannot turn down the retirement of its own workforce.

Where the agents land first

The first agents in a mid-sized company will not be inside the product. They will work in the back office, and the reasons for that are structural, not fashionable.

Agents work where processes are documented, data is structured and results are measurable – and where a system already sets the boundaries of the task. In a German SME, that description fits accounting, order processing, purchasing, collections, scheduling and first customer contact, which mostly already run in an ERP system and in the DATEV ecosystem through which the bookkeeping of most companies reaches their tax advisor. It fits the shop floor far less well, which is why the KfW figure for AI in autonomous machine movement stands at one percent and is likely to stay there for years.

Two things are quietly laying the foundation for this. The first is the e-invoicing mandate. Since January 2025, German companies have had to be able to receive structured electronic invoices, and the obligation to issue them phases in from 2027 for larger companies and from 2028 for everyone. A structured invoice can be processed by machine in a way a PDF cannot; an agent can read it, match it against the purchase order, book it and schedule the payment. The e-invoicing mandate was sold as a tax issue. In fact it is the agent readiness program for the German back office, and almost nobody describes it that way.

The second is that Bitkom number: 61 percent of companies do not use the data they already have. That is less an obstacle for agents than a map of where the gains are to be found. The companies that clean up their master data, define their processes and connect their systems before 2027 will be the ones for which an agent is a matter of configuration rather than a project. The others will discover, as in every previous wave, that the technology was never the hard part.

This is operating expense – and that is a problem for the bank

This is the point where the story leaves the workshop and arrives on the desk of the commercial banker.

In April, KfW Research published a second study that received less attention than it deserved. Bank loans remain the most important source of external financing for German SMEs – and yet only 27 percent of small and medium-sized enterprises say they would consider taking out a loan from a bank or savings bank to finance an investment. In 2017 the figure was 66 percent, and in 2023 it was still 42. Among micro-enterprises with up to ten employees, the willingness to borrow has collapsed from 69 to 23 percent. Service companies bring up the rear at 21 percent. Only among the larger SMEs, where more than half would still consider borrowing, has the figure held up.

The reasons are as revealing as the number itself. Sixty-three percent of companies that rule out borrowing want to avoid debt, 23 points more than in 2017. Half do not seek outside financing as a matter of principle, compared with 15 percent nine years ago. Thirty-six percent say they have enough funds of their own. And 30 percent complain that banks demand too much disclosure – almost twice as many as in 2023. KfW itself concludes that investment needs of SMEs remain high, that internal funds will not cover them and that growth will suffer if companies keep refusing outside capital.

Now lay the agent wave on top of all that. The classic instrument of SME financing is the investment loan: a machine, a production hall, a vehicle fleet, financed over its useful life and secured by the asset itself. Agents are not a machine. They are a monthly, usage-based operating expense with no asset that could be pledged as collateral and no useful life over which it could be depreciated. Leasing does not fit either. And the companies that need them most urgently – the micro and small businesses whose bookkeeper will not be replaced – are exactly the ones that are least willing to borrow and most allergic to disclosure.

So the bank finds itself in a strange position. Its customers face the most consequential operational change in a generation; the change shows up in their cash flows before it shows up in their balance sheets; and the bank’s product shelf has nothing that is shaped for it. Financing the agent decade will have to take the form of a working capital facility: tied to measurable process outcomes rather than to collateral, sized to monthly spending rather than to a purchase price, and probably bundled with what the bank already has and no one else does – the account data. A banker who sees a customer’s payroll lines flatten while a new software debit appears every single month is watching an agent deployment in real time. That information advantage is worth more than any subsidy program, and it is the reason why the house bank, and not the venture capital market, will end up financing most of this.

The counterargument

Fairness requires the other side, and it deserves to be taken seriously.

The IfM’s panel of experts, surveyed in the spring, sees securing innovation and competitiveness as the greatest challenge facing SMEs and digitalization and AI as its precondition – but it also names cyber risks and dependence on external providers as dangers that companies and economic policy have to keep an eye on. The point about dependence is not an abstract one. The agent that runs a company’s collections runs on someone else’s model, in someone else’s cloud and on someone else’s terms, and the SME instinct for financial and operational independence, so visible in the KfW lending data, will apply to that as well.

Nor does the skills problem disappear; it simply shifts. The IfM’s finding: because AI changes job profiles, the demand rises for employees with the right qualifications, and vocational training and continuing education have to catch up quickly, or small businesses will lose the race for the people who are able to operate the machines. Since February 2025, the EU’s AI Act has required every company that uses AI to ensure sufficient AI literacy among its staff – a modest obligation on paper, a very real one in a business with 30 people and no IT department.

And the productivity gains are, for the time being, mostly claimed. Twenty percent of companies use AI; three percent use it for anything analytically demanding. An agent wave on such a narrow base could stall for the same reasons as earlier digitalization programs did – the federal programs go-digital and Digital Jetzt have since expired – with consultants and grants standing in for the process discipline that is missing.

All of that is true. None of it changes the demographic floor. A skeptic can argue that agents will disappoint and be right about the timing. What a skeptic cannot argue is that the retired bookkeeper will be replaced.

Three things to watch

The first is 2027. When the obligation to issue structured invoices takes effect for larger companies, the back office of German SMEs becomes machine-readable by law. That date, and not any product launch, is the moment at which agents in accounting and order processing stop being a pilot project.

The second is succession. The IfM estimates that around 186,000 German companies face a handover in the next five years because their owners are reaching retirement age. Whoever buys a mid-sized company in 2028 will assume that its back office runs on agents and will price the business accordingly – and the bank that finances that succession prices in the same thing, whether or not it already has a name for it.

The third is the 27 percent. If agents do what their advocates claim, they make the cash flow of a small company more predictable: fewer vacant positions, fewer late invoices, fewer errors in the order book. Predictable cash flow is exactly what a bank lends against. The question for the German banking system is whether it can build a product for that before its own research department reports next spring that the number has fallen yet again.

SMEs never bought technology because it was impressive. They bought continuity – the certainty that the business will still be running in twenty years, in the same family, in the same place. Agents are the first technology whose honest promise is continuity. That is why they will be adopted, and that is why the conversation about them belongs less at a tech conference than in a bank.

---

Sources and further reading

• IAB projection: potential labor force shrinks for the first time in 2026, as reported by INSM, October 2025
• IAB: The potential labor force will shrink by 11.7 percent by 2060, May 2023
• IAB: Only with annual net immigration of 400,000 people will the labor supply remain constant in the long term, April 2022
• IAB-Forum: How a demographic shrinking of the labor market can still be averted, January 2024
• KfW Research, Focus on Economics No. 533: Use of artificial intelligence in SMEs, February 11, 2026
• KfW press release: Artificial intelligence is being used more and more often in SMEs, February 11, 2026
• KfW Research, Focus on Economics No. 543: Willingness to use bank loans continues to fall, April 2026
• KfW press release: Fewer and fewer mid-sized companies want to take out a loan, April 14, 2026
• KfW-ifo Skilled Labor Barometer, as reported by UnternehmerJournal, June 2026
• IfM Bonn, IfM-Materialien No. 312: Opportunities of artificial intelligence for meeting the demand for skilled workers in SMEs, March 2026
• IfM Bonn: How artificial intelligence is changing the demand for skilled workers, March 2026
• DATEV magazin: How artificial intelligence is changing the demand for skilled workers, March 17, 2026
• IfM Bonn, expert survey: The future of SMEs 2026, June 2026
• IfM Bonn, press releases 2026 (succession, Future Panel): presseportal.de/nr/160935
• Bitkom survey of 604 companies with 20 or more employees, March 2026, as reported by Münker & Partner