KfW Study 2026: AI Makes Strong German SMEs Stronger

German Mittelstand companies that use artificial intelligence grow faster and earn higher margins than comparable companies without AI, even after size, sector, innovation activity and credit rating are taken into account. That is the result of an analysis published by KfW Research on 15 September 2026 based on the KfW SME Panel: the gap is 1.1 percentage points in revenue growth, 0.9 percentage points in return on sales and 3.4 percent in labour productivity. The study establishes a correlation without proving that AI causes it. It finds the lead in growth and margin only among companies that are already growing or already highly profitable, while the productivity lead can be measured across a broad span of the Mittelstand. For executive teams this means that AI widens existing strengths, provided it is embedded in workflows and the organisation. The dependable first return lies in productivity.

The KfW study compares Mittelstand companies with and without AI that are otherwise alike

KfW measured whether Mittelstand companies that used AI between 2022 and 2024 performed better in 2024 on revenue growth, return on sales and labour productivity than comparable companies without AI. The basis is the 23rd wave of the KfW SME Panel from spring 2025, with around 6,200 usable responses on growth, 4,400 on margin and 5,500 on productivity, reports KfW Research in Fokus Volkswirtschaft No. 558 by Volker Zimmermann.

The controls include headcount, age, sector, the reach of the sales market, employment of university graduates, innovation projects in the past three years and the credit rating. KfW treats the credit rating as an indicator of management quality, because long-term business success would not be achievable without good management. Growth is adjusted for inflation. KfW measures productivity as annual revenue per full-time equivalent. An additional quantile regression tests whether the relationship is the same for shrinking and growing, low-margin and high-margin companies.

AI users in the Mittelstand grow 1.1 points faster and earn a quarter more margin

On average the lead is 1.1 percentage points of growth, 0.9 percentage points of margin and 3.4 percent of productivity, even though 2024 was a weak year for the Mittelstand. According to KfW, a typical Mittelstand company using AI achieved a revenue gain of 0.4 percent, while a comparable company without AI lost 0.7 percent. Return on sales stands at 4.5 percent against 3.6 percent, so AI users keep a quarter more profit per euro of revenue.

Adoption has multiplied within a few years. According to Fokus Volkswirtschaft No. 533 of 11 February 2026, around 20 percent of Mittelstand companies used AI between 2022 and 2024, up from 4 percent between 2016 and 2018. How these figures relate to other surveys is discussed in AI adoption in German companies.

The AI lead is broadly proven in productivity, in margin for just under a fifth

The relationship is statistically proven for a broad group in productivity, only for companies with positive growth in revenue growth and only for a band of highly profitable companies in margin. According to KfW's quantile regression it counts as proven where the 90 percent confidence interval does not touch the zero line.

For revenue growth, the gap is consistently significant from the 60 percent quantile upwards, that is for companies growing faster than 60 percent of the sample. Growth in the sample turns positive at the 54 percent quantile, so the proven range begins almost exactly where companies start to grow. Within that range the gap rises from 0.7 to 7.6 percentage points in the top quantile.

For return on sales, the relationship is significant only between the 69 and 88 percent quantiles and from the 97 percent quantile upwards. The 69 percent quantile already corresponds to a margin of almost 11 percent, so according to KfW the proven lead applies to just under a fifth of AI users. For productivity, the significant range runs from the 2 to the 75 percent quantile, with gaps that include 13 quantiles between 52 and 75 percent. Above that, KfW finds no relationship and suspects trading companies whose revenue per full-time employee is high because of low vertical integration.

KfW concludes that AI is currently spreading company performance further apart. It suspects that growing companies use AI more often for marketing and new markets, shrinking companies more for rationalisation.

The KfW study establishes a correlation, while the cause remains open

The study does not prove causation, as KfW itself states. It refers to a 2023 paper by Daron Acemoglu and co-authors that asks whether a new technology has an effect or whether successful companies are simply more likely to adopt it. First, AI use and performance come from the same period, so a growing company may have adopted AI because it had the money and the reason to do so. Second, the degree of digitalisation is missing from the controls. According to the KfW analysis from February 2026, companies with a digitalisation strategy use AI with a probability of 35 percent, companies without graduates and without innovation with 8 percent. Part of the gap may therefore stem from the general level of digitalisation. Third, the credit rating reflects past success and captures the ability to introduce a new technology only indirectly.

The fact that the gap remains after the usual performance drivers have been removed points towards an effect. KfW also refers to the J-curve from the research literature: new general-purpose technologies first require investment in infrastructure, data and know-how, so performance may fall at first before the benefit becomes visible. According to KfW, many companies are unlikely to have reached the rising part of this curve yet. If that is the case, the 2024 figures understate the later benefit.

The KfW figures are based on booked revenue and profit, which sets them apart from self-assessments. According to a report by retail-news.de of 14 September 2026 on the German results of McKinsey's “The State of AI in 2026”, 43 percent of respondents from companies using AI cannot quantify its contribution to operating profit.

At EUR 20 million in revenue, the AI gap equals up to EUR 592,000 profit over three years

If the 2024 gap persists for three years, it amounts to around EUR 592,000 in additional profit for a company with EUR 20 million in revenue, more than nine tenths of it from the higher margin. Using KfW's model values, revenue with AI grows to EUR 20.24 million in the third year and falls to EUR 19.58 million without AI. Over three years the difference adds up to EUR 1.32 million in revenue.

The margin share of EUR 544,000 arises because 0.9 percentage points apply to the entire revenue of three years, while the growth share of EUR 47,000 rests only on the additional revenue. For a company of this size, the question of whether AI lowers costs or improves prices and quality therefore matters more economically than the question of growth.

For EUR 20 million in revenue, the company without AI needs around 89.8 full-time employees, with AI around 86.8. The difference equals three full-time positions that become available for additional orders or unfilled vacancies. This value must not be added to the revenue lead, because KfW measures productivity as revenue per full-time employee.

How much of this an individual company achieves cannot be quantified, because the three readings do not exclude one another and the study does not separate their shares. If selection explains the gap entirely, adoption alone brings nothing. The EUR 592,000 is therefore an upper limit against which adoption and operation can be measured. How to turn this into a robust business case is described in AI Business Case: Calculating Benefits and Costs Reliably.

The benefit of AI depends on data, skills and embedding in workflows

The benefit depends on embedding because KfW finds the lead where companies bring the prerequisites for use. According to KfW, a profitable deployment requires adapted IT systems, analysable data, skills in the workforce and often interdisciplinary teams of business experts and technologists.

The distribution of use fits this picture. Mittelstand companies with their own research and development use AI almost three times as often, at 53 percent, as companies without their own research, at 18 percent, because they have workflows into which AI can be fitted. KfW names the strategic direction and commitment of management as decisive factors for organisational implementation. Which prerequisites the executive team can shape is described in AI Readiness: Technology, Data and Organisation. Why initiatives without them get stuck in the pilot stage is shown in Why AI Pilots Fail to Reach Operations. The productivity gains that controlled studies have measured for individual tasks are summarised in AI Productivity Gains: What the Studies Show.

Mittelstand executive teams should measure AI first by its effect on productivity

Executive teams can adopt the KfW metrics as a benchmark for their own AI programme and start with productivity, because that is where the relationship is proven across the broadest span:

1. Record baseline values: on behalf of the commercial managing director, controlling collects revenue per full-time employee, return on sales and inflation-adjusted growth per business unit within four weeks, using the KfW definitions. The step is complete when the three values for the current and the previous year are available for every unit.
2. Select workflows with a productivity lever: within six weeks, the executive team and heads of department choose two to three high-volume workflows such as quotation, order entry or customer service. Success is measured in hours per transaction and lead time before and after adoption.
3. Organise the embedding: for each workflow, a head of department takes ownership and, with a team of business experts, IT and data owners, adapts the workflow, rules and training within three months. The metric is the share of transactions that then run in the new workflow.
4. Apply growth use cases where growth happens: over the following six months, the head of sales tests AI in sales and marketing for units that are already growing, because KfW proves the growth lead only there. Success shows in quotation lead time, win rate and revenue per sales employee.
5. Review against the benchmark after twelve months: the executive team and advisory board compare the values from step 1 with actual results and with the KfW gap of 3.4 percent in productivity and 0.9 percentage points in margin. A unit without AI or the trend of previous years serves as a comparison so that market developments are not mistaken for an AI effect.

Which metrics are suitable for ongoing steering is described in AI KPIs for Management Reporting. How to demonstrate the effect of individual initiatives is shown in Measuring the ROI of AI Projects.

Niehues Consulting connects AI management with adoption in the business units

Niehues Consulting supports executive teams and heads of department in building organisations that deliver reliably. Where AI enters the workflows, a business case and a prototype come first, followed by introduction in the business units. Organisation design creates the structures and decision paths into which AI is fitted. In delivery management, a delivery manager or product owner steers the programme until the change has arrived in operations.

The approach follows the Niehues Delivery Model with its phases Understand, Build and Introduce. Before an AI adoption it asks three questions. In Understand: where is the measurable productivity lever and who owns the workflow? In Build: which data, rules and roles does the solution need in operations? In Introduce: are users enabled and workflows adapted? Founder Marin Niehues, a management consultant from the delivery side, builds organisations that deliver. His projects range from venture-backed scale-ups to DAX 40 groups, with assignments in more than 20 countries on six continents. In the free 30-minute initial consultation you assess together in which workflows of your company AI can measurably raise productivity and which prerequisites are still missing: book a consultation.