AI adoption in German SMEs 2026: what studies measure

57%, 54.5% or 20%: the 2026 AI studies only seem to contradict each other. One table with sample, question and value, and what the figures mean for SMEs.

How widespread is AI adoption in German SMEs (the Mittelstand) in 2026? Depending on which study you read, the answer is 20 percent, 47.2 percent, 57 percent or 76 percent. None of these figures is wrong. They simply measure different things: different company sizes, different questions, different periods. Anyone who puts them side by side in a board paper without saying so is comparing apples with oranges.

This article puts the most important surveys of the past twelve months into one table, explains what each of them measures, and draws only the conclusions the figures support.

TL;DR

Adoption reached the majority in 2026. 57 percent of companies with 20 or more employees use AI (Bitkom), 54.5 percent according to ifo. Mid-sized firms are at 47.2 percent in the ifo survey, behind small and large ones.
Agents are still the exception, but in planning. Of the companies that use, plan or discuss AI, 11 percent already use AI agents and 29 percent plan to (Bitkom). 81 percent of the CIOs surveyed by Dataiku lack a full overview of agents that arise outside formal channels.
Skills, legal issues, security and integration are among the central barriers. 85 percent of companies without AI cite a lack of know-how (Bitkom). According to BCG, a clear strategy raises the measurable impact by 25 percentage points, better tools alone by about 5.

The figures at a glance

The table shows exactly one metric per row, with the sample it refers to. Read it row by row: a value applies only to the group named in its row.

Study	Date	Sample	Metric	Value	Source
Bitkom, AI survey	14 Sep 2026	603 companies with 20 or more employees, by telephone, calendar weeks 28 to 33/2026	use AI	57% (2025: 36%)	Bitkom
Bitkom, AI survey	14 Sep 2026	Companies that use, plan or discuss AI	use AI agents	11% (29% plan to)	Bitkom
Bitkom Research, digitalisation of the economy	11 Mar 2026	604 companies with 20 or more employees, by telephone, calendar weeks 2 to 6/2026	use AI	41% (48% plan or discuss)	Bitkom Research
ifo Institute	5 Jun 2026	ifo survey of companies in Germany, May 2026	use AI in business processes	54.5%; mid-sized firms 47.2%	ifo
KfW Research, Focus No. 554	29 Jul 2026	KfW SME Panel (turnover up to EUR 500 million, including micro-enterprises), survey in spring 2025	used at least one AI technology in 2022 to 2024	20%	KfW
Eurostat	11 Dec 2025	Enterprises with 10 or more employees, including self-employed, in selected EU economic sectors, data year 2025	use at least one of the AI technologies asked about	EU 19.95%, Germany 25.97%	Eurostat
techconsult for CANCOM and ServiceNow	2 Jul 2026	IT and business decision-makers from firms with 500 to under 2,000 employees, early 2026	use AI productively	76%	CANCOM
McKinsey, State of AI 2026	25 Aug 2026	1,719 respondents in 97 countries, 4 May to 8 June 2026	scale AI agents: organisations with turnover above USD 1 billion and smaller ones	40% and 22%	McKinsey
BCG, AI at Work 2026	3 Jun 2026	11,749 employees in 14 markets	agents are integrated into workflows	30% (previous year 13%)	BCG
Dataiku, CIO Edition 2026	24 Sep 2026	685 CIOs in eight countries, including Germany, 9 to 29 July 2026	lack a full overview of agents outside formal channels	81%	Dataiku

A note on our own archive: the article Bitkom 2026: 89 percent of German companies are on board with AI relies on the Bitkom survey from the start of the year. Its 89 percent adds 41 percent using AI and 48 percent of companies that plan or discuss use, so it measures engagement with AI, whereas the 57 percent from September counts only companies that actually use AI.

How many companies use AI

The most recent Bitkom survey considered here, from September 2026, shows that 57 percent of companies with 20 or more employees use AI. In 2025 it was 36 percent, in 2024 only 20 percent. A further 38 percent plan or discuss use, and only 4 percent still regard AI as a non-issue. Bitkom President Ralf Wintergerst sums it up: "Künstliche Intelligenz ist keine Sache einzelner Vorreiter, sie ist in der Breite der Wirtschaft angekommen." (unofficial translation: "Artificial intelligence is not the preserve of a few pioneers, it has reached the broad economy.")

Breadth is not depth, however. No company that uses AI says it exploits the potential fully. Only 3 percent say they make substantial use of its potential, while 28 percent make little use of it and 59 percent none at all. Wintergerst on this: "Für die meisten ist damit aber nur der Einstieg geschafft, die eigentliche Arbeit beginnt jetzt." (unofficial translation: "For most, however, that only means the start has been made; the real work begins now.")

The ifo Institute reaches a similar value with a different survey: 54.5 percent of companies used AI in their business processes in May 2026, compared with 40.9 percent a year earlier. AI is most widespread in industry (58.7 percent) and among service providers (56.2 percent); in trade it is around 45 percent. In construction the share rose from 7.1 to 39.8 percent within three years.

Why Bitkom has two different prior-year values

Anyone who looks closely stumbles over an apparent contradiction. The March 2026 Bitkom survey reports that 41 percent of companies use AI, compared with 17 percent the previous year. The Bitkom survey of September reports 57 percent and gives 36 percent for 2025. Both come from Bitkom, and both survey around 600 companies with 20 or more employees. But they are two separate survey series at different points in the year.

The lesson: compare developments only within one series. 17 to 41 percent is a statement, 36 to 57 percent is a statement. "17 to 57 percent" would not be.

Why Eurostat is so much lower

Eurostat reports for 2025 that 19.95 percent of EU enterprises with at least 10 employees use AI, and 25.97 percent in Germany. The survey covers enterprises with at least ten employees, including self-employed persons, in selected economic sectors; agriculture, forestry and fishing, mining and the financial sector, among others, are excluded. Denmark leads with 42.03 percent, ahead of Finland (37.82) and Sweden (35.04). Eurostat counts an enterprise as a user only if it uses at least one technology from a fixed list, such as text mining, speech recognition, natural language generation, image recognition or machine learning. In addition there are the data year 2025 and the lower limit of 10 employees. The values are therefore not directly comparable with Bitkom or ifo.

Where you stand

How AI-ready is your company?

Start the readiness check

How many already use agents

For AI agents the figures diverge even further, because each study counts a different unit: companies, organisations worldwide or individual employees.

Bitkom counts companies in Germany. Of the companies that use AI or plan or discuss its use, 11 percent already use agents, 29 percent plan to and 31 percent are discussing it. For 24 percent the topic is currently not on the agenda. Wintergerst: "KI-Agenten stehen heute etwa dort, wo Künstliche Intelligenz insgesamt vor drei Jahren stand. Sie werden sich aber ähnlich schnell verbreiten." (unofficial translation: "AI agents today are roughly where artificial intelligence as a whole stood three years ago. But they will spread at a similar pace.")

CANCOM and ServiceNow count the upper mid-market. In companies with 500 to under 2,000 employees, 21 percent use or are testing agent-based approaches, and 36 percent plan to introduce them in the next twelve months. The study was commissioned by the two vendors, which you should bear in mind when reading it.

McKinsey counts organisations worldwide that are scaling agents. About two in ten respondents are in the scaling phase with AI agents. Among organisations with turnover above USD 1 billion it is 40 percent, up from 27 percent the year before. Smaller organisations remain unchanged at 22 percent. About two in ten respondents also report scaling coding agents; the figure is 31 percent in larger companies. 32 percent of respondents report that their organisation has forgone buying at least one software product or feature because it was able to build the functionality in-house with agentic coding tools.

BCG counts employees. 30 percent of the employees surveyed say that agents are already integrated into their workflows, compared with 13 percent the year before. 61 percent consider it possible that within three years agents will do at least half of their work. At the same time, 52 percent have only a limited understanding of what agents actually are.

11, 22 and 30 percent therefore do not contradict each other. They answer three different questions. If you want to introduce agents in your own organisation, you will find the basics in our article on human-in-the-loop and agent autonomy.

What is different about the Mittelstand

For AI adoption, the Mittelstand sometimes trails other company sizes. In the ifo survey, 67.2 percent of large companies use AI, compared with 51.2 percent of small companies and only 47.2 percent of mid-sized companies.

KfW: 20 percent, but a different population

In Focus Volkswirtschaft No. 554, KfW arrives at a much lower figure: 20 percent of companies in the KfW-defined Mittelstand used at least one AI technology between 2022 and 2024. The KfW SME Panel covers private companies of all sectors with annual turnover of up to EUR 500 million, which includes micro-enterprises that do not appear at all in the Bitkom survey (20 or more employees). The survey was conducted in spring 2025 and asked about the period 2022 to 2024.

More interesting than the rate is what is used. The most common is natural language generation, for example chatbots or ChatGPT (14 percent), followed by text recognition (10 percent). Data analytics (3 percent) and AI for autonomous machines (1 percent) are rare. Businesses with fewer than 5 employees also use AI: 13 percent use language generation and 9 percent text recognition.

KfW's most important finding for you: company size matters less for AI use than the degree of digitalisation, existing digital know-how and general innovation activity.

KfW: AI users perform better, but unevenly

In Focus No. 558 of September 2026, KfW examined the relationship between AI use and business success in the Mittelstand, taking other success drivers into account. The performance indicators refer to 2024, AI use to 2022 to 2024. As a statistical association: AI-using companies grew their turnover in 2024 by 1.1 percentage points more on average than comparable companies without AI, achieved a return on sales 0.9 percentage points higher and labour productivity higher by EUR 7.6 thousand per full-time employee (about 3.4 percent).

The caveat comes right alongside. For growth and above all for return, the positive association is concentrated among companies that are already growing or profitable. KfW concludes that AI use currently leads to a wider spread in company performance. At the same time it treats the findings as a snapshot: in future, more companies are likely to benefit than can be measured today.

McKinsey: the gap between large and small is widening

Internationally, too, the large ones are pulling ahead. In the McKinsey study, the share of organisations with turnover above USD 1 billion that scale agents rose from 27 to 40 percent. Smaller organisations stayed at 22 percent. A Mittelstand company as defined by KfW (up to EUR 500 million annual turnover) is clearly below this threshold.

Where it fails: skills and control

Skills, legal issues, security and integration are among the central barriers in these studies.

Skills: know-how is missing, training is increasing

At Bitkom, 85 percent of companies without AI cite a lack of technical know-how as the reason. Next come legal uncertainty (66 percent), costs that are hard to calculate (63 percent) and lack of staff (51 percent). Companies that already use AI struggle mainly with data protection requirements (66 percent) and legal uncertainty (56 percent).

66 percent of companies rate their employees' AI skills as low, and only 28 percent as high. Still: 70 percent of all companies now train people in using AI, and among AI users it is 91 percent. But one in four companies (25 percent) offers no training at all, down from 43 percent the year before. Wintergerst: "Kompetenz entsteht mit dem KI-Einsatz, aber sie entsteht nicht von allein." (unofficial translation: "Skills develop with the use of AI, but they do not develop on their own.")

BCG shows where everyday work gets stuck. 42 percent of frontline employees who use AI regularly say they save at least one full working day per week, but 66 percent say they receive little or no guidance on what to do with that time. 47 percent spend more time managing AI than doing the actual work. According to BCG the strongest lever is not the tool: a clear strategy raises the measurable impact by 25 percentage points, better tools alone by about 5. Where workflows are redesigned, the likelihood of a measurable improvement is 24 percentage points higher.

In the upper mid-market, the CANCOM study sees the barriers distributed somewhat differently: IT security and risk management (38 percent), costs and budget (28 percent), lack of expertise (25 percent), data sovereignty (23 percent) and regulatory uncertainty (22 percent). Only 26 percent have fully integrated AI into their core processes, 49 percent pursue isolated solutions in individual departments, and 16 percent use uncontrolled shadow AI. How to get shadow AI under control is covered in our article on shadow AI and governance.

Control: agents arise faster than IT can keep up

The Dataiku survey of 685 CIOs in eight countries, including Germany, shows the flip side of the agent wave:

84 percent agree that employees build AI agents and applications faster than IT can govern them.
81 percent lack a full overview of agents that arose entirely outside approved systems or formal channels.
83 percent have no standardised lifecycle management for agents.
72 percent cannot consistently confirm whether agents deliver the business results they were built for.
47 percent have already decommissioned more than 20 agents this year.

This matches what Bitkom President Wintergerst said about agents: "Ein Agent, der eigenständig handelt, braucht verlässliche Daten, verbundene Systeme und Regeln, was er darf und was er zu lassen hat." (unofficial translation: "An agent that acts independently needs reliable data, connected systems and rules on what it may do and what it must leave alone.")

Free AI policy with an agent annex

Word template in German or English, with an agent register, approval levels and human oversight.

Such rules do not have to start from zero. In Annex A, our template contains an agent register, approval levels for agent actions and sections on connectors, human oversight and logging.

What this means for the next 12 months

Four points follow from the studies. They are not forecasts but rest directly on the figures above.

1. From entry to depth. More than half of companies use AI, but 59 percent of users say they do not exploit the potential at all (Bitkom). The task for the next twelve months is therefore not the first tool but the step from individual experiments to a fixed process. Our article from AI pilot to AI programme explains how to make that transition.

2. Rules for agents before the rollout. Among companies that use, plan or discuss AI, 29 percent plan to use agents (Bitkom), and in the upper mid-market 36 percent want to start within twelve months (CANCOM). At the same time, 81 percent of the CIOs surveyed already lack a full overview of agents that arise outside formal channels (Dataiku). For us it follows that anyone planning agents should first set up a register, approval levels and clear owners, for example with our AI policy template.

3. Budget for skills, data and integration, not just licences. For the AI users at Bitkom, infrastructure (51 percent), data preparation (50 percent) and integration (41 percent) are the big cost items, licences only for 21 percent. Wintergerst: "Wer nur eine Lizenz kauft, hat KI eingekauft, aber noch nichts verändert." (unofficial translation: "Anyone who only buys a licence has bought AI but has not yet changed anything.") BCG points the same way with its strategy lever, and KfW explicitly names targeted upskilling of employees as a starting point. Legally, Art. 4 AI Act requires measures to build AI literacy, but no particular level, no prescribed training and no certificate; the Federal Network Agency (Bundesnetzagentur) recommends documenting the measures well. Details are in our articles on AI literacy under Art. 4 and the Digital Omnibus and on cost and funding of AI training.

4. The gap is currently widening. In the ifo survey, mid-sized firms are behind small and large ones, McKinsey sees large organisations pulling away in scaling agents, and KfW observes a wider spread in performance, mainly in favour of AI users that are already growing or profitable. KfW calls this a snapshot. From our point of view it still argues for starting now with a clearly defined use case rather than postponing the topic to next year.

If you want to know where your company stands in this picture, take the free AI readiness check: 12 questions, about 5 minutes, no sign-up. For teams that want to start right away: EUR 599 per licence, one-off. Training plus Cowork and Claude Code. We are happy to clarify questions in a 30-minute call.

FAQ

How many Mittelstand companies use AI in 2026?

That depends on how the Mittelstand is defined. According to ifo, 47.2 percent of mid-sized companies used AI in May 2026; in the upper mid-market with 500 to under 2,000 employees, 76 percent use AI productively according to the CANCOM study. The KfW SME Panel, which also covers micro-enterprises, arrives at 20 percent for the period 2022 to 2024.

Why do AI studies produce such different figures?

Because they measure different things: the population ranges from firms with 10 or more employees (Eurostat) through firms with 20 or more employees (Bitkom) to all Mittelstand companies including micro-enterprises (KfW). In addition there are different questions, different periods and different units, since BCG surveys employees and not companies. Therefore compare developments only within the same survey series.

Which sector uses AI the most?

In the EU, according to Eurostat, enterprises in information and communication use AI most often in 2025 (62.52 percent), followed by professional, scientific and technical services (40.43 percent). In Germany, industry leads in the 2026 ifo survey at 58.7 percent, just ahead of service providers at 56.2 percent. In the 2025 ifo survey, advertising and market research (84.3 percent) and IT service providers (73.7 percent) were at the top.

Why do large companies use more AI than small businesses?

Eurostat measures a clear gap in 2025: 55.03 percent of large, 30.36 percent of medium-sized and 17 percent of small EU enterprises use AI. As possible explanations Eurostat names the complexity of adoption, economies of scale and cost. For the German Mittelstand, KfW stresses that the degree of digitalisation, digital know-how and innovation activity matter more than size alone.

What are the biggest barriers to AI in the Mittelstand?

For companies without AI, lack of technical know-how is at the top (85 percent), followed by legal uncertainty (66 percent) and costs that are hard to calculate (63 percent), according to Bitkom. Those who already use AI cite mainly data protection requirements (66 percent) and legal uncertainty (56 percent). In the upper mid-market, IT security and risk management lead with 38 percent (CANCOM).

Is AI worthwhile for companies at all?

The data show both. According to KfW, AI use and business success go together in the Mittelstand: AI-using Mittelstand companies grew by 1.1 percentage points more on average in 2024 than comparable companies without AI and were about 3.4 percent more productive (a statistical association, not proof of cause and effect). Two thirds of the AI users at Bitkom (66 percent) have improved their competitive position. At the same time, in the Bitkom survey from the start of the year, one third (33 percent) of AI-using companies report significantly higher costs than expected, and at McKinsey 80 percent of respondents see individual productivity gains but only 37 percent see an impact on company earnings (EBIT).