Skip to content
Chat with AI Agent
Business & governance

Swiss SMEs Use AI, but Not to Change the Business: Reading the 2026 Data

Orange ITS — AI engineering team 9 min read

Take an illustrative case: you run a 45-person engineering supplier near Winterthur. Your office staff use an AI assistant to translate quotes and tidy up emails, someone in marketing tries it for advertising copy, and nobody has decided what the company wants from it. When a supplier association asks whether you “use AI”, you answer yes, and you are telling the truth.

That honest yes is the number most Swiss adoption headlines are built on, and it says little about whether AI has changed how a firm earns money. This article reads the 2026 Swiss surveys closely: what each one measured, who answered, who paid for it, and where the evidence stops. If you are already past reading and want to decide where AI should start in your company, our AI strategy service is the place to begin.


What the headline 74% actually counts

The most cited recent figure comes from the AXA KMU-Arbeitsmarktstudie 2026, published on 15 September 2026. Sotomo ran the fieldwork between 11 and 18 March 2026 with 336 SMEs in German- and French-speaking Switzerland. AXA is an insurer with a commercial interest in the SME market, and the release does not state the size threshold or the sector mix of the sample.

The result splits three ways:

  • 35% integrate AI deliberately into their business processes
  • 39% are testing the technology
  • 26% do not use it yet

Add the first two and you get 74%, which is the figure that travels. The release gives the total only as “about three in four”; it is a sum of two categories that describe quite different situations. A firm with an AI assistant in the finance workflow and a firm where one employee tried a chatbot last spring both land in the 74%.

The earlier editions use the same three-way question. In the 2025 edition (300 SMEs, fieldwork 3 to 10 March 2025), 34% integrated AI deliberately, 37% were testing and 29% were not using it, which makes 71% in total. The 2024 edition (300 SMEs with five or more employees) put the combined share at 55%, of which 22% were integrating deliberately.


Up from 2024, level since 2025

Read as a trend, the three AXA releases show a jump and then a plateau. Use-or-test went from 55% in 2024 to 71% in 2025 and 74% in 2026. Deliberate integration went from 22% to 34% to 35%. AXA’s own text describes the 2026 intensity as practically at the previous year’s level.

Two cautions apply before anyone calls the 2025-to-2026 change growth. With samples of 300 to 336, a simple random sample would carry a margin of roughly five percentage points either way (our arithmetic, since the releases publish none), so 71% against 74% sits inside the noise. And the 2024 sample was defined as firms with five or more staff. The 2025 release implies the same floor, but the 2026 release does not restate the rule, so we cannot confirm that all three samples match.

The fair summary is that adoption rose quickly until 2025 and the latest data gives no reason to say it is still rising. A third of SMEs working with AI deliberately is a real base. It is also a lower number than the headline implies.


What Swiss SMEs do with AI

The uses are the more informative part of the AXA release. In 2026 the most common were translation (47%), correspondence (42%), advertising texts (35%) and optimising work steps (35%). Customer service and support reached 17%. The 2025 edition put data analysis at 32%, which the 2026 release restates as 34%.

Look at the shape of that list. Translation, letters and ad copy are tasks where one person asks a tool for help and moves on. They are valuable, and they are easy to start with. They also leave the surrounding process untouched: the same people, the same approvals, the same handoffs.

Optimising work steps is the closest item to process work, and it rose from 23% in 2024 (as restated in later releases) to 35% in 2026. It shows that a growing minority of SMEs look at how work flows. It is still a minority use, and the survey does not say what “optimising” meant in each firm. Figures for the same use also move between releases (translation was 48% in 2024, 52% in 2025 and 47% in 2026), so single-year differences deserve little weight.


AXA’s own reading, and what it can carry

Kathrin Braunwarth of AXA draws a conclusion from the data: AI optimises existing processes but does not fundamentally question the business model. The release adds that 31% of SMEs think their business model needs adjusting and that nearly two thirds see little or no need. By size, 42% of large firms and 23% of small firms say the model needs to change.

This is AXA’s interpretation of its results. The survey asked about perceived need for change and about uses. It did not ask whether firms rebuilt a process around AI, so the sentence is a reasonable reading that the data supports in direction only. We find it plausible, and we flag it as a reading.


Why the other surveys give other numbers

Anyone searching for the Swiss figure will find several, and they look like they contradict one another. They answer different questions.

SurveyYear and sampleSponsorWhat it measuresHeadline
AXA / Sotomo2026, 336 SMEsAXA (insurer)Use or test74% (35% integrated)
UBS Outlook SchweizReleased 19 May 2026, about 2,500 firms of all sizes, fieldwork by IntervistaUBS (bank)AI in use at the firmabout 60%
EY AI Survey2026, 604 respondents, mostly from large firmsEY (consultancy)Daily use of AI solutions at work89%
localsearch / HSLU KMU Digital Pulse2026, 504 marketing decision-makers at SMEslocalsearch (directory publisher), with HSLUAI tools in marketing21%

The EY sample deserves a second look. Only about 31% of its respondents come from firms with fewer than 250 staff, by our arithmetic on the reported shares, and 30% come from firms with more than 10,000. The 89% therefore describes mostly large organisations, and quoting it as an SME figure is a mistake some portals make. Within the same release, 9% of respondents say AI has transformed their business model.

UBS reports about 60% for firms of all sizes. The localsearch figure covers one function, marketing, and rises with size: 49% of medium firms, 39% of small and 18% of micro firms use AI tools there, according to the release as reprinted by Handelsverband. None of these numbers is wrong. They are answers to separate questions, so we quote one source per claim and never average them.


Efficiency is common, revenue is not

Two further studies speak to depth rather than breadth. The ti&m and HSLU AI Maturity Study, published in February 2026, reports around 75% of organisations with productivity improvements and describes the outcome as efficiency gains with little revenue growth. ti&m is an IT consultancy that sells AI services, and it co-authored the study. The study covers more than 200 companies from more than 15 industries.

The NZZ-KMU-Barometer 2026 points the same way. According to Netzwoche’s report on it, 87% of 495 managers are experimenting with AI or building skills, while only 27% use data systematically for strategic decisions. The barometer is commissioned by NZZ and run by Kalaidos, and its sample is managers rather than firms.

Put the three findings together carefully. They come from different samples and different methods. What they share is direction: broad activity, narrow depth, and gains that show up as saved time rather than new income. That is consistent with AXA’s reading and stops short of proving it.


What holds firms back

The barrier data is patchy, and the sources again differ in population. Three results are useful for an SME owner.

Know-how. The ZKB and ZHAW KMU ZH Monitor 2026 surveyed 1,287 SMEs in the canton of Zurich between 9 March and 9 April 2026. 40% of firms cite a lack of know-how, rising to 51% among firms that find recruiting difficult. The survey covers Zurich only and publishes attitudes more than usage shares.

Data protection. In the same ZKB study, 46% of medium-sized firms raise data-protection concerns, against 38% of smaller ones. AXA’s 2025 edition found that only about a third of AI-using SMEs had clear rules on what data may go into AI tools. That gap between concern and written rules is the practical problem. If you want the legal side, our article on AI agents and the Swiss data protection act covers what the law requires.

Trust in output. AXA’s 2026 release says worry about hallucinations and uncontrolled AI behaviour still seems large. The ti&m study names hallucinations, data protection and data loss among the biggest concerns. Neither source gives a percentage for hallucination as a barrier, so we cannot say how many firms hold back for that reason.

EY’s barrier list (data quality and silos at 20%, security and privacy at 19%, shortage of qualified staff at 18%) comes from the large-firm-skewed sample and should not be read as the SME profile.


What the data cannot tell you

Three limits matter before anyone builds a plan on these numbers.

  1. No Swiss survey measures process redesign. None of the studies above asks whether a firm rebuilt a process around AI, or what that did to cost, speed or revenue. The claim that redesign is the open advantage for Swiss SMEs is therefore an argument. It rests on international work, such as McKinsey’s finding on workflow redesign among high performers, which is an association and not proof of cause.
  2. Official statistics give a population figure. The Federal Statistical Office reports that 47% of the Swiss population used generative AI in its 2025 Omnibus survey. That counts individuals, so it says nothing about what firms do.
  3. Most of the sponsors sell something. An insurer, a bank, a consultancy, an IT provider and a directory publisher each have a reason to find that AI matters for SMEs. That does not make the fieldwork unreliable. It is a reason to read the wording of the question before the headline.

If the redesign argument turns out to hold, the open position favours smaller firms: with a few dozen staff, one owner can change how a process runs in weeks rather than quarters. If it does not hold, a firm that tried loses a few weeks on one process. The asymmetry is the case for testing it.


What an owner can do with this

You do not need a firm-wide programme to move from the 74% into the 35%. The AXA split suggests that deliberate integration is mostly a decision rather than a technology purchase. A workable first step looks like this:

  1. Pick one process. Choose something that runs weekly or daily, has a clear owner and produces a measurable output, such as quote handling, invoice entry or support triage. Our article on the AI readiness assessment helps with the shortlisting.
  2. Take a baseline before you change anything. Record cycle time, error rate and cost per case for a few weeks. Without it you will not be able to say whether the AI helped.
  3. Name the owner. One person is accountable for the result of the process, whichever tool sits inside it.
  4. Write the data rules. Say what may and may not go into AI tools, in one page. This closes the gap AXA found in 2025.
  5. Decide what stays with a person. Fix where a human reviews, and what happens when the output is wrong.

For the difference between adding a tool and rebuilding the work around it, see AI-native versus AI-enabled. If your first test shows nothing, you have still learned something the surveys cannot tell you: what AI does for your process, with your data.

Frequently asked questions

How many Swiss SMEs use AI in 2026?

According to the AXA and Sotomo survey of 336 SMEs, 74% use or test AI, made up of 35% who have integrated it deliberately and 39% who are testing it. The other 26% do not use it yet. Other surveys give different shares because they define use differently and sample different firms, so quote one source per figure and state what it measures.

Why do Swiss AI adoption surveys disagree?

They count different things on different populations. AXA counts use or testing among SMEs, UBS counts use among firms of all sizes, EY counts daily use among respondents who are mostly from large companies, and localsearch counts AI tools in marketing only. The percentages are answers to different questions, so averaging them produces a number that describes nothing real.

What are Swiss SMEs mainly using AI for?

Mostly everyday tasks on individual tools. In the 2026 AXA survey the top uses were translation at 47%, correspondence at 42%, advertising texts at 35% and optimising work steps at 35%. Customer service and support came in at 17%. These are uses where one person gets faster, rather than uses where a whole process is rebuilt.

What holds Swiss SMEs back from using AI more?

The surveys name data protection, worry about hallucinations and a lack of know-how. In the ZKB and ZHAW monitor of 1,287 Zurich SMEs, 40% cite missing know-how, and 51% among firms that struggle to recruit. AXA reported in 2025 that only about a third of AI-using SMEs had clear rules on what data may enter AI tools.

Does the 2026 data show that redesigning processes beats adding AI tools?

No. No Swiss survey found measures process redesign or compares returns between redesigners and tool users. AXA reads its own data as AI optimising existing processes, which is an interpretation. International research associates redesign with larger gains, but those studies cover other populations, so for a Swiss SME it is a hypothesis to test on one process.

Insights

Put these ideas to work

A 30-minute call is enough to find out whether an AI agent fits your workflow — and what it would return.