On 15 September 2026, AXA published its annual survey of Swiss SMEs: three out of four use artificial intelligence, but only 35% have built it deliberately into their processes, and adoption has stopped growing. AI for small business is no longer new; it has become an everyday tool that, more often than not, brings in very little. The short answer: pick one or two processes, measure them before and after, write down the rules and train your team to apply them.
Key takeaways
- According to the AXA survey carried out by the Sotomo institute, 35% of SMEs use AI in a targeted way, 39% are testing it and 26% do not use it: practically unchanged in a year.
- Use cases remain mostly about writing: translation (47%), correspondence (42%), marketing copy (35%).
- BDO observes that many SMEs remain stuck at the pilot stage, and advises them to focus on a few use cases with measurable benefits.
- Time saved by each employee does not turn into a gain for the company on its own: it is a management matter, not a tool matter.
What is happening
The SME labour market survey is carried out every year by the Sotomo institute for the insurer AXA. The 2026 edition is based on 336 SMEs in German- and French-speaking Switzerland, surveyed from 11 to 18 March 2026. Its finding fits in one word: plateau. Between 2024 and 2025, the share of companies testing or using AI had jumped from 55% to 71%. A year later, 35% use it in a targeted way and 39% are testing it, practically the same level. According to AXA, the initial adoption phase is largely over.
Use cases, meanwhile, are slowly broadening. Language tasks still dominate, but the share of SMEs using AI to optimise their work processes has risen from 23% in 2024 to 35%. Customer service and support (17%), on the other hand, is not growing: companies worry about AI mistakes in front of their customers. Almost half of the SMEs that have tried AI (47%) see it as a time saver with no effect on headcount, and only 12% think they could do without some of their staff.
Kathrin Braunwarth, a manager at AXA Switzerland, raises the real question: how can companies gather the "fragments of efficiency" gained by each person and turn them into genuine added value? In her view, this will be one of the biggest management challenges of the coming years.
BDO reaches the same conclusion by another route. In its report AI Toolkit: Switzerland Snapshot, presented on 26 June 2026 and based on a meta-analysis of national and international studies, the audit and advisory firm sees a widening gap: large companies have built AI into their processes, while many SMEs stick to occasional use and struggle to move beyond pilot projects. The reasons given: few resources, insufficient data quality, no time to test methodically, and doubts about data protection and real benefits. Its recommendation: no massive investment, but a few concrete use cases with measurable benefits, carried out rigorously, and AI treated as a leadership issue rather than an IT question.
What it means for you: AI for small business is a leadership matter
If your team "uses ChatGPT" and you see no difference in your results, you are in good company. This is not a failure of the tool; it is the consequence of individual use: everyone saves a few minutes on an email or a translation, and that time dissolves into the day. Nobody measured it, nobody reallocated it.
Moving to the next stage takes three changes, and all of them are the owner's job.
Start from a process, not a tool. The question is not "which AI tool should we adopt?" but "which repetitive, costly and measurable piece of work do we want to do differently?". Entering supplier documents, reconciliations, standard replies, preparing quotes, follow-ups: a process has a start, an end, a volume and a cost.
Measure before you automate. Without a baseline, there is no way to tell whether AI pays off. Time spent each month, turnaround time and the number of errors are often enough.
Write the rules and train the team. AI applies clear rules very well and implicit rules very badly. If you could not explain it to a new employee, you will not be able to hand it to an AI.
What I have learned
I use AI agents every day on my own processes: my business management system, my notes, my publications and the administration of my servers. To keep them running around the clock, I built claudeVPS, an environment that runs them on a server. But the tool is not what made the difference.
The clearest example is my bookkeeping, which I describe in AI bookkeeping in practice. I handed it to an agent within a precise framework. The rules are mine, and they are written down: the accounting date is the payment date, a purchase paid with my private money follows a defined path, an expense without a receipt is only recorded with my approval. The agent prepares a plan, runs it as a dry run first, I review it, and then it applies it after a backup. The result: January to October 2026 has been fully entered and reconciled, and every month-end close now follows a written procedure. The agent flags doubtful cases; I make the call.
I have also learned to say no. On TSFind, a service directory I developed, I dropped an automatic document check: hours of development and a security risk, to save less than one hour of manual work per month. Measuring before building keeps you from automating what is not worth it.
Finally, the advice I give most often is the one I gave Justin Gary, CEO of Stone Blade Entertainment, at an AI hackathon. He told the story in an article published in June 2025: AI is powerful, but it will not save you from the hard thinking needed to define clearly what you want. You have to start with a product requirements document. For an application as for a process, that document is what separates an experiment from real use.
What to do in practice
1. Pick one or two processes
List with your team the repetitive tasks that take time every week. Keep one or two: frequent, with known rules, where a mistake can be corrected. Leave aside for now anything that directly touches your customers.
2. Measure the baseline
How many hours a month, what turnaround time, how many errors or redos? Write it down. Also set what you expect: without a numeric target, you will not know whether to continue.
3. Write the requirements
Describe the process as it should work: the inputs, the rules, the special cases, and what requires human approval. This document is used to choose the tool, configure it and train the team.
4. Put safeguards in place
A dry run before anything is written, human approval of decisions, a backup, and control of what leaves the company: data entrusted to an external service, documents sent to customers.
5. Train the team on the process, not the tool
Show the people who do the work what the AI does, what it does not do and what they must check. Appoint a process owner who follows the measurements.
6. Measure again, then decide
After one or two months, compare with the baseline. If the gain is real, decide what the freed-up time is for and move on to the next process. If not, stop: that is a result too.
Frequently asked questions
Which process should a small business start with when using AI?
An internal, repetitive and measurable task with known rules: processing supplier documents, reconciliations, standard replies, preparing documents. Keep direct exchanges with your customers for later, once the method has proven itself.
Do we need to buy a dedicated AI tool?
Not at first. Describe the process and measure its current cost first. The requirements will then tell you whether a general-purpose assistant is enough, whether your business software already has what you need, or whether a custom automation is justified.
How do you know whether AI really pays off?
By comparing before and after, on the same process: hours spent, turnaround times, errors. If you did not measure before, you will not be able to tell. And time saved only pays off if you decide what it is used for.
Going further
I help business owners choose which processes to hand to AI, write the requirements and measure the result: see the Strategy and guidance page. When the tool has to be built, that is the job of custom applications and automations. To work out where to start in your company, let's talk during a 20-minute call.
Sources
- Intelligence artificielle et travail : les PME suisses ne s'attendent pas à des changements radicaux (in French), AXA Switzerland (Sotomo survey), press release of 15 September 2026.
- Intelligence artificielle : le fossé se creuse entre grands groupes et PME (in French), BDO Switzerland, press release of 26 June 2026.
- A Game Designer's Guide to Vibe Coding, Justin Gary, Substack, 3 June 2025.
Photo: Paymo / Unsplash (unsplash.com/photos/kjM2AxNsxNo).
Tags: AI for small business, artificial intelligence, processes, requirements, change management, strategy, automation