AI automation for SMEs: where to start and what to avoid.

At a glance

• In 2025, Swiss businesses were subject to around 36,000 federal obligations, half as many again as in 2005. And nine in ten businesses employ fewer than ten people.
• SMEs mostly use AI to translate and write. The real time savings are in the office work that comes round every week.
• Three questions are enough to decide where to start: does it repeat? Does it follow clear rules? Does it start from data that’s already written down?
• Many tasks don’t need AI at all: fixed rules will do. And before any quote, you count the hours.

Picture this: half past four on a Friday at a three-person engineering firm. Someone is still opening the PDFs that came in during the week, renaming them one by one and filing each in the right site folder. Nobody has ever counted how long it takes. AI automation for SMEs starts with jobs like this, not with the chat tool everyone has already tried.

More obligations, same people

In 2025, Swiss businesses were subject to around 36,000 federal legal obligations. In 2005 it was around 24,000, so the number has grown by half in twenty years1. The count was carried out by BAK Economics for the State Secretariat for Economic Affairs (SECO), using an unusual method: twenty years of laws and ordinances were read by a language model, the same kind of artificial intelligence that powers ChatGPT.

The perceived burden, at least, isn’t getting worse. In SECO’s Bürokratiemonitor (bureaucracy monitor), 60% of more than 1,500 businesses rate the administrative burden as high or fairly high2. That figure dates from 2022, was published in 2023 and is slightly lower than in 2018: according to SECO, thanks to public services going digital3.

The real question is who does this work. Fewer than ten people work at 89.8% of Swiss businesses4. There’s no admin department: there’s the owner in the evening, or the same person who answers the phone, prepares the invoices and orders the materials. Every hour spent on paperwork is an hour taken away from customers.

What Swiss SMEs use AI for today

The most recent study comes from AXA: in March 2026, the research institute Sotomo surveyed 336 SMEs with at least five employees in German- and French-speaking Switzerland (Ticino was not in the sample). Some 35% say AI is already built into their processes, 39% are testing it and 26% don’t use it yet5.

Today, AI lives where the writing happens: translating, answering a letter (42%), drafting advertising copy (35%). In processes and customer work, far less so. Yet that’s one of the fastest-growing areas: the share of SMEs using AI to improve their workflows has risen from 23% in 2024 to 35%5.

And what about jobs? Almost half of SMEs with AI experience expect to save time without cutting jobs; only 12% think they could do without someone5. We see it the same way: the time freed up goes back to customers and to the tricky cases.

Chat or automation: the difference that matters

A chat tool like ChatGPT does what you ask, one request at a time. Someone has to open it, paste in the text, check the answer and copy it to wherever it’s needed. You save minutes, but you’re still the one carrying the work.

An automation, on the other hand, starts by itself when an email, a PDF or an order arrives, and leaves the result in the software you already use, ready to check. Traditional automations follow fixed rules and work as long as the data always arrives in the same format. AI comes in when everyone writes in their own way: the order typed into the body of an email, invoices laid out differently by every supplier, the voice message on WhatsApp.

Three ways to take a job off your plate

Aspect	Fixed rules	Chat	AI automation
What starts the job	A date or an event	You, with a question	The incoming email or PDF
Understands free-form text	No, it needs fixed forms	Yes	Yes
Where the result ends up	In your software	In a chat window, to copy over	In your software, ready to approve
Example	The payment reminder after 30 days	Translating a quote	A PDF order turned into a confirmation

First, cut out the copy-and-paste. Then add AI, if you need it.

AI automation for SMEs: the three-question test

To find out whether a task can be automated, we ask three questions. If the answer to all three is yes, it can almost always be done.

1. Does it repeat every week? A task that comes up twice a month won’t pay back the time it takes to set up. One that comes round every day will.
2. Does it follow clear rules? If three different people would do it the same way, the rules exist, even if nobody has ever written them down. If it takes judgement every time, AI can prepare the work, but not make the call.
3. Does it start from data that’s already written down? Emails, PDFs, Excel sheets, your business software. If the information only exists in someone’s head, it needs to be put in writing first.

The test on six office tasks

Task	The three answers	Best approach
Recording supplier invoices	Yes, yes, yes (as PDFs)	AI automation: every supplier lays out invoices differently
Sending reminders for overdue invoices	Yes, yes, yes	Fixed rules are enough
Preparing quotes from enquiries	Yes, yes (the price list), yes (emails and attachments)	AI automation, with a person approving
Reminding patients of appointments	Yes, yes, yes (the appointment book)	Fixed rules are enough
Negotiating with a key client	Occasionally, no, partly	Stays with a person
Closing the annual accounts	Once a year, yes, yes	Not worth it: your accountant handles it

For invoices and quotes, we have ready-made foundations that we adapt to your business: Digitam Invoice and Digitam Quote. For phone and WhatsApp enquiries, there’s Digitam Voice.

Why so many AI projects don’t pay off

In August 2025, an MIT report made headlines around the world: only about 5% of generative AI pilots lead to rapid revenue growth, while almost all the rest have little or no impact on the bottom line6. It’s a report, not a peer-reviewed study, and it’s about large companies: treat it as a signal, not a measurement.

Two of its findings, though, apply to SMEs too. More than half of generative AI budgets go on sales and marketing, yet the biggest returns are in back-office work. And solutions bought in from specialist vendors succeed far more often than those built in-house6.

In Switzerland, too, many are still experimenting: in EY’s May 2026 survey of 604 people at Swiss companies, 31% said their company is still at the pilot stage7. The sample, however, is made up mainly of large companies.

What successful projects have in common

The projects we have running in production always share the same ingredients: a single task, data that’s already written down, an automation that prepares the work and stops when in doubt, and a person who approves. You can read about them on our automation projects page.

• A five-person professional practice near Lugano. Every email is read, sorted and given a draft reply. Of 42 emails in one morning, only 5 needed a person.
• An industrial company in the Mendrisio district. Website enquiries become draft quotes priced from the list: two minutes instead of two days.
• Short-term rentals, six properties. The monthly statements for the owners used to take a full day’s work every month. Now they’re approved with a single click.

The mistakes we see most often

• Tackling ten processes at once, without measuring the starting point. Start with one and time how long it takes today, or you’ll never know whether it worked.
• Automating a task everyone does their own way. Agree on the rules first, otherwise you’re just automating the mess.
• Building on scattered data. If customers and orders are spread across ten different files, automation just copies the mistakes faster.
• Letting AI make decisions about money and customers. A payment to an IBAN that has just changed, a quote with the wrong price: that’s where a person decides.
• Putting customer data into a tool without knowing where it ends up. Under the revised Swiss Federal Act on Data Protection (FADP, nLPD in Italian), in force since 1 September 2023, anyone processing personal data on your behalf must be bound by a contract and must guarantee its security (Art. 9)8. We cover this in ChatGPT and the FADP. For your specific case, ask your fiduciary or a lawyer.

Run the numbers before the quote

Before asking anyone for a quote, us included, do this calculation. It takes ten minutes.

Don’t use net salary for the hourly cost. As a rough guide, in 2024 the full-time median wage was CHF 7,024 gross a month in Switzerland and CHF 5,708 in Ticino9. At 42 hours a week over 46 weeks, that’s roughly 44 francs an hour in Switzerland and 35 in Ticino, before social charges.

Then set that against the cost of automating it: development, paid once, and any monthly fee. With us, you get a fixed price for both before you sign, and you see a working version on your real cases before going any further. If the numbers don’t add up, we tell you straight away and we don’t go ahead.

If you’d like to start on your own, our free PDF guide (in Italian for now) takes you, in 15 minutes, to the first two or three tasks worth automating, with the figures in francs. If you’d rather talk it through, in a free 30-minute consultation we’ll look together at the work that weighs on you most.

Sources

1. Wie viele Pflichten gelten für Unternehmen? (in German). Rafaela Schinner and Andreas Streich (BAK Economics), Die Volkswirtschaft, SECO’s economic policy magazine, 3 June 2026.
2. Administrative Entlastung und Bürokratiemonitor (in German). SECO, the federal SME portal. 2022 survey, published on 1 March 2023.
3. SECO veröffentlicht neues Monitoring zur Regulierungsbelastung von Unternehmen (in German). SECO press release on the new regmonitor.ch monitoring tool, 31 March 2026.
4. Schweizer KMU: Eine Analyse der aktuellen Zahlen, Ausgabe 2026 (in German). OBT and the University of St. Gallen (KMU-HSG), 2026. Provisional Federal Statistical Office STATENT data for 2023.
5. Künstliche Intelligenz: Schweizer KMU erwarten keine fundamentale Veränderung der Arbeitswelt (in German). AXA Switzerland, Sotomo study of 336 SMEs (survey conducted from 11 to 18 March 2026), 15 September 2026.
6. MIT report: 95% of generative AI pilots at companies are failing. Sheryl Estrada, Fortune, 18 August 2025, on the MIT NANDA report “The GenAI Divide”.
7. Artificial intelligence widely established in Swiss companies, but many are still in the early stages of scaling. EY Switzerland, survey of 604 respondents, 27 May 2026.
8. Federal Act on Data Protection (FADP), SR 235.1. Fedlex, in force since 1 September 2023. Art. 9.
9. In 2024, the gross median wage was CHF 7024. Federal Statistical Office (FSO), Swiss Earnings Structure Survey 2024, 25 November 2025.

Written by Alexandru Ciobanu. Sources checked on 16 September 2026. If you spot a figure that doesn’t add up, email us at info@digitam.ch.