Most automation projects in SMEs do not fail because of the technology. They fail because of the order in which things are tackled. The most exciting or loudest process gets dealt with first, rather than the one that makes the most economic sense. The result: a lot of effort, a showcase project, and little changes in day-to-day operations. This list helps you choose the right task first, before you invest money and time.
SMEs are the backbone of the Swiss economy: 99.7 per cent of all roughly 600,000 companies have fewer than 250 employees, and around 90 per cent of those are micro-firms with fewer than ten staff (source: Federal Statistical Office, STATENT). It is precisely these businesses that rarely have their own IT department and can least afford bad investments. That makes a sober selection all the more important.
The Four-Criteria Framework: Which Task Deserves to Go First?
Before you automate anything, assess the process against four criteria. Answer each question with high, medium or low.
1. Frequency – How often does the task occur? An activity that comes up several times a day delivers a lot once automated. Something that happens twice a year is rarely worth the effort.
2. Rule-based nature – Does the task follow clear, describable rules? If you can explain it to a new colleague in five sentences, it is well suited to automation. If the answer is often “it depends”, it is not.
3. Time required – How much time does a single run take? Multiply this figure by the frequency. A two-minute task that occurs 50 times a day is more valuable than an hour of work per month.
4. Risk – What happens if the automation makes a mistake? With an incorrect appointment confirmation, you write a corrective email. With incorrectly posted VAT or a customer list sent out in error, it can become expensive and sensitive.
The rule of thumb is: start with processes that score high on frequency, rule-based nature and time required, and low on risk. This is the unspectacular but reliable route to genuine benefit. High-risk processes are automated later and always with human oversight.
| Criterion | Good to start with | Better to wait |
|---|---|---|
| Frequency | daily / several times a day | a few times a year |
| Rule-based nature | clear, fixed rules | much judgement, gut feeling |
| Time required | a noticeable sum per month | barely measurable |
| Risk if it goes wrong | low, easily corrected | legally or financially sensitive |
Six Typical Entry-Level Processes for Swiss SMEs
These six processes meet the framework in most businesses and are therefore good first candidates.
-
Incoming invoices and receipt capture. Reading out incoming supplier invoices, checking them, assigning them to the correct cost centre and filing them. High frequency, clear rules. Since 1 October 2022, the QR-bill has been the standard in Switzerland; the old orange and red payment slips have not been processed since then (source: Federal Tax Administration). This makes machine reading easier than it used to be.
-
Quotations and order confirmations. Generating a quotation or a confirmation from standardised enquiries. With recurring standard line items, this is easy to plan for.
-
Appointment confirmations and reminders. In the trades, in the plumbing business or in the service sector, automatic confirmations and reminders save time on the phone and reduce idle periods.
-
Standard email enquiries. Having recurring questions pre-qualified, sorted and answered with drafts – with a person who approves them. This is relief, not a replacement.
-
Data transfer between systems. Typing the same data by hand into two or three programs is classic, error-prone double work. That is exactly what can be automated cleanly.
-
Document filing and archiving. Naming and filing receipts in line with the rules. Important: in Switzerland, business records and accounting documents must be retained for ten years (Art. 958f of the Swiss Code of Obligations). Electronic filing is permitted if consistency with the business transaction is guaranteed and immutability is demonstrable (Ordinance on the Maintenance and Retention of Account Books, GeBüV). An automation must meet these requirements, not circumvent them.
At Vollmer Labs, building blocks of this kind are already running in production. We operate accounting agents that work alongside the team of a fiduciary firm in day-to-day operations, and our ballistic.club platform is live in use. Further in-house tools are in development: rfqbuddy.com for tender and enquiry processes is being built, and jeffri.ch, a workflow and automation tool for kitchen studios with Swiss data hosting in Zurich, is in a Swiss pilot. We say this openly: the building blocks are proven, but the application is at a different stage in each individual sector – some things run in production daily, others are still at the pilot or build-up stage.
The Do-Not-Automate List
Just as important as the starting list is the list of what you should leave alone for now. Ignore it and you burn budget or create problems for yourself.
- Rare special cases. If a process occurs in only 5 per cent of cases and looks different every time, the programming effort usually clearly exceeds the benefit.
- Genuine judgement. Price negotiations, complaints with an emotional charge, the assessment of a new supplier. Here, human judgement is the actual value.
- Sensitive personnel decisions and customer relationships. A dismissal conversation or the relationship with your most important customer does not belong in a workflow.
- High-risk decisions without human approval. Anything that is legally or financially binding needs a checkpoint. Language models demonstrably invent facts; depending on the task, error rates are considerably higher than laypeople assume. In practice, serious solutions therefore deliberately build a human into the loop before an AI result takes effect.
- Processes that are not yet stable. The single most important rule: first simplify, then standardise, then automate. If you automate a chaotic workflow, you get faster chaos – and make the mistake on a large scale.
A concrete Swiss point of diligence: when handling personal data, the revised Federal Act on Data Protection (revFADP) has applied since 1 September 2023. Deliberate breaches of certain obligations can be penalised with fines of up to 250,000 francs, which in principle target the responsible natural person rather than primarily the company (source: revFADP, Art. 60 ff., FDPIC). So anyone who processes or sends customer data in an automated way should not set this up on the side.
Anti-Hype: What You Really Should Know About ROI
Snappy promises circulate online: “25 to 40 per cent time savings”, “amortisation in three to six months”. Such figures almost always come from providers who want to sell something. They are not neutral measurements but best-case estimates under ideal conditions. Treat them as advertising, not as a basis for planning.
The sober counter-position is borne out by independent market observers: various industry analyses and analysts put the share of first automation projects that fail to meet expectations at around 30 to 50 per cent, depending on the source, and a considerable proportion never gets beyond the pilot stage. This is rarely down to poor software – usually it comes from the wrong choice of process, unstable workflows and underestimated maintenance.
Here is how to calculate it seriously yourself, without blanket figures:
- Note the frequency per month (for example, 200 supplier invoices).
- Honestly estimate the time saved per run (for example, 4 minutes instead of doing it by hand).
- Multiply the two to get the monthly time saved (here, around 13 hours).
- Set this against the one-off setup and ongoing maintenance costs.
Only this calculation – per specific process, with your real figures – tells you whether the step is worthwhile. Everything else is gut feeling with a slide deck. And do not forget maintenance: an automation is not a piece of furniture you set up once. It needs care when forms, suppliers or regulations change.
The honest core of this topic in one sentence: automate the frequent, rule-based and time-consuming with low risk first – and leave the rare, sensitive and judgement-dependent to people. Those who proceed in this order see a measurable benefit early and avoid the expensive mistakes on which most projects fail.