Contents
Process automation for SMEs in Switzerland does not start with a platform but with a single, clearly scoped workflow that combines high volume, clear rules, digitally available data and a tolerable cost of error. The standard case then runs on its own, exceptions go to a person, and the effect is measured from day one. Start like this and after around 90 days you have one working workflow and a template for the next.
The question is rarely whether automation pays off but where you start. That choice decides whether automation brings relief or turns into a project that never finishes. This guide is what we hand to owners and managing directors of Swiss SMEs in a first conversation: a test for the right process, a scoring sheet you can fill in yourself, typical first workflows, the tool question, data protection under the revDSG, a 90-day plan and the mistakes we see most often.
What is process automation, and what separates a system from manual work?
Process automation means a recurring workflow runs through without manual copying, retyping or chasing. An enquiry lands in the CRM, the quote is generated from a template, the order goes to accounting, the invoice is created and the status is visible to everyone. Every handover a person does today is a candidate.
Two building blocks complement each other:
- Rule-based automation follows fixed if-then rules and connects systems via interfaces. It is reliable but fails as soon as input is unstructured.
- AI-assisted automation understands language, documents and context. It reads an email, extracts the fields from a PDF invoice or classifies a customer enquiry, and then hands over to a rule-based workflow.
This combination is what most people mean by AI automation for mid-sized companies. It opens up workflows that used to be "too fuzzy" to automate. The bar stays the same, though, and it is simple:
In day-to-day operations the difference shows up in a few places:
| Area | Manual work | System |
|---|---|---|
| Data between tools | Copy-paste, media breaks, typos | Flows automatically via interfaces |
| Status of a case | Chasing by email or phone | Visible to everyone at any time |
| Errors and exceptions | Noticed late or not at all | Detected and reported to a person |
| Knowledge of the workflow | Lives in one person's head | Documented and monitored |
| Scripts and workarounds | Rot when nobody is watching | Monitored, self-reporting flows |
| Growth | Needs proportionally more staff | Scales with volume |
The goal is not to replace people. It is to move them from processing volume to handling exceptions, the place where human judgement really counts.
Why is automation especially worthwhile for Swiss SMEs?
According to the federal government, SMEs make up the overwhelming majority of companies and provide two thirds of jobs in Switzerland. At the same time, labour costs are high and qualified people are hard to find. Every hour a skilled employee spends entering receipts, transferring data or assembling a quote is expensive and missing elsewhere.
Automation works twice over here: it lowers the cost of recurring tasks and decouples growth from the question of whether you can still hire anyone. More on this in our article on the skills shortage in Switzerland and automation. For SMEs, automation is therefore not a question of company size but of process selection.
How do you pick the first process?
Collect candidates before you score them. A simple exercise works best: ask your team to note, for one week, every task where something is copied, retyped, chased or assembled by hand. This gives you a list of 10 to 20 workflows. Each of them then goes through four questions:
- Volume: Does the workflow run often enough for the setup to pay off? Something that happens daily almost always beats something that happens twice a year.
- Rules: Can the standard case be described in traceable rules? If three employees would solve the same case differently, the workflow is not ready yet.
- Data: Is the information needed available digitally and accessibly? Paper, scattered Excel lists or tools without an interface slow down any automation.
- Cost of error: What happens if a single case goes wrong? A misrouted lead is annoying, a wrong payment is critical.
The clearer the volume and rules, the better the fit. A high cost of error does not mean "don't automate" but "automate with a review step": the system prepares, a person approves.
Scoring sheet: how to prioritise your candidates yourself
To turn the list into an order, we score every candidate with a simple grid. Give 1, 3 or 5 points per criterion:
| Criterion | Question | 1 point | 3 points | 5 points |
|---|---|---|---|---|
| Volume | How often does the workflow occur? | Monthly or less | Weekly | Daily or more |
| Rule clarity | Can the standard case be described clearly? | Every case is different | Mostly clear, some exceptions | Clear rules, few exceptions |
| Data | Is the data digital and structured? | Paper or scattered | Digital but unstructured (email, PDF) | Structured in one system |
| Cost of error | How costly is a wrong case? | Critical, hard to correct | Annoying, correctable | Low, easy to correct |
| Integration | Do the tools involved have interfaces? | No | Partly | Yes, documented |
| Pain | How much does the workflow hold you back today? | Barely noticeable | Noticeable | A real bottleneck |
How to read the result:
- Add up the points. The maximum is 30.
- Remove every candidate with a 1 for rule clarity or data from the starting list. It comes back later, once the workflow has been tidied up.
- Estimate the effort roughly as small, medium or large. Small means few systems and hardly any special cases.
- Start with the highest score at small effort. Not with the biggest project, but with the safest win.
A filled-in example of how this could look at a fiduciary or service business:
| Candidate | Volume | Rules | Data | Cost of error | Integration | Pain | Total | Effort |
|---|---|---|---|---|---|---|---|---|
| Capturing incoming invoices | 5 | 5 | 3 | 3 | 5 | 5 | 26 | small |
| Weekly reporting | 3 | 5 | 5 | 5 | 3 | 3 | 24 | small |
| Quotes for special projects | 1 | 1 | 3 | 3 | 3 | 5 | 16 | large |
Incoming invoices win. The special quotes stay on hold despite the pain, because every one is different.
For a rough business case, a simple calculation is enough: cases per month times minutes per case gives today's manual work. An example: 200 invoices times 6 minutes is 20 hours a month. Subtract the time that remains for review and exceptions after automation and multiply by your internal hourly rate. Compare that with the cost of build, licences and maintenance. How to make such estimates honestly is shown with worked examples in How much time does AI really save?.
Typical first processes: what runs automatically and what stays human
The most effective candidates are usually the least glamorous: not the big showcase project, but the daily routine nobody misses once it is gone.
| Workflow | What runs automatically | What stays with people |
|---|---|---|
| Incoming invoices and receipts | Fetch the invoice from the inbox, extract fields, match supplier and account, booking proposal in Bexio or Abacus | Approval, unknown suppliers, releasing payments |
| Quote, order, invoice | Quote from template and CRM data, create order on acceptance, create and book invoice | Prices outside the rules, sending large quotes |
| Lead intake and routing | Capture, enrich, score and assign the enquiry, send confirmation | Edge cases, first conversation, relationship |
| Standard enquiries | Classify the enquiry, draft a reply from the knowledge base | Approving the send, complaints, goodwill |
| Reporting | Pull figures from accounting, CRM and ad accounts, update the dashboard | Interpretation and decisions |
| Onboarding new staff | Create accounts, send documents, assign tasks | Welcome, contract, training |
| Scheduling | Booking, confirmation, reminder, rescheduling | Special requests and conflicts |
Two Swiss specifics help at the start. First, the QR-bill, which replaced the old payment slips in autumn 2022, already contains the payment data in structured form in the QR code. That makes capturing incoming invoices much easier. Second, widely used accounting systems such as Bexio and Abacus offer interfaces through which receipts, contacts and bookings can flow automatically.
Mind the retention duty for receipts: under the Swiss Code of Obligations, accounting records and vouchers must be kept for ten years (Art. 958f CO), and receipts stored electronically must be kept tamper-proof and traceable under the Accounting Records Ordinance (GeBüV). A good automation therefore always archives the original receipt securely before extracting anything from it.
What this looks like for sales and marketing, from the form to the qualified meeting, is covered in our article on automated lead generation in B2B.
Human in the loop: the principle behind every good automation
Good automation does not replace judgement, it frees it from volume. The standard case runs through automatically. Exceptions, edge cases and anything above a defined threshold go to a person.
For lead routing, it looks like this:
// Enrich, score, route: no human touches the standard case.
async function routeLead(lead: InboundLead) {
const enriched = await enrich(lead);
const score = scoreFit(enriched);
if (score > 0.8) return assignTo("senior", enriched);
if (score > 0.4 && enriched.hasConsent) return assignTo("nurture", enriched); // nurturing only with consent
return flagForReview(enriched); // the exception path stays human
}
Everything clear-cut runs untouched. Everything unclear lands exactly where a person should look. In practice we use three patterns, depending on the cost of error:
- Approval before execution: the system prepares, a person confirms. Right for payments, quotes above an amount or replies to customers.
- Sampling after execution: the system acts on its own, a person regularly checks a share of cases. Right for filing, matching and internal notifications.
- Escalation when uncertain: if the system is unsure or data is missing, it stops and reports. This belongs in every workflow.
This pattern becomes especially important when systems carry out several steps on their own. What to watch out for is covered in our article on AI agents and agentic AI in the enterprise.
Build it yourself, buy a tool or have it developed?
There are four levels of implementation. They are not mutually exclusive; most SMEs combine them.
| Option | Fits when | Watch out for |
|---|---|---|
| Features in existing software (Bexio, Abacus, CRM, Microsoft 365) | The workflow lives mostly in one system | Often underestimated, usually already paid for |
| Power Automate | You work in Microsoft 365 (Outlook, Teams, SharePoint, Excel) | Premium connectors and external interfaces need an additional licence |
| Integration platforms such as Zapier, Make or n8n | Several cloud tools need connecting | Data location, data processing agreement, cost as volume grows; n8n can also be self-hosted |
| Custom development | High volume, your own business logic or standard tools hit their limits | Documentation, monitoring and clear ownership of maintenance |
AI components, for example for reading documents or sorting emails, can be built into any of these levels.
Our rule of thumb: start with what you already license and only move up a level when the current one demonstrably falls short. Many Swiss SMEs already use Microsoft 365, and Power Automate covers many internal workflows with its standard connectors. As soon as data flows between different vendors, integration platforms are often the faster route. Custom development pays off when the workflow is core to your business or when you want full control over data and logic.
More important than the tool is the question of who will look after the workflow in two years. An automation without an owner, documentation and monitoring becomes the next workaround that rots. How we build our own systems and AI components where standard tools are not enough is shown on our AI automation page.
Data protection: what the revDSG means for automation
As soon as an automated workflow processes personal data, meaning names, email addresses, customer data or employee data, the revised Federal Act on Data Protection (revDSG, in English FADP) applies, in force since 1 September 2023. For automation projects these points are decisive:
- Data minimisation: only transfer the fields the workflow really needs. A report rarely needs names.
- Processing on your behalf: every tool that processes personal data on your behalf needs a contractual basis (often called a data processing agreement, DPA). For AI services this includes ensuring your data is not used for training without your consent.
- Data flows abroad: many cloud tools process data outside Switzerland. EU countries are generally unproblematic. For the US, since 15 September 2024 the Federal Council has recognised an adequate level of protection only for companies certified under the Swiss-US Data Privacy Framework. For all others you need additional safeguards such as standard contractual clauses.
- Automated individual decisions: if a system makes a decision exclusively by automated means and it has a legal effect on the person concerned or significantly affects them, you must inform the person, and they can ask for the decision to be reviewed by a human (Art. 21 FADP). A clean human-in-the-loop step is therefore not just good practice, it also simplifies compliance.
- Documentation: companies with fewer than 250 employees are exempt from keeping a record of processing activities as long as the processing carries no high risk. If you process sensitive personal data on a large scale or carry out high-risk profiling, you still need the record and possibly a data protection impact assessment.
- Transparency: new tools and data flows belong in your privacy notice.
The GDPR comes into play additionally when you specifically offer goods or services to people in the EU or monitor their behaviour. For an SME with customers in Germany or Austria, that is quickly the case.
Settle these points in the concept phase, not as an afterthought. A compact overview is given in our article on AI, data protection and the revDSG. Which tools and data your team may use at all is best set out in a one-page internal AI policy. This section does not replace legal advice.
What does the EU AI Act mean for Swiss SMEs?
In short: not much for most back-office automation. The EU AI Act can apply to Swiss companies if they offer AI systems in the EU or their output is used there. Strict obligations cover high-risk applications such as recruiting or credit scoring, after the Digital Omnibus from 2 December 2027, including human oversight. If you automate today and plan the human in the loop from the start, you are well prepared. The details are in our article on the EU AI Act 2026 for SMEs.
Where you should not automate (yet)
Not every workflow should be automated. Be careful with:
- Rare decisions with high consequences. An investment, a major contract, a dismissal: automation pays off neither economically nor professionally here.
- Workflows without clear rules. If every case is solved differently, you automate the chaos. Tidy up and standardise first, then automate.
- Regulated, sensitive areas. Recruiting, credit scoring or medical assessments need human decisions and particular care with data protection.
- Relationship work. Complaints, negotiations and important customer conversations gain nothing from automation. The system can prepare; a person leads the conversation.
- Workflows about to change. If you are switching your ERP or accounting system soon, automate in the new system.
The 90-day plan for your first automation
A first complete cycle of selection, build and measurement is feasible in around 90 days for a clearly scoped workflow. This is how we approach it:
Weeks 1 to 2: select and measure the current state
- Collect candidates and prioritise them with the scoring sheet
- Map the chosen workflow step by step: who, which tool, which handover
- Measure baselines: cases per week, minutes per case, lead time, errors
- Name one person in the business who owns the workflow
Weeks 3 to 6: tidy up and build
- Define the target workflow and remove unnecessary steps before automating
- Build the standard case, define exceptions and escalation paths
- Check data protection: data flows, contracts, storage locations
- Test with real, historical cases
Weeks 7 to 10: pilot with approvals
- Go live, at first with a person approving every case
- Log errors and exceptions, sharpen rules and thresholds
- Reduce approvals to sampling step by step where quality holds
- Brief the team and document the workflow
Weeks 11 to 13: measure and decide
- Compare before and after
- Decide: expand, adjust or stop
- Pick the next candidate from the scoring sheet
The first success funds the next step. The first workflow becomes the template for all others: same structure, same monitoring, same escalation logic.
How do you measure the effect?
Without before-and-after figures you can neither prove nor improve the benefit. Measure the same values before the start and afterwards:
| Metric | What it shows |
|---|---|
| Handling time per case | How much manual work remains |
| Lead time | How fast a case moves from intake to completion |
| Share without manual intervention | How many cases run through fully automatically |
| Escalation rate | How often the system hands over to a person, and whether that falls |
| Errors and rework | Whether quality holds |
| Business impact | For example faster quotes, shorter response time to leads, fewer open items |
Two notes from practice. First, saved time only becomes value when it is redirected, for example to sales, customer care or projects that have been left lying. Decide beforehand what the time will be used for. Second, an escalation rate that never falls is a sign that rules or data are off, not that the human is wrongly involved.
Which mistakes should you avoid?
- Starting too big. A platform before the first working workflow ties up budget without showing an effect.
- Picking the wrong process. Prestige does not replace high volume and clear rules.
- Automating broken workflows. Tidy up first, then automate; otherwise you scale the chaos.
- Taking the human out. Without an escalation path, exceptions become errors.
- Underestimating exceptions. The standard case is built quickly; the special cases are the real work.
- Forgetting measurement. Without baselines, the benefit remains a claim.
- No owner. A workflow without an owner and monitoring fails unnoticed at the first tool update.
- Leaving the team out. The people who run the workflow today know the special cases. Without them, the rules are missing.
- Tool sprawl. Every department with its own automation platform leads to data flows nobody can oversee.
Keep these points in mind and you treat automation as a system with a measurable result, not as a one-off tool experiment.
How do you get started?
Don't look for the perfect plan, look for the first good workflow. Fill in the scoring sheet with your team, take the candidate with the highest score at small effort, define the target outcome and the escalation path, and measure from day one.
How we implement such workflows end to end, from connecting Bexio, Abacus or Microsoft 365 to the monitored workflow, is shown on our service page Process automation for SMEs. If you want to find out which workflow pays off first for your SME, get in touch: together we assess your most promising candidates and a realistic first step. And if the bottleneck is not in operations but at the front, in demand, the free AI visibility check shows how well you are found on Google and in AI assistants. How we run acquisition, automation and AI as one system is shown on our Marketing agency Zurich page.
