AI for Swiss SMEs: where to start?

Artificial intelligence has been making headlines for two years. Solution vendors promise productivity gains of 40%, total process automation and "AI agents" that run your business for you. The reality for a Swiss SME of 10 to 100 people is far more nuanced. and that's a good thing, because it means you can move forward pragmatically.

What AI does well for an SME

Generative AI (language models such as ChatGPT, Claude, Gemini) is genuinely useful for:

• Writing and rephrasing: emails, sales proposals, meeting minutes, translations. Not a human replacement, but a significant accelerator for people who write a lot.
• Document summaries: quickly analysing a contract, a proposal or an audit report to extract the key points.
• Structured data processing: extracting information from invoices, forms or tables, with human validation.
• Internal support: an assistant that knows your internal procedures and answers common questions from new employees.

What these use cases have in common is that they are defined, verifiable and supervisable. A person stays in the loop to validate the result.

What AI does badly. or not at all

Several common vendor promises don't hold up to scrutiny:

• Autonomous decisions: an AI system should not make decisions that commit your company (validating contracts, responding to dissatisfied customers, financial transactions) without human supervision.
• Undocumented processes: AI cannot automate what you yourself cannot describe precisely. If your process is vague, the automation will be vague, only faster.
• Unstructured data: generative AI can produce plausible but false results (this is called "hallucinations"). On factual data (figures, dates, names), the error rate is unacceptable without systematic verification.
• Replacing customer contact: a chatbot can handle simple, repetitive requests. It does not replace a commercial relationship, and a poor customer experience via a chatbot costs more than it saves.

The real questions to ask before getting started

What specific problem do you want to solve?

"Using AI" is not a goal. "Halving the time spent re-entering supplier invoice data into our ERP" is a goal. Starting from a concrete, quantified problem with an identified owner is the basic condition for an automation project to succeed.

Who will supervise and correct errors?

Any automated system produces errors. The question is not to eliminate them (that's impossible), it's to detect and correct them. Before deploying anything, define who checks the results, how often, and against what criteria.

Where does your data go?

If you use a cloud AI service (ChatGPT, Claude, Copilot), your data may transit through servers in the United States. For confidential customer data, HR data or data subject to professional secrecy, check the service's terms of use and their compliance with the Swiss Data Protection Act (FADP). Some vendors offer European deployment options or suitable contractual guarantees.

Concrete use cases and indicative costs

Automating invoice entry

Automatic extraction of data from a PDF invoice (supplier, amount, date, line items) and injection into your accounting software. Available via solutions such as Dext, Klara or custom connectors. Cost: CHF 50 to 200/month depending on volume. Estimated gain: 2 to 5 hours per month for an SME processing 100 to 300 invoices.

Automatic summaries of emails and documents

Microsoft Copilot (included in certain Microsoft 365 Business Premium licences) or similar tools let you summarise long discussion threads or Word documents. Cost: included in the licence or ~CHF 30/user/month as an option. Useful for teams that handle a lot of documents.

Internal FAQ chatbot

An assistant that answers employees' common questions (HR procedures, product information, internal policies) from a document base you provide. Setup cost: CHF 2,000 to 8,000. Monthly cost: CHF 100 to 300 depending on the platform. Relevant from 30 to 40 employees with recurring, documentable questions.

How to get started without getting lost

Start small and measure. Identify a repetitive task you already do, with structured data, a volume large enough to justify automation, and a person to validate the results. Deploy a simple solution, measure the real gain after 4 to 6 weeks, and decide whether to continue or stop.

Don't invest in an "enterprise AI platform" before validating a concrete use case. Generic platforms often require months of integration and configuration to produce anything useful.

Get support from someone who knows both your business processes and the available technical tools, not from a vendor whose pay depends on the size of the project they sell you. This is the approach of our AlpenFlow (business automation and AI) offering: we start from a concrete process, we cost it, then we automate it without overhauling your existing tools.

In summary

AI can create real value for a Swiss SME in 2026, but not in the way vendors present it. It is useful for well-defined tasks, with structured data, supervised by humans. It is not ready to manage complex processes, customer relationships or decisions that commit the company autonomously.

The best first step: identify your team's most repetitive and most tedious task, and ask an expert whether it can be automated. The honest answer will give you more information than any vendor demo.