AI for SMEs: where artificial intelligence really helps in the office

Where AI reliably helps in the office today

Language models are good at understanding, summarising and writing text. In an SME office, that fits many recurring tasks:

• Sorting emails: enquiry, appointment request, complaint or invoice? AI recognises the request and how urgent it is.
• Drafting replies: for questions about prices, availability or process, a draft appears in your tone.
• Preparing quotes: AI reads from an enquiry what the customer wants and suggests suitable line items.
• Carrying over invoice data: details from the quote and the email thread end up in the invoice instead of being retyped.
• Spotting appointments: appointment requests in emails are recognised and suggested as calendar entries.
• Summarising and translating: getting to the point of a long thread and answering customers in their national language.

Where you should decide yourself

AI writes convincingly even when it’s wrong. Experts call this hallucination. So the rule is: AI suggests, a person approves. Especially for these points:

• prices, discounts and dates you commit to
• legal statements such as warranty, liability or cancellation
• sensitive situations such as complaints or payment disputes
• figures and amounts, before they go into a quote or invoice

Good AI software is built so that you check every suggestion before it leaves the business.

General chatbot or AI inside specialised software?

General chatbots like ChatGPT are versatile and good for getting started. Day to day, though, it means: copy the email, explain the context, copy the answer back, look up prices yourself. The chatbot knows neither your services nor your customers.

AI inside specialised software works where the work happens, for example directly in the inbox. It knows the company knowledge you have stored, and the result is a reply, quote or invoice right away. For recurring tasks that is usually the more direct route; for one-off texts a chatbot is often enough.

Data protection: what the FADP requires

Switzerland has no AI-specific law. The Federal Data Protection and Information Commissioner (FDPIC) states that the Data Protection Act applies directly to AI-based data processing. For an SME this mainly means:

• Transparency: disclose purpose, functioning and data sources. Anyone communicating with a chatbot has a right to know.
• Only necessary data: give the AI only what it needs for the task.
• Processing on your behalf: you need a contract with the AI provider that governs how it processes the data (Art. 9 FADP).
• Transfers abroad: many AI providers process data in the USA. This is permitted if adequate protection is ensured, for example through the Swiss-U.S. Data Privacy Framework or standard contractual clauses (Art. 16 FADP).
• No training on your data: check whether the provider uses inputs for training. With business accounts and interfaces (API) this is often excluded.

This guide is not legal advice. For sensitive data, such as health data, it is worth consulting a specialist.

How to start small

1. Pick one task: choose work that repeats often, for example replies to price enquiries.
2. Store your company knowledge: services, prices, frequent questions and your tone. The better the AI knows your business, the less you need to correct.
3. Test for two weeks: check every suggestion. You quickly see where the AI is reliable and where it isn’t.
4. Then expand: quotes, invoices and reminders follow once the first task runs smoothly.

Velro is SME software that works exactly like this: the AI reads along in the inbox and drafts replies, quotes and QR invoices. You check and send.