Data protection and AI: is my customer data really safe?

As soon as a CRM puts "AI" on the box, the same question comes up in every sales team: where do our customer data actually go when the software summarises a conversation or drafts an email? The concern is justified. Anyone working with quotes, call notes and contract details is handling sensitive information about people and companies who trust that it won't leak out somewhere on the internet. The good news: you don't have to choose between useful AI and data protection. You just need to understand which questions to ask.

What really happens with an AI feature

When a CRM generates text or summarises a conversation, it sends a slice of your data in the background to a language model, for example from Anthropic (Claude) or OpenAI. That slice is processed and a response comes back. What matters is what happens to the transmitted data afterwards. Here it pays to keep three things apart:

• Transmission: are the data transported in encrypted form? That is standard today, but it's worth verifying.
• Storage: does the content stay with the model provider, or is it discarded after processing?
• Reuse: can your data be used to train future models?

This last point decides a great deal. Reputable providers use what are known as enterprise or API contracts to explicitly rule out business data ever flowing into model training. A customer note submitted through the interface is therefore something quite different from a text someone privately types into a free chatbot.

Three terms you should know

You don't have to become a lawyer, but a few concepts help in any conversation with a software provider:

Data hosting and server location

Where does the database with your contacts, deals and notes sit? If it's in Switzerland, Swiss law applies, and the path to your data is short and traceable. Important: the storage location of your CRM and the processing location of an AI model are two different things. A provider can keep your master data in Switzerland and still use an external model for individual AI tasks.

Data processing agreements

The revised Swiss Data Protection Act requires that the relationship between you and every service provider who processes personal data on your behalf is governed by contract. Ask specifically: is there a data processing agreement, and are the AI sub-processors transparently listed in it?

Data minimisation

The most effective safeguard is often the simplest: give the AI only what it needs for the task. A conversation summary needs the content of the conversation, not necessarily the customer's social security number.

Data protection with AI is not a single setting you switch on, but a chain of decisions that is only as strong as its weakest link.

A short checklist for choosing a provider

Before you put an AI-powered CRM into production, clarify these points in writing:

1. Where are my master and transactional data stored, and under which law?
2. Which AI providers are used for which features?
3. Is it contractually ruled out that my data flows into model training?
4. Can I disable AI features individually if a customer asks me to?
5. How long is data retained, and how do I delete it completely?

If a provider answers these questions clearly and without evasion, that's a good sign. Vague wording or a pointer to a 40-page policy with no concrete statement is a bad one.

Trust comes from transparency, not from doing without

Some teams conclude from the uncertainty that it's better to leave AI out altogether. That's understandable, but usually the more expensive option: employees then quietly reach for private tools, and that's exactly where you lose control over your data. A deliberately chosen tool that regulates the processing cleanly and makes it traceable is the better path.

At Advanzo we have taken exactly this route. Customer data is hosted in Switzerland, AI features such as email generation, "deal scoring" and conversation summaries run on established models from Anthropic and OpenAI with clear contractual guardrails, and every feature stays traceable for you. That fits our guiding idea of "remove complexity, not add it": you should be able to use the benefits of AI without needing a law degree or a guilty conscience. Data protection then isn't an obstacle, it's simply part of a tool that you and your customers can trust.