When Teamdock drafts a reply to a customer, an AI provider reads the question. That is how these features work. What it doesn't need to read is who the customer is. Here is what we send, what we replace first, and what we can't promise.
Your workspace runs on its own server space in Germany. Data only goes to an AI provider (Mistral, on models hosted in the EU, through a gateway Teamdock runs for your workspace) when you use an AI feature. These are the features that can involve customer data, and what happens to it:
| Feature | What it reads | Personal data |
|---|---|---|
| Ticket analysis and reply drafts | The message, the customer's record and their recent orders | Replaced, then put back in the draft |
| The assistant | Your question and what its tools look up (orders, customers) | Replaced, then put back in the answer and in tool lookups |
| Order from an e-mail | The forwarded e-mail | Replaced, then put back in the order fields |
| FAQ suggestions | Closed tickets | Replaced, and removed from the suggestion |
| Product copy and translations | Product fields and your notes | Nothing to replace: no customer data |
Before a request leaves the workspace, Teamdock looks for personal data and swaps each value for a placeholder: names, e-mail addresses, phone numbers, street addresses, postcodes, company names, IBAN and VAT numbers, and links. This is the actual output for a typical e-mail (the person is made up):
From: Jan de Vries <jan.devries@example.nl> Subject: Question about order 10482 My order 10482 hasn't arrived. Can you send it to Dorpsstraat 12, 1234 AB Utrecht? Call me on 06 12345678. Kind regards, Jan de Vries
From: [NAME_1] de [NAME_2] <[EMAIL_1]> Subject: Question about order 10482 My order 10482 hasn't arrived. Can you send it to [ADDRESS_1], [POSTCODE_1] Utrecht? Call me on [PHONE_1]. Kind regards, [NAME_1] de [NAME_2]
Three details make this work in practice:
Order numbers, amounts, dates, products and the town or city are not replaced. Without them the AI can't answer "where is my order?", and on their own they don't identify anyone.
Teamdock can suggest new FAQ entries based on questions your team has answered. An FAQ is public text, so a customer's details should never appear in it, not even correctly. For this feature, placeholders in the answer are deleted instead of filled back in.
The replacement is automatic. It works from patterns, from the fields in your shop's records, and from where names usually appear (greetings, sign-offs, the sender line). That catches most personal data in a typical support e-mail. It does not catch everything:
"My neighbour Kees took the parcel in." Kees is not the sender, not in a field and not in a greeting, so nothing marks him as a name.
Street addresses are matched on Dutch patterns ("Dorpsstraat 12"). "12 rue de la Paix" or "Hauptstr. 5" can get through.
The town or city, the products, and the question itself are sent as written, because the AI needs them. A complaint about a medical product still says so.
The assistant is told who is signed in, so that "send it to me" works.
We call this pseudonymisation, and not anonymisation, on purpose. Under the GDPR, data is pseudonymised when it can no longer be linked to a person without extra information that is kept separately (Article 4(5)). We hold that extra information: it's how the draft gets the customer's name back. Recital 26 is clear that pseudonymised data is still personal data.
That is why nothing else changes on paper. The AI providers are still listed as sub-processors in Annex A of our data processing agreement, transfers outside the EU still run under Standard Contractual Clauses, and the replacement is described as a security measure in Annex B. What does change is how much a provider can see about your customers: in a normal request, their name, e-mail address, phone number and street address no longer reach it at all.
The full list of companies that process data for Teamdock, and where, is on our security page.
Ask us before you sign up. We'll tell you exactly what is sent where, including for features you're not sure you'll use.