Key Takeaways
- Microsoft Copilot polarizes: some consider it weak, others value it because the data stays in the EU – the truth lies in between.
- Microsoft does not rely on its own AI model for Copilot, but uses models from providers such as OpenAI – and now also Anthropic's Claude models.
- If you work with Microsoft 365 in Germany, your data is already stored with Microsoft in Frankfurt; AI processing currently runs in Ireland – within the EU, so GDPR-compliant use is possible.
- The argument “Microsoft 365 yes, but AI no” is often inconsistent – anyone using Teams, Outlook and SharePoint is already deep in the Microsoft universe.
- With Purview, companies can configure in detail which models and data uses are allowed per department.
- The paid license adds file uploads with company context, the powerful Researcher agent, meeting preparation and much better Office integration – especially in Word.
- Copilot agents automate recurring tasks, for example a chatbot answering questions on internal guidelines and regulation (DORA, NIS 2).
- For very demanding tasks Copilot is not yet on Claude's level – but in everyday Microsoft 365 work it is increasingly useful and improves every month.
Two camps – and a differentiated view
Opinions on Microsoft Copilot are split. One camp is frustrated (“it can't do anything”), the other loves it because the data stays with Microsoft in Frankfurt. The realistic view lies in between: compared with Claude, Copilot is not yet at the same level for demanding tasks – but its integration into Microsoft 365 and its clear data location within the EU are strong arguments.
Data protection: where does the processing happen?
If you use Microsoft 365 in Germany, your data already lives on Microsoft's servers in Frankfurt. Using Copilot does not throw it into a completely different pot: the AI processing currently happens in Ireland – not in Germany, but within the EU, where GDPR-compliant work is possible. That is why the frequent position “Microsoft 365 yes, but AI no” falls short: whoever works with OneDrive, SharePoint, Teams and Outlook is already in the Microsoft universe. And an unencrypted email is far easier to intercept than anything Copilot does – sending one is like sending a postcard.
No own model – and fine-grained control via Purview
Microsoft does not build its own frontier models for Copilot but uses established providers. The OpenAI (ChatGPT) models run inside Microsoft's own data centers; if you choose Anthropic's Claude models – including the Opus models – requests go to American servers. The good news: with Purview in the compliance area you can configure precisely which department may use what. Marketing, which works with public, non-personal content anyway, may upload data and use Anthropic models; controlling may be restricted to EU-hosted processing only.
What cooperative banks (Volksbanken) need to know
In the Atruvia environment of the German cooperative banks these configuration options are more restricted – understandable given the complexity and risk. The important news (thanks to Trutz Winter for the update): cooperative banks can now book the paid Copilot license through Atruvia. The free version included in Microsoft 365 already offers the OpenAI models and file uploads for individual prompts; the paid version adds access to company data, agents and AI-generated meeting summaries. Copilot Studio is not yet available there.
What the paid license adds
The paid Copilot license connects the AI with your own work context: SharePoint libraries, OneDrive, emails, chats and meetings. In Teams, the Facilitator creates not just transcripts but AI-generated meeting summaries, which the Researcher agent can use later. Preparing for an appointment (“use our history and this file, draft a reply”) delivers remarkably good quality – the result of the AI knowing your context. The Researcher (German: “Rechercheagent”) lets you choose your sources precisely: web on or off, meetings, chats, emails, SharePoint, even specific Teams sites – and it can output finished, properly formatted Word documents. A real-world example: a complete 10-part AI video series for a client – planned with personas, scripts and learning goals – required no manual post-formatting at all.
Agents: automating recurring questions
Copilot agents are configurable assistants. Point an agent at a Teams/SharePoint folder containing all internal guidelines – DORA, NIS 2, MaRisk, travel expense policies – and you get a chatbot that answers recurring questions automatically, shareable via link, intranet or Teams. Not everyone in the company needs the full license, but for certain roles it clearly pays off. The interface keeps evolving: an agent activity view, a Prompt Lab with prompt suggestions by topic, and Notebooks – Microsoft's answer to NotebookLM – which now generates audio overviews, mind maps, study guides, infographics and even presentations. OneNote content is now AI-searchable too, and in SharePoint and Teams you can select several files and have them compared.
Frequently Asked Questions
Can Microsoft Copilot be used in a GDPR-compliant way?
In many cases, yes. With Microsoft 365 your data is already stored in Frankfurt, and AI processing currently runs in Ireland – within the EU. The specific configuration and involving your data protection officer remain decisive.
Does Microsoft use its own AI model for Copilot?
No. Microsoft relies on partner models – OpenAI's models run in Microsoft's own data centers, and Anthropic's Claude models are available as an option.
Do I need to assess Copilot separately if I already use Microsoft 365?
Anyone using Teams, Outlook, SharePoint and OneDrive is already working in the Microsoft universe. A blanket “Microsoft 365 yes, AI no” falls short – what matters is a differentiated look at where which data is processed.
What does the paid Copilot version offer?
Access to your own company context (SharePoint, OneDrive, emails, meetings), the Researcher agent with selectable sources and Word export, AI meeting summaries in Teams, agents, and significantly extended Copilot features in Word, Excel and PowerPoint.
What are Copilot agents?
Configurable assistants that automate recurring tasks – for example a chatbot on a policy folder that answers questions about DORA, NIS 2 or travel expense guidelines.
Is Copilot as capable as ChatGPT or Claude?
For very demanding tasks Copilot is not yet on Claude's level. Its big advantage is the tight integration into Microsoft 365 and the clear data location within the EU – and it improves every month.
Conclusion
Microsoft Copilot deserves neither blanket praise nor blanket criticism. Its strengths are integration, data location and monthly progress – especially in Word and with the Researcher agent. If a feature is not good enough yet, don't discard the tool: look again a month later. And if you want in-house support for productive AI use with Microsoft 365, feel free to get in touch.
