Microsoft 365 Copilot: AI in the Microsoft ecosystem

Copilot is the umbrella term for Microsoft's AI assistants. What matters for businesses is Microsoft 365 Copilot: AI in Word, Excel, PowerPoint, Outlook, and Teams, with access to emails, documents, and chats in your own tenant. Copilot summarises meetings, drafts documents, analyses spreadsheets, and answers questions across your company data.

Add Copilot Studio for custom agents and, since 2026, features like Work IQ, which factors in work context and preferences. Under the hood run OpenAI models; since early 2026, Microsoft supplements them with Claude models from Anthropic for selected features.

Important distinction: the free Copilot and Copilot Pro are consumer products without enterprise data protection. For company use, only the business licence counts.

Microsoft 365 Copilot costs about 28 EUR per user per month on top of the existing Microsoft 365 licence (as of July 2026). A qualifying base licence is required, typically Business Standard, Business Premium, E3, or E5.

With 100 employees, that is roughly 34,000 EUR per year on top of the existing Microsoft costs. Agents and consumption-based features are billed additionally via Copilot Credits. At the same time, Microsoft is gradually folding Copilot into its bundle prices, so the pricing structure is in flux.

For comparison: for multi-model platforms such as Langdock the vendor quotes a list price of 25.00 EUR per user per month excluding VAT for its Business Standard plan (langdock.com, retrieved 7 August 2026), without a Microsoft licence requirement, but also without the deep Office integration.

The fundamentals are in place: company data stays in the tenant, is not used to train public models, and the EU Data Boundary contractually guarantees processing in Europe. The Hessian data protection authority confirmed fundamental GDPR compliance in a detailed report in late 2025.

Two points still demand attention.

Since April 2026, Microsoft may process AI requests in data centres outside the EU during load peaks. The feature is called flex routing and is on by default for new tenants. Anyone who has to adhere strictly to the EU Data Boundary disables it in the Admin Center and adjusts the data protection impact assessment.

The second point is permissions. Copilot sees everything the person asking is allowed to see. Overly broad shares in SharePoint and Teams thus become an AI problem: the assistant finds the salary document that never should have been shared. So the permission model belongs before the rollout, not after it.

We summarised the general criteria for legally sound AI use in our article on GDPR-compliant AI use.

Copilot fits when the company is deep in the Microsoft ecosystem and AI is mainly meant to speed up office work: meeting summaries, email drafts, document analysis. As long as that workflow consists of Microsoft tools, the integration is unmatched.

Something else fits better as soon as one of three conditions applies. When model diversity matters: Copilot only knows Microsoft's own selection, while platforms like Langdock bundle over 40 models from six providers by the vendor's own account. When dependency on Microsoft is a strategic concern anyway. Or when AI should go into your own product: for AI features in your own website or customer platform there is no way around a direct AI integration via model APIs, and Copilot is not built for that.

Here too: both can run in parallel. Copilot for internal office work, a model-agnostic integration for your own platform.