Microsoft’s Copilot: adoption, features and EU data change
Microsoft’s workplace AI faces two practical problems: slower-than-advertised seat uptake and tricky data-governance choices as Copilot scales inside organisations. (seekingalpha.com) Microsoft is rolling out new Copilot features like “Researcher” and “Council Mode” to help teams vet outputs, but it has also said Copilot may send data outside the EU during peak demand — a shift that raises compliance questions for multinational buyers. (hubsite365.com) (cybernews.com)
Microsoft is trying to sell an office assistant that sits inside Word, Excel, Outlook, and Teams, but many companies are still treating it like a pilot program instead of a standard software roll-out. Microsoft’s listed price for Microsoft 365 Copilot is now “starting from $18.00 user/month, paid yearly” in the United States, after earlier pricing that was widely framed around $30 per user, and that still leaves buyers asking which employees actually save enough time to justify it. (microsoft.com) The adoption problem is visible in Microsoft’s own tooling. Microsoft now offers a Microsoft 365 Copilot adoption report inside Viva Insights so managers can see which teams use Copilot often, which is the kind of dashboard vendors add when they know many paid seats are going quiet after launch. (learn.microsoft.com) Outside analysts have been pressing Microsoft on the same issue. Computerworld reported in January that Microsoft was claiming 15 million paid Microsoft 365 Copilot users, while analysts still said paid seat growth was lagging expectations for a product Microsoft has put at the center of its artificial intelligence strategy. (computerworld.com) Microsoft’s answer has been to make Copilot look less like a single chatbot and more like a team of reviewers. On March 30, 2026, Microsoft said its Researcher agent was adding two new multi-model features called Critique and Council to improve “accuracy, depth, and confidence” on harder work tasks. (techcommunity.microsoft.com) Critique works like handing a draft to a second editor before it leaves the building. Microsoft says one model can challenge another model’s assumptions, check gaps, and push the system to produce a stronger final report instead of giving the first polished answer it finds. (techcommunity.microsoft.com) Council goes one step further and asks multiple models to answer the same question in parallel, then compares where they agree and where they split. That is Microsoft admitting a practical fact about workplace artificial intelligence: companies trust an answer more when they can see disagreement, not just fluency. (techcommunity.microsoft.com) At the same time, Microsoft is loosening one of the promises that made Copilot easier to buy in Europe. Microsoft Learn says a new feature called flex routing lets customers in the European Union and the European Free Trade Association allow large language model inferencing outside the European Union Data Boundary during periods of peak demand. (learn.microsoft.com) Inferencing is the moment when the model actually processes a prompt and generates an answer, like the instant a call center agent picks up the phone. Microsoft says stored customer data will remain inside the European Union Data Boundary, but some prompt processing can move outside that zone when demand spikes, with data encrypted in transit and at rest. (learn.microsoft.com) That change is not theoretical. Microsoft-focused administrators have been told flex routing will be enabled for European Union and European Free Trade Association tenants starting April 17, 2026, and reports from Microsoft message center posts say the setting is on by default unless an organization changes it in the Microsoft 365 admin center. (office365itpros.com) So Microsoft now has two sales jobs instead of one. It has to convince companies that Copilot’s newer “committee of models” features make the answers reliable enough for real work, and it has to convince compliance teams that occasional processing outside the European Union will not break internal rules that were written assuming the opposite. (techcommunity.microsoft.com) (learn.microsoft.com)