Insights
Claude vs Copilot for Microsoft 365 organisations (2026 guide)
You already pay for Microsoft 365, and Copilot is either in your estate or on the table. So the question in front of you is rarely "Claude or Copilot?". It is where Claude fits in an estate that runs on Microsoft, what it adds, and how to avoid paying twice for the same work. This guide answers that for a UK CIO. Everything below is correct as of September 2026.
We are Kowalah, a UK Anthropic Implementation Specialist, so we have a view. We have kept the facts dated and sourced so you can check them.
The short answer
- Keep Copilot for everyday work inside Outlook, Teams, Word and Excel. It sits inside your Microsoft security boundary and draws on your Microsoft Graph data.
- Add Claude where the work is long, careful and document-heavy, where you want an agent that carries out multi-step tasks, or where you plan to build agents of your own.
- Claude is already inside Copilot. Microsoft now offers Anthropic's Claude models in its own Copilot products, so part of the decision is which models your users can reach.
First, the names have changed
In September 2026 Microsoft renamed Microsoft 365 Copilot to Microsoft Copilot, and Microsoft 365 Copilot Chat to Microsoft Copilot Chat. Security, compliance and privacy terms did not change. We use the new names below.
What Microsoft announced on 25 September 2026
Microsoft rebuilt Copilot around three new parts:
- Home, the new starting point, where Chat and Cowork come together, with Word, Excel and PowerPoint built into the experience ("Office in Copilot"). It rolls out to Microsoft's Frontier programme in the coming weeks.
- Code, which lets people who are not developers build small apps, trackers and automations by describing them, running in a sandbox. It is built on the same technology as GitHub Copilot and reaches Frontier from the end of September.
- Autopilot, a persistent, cloud-hosted agent that its owner gives a name, a role and a goal. It watches channels, follows up threads and runs recurring work without being prompted. It enters private preview at the end of September.
The change that matters most for your budget is billing. The per-user licence now covers Chat and the Office apps, with Microsoft choosing the model automatically. Agentic work (Cowork, Code and Autopilot) moves to usage-based billing, where administrators choose which model families each group can use and set spending policies.
Three ways to run Claude in a Microsoft estate
| Claude inside Microsoft Copilot | Claude Enterprise with the Microsoft 365 connector | Claude models in Microsoft Foundry | |
|---|---|---|---|
| What it is | Anthropic's models as a choice inside Copilot, Researcher, Copilot Studio, Power Platform and Copilot in the Office apps | Anthropic's own app (chat, Claude Cowork, Claude Code) reading and acting on your Microsoft 365 data | Claude models running in your Azure environment for the agents and applications you build |
| Whose terms | Microsoft's Product Terms and Data Protection Addendum, with Anthropic as a Microsoft subprocessor | Your agreement with Anthropic | Your Azure agreement |
| UK tenants | Off by default; an AI Administrator or Global Administrator opts in | Needs one-time consent from an Entra Global Administrator, and a Claude organisation owner to enable it | You choose the region |
| Best for | Getting Claude's reasoning without a new vendor or licence | The full Claude product, including agents that do work across your systems | Production agents with regional data requirements |
1. Claude inside Microsoft Copilot
Microsoft has onboarded Anthropic as a subprocessor, so Claude models can power Copilot experiences under Microsoft's own terms: the Product Terms, the Data Protection Addendum and Microsoft's Customer Copyright Commitment all apply. In Copilot Studio, agent builders choose the model; in Researcher and in Edit with Copilot, users can select Claude. Microsoft's Copilot Cowork, its agent for multi-step work across Microsoft 365, was built with Anthropic.
The UK detail that matters: Anthropic models in Copilot are excluded from the EU Data Boundary and from in-country processing commitments, and for that reason they are off by default in the UK and the EU. An AI Administrator or Global Administrator turns them on, for everyone or for named groups. Since April 2026 there is also a separate setting to make Anthropic the default model for Copilot in the Office apps in the UK.
2. Claude Enterprise with the Microsoft 365 connector
Anthropic's Microsoft 365 connector is available on every Claude plan. It searches across SharePoint, OneDrive, Outlook and Teams, and reads Word, Excel, PowerPoint and PDF files. Since July 2026 it can also act: sending email, managing calendar events, creating and updating files, and sending Teams messages, with each send action set to Ask or Blocked.
Two points for your Microsoft administrators:
- Nobody in the tenant can connect until an Entra Global Administrator grants one-time consent, and write tools stay off until a Claude administrator turns them on.
- SharePoint search needs the tenant-wide Sites.Read.All permission; site-specific permissions are not supported. Claude only sees what each user can already see, so fix oversharing in SharePoint before you switch it on.
Content Claude retrieves through the connector is processed by Anthropic under your Anthropic agreement, not Microsoft's. Review it as you would any third-party app with Microsoft Graph permissions.
3. Claude models in Microsoft Foundry
For agents and applications you build, Anthropic serves Claude models through Microsoft Foundry, where you choose the region. This is the route for UK or EU data residency, and it keeps the build inside your Azure governance. It does not replace the Claude app your employees use day to day.
What each does well
Copilot is strongest where your people already work. It lives in the ribbon of every Office app, grounds answers in your Graph data (email, meetings, files and the relationships between them) and runs inside identity, permissions and compliance controls your IT team already manages. For everyday drafting, meeting summaries and quick answers about your own documents, that closeness is hard to beat. With Autopilot and Code, Microsoft is now moving into always-on agents and app building too, both in preview.
Claude Enterprise is strongest on depth. It handles long, careful work well: contracts, policies, analysis and board papers. Claude Cowork carries out multi-step tasks on the desktop with access to files and connected systems, Claude Code serves engineering teams, and Anthropic's Managed Agents platform already runs always-on agents with their own tools and memory; we run Kowalah's own delivery on it. New Anthropic models also reach Claude first.
The cost overlap
Microsoft Copilot is a per-user add-on to a qualifying Microsoft 365 licence. Microsoft's US list price is $30 per user per month on an annual commitment for enterprise plans, and $21 for Business plans; UK pricing comes through your Microsoft agreement. Since the September 2026 announcement, that licence covers Chat and the Office apps, and agent work (Cowork, Code and Autopilot) is billed on usage. Claude Team is listed at $20 per seat per month billed annually, and Claude Enterprise is a seat price plus usage at API rates, agreed with Anthropic.
Both vendors now charge for agent work by consumption. Put spend controls and usage reporting in place on both before agents go live, not after the first invoice. For Claude, our 10 ways to take control of your Anthropic spend covers the levers, from organisation-wide limits to default models.
Licensing both for everyone is rarely the answer. Instead:
- Give Copilot to the roles whose work lives inside Outlook, Teams and Office.
- Give Claude Enterprise to the roles doing long analysis, drafting and building: finance, legal, strategy, product and engineering.
- Turn on Claude inside Copilot for everyone else, which adds Anthropic's models without adding a licence.
- Model the first year on expected active users, not headcount, and review usage at 90 days.
If your Copilot rollout stalled
A common story: Copilot licences went out, usage peaked in the first month, and now a fraction of seats are active. Adding Claude will not fix that on its own, because the cause is rarely the model. Adoption stalls when people are given a tool without a use case, a champion or a reason to change how they work.
Before buying more licences, pick two or three workflows with a measurable outcome, put a named owner and a network of internal champions behind them, and measure adoption from launch day. That applies whether the workflow runs in Copilot, in Claude, or across both. Our Change Enablement work is built around this.
How to decide
Answer these five questions:
- Which work matters most to the business? Quick tasks inside the Office apps favour Copilot; long documents, analysis and agents favour Claude.
- Who will build agents, and where? Copilot Studio suits agents that live inside Microsoft 365. Anthropic's platform or Foundry suits agents that reach across other systems.
- What are your UK data requirements? Agree with your DPO which data classes each route is allowed to process, before switching anything on.
- Is SharePoint permissioning in order? Both Copilot and Claude surface whatever a user can already reach.
- What does success look like in 12 months? Define the business outcome first; the licences are a means to it.
Where Kowalah fits
We implement Claude in Microsoft-standardised organisations, working inside your Entra identity, Purview governance and Azure security from day one. Our own Kowalah Agent runs in Microsoft Teams, so we work the way your people already do. For the other decisions around Claude Enterprise, read our Claude Enterprise guide for UK organisations. When you are ready to put Claude into production, see Claude implementation for UK enterprises or book a conversation.
Frequently asked questions
Can we use Claude inside Microsoft Copilot?
Yes. Microsoft offers Anthropic's Claude models in Copilot, Researcher, Copilot Studio, Power Platform and Copilot in the Office apps, under Microsoft's terms with Anthropic as a subprocessor. In the UK they are off by default, and an AI Administrator or Global Administrator turns them on.
Is Claude in Microsoft Copilot covered by the EU Data Boundary?
No. Microsoft excludes Anthropic models in its Copilot products from the EU Data Boundary and from in-country processing commitments, which is why they are off by default in the UK and the EU.
Does Claude work with Teams, Outlook and SharePoint?
Yes. Anthropic's Microsoft 365 connector, on every Claude plan, searches SharePoint, OneDrive, Outlook and Teams. Since July 2026 it can also send email, manage calendars, update files and send Teams messages, once an administrator enables write tools.
Should we replace Copilot with Claude?
Rarely. For a Microsoft organisation we recommend both: Copilot for everyday work inside the Office apps, Claude for deeper analysis and agents, with licences assigned by role rather than to everyone.
How much do Copilot and Claude cost?
Microsoft Copilot lists at $30 per user per month for enterprise plans (US list price, annual commitment), on top of a Microsoft 365 licence; that covers Chat and the Office apps, and agent work such as Cowork, Code and Autopilot is billed on usage. Claude Team lists at $20 per seat per month billed annually; Claude Enterprise is a seat price plus usage, agreed with Anthropic.
What is Microsoft Copilot Autopilot?
Autopilot is a persistent, cloud-hosted agent Microsoft announced on 25 September 2026. Its owner gives it a name, a role and a goal, and it keeps working inside Microsoft 365, watching channels, following up threads and running recurring work. It enters private preview at the end of September 2026 and is billed on usage.
Enjoyed this?
Weekly insights for executives navigating enterprise AI. A five-minute read.