We already have Microsoft Copilot. Why would we add Claude?

What your Microsoft 365 license actually includes

Start here, because this is where the conversation usually goes wrong. Somebody in the room says the AI is already included. That is half true, and the half that is not true is the half that decides the whole question.

What is included is Microsoft 365 Copilot Chat. On Microsoft's own Copilot Chat overview page, Microsoft writes that unlike Microsoft 365 Copilot, Copilot Chat is not grounded in organizational content like files, emails, or chats as part of the chat experience, and that it is grounded in data from the web only. Microsoft then lists three documented ways a user can get company content into it anyway: putting it in the prompt by pasting, uploading, or picking a file; using Copilot Chat inside select Microsoft 365 apps like Teams and Outlook, where it is aware of the content the user has open; or using a pay-as-you-go agent that has access to organizational content.

So the free tier is real and it is useful. It is secure AI chat with enterprise data protection, IT controls, file upload, image generation, and Copilot Pages, at no extra cost, for users on the license list Microsoft publishes. It is not the thing most people picture when they picture Copilot. Microsoft even labels this inside the apps: a user with the add-on license sees M365 Copilot (Premium), a user without it sees M365 Copilot (Basic) or Copilot Chat (Basic), and Microsoft notes that Copilot Chat (Basic) means no access to Copilot chat in Word, Excel, PowerPoint, and OneNote at all.

The paid Microsoft 365 Copilot add-on is what adds work-grounded chat through Microsoft Graph, Copilot inside Word, Excel, PowerPoint, Outlook, and Teams, the prebuilt Microsoft agents including Researcher, Analyst, and Facilitator, and the analytics an admin uses to prove any of it worked. Microsoft publishes two add-on families with different prerequisite lists, Microsoft 365 Copilot Business for the Business Basic, Standard, Premium and Apps for business plans, and Microsoft 365 Copilot for a much longer list that includes E3, E5, E7, F1, F3, and the Office 365 plans.

On price, read the live page rather than trusting anyone's summary, including this one. When we checked Microsoft's plans and pricing page it listed Microsoft 365 Business Premium with Copilot at 32.00 US dollars per user per month paid yearly and Microsoft 365 Business Standard with Copilot at 23.50 US dollars per user per month paid yearly, both as annual auto-renewing subscriptions. The third column, the standalone Microsoft 365 Copilot Business add-on, rendered a 0.00 placeholder rather than a real figure, which is a page bug rather than a free product. Get the add-on number from that page or from your own admin center quote, not from a blog.

• ·Included at no extra cost with an eligible plan: Copilot Chat, web-grounded, with enterprise data protection and admin controls. Microsoft's eligibility list includes Microsoft 365 Business Basic, Business Standard, Business Premium, Microsoft 365 E3 and E5, F1 and F3, G3 and G5, Office 365 E1, E3 and E5, and the Microsoft Teams plans
• ·Paid add-on only: chat grounded in work data through Microsoft Graph, Copilot in Word, Excel, PowerPoint, Outlook and Teams, prebuilt agents like Researcher and Analyst, Copilot Analytics, SharePoint Advanced Management
• ·Also paid add-on only: model choice in the Microsoft 365 Copilot app, which is the setting that lets a user pick Claude. Microsoft's support page for Claude in Researcher states you must have a Microsoft 365 Copilot license to use model choice
• ·Worth knowing before you promise anyone anything: Copilot Chat gets standard access to capabilities and models, which Microsoft says is subject to service capacity. Priority access comes with the paid license
• ·Copilot Chat cannot be fully switched off with one setting. Microsoft documents that as of January 28, 2026 the pinning control no longer removes Chat from the Microsoft 365 Copilot app navigation for any tenant worldwide, and that blocking it properly means managing the Copilot integrated app, Teams app policies, and an Edge policy

What Copilot is genuinely better at, and it is not a short list

If you are looking for a page that tells you Microsoft is bad at this, close this tab. Copilot has a structural advantage that no outside tool can copy, and pretending otherwise wastes your time.

Copilot lives inside the apps your staff already have open. It reaches your tenant data through Microsoft Graph, and Microsoft's architecture documentation states plainly that Copilot only accesses data that an individual user is authorized to access based on existing Microsoft 365 role-based access controls, and that it does not access data the user does not have permission to access. It honors your Conditional Access policies and the multifactor authentication you already configured. That means the access review you did years ago is the access review that governs Copilot, with no new identity surface to argue about.

The commercial terms are the same ones you already signed. Microsoft's enterprise data protection page states that use of Microsoft 365 Copilot and Copilot Chat is covered by the Microsoft Products and Services Data Protection Addendum and the Microsoft Product Terms, with Microsoft acting as a data processor, that Copilot inherits your sensitivity labels, applies your retention policies, and supports audit of interactions, and that prompts, responses, and data accessed through Microsoft Graph are not used to train foundation models. For a 200 person services firm, skipping a fresh vendor security review is worth real money.

Two caveats on that same page deserve to be read out loud, because they are the parts a vendor summary drops. Web search queries are handled by the Bing search service, which Microsoft describes as operating separately from Microsoft 365 with Microsoft acting as an independent data controller under the Microsoft Services Agreement and Privacy Statement. Microsoft's footnotes then state that HIPAA compliance does not apply to web search queries, and that the EU Data Boundary does not apply to web search queries either. If you handle protected health information, that footnote is a policy decision you have to make, not a detail.

• ·Copilot wins on anything that lives in a Word, Excel, PowerPoint, Outlook, or Teams surface and needs your tenant context
• ·Copilot wins on permission inheritance. No new access model, no new deprovisioning path, no new argument with IT
• ·Copilot wins on procurement. One agreement, one DPA, one invoice, no second security questionnaire
• ·Copilot wins on meeting recaps, because the transcript is already in Teams and no outside tool can get to it as cleanly
• ·Copilot wins on governance tooling if you already own Purview, since sensitivity labels, retention, and audit carry over

Microsoft already ships Claude inside Copilot, and that changes the question

This is the fact that reframes the whole debate, and most people in the room do not know it yet. Microsoft has onboarded Anthropic as a Microsoft subprocessor. Microsoft's own documentation states that unless models are labeled Preview models with Data Retention, the Microsoft Product Terms and the Microsoft Data Protection Addendum apply to use of Anthropic models through Microsoft's enterprise Online Services, that such use is also covered under Enterprise Data Protection, and that the Microsoft Customer Copyright Commitment applies to Anthropic models used within products covered by it, including Microsoft 365 Copilot and Copilot Studio.

The default is more aggressive than most admins realize. Microsoft states that it enables Anthropic models on by default for most customers in commercial cloud, excluding the EU, EFTA, and the UK, where the setting appears but defaults to No users. The same page lists where those models show up: Microsoft 365 Copilot, Researcher, Copilot Studio, Power Platform, and Copilot in Microsoft 365 apps, with UI indicators when Claude models are in use. Microsoft's support article for Claude in Researcher says a user selects Claude from the model dropdown, that an admin must first allow access to Anthropic AI models in the Microsoft 365 admin center, and that model choice requires a Microsoft 365 Copilot license.

So before you buy anything, have someone with the AI Administrator or Global Administrator role open the Microsoft 365 admin center, go to Copilot, Settings, View all, and look at AI providers operating as Microsoft subprocessors. You may find Claude is already available to your staff and nobody told the leadership team. That is a five minute check that has saved more than one client a purchase order.

Now the inconvenient parts, in the order your IT lead will find them. Microsoft states that Anthropic models deployed in Microsoft offerings are currently excluded from the EU Data Boundary and, where applicable, in-country processing commitments. Anthropic models are not available for federal customers in GCC, or for any customers in GCC High or Department of Defense environments, or in other sovereign clouds. Non-federal GCC customers got an opt-in setting on July 22, 2026, and Microsoft warns directly that when enabled, those models process customer data outside Microsoft's FedRAMP-authorized US Government cloud.

The sharpest distinction is the one almost nobody reads. Microsoft carves out a category it calls Anthropic Preview models with Data Retention, and on that page names Claude Fable 5 and Claude Mythos 5 as examples. For those models Microsoft states that Anthropic acts as an independent data processor, not a Microsoft subprocessor, that use is subject to Anthropic's own Commercial Terms of Service and Data Protection Addendum rather than your Microsoft Customer Agreement, that the setting is default off for every tenant even where other Anthropic models are on by default, and that Anthropic stores most inputs and outputs for up to 30 days, may retain flagged content for up to two years, and may keep trust and safety classification scores for up to seven years. If your compliance posture depends on data staying under your Microsoft paper, leave that toggle off and say so in writing.

One first-party conflict to know about, because you will hit it if you research this yourself. Anthropic's September 24, 2025 announcement about Claude in Microsoft 365 Copilot says Anthropic models are hosted outside Microsoft-managed environments and subject to Anthropic's Terms and Conditions. Microsoft's subprocessor documentation, updated July 22, 2026, describes a different arrangement for the general case: Anthropic as a Microsoft subprocessor with Microsoft Product Terms and the Microsoft DPA applying, with the older arrangement now surviving only for Preview models with Data Retention. Both pages are still live. The Microsoft page is the newer one and the one that matches the admin center you are looking at, and it is the one your reviewer should read. We have linked both below rather than quietly picking one.

The practical consequence: vendor loyalty is the wrong axis. You are not choosing between two companies. You are choosing which surface a piece of work happens on and which model runs it, and Microsoft has already made those two decisions separable.

What buying Claude directly still gets you

Given all of that, there is still a real case, and it is narrower and more specific than a salesperson will tell you.

The integration runs in both directions, which most comparisons miss. Anthropic's own plans page lists Claude for Microsoft 365 and Claude for Microsoft Outlook as plan features, and lists Claude for Microsoft 365 in the Pro plan feature list. In other words, Copilot can call Claude models, and Claude can reach into Microsoft 365. Which direction is right depends on where the work starts. If the work starts in a Word document, Copilot is closer. If the work starts as a question, a pile of PDFs, or a client problem that has not become a document yet, Claude is closer.

Where we see the difference hold up inside real mid-sized firms, the jobs cluster tightly. Long-document reasoning where you need to see the argument, not just the summary: a 90 page contract, a payer policy update, a stack of inspection reports. Drafting in one specific human voice, sustained across dozens of pieces, which is the difference between copy you send and copy you rewrite. Work that never becomes an Office file, like triaging an inbox into a decision list or turning a messy spreadsheet export into a repeatable process. And anything adjacent to code or data scripting, where Claude Code is a category of its own and Copilot in Excel is not trying to compete.

On Anthropic's pricing, be careful, because the consumer and business plans are different products with different governance. Anthropic lists Free, Pro, Max, Team, and Enterprise. When we read the page it showed Free at 0 dollars, Pro at 17 US dollars per month with the annual discount at 200 dollars billed up front or 20 dollars billed monthly, and Max starting at 100 dollars per month, with a footnote that prices and plans are subject to change at Anthropic's discretion. The detailed feature comparison table on that page covered Free, Pro, Max 5x, and Max 20x only, so read Team and Enterprise numbers off the live page. This matters operationally: Pro and Max are individual subscriptions, so a firm that buys Pro seats for staff has bought consumer accounts, not an administered organization with SSO and central control.

And the honest cost nobody prices in: adding Claude directly means a second vendor security review, a second identity integration, and a second place your client data can go. That is not a reason to skip it. It is a reason to know what you are signing up for before you start, and to decide on Team versus Enterprise before IT decides for you.

• ·Buy Claude for a named list of jobs and name them out loud. Buying it as a general upgrade produces a graveyard of unused seats
• ·Give it to the five or ten people whose day is reading and writing, not to everyone. The heaviest users make the case for you
• ·Decide the plan before the pilot. If your reviewer requires automated deprovisioning or an audit trail, that requirement points at Enterprise, not Team
• ·Keep the Microsoft admin center check in the plan. If Claude in Copilot covers 80 percent of what your people wanted, buy for the other 20 percent, not for all of it

A decision rule you can apply in one sitting

Here is the version we use on a whiteboard with an ops leader. It takes about twenty minutes and it settles the argument, because it sorts by the shape of the work rather than by the logo.

Ask one question about each recurring piece of work: does this job start inside a Microsoft file or meeting, and does it need my company's data to be any good? If yes to both, it belongs on Copilot, and adding Claude will not help. If the job starts as a question or a pile of source material, and the quality of the reasoning or the writing is the whole point, it belongs on Claude. If the job needs both, do the thinking in Claude and land the artifact in the Microsoft file, which is how most of our clients actually run.

• ·Summarize this meeting, find that email thread, build this Excel formula, clean up this deck, draft a reply in Outlook: Copilot, paid add-on, done
• ·Read these 90 pages and tell me what I am exposed to: Claude, and check whether Claude in Researcher already covers it before you buy
• ·Write six weeks of client-facing content that sounds like our founder, not like a robot: Claude
• ·Turn our intake process into something a new hire can follow: Claude to build it, Microsoft 365 to house it
• ·Anything touching code, scripts, or data plumbing: Claude Code, and it is not close
• ·Anything that must inherit SharePoint permissions and land in an audit log your reviewer already accepted: Copilot
• ·Anything in GCC High, DoD, or another sovereign cloud: Copilot without Anthropic models, because Microsoft states those models are not available there

When the answer is no, or not yet

We have told firms not to buy Claude. Here are the cases where that is the right call, and the thing to do instead.

If almost nobody at your company uses Copilot today, adding a second tool makes adoption worse, not better. Most Copilot disappointment we see is not a model problem, it is that nobody was ever shown a job worth doing with it and no manager was ever accountable for the result. Two unused tools cost more than one unused tool and teach your staff that AI is theater. Fix adoption on the surface you already pay for, then revisit this in a quarter.

If your data has to stay inside the EU Data Boundary, read Microsoft's own footnote before you plan anything. Microsoft states that Anthropic models in Microsoft offerings are currently excluded from the EU Data Boundary and, where applicable, in-country processing commitments, and that EU, EFTA, and UK tenants have those models off by default. That is a real constraint on the Copilot side, and buying Claude directly does not solve it either. That is a conversation with both vendors about regional processing, not a checkbox.

If nobody will own it, do not buy it. Not IT, not a committee. One person who can name the five jobs, is measured on whether those jobs got faster or better, and has authority to tell people to stop doing it the old way. Without that, the outcome is predictable and the budget is gone.

And the case people find hardest to hear: for a lot of firms Copilot with the paid add-on is enough. If your team's real AI work is meeting recaps, Outlook drafts, deck cleanup, and Excel formulas, Copilot is the better tool for all four, model choice already lets a user pick Claude for the parts that benefit from it, and adding a second product adds a login and a review for no gain. That is a good outcome. Spend the money on training instead, and you will get more out of it than another subscription.

• ·Not yet: your Copilot usage report shows thin adoption and no manager owns changing that
• ·Not yet: no named owner, no named jobs, no measure of done
• ·Probably not: your AI work is entirely inside Office files and Teams meetings
• ·Not available as described: GCC High, DoD, federal GCC, and other sovereign clouds, where Microsoft states Anthropic models are not offered
• ·Blocked until resolved contractually: a hard EU or in-country data residency requirement, since Microsoft states Anthropic models in Copilot are currently outside the EU Data Boundary
• ·Yes, narrowly: a small group whose day is long-document reasoning, sustained writing, or code and data work, with one owner and five named jobs

What people ask next

• How much does Claude cost for a team?
• How do I roll out Claude to my whole team?
• What is the difference between Claude Team and Enterprise?
• Can I control what my staff put into Claude?

Worked examples