Claude Enterprise and Microsoft Copilot are both per-seat AI subscriptions sold to organisations, but they sit in different places. Copilot runs inside Word, Outlook, Excel and Teams, on top of a Microsoft 365 licence you already hold. Claude Enterprise runs as a separate workspace your team signs into.
For most Microsoft-house organisations, Copilot alone is the right answer. Add Claude Enterprise when a defined group does long-document analysis or sustained research that Copilot handles badly, and buy it for that group only. Before you buy anything, check whether Anthropic's models are already sitting switched off inside the Copilot you pay for.
That last point changes the shape of the decision, and almost nobody mentions it. This post answers the organisational question only: what you commit to contractually, what your admins can control, what a seat costs once you count the licence underneath it, and who owns the rollout. For the feature-by-feature verdict on writing, research and coding, and for individual subscription prices, read our comparison of whether Claude is better than Copilot on everyday work.
Criterion | Microsoft Copilot | Claude Enterprise |
|---|---|---|
Seat price | US$30 per user per month, billed annually | US$20 per seat per month, plus usage at API rates |
What sits underneath | A qualifying Microsoft 365 or Office 365 licence, bought separately | Nothing. The seat is the whole product |
Minimum commitment | None stated beyond the annual term | 20 seats, annual term |
Training on your content | No, per Microsoft's published commitment | No, per Anthropic's commercial terms |
Identity and provisioning | Microsoft Entra ID, already deployed | SAML or OIDC single sign-on, plus SCIM |
Audit and eDiscovery | Microsoft Purview, Content search, retention policies | Audit logs plus a Compliance API |
Where it lives | Inside the apps people already have open | A separate workspace, web and desktop |
What have you already bought without switching it on?
If you pay for Microsoft Copilot, Anthropic's Claude models are already part of the product, and in the UK they are off by default. Microsoft onboarded Anthropic as a subprocessor and turned its models on by default across most of the commercial cloud, but its own documentation states that "Customers within the EU Data Boundary and customers in the UK have Anthropic models disabled by default".
Turning them on is a setting, not a purchase. An AI Administrator or Global Administrator goes to the Microsoft 365 admin centre, opens Copilot, then Settings, then View all, then "AI providers operating as Microsoft subprocessors", and selects Anthropic and the users or groups who should get access. Claude then appears as a model choice in Copilot, Researcher, Copilot Studio and Copilot in Microsoft 365 apps.
There is a real trade-off attached, and it is the reason the toggle exists. Anthropic models inside Microsoft's products are, in Microsoft's words, "currently excluded from the EU Data Boundary, and when applicable, in-country processing commitments". A UK firm that has told its customers where their data is processed needs to read that sentence before an administrator flips the switch.
Two categories are worth telling apart. Standard Anthropic models run with Anthropic as a Microsoft subprocessor, covered by the Microsoft Product Terms and the Data Protection Addendum you already signed. Anthropic Preview models with Data Retention are different: Anthropic acts as an independent data processor, your Microsoft Customer Agreement does not cover the interaction, and those models stay off until an admin opts in specifically. If you want the practical walkthrough for end users once it is enabled, we have covered how to use Claude inside Microsoft Copilot.
What does each vendor commit to on your data?
Both vendors commit in writing not to train their models on your content, and both commitments are strong enough to hand to a legal team. Microsoft's position is published on its privacy page for Copilot: "Prompts, responses, and data accessed through Microsoft Graph aren't used to train foundation LLMs, including those used by Microsoft Copilot." Anthropic's commercial terms state that "Anthropic may not train models on Customer Content from Services". Neither is a settings toggle. Both are contractual.
Retention is where they separate. Copilot stores prompts and responses as your users' Copilot activity history, encrypted, and treated in line with your other Microsoft 365 content, which means your existing Purview retention policies, Content search and eDiscovery reach it. For a regulated firm that already has a retention schedule signed off, this is the single strongest argument for Copilot: AI interactions land inside the compliance machinery you have already built rather than beside it.
Claude Enterprise offers custom data retention controls, customer-managed encryption keys and a Compliance API giving programmatic access to activity logs, chats, files and projects. That is a capable set. It is also a second system to configure, staff and evidence.
The one number worth writing down concerns those Preview models with Data Retention. Where they are enabled, Anthropic stores most inputs and outputs for up to 30 days, may retain content for up to two years if its trust and safety classifiers flag a potential usage policy violation, and may keep classification scores for up to seven years. Those are Anthropic's retention periods operating outside your Microsoft agreement, which is precisely why the setting is default-off. If you have not yet drawn the line between what staff may and may not put into any of these tools, that work comes first: we have set out which company data can go into a tool like ChatGPT and how to classify the rest.
What admin and audit controls do you actually get?
Copilot inherits every control you have already deployed, and Claude Enterprise asks you to build an equivalent set from scratch. That is the honest summary, and how much it matters depends entirely on how good your existing Microsoft estate is.
On the Microsoft side, Copilot surfaces organisational data only where the individual user already has at least view permissions, so your SharePoint and Teams permissions become your AI permissions. That cuts both ways. Firms with a decade of over-shared SharePoint sites find that Copilot is an excellent tool for discovering exactly how over-shared they are, usually in week one, usually via someone finding the salary file. Fixing that is a prerequisite, not a Copilot problem.
Claude Enterprise ships SAML 2.0 and OIDC single sign-on, SCIM provisioning from your identity provider, audit logs, role-based access with fine-grained permissioning, network-level access control, IP allowlisting and admin-set spend limits at both organisation and user level. Anthropic lists customer-managed encryption keys and a US-only inference option as well. A competent IT function can stand this up. It is a genuine second control plane, with its own joiners-movers-leavers process, its own audit export and its own review when the auditor asks who had access to what.
What does each cost per seat, all in?
Copilot is US$30 per user per month billed annually, and Microsoft's pricing page is explicit that "A separate license for a qualifying Microsoft 365 plan is required". Claude Enterprise is US$20 per seat per month with a 20-seat minimum on an annual term, and usage billed on top at API rates. Both are list prices in US dollars; UK buyers should check local pricing and their reseller agreement.
Two things about that comparison are widely reported wrong. The first is the Copilot prerequisite. Almost every comparison page states you need Microsoft 365 E3 or E5, and Microsoft's current licensing documentation lists a far broader set: Microsoft 365 E7, E5, E3, F1 and F3, Business Basic, Standard and Premium, Apps for business and enterprise, Office 365 E1, E3, E5 and F3, and standalone Teams, Exchange, SharePoint, OneDrive, Visio and Project plans. If you were told you had to upgrade the whole estate to E5 first, check that against the list.
The second is the Claude Enterprise number. The US$20 seat fee is not the bill. Usage is charged separately at API rates, so a team running heavy document analysis produces a very different invoice from a team drafting occasionally, on the same seat count. Claude Team is US$20 per seat billed annually or US$25 monthly for 2 to 150 people with no separate usage charge, which makes it the predictable line item and the better starting point for most organisations buying either for the first time.
The cost nobody models is the unused seat. Copilot bills per assigned licence whether or not anyone opens it, and Microsoft gives you the tool to check: the Copilot usage report in the admin centre reports enabled users, active users and an active users rate, which is simply active divided by enabled. Run that report before you buy a second AI product. If your current active users rate is poor, a new vendor will not fix it, and you will be paying two bills for one habit.
Will people use a tool that is not already open?
Adoption follows proximity, and this is the strongest structural argument for Copilot. It appears in Outlook while someone is already writing the email. Claude Enterprise requires a person to leave the document they are working on, move to another window, paste context in, and come back. That sounds trivial written down. Across 200 people and a busy quarter it is the difference between a tool that becomes a habit and a tool that becomes a renewal conversation.
The counter-argument is real. A separate window suits work that is not a quick edit: reading four contracts against each other, working through a long research question, building an analysis over an afternoon. Nobody doing that minds switching windows. The mistake is buying the separate tool for everyone when the work justifying it belongs to 15 people in legal, strategy and finance.
Neither product solves adoption on its own. If your Copilot rollout has stalled, the cause is usually that nobody changed what the team is measured on, and we have written separately on getting employees to actually use the AI tools you bought.
Who owns the rollout?
Copilot has an obvious owner and Claude Enterprise usually does not, and that gap is where second-tool rollouts fail. Copilot lands with IT because the licences, the admin centre, the Entra groups and the Purview policies are already theirs. The path is well worn.
Claude Enterprise arrives without a natural home. It is usually championed by the function that wanted it, which is rarely the function that can run identity, audit and offboarding for it. Before signing, name three people: who administers the workspace, who reviews the audit logs and on what schedule, and who removes a leaver's access on their last day. If those names are not available, the tool gets bought and then orphaned. A subscription sitting outside the IT budget also escapes the annual licence review that would have caught it going unused.
Which should you actually buy?
Copilot alone, for most Microsoft-house organisations. If the work is email, documents, spreadsheets, meetings and Teams, and nobody is doing sustained research or long-document analysis, Copilot covers it and adds nothing new to govern. Spend what you saved on making people competent with it.
Copilot plus Claude Enterprise for a named group. Where a defined team does deep research, long-context analysis or heavy drafting, buy Claude Enterprise for that team at its 20-seat minimum and leave everyone else on Copilot. This is the most common right answer at 200 people and above, and it is a deliberate two-tool position rather than a migration.
Claude Enterprise alone, if you are not a Microsoft house. On Google Workspace or Slack, the Copilot integration argument does not apply and this is a different comparison.
Neither, yet, if your active users rate is poor. An organisation with 300 Copilot licences and 60 active users has an adoption problem, not a tooling problem. A second product makes it a more expensive adoption problem.
Where this advice comes from: ivee is not a licence reseller. We take no commission on Copilot or Claude, we cannot get you a better price on either, and we have no stake in which you pick. What we do is make bought seats get used, which is why the last recommendation is one a reseller would never make.
Is this really a choice between two products?
Usually it is not, and treating it as one is the most expensive mistake here. Three framings cause most of the damage.
"We need to pick a winner." Most organisations getting value from both run a two-tool position on purpose: Copilot everywhere for work inside Microsoft 365, Claude for a small group doing work that is not. Picking one for everyone optimises for procurement simplicity and against how the work is distributed.
"Claude is not available to us." It may already be, inside Copilot, waiting on an admin setting. Confirm what your tenant has enabled before running a procurement exercise for something you own.
"The model is the decision." Model quality moves every few months and both vendors ship upgrades continuously. Contractual terms, retention behaviour, admin control and rollout ownership move slowly, and those are what you sign up to for three years. Choose on the slow-moving things.
Working out what to do with the seats you already pay for?
ivee runs AI training and adoption programmes for UK organisations, and we are vendor-neutral on which tools you licence because we do not sell any of them. If you have Copilot deployed and an active users rate you would rather not put in front of the board, that is the problem we are usually called in to fix. Tell us what your rollout looks like and we will tell you whether you have a tooling problem or a habit problem.
