Does Anthropic train its models on our company's Claude chats?

Not by default. Anthropic's privacy centre says it does not use inputs or outputs from commercial products, which include Claude for Work plans, to train its models. The exception is feedback: if a user rates a response with thumbs up or down or reports a bug, that conversation may be used. Tell staff this when you roll Claude out.

What is the difference between Claude Team and Enterprise?

Team is for 2 to 150 seats and includes Projects, connectors, memory, skills, Claude Code, Cowork, single sign-on and spend controls. Enterprise starts at 20 seats and adds SCIM provisioning, audit logs, a Compliance API, custom data retention, fine-grained role-based permissions, network-level access control and IP allowlisting. Larger or regulated organisations usually need the Enterprise controls.

Where is Claude data stored for UK organisations?

Anthropic's privacy centre states that data is stored in the US and that traffic may be routed to select countries in the US, Europe, Asia and Australia. Enterprise customers on usage-based billing can select a routing location. Anthropic's Data Processing Addendum includes the UK Addendum to the Standard Contractual Clauses, which your data protection officer will want to review.

Can Claude read our SharePoint and OneDrive files?

Yes, through the Microsoft 365 connector, once an owner enables it for the organisation and each user signs in. Claude only sees what that user can already access in Microsoft 365, so tidy permissions matter. Owners can also set action limits per connector, such as requiring approval before Claude takes any action, which is a sensible default for a pilot.

How many people should be in a Claude pilot?

There is no official number. A practical pilot is small enough to support closely and varied enough to test real work: a handful of people from two or three teams, each with a champion and a clear use case. The point is to learn which tasks genuinely improve, so pick people who do those tasks every week.

Claude for Business: A Practical Guide for Teams in the UK

Claude for business means the Team plan, for 2 to 150 seats, or Enterprise, which starts at 20 seats. Both give an organisation shared Projects, artifacts, connectors to tools such as Microsoft 365, Google Drive and Slack, and central admin controls. Anthropic says it does not train its models on commercial inputs and outputs by default. Start with a two-week pilot on three real tasks.

At a glance

	Team	Enterprise
Seats	2 to 150	20 minimum
Single sign-on	Yes	Yes
SCIM and audit logs	No	Yes
Custom data retention	No	Yes
Compliance API, IP allowlisting	No	Yes
Projects, connectors, memory, skills	Yes	Yes
Claude Code and Cowork	Yes	Yes

Feature lists change often. Check Anthropic's pricing page for the current comparison and prices before you buy.

What Team and Enterprise add over personal accounts

The main reason to move from staff using personal Pro accounts to a business plan is control. You get central billing, single sign-on, spend limits at organisation and user level, and owners who decide which capabilities and connectors exist at all. Team offers Standard and Premium seats, and Anthropic says Premium seats give five times the usage of Standard, so you can mix them: Premium for the handful of heavy users, Standard for everyone else.

Enterprise adds the controls that larger or regulated organisations ask for: SCIM provisioning, audit logs, a Compliance API for monitoring, custom data retention, fine-grained role-based permissions, network-level access control and IP allowlisting.

The features your team will actually use

Projects

A project is a self-contained workspace with its own chat history, knowledge base and instructions. Upload the documents a task depends on, such as a style guide, a product sheet or last year's report, and write instructions once instead of repeating them in every chat. On Team and Enterprise you can share a project with colleagues as Can view (chat in it, but not change it) or Can edit (change knowledge, instructions and members). On paid plans, when project knowledge nears the context limit, Claude switches to retrieval, which Anthropic says expands capacity by up to ten times.

Artifacts

An artifact is something Claude makes that you would put in front of someone: a document, a deck, a dashboard, a diagram or a small interactive tool. Artifacts open in their own window, start private to you, and can be shared within the organisation. On Team and Enterprise, they need the code execution and file creation capability switched on in organisation settings.

Connectors

Connectors let Claude search, read and act in tools such as Microsoft 365, Google Drive, Gmail, Google Calendar, GitHub and Slack. An owner enables each connector for the organisation, and then each person signs in to it individually. Claude only sees what that person can already see in the connected service. Owners can also set action limits per connector: always allow, needs approval, or blocked.

Memory and skills

Memory is off for Team and Enterprise members until an owner makes it available, and each member then chooses to turn it on. Each project keeps its own memory, separate from other projects. Skills are folders of instructions, scripts and resources that teach Claude a repeatable task, such as producing a report in your house format. Owners can provision approved skills to everyone.

Cowork and Claude Code

Claude Cowork hands Claude multi-step work. In the desktop app, it can read, edit and create files in folders you choose, and it asks for approval before deleting anything. Anthropic is folding the Cowork name into Claude itself, so the label may differ in your app. Claude Code is the developer tool for working in codebases. Admins can turn Cowork off for the organisation and control tool permissions by department.

Data handling: what UK teams should check

Your data protection officer will ask these questions first, so answer them before the pilot.

Training. Anthropic's privacy centre says it does not use inputs or outputs from commercial products for model training by default. The exception is feedback: if someone presses thumbs up or down, that conversation may be used, and feedback data is kept for five years.
Retention. Deleted chats leave your history immediately and Anthropic's back-end storage within 30 days. Chats flagged by trust and safety systems may be kept for up to two years. Enterprise adds custom retention controls.
Location. Anthropic states that data is stored in the US, and that traffic may be routed to select countries in the US, Europe, Asia and Australia. Enterprise customers on usage-based billing can select a routing location.
Transfers. Anthropic's Data Processing Addendum includes the UK Addendum, the international data transfer addendum to the EU Standard Contractual Clauses issued by the Information Commissioner.
Connected services. Data reached through a connector is processed by that service under its own terms, so your Microsoft 365 or Google settings still apply.

Record these answers in a data protection impact assessment if your use involves personal data. If staff use Claude for work delivered to EU clients, read our guide to the EU AI Act's AI literacy duty as well.

Admin settings to decide before rollout

Setting	Pilot recommendation
Single sign-on	Turn on before inviting anyone
Code execution and file creation	On, so artifacts and skills work
Connectors	Enable two or three, not all
Connector action limits	Needs approval for anything that writes
Memory	Off until your policy covers it
Organisation instructions	Short: tone, spelling, what not to share
Spend limits	Set per user from day one
Cowork	On for the pilot group only

Write a one-page AI use policy before the pilot, not after. Our guide to what an AI use policy should include gives a starting structure.

Good first use cases

Pick tasks that happen every week, have a clear owner and are easy to check.

Use case	Feature	Good result
Meeting notes to actions	Project with team template	Actions with owners, checked in minutes
Policy or proposal first drafts	Project knowledge	Draft in house style, then human edit
Spreadsheet analysis	Code execution	Summary plus the steps it took
Research brief	Research	Sources you can open and verify
Small internal tool	Artifact	A calculator or tracker the team reuses
Customer reply drafts	Skill or project	Consistent tone, reviewed before sending

Avoid starting with anything where a wrong answer is costly and hard to spot, such as final legal wording or figures sent straight to a client.

A two-week pilot plan

Days 1 and 2, set up. Configure SSO and the settings above. Choose a small pilot group from two or three teams, with one champion per team. Create one shared project per use case.
Day 3, train. Run a short session on prompting, checking outputs and your data rules. Show one worked example per use case.
Days 4 to 8, use it daily. Each person logs three things per task: roughly how long it took before, how long it took with Claude, and what they had to fix.
Day 9, check quality. A reviewer from each team samples outputs against the original standard.
Day 10, decide. Keep the use cases that saved time without lowering quality, drop the rest, and agree the rollout scope, seat mix and settings.

Measure your own results rather than relying on anyone's headline figures. A pilot that ends with "two use cases worked, one did not, here is why" is a success.

Common mistakes

Enabling every connector on day one, before anyone has thought about permissions.
Buying Premium seats for everyone before you know who the heavy users are.
Skipping the checking step because early outputs looked good.
Treating the pilot as a software trial rather than a change in how people work.

How CertTulen Academy can help

CertTulen Academy runs live, instructor-led Claude training. Claude 101 suits a pilot group that is starting out, and Introduction to Claude Cowork suits people ready to hand Claude multi-step work. For a pilot, a private corporate cohort keeps the examples focused on your team. Contact us for a batch that suits your UK time zone.