Setting up Claude and Anthropic workspaces for teams, the right way

Setting up Claude for a team means putting everyone on a shared Claude Team or Enterprise workspace instead of separate free accounts, then configuring the controls that make it safe for company use: central billing and user management, single sign on, permissions, and the assurance that your conversations are not used to train the models. That is the setup. The other half of the job is a plan to get people actually using it, because access on its own does not change how a team works.

What does setting up Claude for a team involve?

Two things, in order. First the workspace: a shared Claude Team or Enterprise account that the business owns and controls, rather than a scatter of personal logins the company cannot see. Central billing, so it is one line item and not expenses claims. User management, so joiners and leavers are handled properly. And the data assurance most businesses ask about first, that your conversations and uploads are not used to train the underlying models.

Second, and the part most rollouts skip, a plan for use. A licence is not adoption. Setting up the workspace is a morning's admin; getting a team to change how it works is the real project, and it needs an owner, a few proven use cases, and a way to spread what works.

Adoption of the technology is already wide. McKinsey's State of AI survey, published in 2025, found 71 percent of organisations regularly use generative AI in at least one business function. The gap now is between companies where a few people use it privately and companies where it runs through the team on controlled infrastructure.

Why not just let people use free accounts?

Because they already are, usually without you knowing. Microsoft and LinkedIn's 2024 Work Trend Index found 75 percent of knowledge workers use AI at work, and 78 percent of those users bring their own AI tools rather than wait for the company to provide one. That is a live situation in most businesses, not a future risk.

The risk with personal free accounts sits in the absence of control around them. Company information goes into an account the business cannot see, govern or switch off when someone leaves. There is no record of what was used or shared. And on consumer tiers, data handling is weaker than on the business plans. A shared workspace does not stop people using AI. It brings the use you already have into something you can see and secure.

What is the difference between Claude free, Team and Enterprise?

Free and individual Pro accounts are built for one person. They have no central admin, no shared billing, no governance, and business data protections are not the default. Fine for an individual, wrong for a company.

Claude Team adds the shared layer: one workspace for the group, central billing and user management, collaboration through shared projects, and business grade data protection where your content is not used for training. It suits a team that wants controlled, collaborative use without heavy IT requirements.

Claude Enterprise adds the controls a larger or more regulated organisation needs. Single sign on so access runs through your existing identity system, fine grained roles and permissions, more security and audit features, and higher usage limits. The honest way to choose is by team size and by how strict your security and compliance requirements are, rather than by reaching for the biggest plan on principle.

How do you keep company data safe in a team workspace?

Start from the concern that stops most rollouts: will our information end up training someone's model. On Claude Team and Enterprise plans, Anthropic does not use your conversations or uploaded content to train its models by default. That removes the objection that usually sends people back to unsanctioned personal accounts, which are less safe.

Around that, a few controls do the heavy lifting. Single sign on ties Claude access to your existing identity system, so leavers lose access automatically and you are not managing a separate password list. Admin roles decide who can do what. Retention settings are yours to set. And where data residency matters, for example under sector rules or a cautious client, processing can run through a UK region such as AWS Bedrock UK South, keeping data in the country under the ICO. Governance here is the thing that lets you say yes to broad use rather than quietly tolerating narrow, hidden use.

How should you structure workspaces, projects and permissions?

Keep it simple enough that people use it and structured enough that it stays clean. Projects are the unit that matters: a project holds the context, instructions and files for a recurring piece of work, so the team gets consistent output without pasting the same background in every time. Set up projects around the jobs the team actually does, a project for proposals, one for client reporting, one for a given account, rather than around the org chart.

Permissions follow sensitivity. Most projects can be open to the team; a few, holding client confidential or commercial material, are restricted to the people who need them. The mistake to avoid is locking everything down so hard that people go back to a personal account to get their work done, which puts you back where you started with less visibility. The design behind this, from readiness to governed rollout, is the work in our practical 2026 framework for implementing AI in a B2B business.

How do you drive real adoption once it is set up?

This is where most rollouts stall, and it is a people problem more than a technical one. A company wide licence with no follow through produces a handful of enthusiasts and a majority who logged in once. What changes that is narrow and practical.

Pick three or four jobs the team does every week and are a poor use of skilled time: first draft proposals, weekly reporting, meeting notes into actions, research summaries. Show people, in their own workflow, how Claude does those faster. Then share the prompts and projects that work, so the second person to try a task starts from the first person's proven setup rather than a blank box. A named owner keeps momentum, answers questions and curates the shared library.

The aim is capability the team keeps, not reliance on one power user. That distinction, building capability rather than dependency, is the subject of our companion piece on a Claude first training approach, and where an organisation wants senior ownership without a permanent hire, the fractional AI director model carries the same logic into leadership.

What does good governance look like under the EU AI Act?

A shared workspace makes governance easier, not harder, because it gives you the things the rules keep asking for. You can produce an inventory of who uses AI and for what, because it runs through one system. You can name an owner. You can document the controls and the acceptable use. And you can point people at training, so AI literacy is real rather than assumed.

The deadline gives this urgency. The EU AI Act reaches full high risk enforcement on 2 August 2026, with penalties up to 35 million euro or 7 percent of global turnover. Most organisations still carry the same four gaps: no AI inventory, no governance owner, no documentation, no AI literacy. A properly set up team workspace, with the AI governance and readiness work around it, is a direct way to close all four.

How should a B2B team start?

Do the admin, then do the adoption, and do not stop at the admin. Choose the plan that matches your team size and risk, Team for most, Enterprise where security and compliance demand it. Turn on single sign on and set your data controls. Then, before you announce it widely, build two or three projects around real recurring work and prove they save time with a small group.

Roll out from that proof, with a named owner and a shared library of what works, rather than from a launch email and hope. Measure the hours it gives back and the jobs it now does by default. Set up this way, a Claude workspace becomes part of how the team works, safely and in the open, instead of a licence nobody remembers buying.

FAQ

What does setting up Claude for a team involve? It means putting the whole team on a shared Claude Team or Enterprise workspace instead of separate free accounts, then configuring the controls that make it safe for company use. In practice that is central billing and user management, single sign on, permissions that decide who can do what, and the assurance that your conversations are not used to train the underlying models. The setup is only half the job; the other half is a plan to get people actually using it.

Is company data safe when a team uses Claude? On a Claude Team or Enterprise plan, Anthropic does not use your conversations or uploaded content to train its models by default, which is the main data concern most businesses raise. You add to that with single sign on so access follows your existing identity system, admin controls over who can use what, and retention settings you choose. For stricter data residency, processing can run through a UK region such as AWS Bedrock UK South to keep data in the country under the ICO.

What is the difference between Claude free, Team and Enterprise? Free and individual Pro accounts are built for one person, with no central admin, billing or governance. Claude Team adds a shared workspace, central billing, user management and collaboration for a group, with business data protections. Claude Enterprise adds the controls larger organisations need: single sign on, fine grained roles and permissions, more security and audit features, and higher usage limits. The right choice depends on team size and how strict your security and compliance requirements are.

How do you get a team to actually use Claude once it is set up? Access alone does not create use. What works is picking a few real jobs the team does every week, showing people how Claude does them faster, and sharing the prompts and projects that work so nobody starts from a blank page. A named owner, a shared library of proven uses, and light training beat a company wide licence with no follow through. The goal is capability the team keeps, not dependence on one enthusiast.