Claude for business teams

A guide for small and mid-sized teams adopting Claude - Team plan seats and costs, Cowork for non-developers, connectors, admin controls and a realistic 30-day rollout.

For a small or mid-sized team, the answer is the Claude Team plan: $20 per standard seat per month on annual billing, $25 monthly, from 2 to 150 seats, with Claude Code and Cowork included on every seat. It adds central billing, an admin console, SSO with domain capture, enterprise search and a contractual promise that Anthropic will not train on your content by default.

If you have twenty people and no compliance department, this page is for you. If a security questionnaire is already in flight, read Claude for Enterprise instead - the controls that decide that purchase are not on the Team plan.

#How does the Claude Team plan work?

Team is a per-seat subscription with two seat types you can mix freely inside one workspace. A standard seat carries 1.25× the usage of a Pro subscription; a premium seat carries 6.25×. Seats can be added mid-term, and the whole plan tops out at 150 people.

Every seat includes the full product surface: chat with Projects and Research, Claude Code, Cowork, Claude Design, Claude in Chrome and the Microsoft 365 add-ins for Excel, PowerPoint and Word. Premium seats do not unlock different features - they buy more of the same allocation, plus Fable 5 included rather than credit-funded.

Admin capabilities that arrive with Team and do not exist on individual plans: central billing, an admin console with an analytics dashboard, SAML SSO through providers including Okta, Entra ID and Google, domain capture so new hires on your domain land in the workspace, server-managed settings for Claude Code, organization-provisioned skills, project sharing, and enterprise search across Slack, Microsoft 365 and custom connectors. Team and Enterprise also get Claude Code's Code Review, and Claude Tag - the Slack-native agent that works under the organization's identity - is in beta on both.

#What does Claude cost for a 10-person team?

Four configurations for the same ten people, at list price, annual billing where available.

The mixed row is usually the right answer. Two premium seats cost $200 a month more than two standard ones and buy an extra 10× of Pro-equivalent allocation - which lands with the two people running long Claude Code or Cowork sessions all day, not spread thinly across ten people who mostly chat. Start everyone on standard, watch the analytics dashboard for a month, then upgrade the seats that are actually constrained. Downgrading someone who never needed a premium seat is a slightly awkward conversation; upgrading someone who is clearly blocked is an easy one.

For comparison, ten individual Pro subscriptions also cost $200 a month, with none of the admin console, SSO, domain capture, enterprise search, project sharing or central billing - and with each person separately deciding whether their chats train a model. Team at the same price is strictly better for an organization. Ten Max 5× subscriptions would cost $1,000 a month and still lack all of it.

Annual billing saves $5 per standard seat per month and $25 per premium seat, roughly 20% either way. Monthly billing costs more but lets you resize a team that is still forming. Neither choice is wrong at ten people.

#Cowork: the part non-developers actually use

Cowork is the same agentic architecture that powers Claude Code, with no terminal required. That single sentence is the whole product. Where Claude Code reads a codebase, runs commands and edits files for engineers, Cowork does the equivalent for documents, spreadsheets, research and browser work - for people who will never open a shell. It went generally available on all paid plans on 9 April 2026.

The framing Anthropic uses is “describe an outcome, step away, and come back to finished work”. In practice that means file management across a folder of documents, research synthesis across many sources, document creation, spreadsheet generation with working formulas rather than pasted values, and browser automation. Sessions run on Anthropic's cloud infrastructure, so work continues when your laptop is shut and progress is visible from another device.

#Where Cowork runs

#The Team controls that came with GA

The April 2026 release added the governance layer that makes Cowork deployable rather than merely available: role-based access control with groups, group spend limits, usage analytics in the admin dashboard covering Cowork sessions, active users, per-user activity, skill and connector invocations and daily/weekly/monthly actives, expanded OpenTelemetry events for tool calls, file operations, skills usage and approvals that are SIEM-compatible on Team and Enterprise, a Zoom connector, and per-tool connector controls - so an admin can allow reads from a system while disabling writes to it. That last one is the most useful lever on this list and the least discussed.

#Which roles get value from Claude first?

Adoption is uneven and predictably so. These functions tend to produce a result in the first fortnight:

The functions that struggle are those whose work is mostly judgement in a live conversation, and those handling data that should not be pasted anywhere. Name both groups out loud at the start; pretending a tool is universal is how internal credibility gets spent.

#A realistic 30-day rollout for a small team

No steering committee required. One owner, four weeks.

#Days 1–5: set it up properly once

Buy seats for everyone rather than a pilot subset - at $20 a seat the cost of over-buying is trivial next to the cost of two-tier access. Turn on SSO and domain capture. Decide which connectors are approved and connect only those. Write a one-page internal note covering what may not go into a prompt. Confirm the Cowork cloud-sessions default is what you want.

#Days 6–15: two Projects, not twenty

Build one shared Project per function with real reference material in it and clear custom instructions - this is where quality comes from, far more than prompt wording. Have each function nominate one recurring task they will actually try. Ask one person per function to run that task in Cowork rather than chat, so somebody discovers what delegation feels like.

#Days 16–25: make the good work reusable

Whatever worked becomes an organization-provisioned Skill or a shared Project so it stops living in one person's history. Check the analytics dashboard: who is active, who has not logged in, which connectors are being invoked. Fix the connector gaps that are blocking people rather than the ones that sound impressive.

#Days 26–30: right-size and write it down

Upgrade the one or two people who are consistently hitting limits to premium seats. Decide whether to enable usage credits for the workspace, which an owner has to turn on. Publish the three prompts and two Projects that are working, with names attached. Book the same review in 90 days - the model lineup will have changed by then.

What to skip: benchmark bake-offs, a bespoke internal wrapper, and mandatory training sessions. What to insist on: someone owning the shared Projects. Starter prompts worth stealing are on the examples page, and using claude.ai covers the interface itself.

#When does a team outgrow the Team plan?

Three signals, in the order they usually appear. First, somebody asks for audit logs - that is Enterprise, immediately, with no workaround. Second, you need a data retention policy, because Team offers no retention controls at all. Third, you pass roughly 150 people, which is the hard seat ceiling.

Everything else is negotiable. Usage pressure is solved with premium seats or by enabling usage credits, not by changing plans. The migration to Enterprise brings a different commercial model - a $20 platform seat plus metered usage at API rates, annual only, minimum 20 seats - so run both numbers rather than assuming the upgrade is a pure addition. On ownership of what your team produces with Claude, which is the other question that comes up during a business rollout, see commercial use rights. If you are wiring Claude into internal systems, MCP and Claude agents and automation patterns are the next stops, and the feature and plan matrix settles any per-plan question this page leaves open.

#Frequently asked questions