How do you roll out Claude company-wide? A phased plan with owners, settings and checkpoints

Roll out Claude in five phases: choose the plan and data rules, set up identity (verified domain, SSO, then SCIM), configure connectors, skills and spend limits before anyone logs in, pilot with 50 to 100 users for two to four weeks, then expand one department at a time with champions and training. Identity, data access and enablement are most of the work. The software is the easy part.

This guide is for the person who has to make Claude, powered by Anthropic, work for a whole company, usually a CIO, COO or IT director at a business of 100 to 5,000 people. It follows the order Anthropic's own Enterprise administrator guide uses, adds the admin settings that sit behind each step, and names who should own each piece. Product details are as of October 2026 and come from Anthropic's pricing page, help center and documentation, linked at the end.

Should you roll out on Team or Enterprise?

Decide this first, because it sets which controls you will have during the rollout. Team covers 2 to 150 seats. Enterprise has a 20-seat minimum on the self-serve route and adds the identity, audit and data controls most security teams ask for. If you are above 150 people, need SCIM, or have to produce audit logs, the answer is Enterprise.

Billing matters too. Team is paid by credit card with no invoicing. Enterprise self-serve supports ACH; invoicing, net terms and AWS Marketplace purchase come through the sales-assisted route. Check current billing terms with Anthropic before you sign. For a fuller comparison, see our guide to Claude Team vs. Enterprise.

What are the phases, and who owns each one?

How do you set up identity without locking people out?

Identity work has a fixed order, and doing it out of order is the most common cause of a bad first week. Anthropic's administrator guide sets out these steps:

• Verify your email domain first. SSO cannot be configured until the domain is verified.
• Configure SSO with SAML 2.0 or OIDC. Test it with a small pilot group, then enforce SSO only after the configuration is validated.
• Turn on restricted organization creation so employees cannot spin up separate Claude organizations on your domain.
• Choose provisioning. SCIM syncs users and groups from your identity provider and is the recommended route. JIT creates users at first SSO login but gives you less control. Manual invites are fine for a small pilot.
• Treat domain claiming as a separate, deliberate decision. It is optional and permanent, it moves personal Claude accounts on your domain into the company organization, and it requires restricted organization creation, enforced SSO, and SCIM or JIT. Tell employees before you do it.
• Phase the SCIM rollout: start with 50 to 100 users, monitor for 2 to 4 weeks, expand by department, then open access org-wide.

Set up groups in your identity provider that match how you will grant features: for example a pilot group, one group per department, a Claude Code group for developers, and a small admin group. On Enterprise, custom roles are assigned to groups, and plugin availability and spend limits can also be set per group, so clean groups save rework later.

Which admin settings should be decided before launch?

Every row below is a real setting in the Claude admin console. Decide each one before the pilot, write the value down, and name an owner. Starting values are our recommendation for a cautious first phase, not Anthropic defaults.

How should you run the pilot?

Pick 50 to 100 people across three or four departments, not one enthusiastic team. Give each department two or three named workflows with a before-and-after measure, such as drafting a monthly variance commentary, summarizing a support escalation, or preparing an account brief from CRM notes and email. Appoint two or three champions per department, give them access a week early, and open a peer support channel.

Anthropic's guide suggests three success targets for the pilot: 70% weekly active users among licensed seats, 25 or more messages per user per week, and a satisfaction score of 4.0 or higher out of 5. Use them as a gate for expansion. Also watch two numbers on the Usage page: members with very low usage, who probably were never activated, and time at limit, which rises when people are being blocked mid-task by their spend cap.

• Week before launch: champions trained, connectors tested by a champion in each department, support channel open.
• Launch day: executive announcement, 30 to 60 minute all-hands 101 session, links to the acceptable-use policy and the training schedule.
• First week: share tips and early wins from champions; hold the first office hours.
• Every week of the pilot: review usage by group, collect friction, fix the top three issues.
• End of pilot: compare results to the targets, write down which prompts people repeated most, and turn the best of them into the first organization skills.

How do you expand to the rest of the company?

Expand by department, not by date. Before each department goes live, confirm it has a champion, a department-specific session with its executive sponsor in the room, a follow-up session two to four weeks later, and at least one skill or connector that makes Claude useful for its actual work on day one. Hold weekly or biweekly office hours for the first 30 to 60 days. Package the 101 and department sessions as courses in your learning management system so you can see who has completed them, and add Claude onboarding to new-hire orientation.

Introduce features in stages: core chat and documents first, then connected data, then integrations such as Microsoft 365 write tools, and automation last. On the spend side, Anthropic's consumption guide suggests defining light, standard and power-user tiers by role, starting conservatively because a cap is easier to raise than to lower, reviewing group consumption weekly during rollout and monthly afterward, and noting that org-level threshold alerts fire at 75% and 90% of the limit.

Where do company-wide Claude rollouts go wrong?

• Enforcing SSO before testing it, which locks out the pilot on day one.
• Claiming the domain without warning employees. Claiming is permanent and pulls personal accounts into the company organization.
• Connecting SharePoint without cleaning up permissions. The Microsoft 365 connector mirrors each user's existing permissions and its search spans the whole tenant, so overshared sites become easy to find. Fix sharing before launch.
• Setting only an org-wide spend limit. When it is reached, everyone stops at once. Use group and user limits.
• Letting connectors grow without an allowlist. Keep a list of approved connectors and route new requests through IT. Community connectors in the directory are screened, not reviewed in depth.
• Giving developers Claude Code with no managed settings. Deploy policy first; see our Claude Code rollout guide.
• Launching with no use cases. People open a blank chat box, try one thing, and stop. Every department needs named workflows on day one.

What does a rollout cost in effort?

Licensing is the visible cost. The larger cost is staff time: identity work, permission cleanup in your document systems, building the first skills, and training. A company that already runs SSO and SCIM for other SaaS tools, has a written data classification policy, and has tidy SharePoint or Drive permissions will move much faster than one that does not. If you bring in outside help, Abstrakt Solutions is a Certified Services Partner in the Claude Partner Network; we bill $180 to $225 per hour and usually quote rollouts as a fixed fee or in phases after discovery. Our Claude enterprise rollout and training services cover the phases above, and the free AI maturity quiz is a quick way to see which phase you are ready for.