Short answer
Roll out Claude in this order: choose the plan, connect single sign-on, set data retention and connector permissions, publish a usage policy, train people on real tasks, then measure use. Regulated firms should settle recordkeeping and supervision questions with compliance before the first user logs in.
1. Plan and contract
- Choose Team or Enterprise. Enterprise adds controls such as SSO with SCIM provisioning, audit logs and configurable data retention; confirm the current feature list with Anthropic before you sign.
- Review the commercial terms on training: Anthropic's business plans don't use customer content to train models by default.
- Get the data processing terms and security documentation to your vendor risk team early.
2. Identity and access
- Verify your domain and turn on SSO through your identity provider.
- Provision and remove users automatically with SCIM where your plan supports it.
- Decide who may use Claude Code, connectors and file uploads, by group.
3. Data controls
- Set the conversation retention period to match your records policy.
- Allow only approved connectors (for example Salesforce, HubSpot, Google Drive) and only for the groups that need them.
- Write down which data classes may go into Claude and which may not.
4. Policy and supervision
- Publish a one-page usage policy: allowed uses, banned uses, review rules for client-facing output.
- Agree with compliance how AI-drafted communications are reviewed before sending, and whether conversations must be archived.
5. Training and adoption
- Train by role on real tasks: meeting prep for advisors, report drafting for operations, code review for developers.
- Set up shared Projects with approved reference material, such as your style guide and product facts.
- Name a champion in each team and hold a 30-day check-in.
6. Measure
- Track active users weekly and the tasks people use Claude for.
- Collect two or three time-saved examples per team at 30 and 90 days.
Official documentation
Frequently asked questions
Does Anthropic train its models on Claude Team or Enterprise data?
Anthropic states that it does not use content from its commercial plans to train models by default. Confirm the current terms in your agreement before rollout.
What should a regulated firm settle before rolling out Claude?
Retention of conversations, which data may be entered, which connectors are allowed, and how AI-drafted client communications are supervised. Get compliance sign-off on all four before users get access.
Salesforce, HubSpot, Anthropic and OpenAI change their products often. Check the official documentation before you rely on a specific setting, limit or price.