How to Deploy Claude Across Your Team of 50 to 150

The Claude deployment decision at a 50-person company has three components a personal subscription does not: the account structure, the project architecture, and the governance framework.

Get these three right before the first team member logs in, and the deployment compounds from week two.

Get these three right before the first team member logs in, and the deployment compounds from week two. Get them wrong, and the deployment produces inconsistent individual use that looks like the personal browser tab experience — but at ten times the subscription cost.

This article describes the specific deployment decisions, configuration choices, and governance steps that 50 to 150 person companies need to make to deploy Claude as an operational system rather than a collection of individual subscriptions.

The technical setup is simpler than most IT teams expect. The governance decisions are more important than most teams anticipate.

Decision 1: The account structure

Claude Teams vs individual Pro subscriptions

For a 50 to 150 person company deploying Claude as an operational system, the Claude Teams plan provides features that individual subscriptions do not:

• Team management: add, remove, and manage team member access from a central admin console
• Shared Projects: create Projects that all team members can access, with shared context documents
• Organisation-wide Project sharing: make Projects available to everyone in the organisation via the Team tab
• Business data handling: data handling terms appropriate for business use

Verify current Teams pricing, seat minimums, and feature inclusions at claude.ai before finalising the account decision.

Who gets access — three access tiers

Full Project access (30 to 50% of the company)

Team members who run AI workflows on company-standard tasks multiple times per week:

• Billing coordinators, account managers, operations staff, customer service
• Development directors, programme coordinators, grant writers
• Sales reps, inside and outside

These are the team members for whom the Project architecture produces the most value.

Read-only or limited access (20 to 30% of the company)

Team members who occasionally need to reference AI outputs or run simple tasks but do not have dedicated workflows: administrators, part-time staff, managers who primarily review rather than produce.

Admin access (2 to 5 people)

• The AI system owner
• The COO or managing director
• The IT manager

Admin access allows Project creation, context document upload, and custom instruction configuration.

The access decision principle

Err toward broader access in the first 30 days. Restricting access too early prevents the organic adoption spread that produces the peer advocacy moments that drive fluency. Restrict access in month two or three if usage data reveals that certain roles are not producing useful workflows.

Decision 2: The Project architecture

The anti-pattern — one general-purpose Project

The company that creates a single “Company AI” Project for all team members produces a shared context that is simultaneously too general for specific function use and too large for any one team member to navigate efficiently.

The billing coordinator who needs the payer communication vocabulary and the operations manager who needs the scheduling communication standards are drawing from the same Project.

The context that matters for each workflow is diluted by the context that does not.

The recommended architecture — function-specific Projects

Create a Project for each primary function in the company.

Example: mid-market manufacturing company

Shared Knowledge Project (all-team access)

Uploaded documents:

• Company overview (200 words: what the company does, who its customers are, what makes it distinctive)
• Brand voice guide (tone, vocabulary to use, vocabulary to avoid)
• Key customer descriptions
• Regulatory vocabulary guide

Custom instructions: none (shared reference only).

Operations Project (operations team access)

Uploaded documents:

• Scheduling communication standards
• Back-order notification vocabulary
• Supplier communication standards
• Management briefing format

Custom instructions: “when scheduling data is pasted, produce the Monday briefing in the standard format. When exception details are provided, draft the customer notification in the appropriate tier standard.”

Sales Project (sales team access)

Uploaded documents:

• Quoting standards guide
• Customer account intelligence layer
• Proposal format standards

Custom instructions: the quoting workflow inputs and outputs, the proposal section sequence, the follow-up communication cadence.

Quality/Compliance Project (quality team access)

Uploaded documents:

• Quality language guide
• NCR/CAR format standards
• Customer-specific quality documentation requirements

Custom instructions: “when defect details are provided, draft the NCR in the standard format.”

How many Projects

The guiding principle: each Project should have a coherent set of context documents and a clear set of workflows that belong to it. If a Project’s context documents span multiple unrelated function areas, split it.

For a deeper look at what makes a shared workspace effective, see how to build a shared AI workspace in Claude. Before finalising the Project architecture, it is worth understanding what Claude Projects does at the feature level — the shared knowledge base, custom instructions, and visibility controls are all Projects capabilities that shape the architecture decisions above. For a concrete picture of what companies using Claude operationally actually look like at month six, how mid-market companies use Claude covers the sector examples in detail.

Decision 3: The data handling standard

Before any team member uses Claude on company work, the data handling decision must be made and documented.

By industry type

Standard data handling (most non-regulated industries)

For manufacturing, distribution, professional services (non-legal/non-medical), real estate, agencies, and non-profits without specific healthcare data obligations: Claude Teams’ standard data handling terms are appropriate for operational AI use.

Verify the current terms at anthropic.com before finalising.

Healthcare (BAA required for PHI-adjacent workflows)

For healthcare companies using Claude for payer appeal letters, care coordination communications, or any workflow that may involve Protected Health Information: a Business Associate Agreement (BAA) is required.

Verify Claude’s current BAA availability and the tier required at anthropic.com or by contacting Anthropic directly.

Configure Zero Data Retention for PHI-adjacent workflow Projects.

Legal and financial services (specific review required)

For law firms (client confidentiality obligations), financial advisory firms (client financial information), and regulated financial institutions: a specific review of Claude’s data processing terms by the firm’s attorney or compliance officer is appropriate before finalising the deployment.

The data handling governance document

Regardless of industry, produce a one-page document before the first team member logs in.

Contents:

• Approved Claude account and tier
• Data categories by Project:
  • What can be uploaded to the Shared Knowledge Project
  • What is appropriate for function Project context
  • What cannot enter any Claude session
• De-identification standard: how team members handle sensitive personal or client data before AI input
• Review requirement: every AI-assisted output is reviewed by a qualified staff member before use

This document is reviewed by the operations director and, for regulated industries, by legal counsel. Filed with the IT governance documentation.

Decision 4: The AI system owner role and the 30-day setup sequence

The AI system owner for a 50 to 150 person deployment

Who they are: one person, operations-minded, with knowledge of how the company’s workflows actually run.

Not the IT manager: the IT manager handles account setup and access controls. The AI system owner handles the knowledge: what goes in the Projects, how the custom instructions are configured, how the output quality is maintained.

Responsibilities:

• Project knowledge base maintenance: updating context documents as the company evolves, as quality feedback comes in, and as new workflows are added
• New team member onboarding: configuring appropriate Project access for new hires, running or supervising the onboarding AI session
• Output quality monitoring: reviewing a sample of Project conversation activity weekly, identifying workflow improvement opportunities, updating custom instructions
• Governance documentation: maintaining the data handling standards document, updating it when Projects or data categories change

Time required:

The 30-day deployment sequence

Week 1: Account setup and architecture

IT manager:

• Configure Claude Teams account
• Set up admin access
• Configure team member access groups

AI system owner and COO:

• Define the Project architecture (how many Projects, which functions, which team members in each)

AI system owner:

• Draft the data handling standards document (legal or compliance review for regulated industries)
• Produce the initial context document drafts for each Project (the Foundation build from the relevant sector article)

Week 2: Context deployment and testing

AI system owner:

• Upload context documents to each Project
• Configure custom instructions for the two highest-priority function Projects

AI system owner and Operations team lead:

• Test the custom instruction configuration with five real current tasks
• Evaluate output quality against the quality standard
• Adjust custom instructions where needed
• Configure the remaining function Projects using the same test-and-adjust process

Weeks 3 and 4: Team training

• Individual anchor workflow sessions for each team member, using the deployed function Projects on real current work (not generic demonstrations)
• Day-seven follow-up sessions scheduled in advance for all week-three session participants
• AI system owner: first output quality review using the Project conversation activity, and first context document updates based on quality feedback

End of week four:

• All team members with full access have had their anchor workflow session
• Two to three function Projects are running at quality
• The first improvement loop cycle has run (session reviews completed, context updates made)
• The data handling document is filed

Common questions on Claude team deployment

”Can we integrate Claude Teams with our existing Slack or Microsoft Teams environment?”

Claude has workplace connectors that integrate with common business tools, and the API enables custom integrations for Phase 3 automation builds. For native Slack or Teams integration, verify the current connector availability at claude.ai.

For most 50 to 150 person companies in the first 30 days: the Projects interface is the primary access point. The integration layer is a Phase 3 investment after the Foundation is proven.

”What happens to Project conversation activity when a team member leaves?”

Verify the current behaviour at support.claude.com. In general, shared Project knowledge and the Project’s uploaded context documents remain in the Project when a team member’s access is removed.

The AI system owner removes the departed team member’s access from the admin console and reviews whether any Project-specific context they maintained needs to be updated or transferred.

”How do we handle team members who are already using Claude individually — do they need to migrate to the Teams account?”

Yes, for any work that uses the company’s operational context. Individual Pro subscriptions do not support the shared Projects, organisation-wide context, or admin management that operational deployment requires.

The individual user’s personal workflows can continue on a personal subscription if desired. The company-standard operational workflows move to the Teams account Project structure.

The transition: the AI system owner exports the context documents from any individual setups the team member has built, uploads them to the appropriate function Project, and runs a 20-minute session to familiarise the team member with the Project interface rather than the personal session interface.

Want the Project architecture designed, the context documents built, and the 30-day deployment sequence run for your company?

Deploying Claude across a 50 to 150 person company requires four decisions made before the first team member logs in: the account tier, the Project architecture, the data handling standard, and the AI system owner.

The company that makes these decisions correctly is not deploying an AI tool. It is deploying an operational system that knows what the company is, how it works, and what good output looks like here.

Path one: make the four decisions this week. Define the account tier (Claude Teams). Sketch the Project architecture for your primary functions. Draft the data handling governance document. Designate the AI system owner. These four decisions, made before any technical setup, determine whether the deployment compounds or drifts.

Path two: bring in a partner. Phos AI Labs designs the Project architecture, builds the context documents, and runs the 30-day deployment sequence that produces an operational Claude system within one month. Thirty minutes, no deck. Start here.