Claude for business: how to roll out Claude across your company
By the Codivated team · Updated
A good Claude rollout gives every employee the right plan, connects it to company tools with each person's own permissions, sets a one-page policy, and trains each department on two or three real tasks from their own week. Then you rebuild the workflows where people use it most. Licenses alone change little; the training and the workflow redesign are where the value is.
Which Claude plan does a company need?
Most companies choose between Anthropic's Team and Enterprise plans. Both give you a shared, admin-managed workspace with single sign-on and admin controls, and both include Claude Code for technical staff. Enterprise adds controls that larger or regulated companies tend to need, such as user provisioning, audit logs and more granular permissions. Prices and seat types change, so check Anthropic's current pricing before you budget.
| You have | Usually start with | Why |
|---|---|---|
| Under about 150 people, standard security review | Team | Fast to set up, admin controls and SSO included |
| Security or compliance team that needs audit logs and provisioning | Enterprise | Logging, user management and data controls at scale |
| Engineers who will use coding agents daily | Premium seats for them, standard seats for others | Heavy users need higher limits; most staff don't |
What should you connect Claude to?
Start with the places people already look for answers: your document store, your wiki, your tickets and your chat. Claude connects to tools through connectors built on the Model Context Protocol (MCP). The rule that matters: every connection should use each person's own login and permissions, so Claude can only see what that person could already open. Avoid one shared admin connection that can see everything; it is the fastest way to get a rollout stopped by security.
What goes in the usage policy?
Keep it to one page that people will actually read:
- What data is fine to use. Company documents and customer data inside approved tools, yes. Personal accounts and unapproved apps, no.
- What must be checked by a person. Anything sent to a customer, anything legal or financial, any code that ships.
- What Claude may do on its own. Drafting and research, yes. Sending, deleting or changing records only through approved, logged workflows.
- Where to ask for help and how to suggest a new use.
How do you train people so they actually use it?
By department, on their own work. A generic "intro to AI" session fades in a week. A session where finance builds the month-end summary they write every month, or support drafts the replies to last week's tickets, sticks. Our team trained a 40-person senior leadership team on Claude and AI tools as part of the AI transformation of a 700-person company, sitting with each department to rebuild its workflows. What worked:
- Start with leadership, so managers use it before asking their teams to.
- One session per department, built on three tasks from that team's real week.
- Office hours for the first month, and a channel where people share what worked.
- Named champions in each team who collect the next workflows to improve.
What do you do after everyone has access?
Look at where people use it most and rebuild those workflows on purpose. Some become shared projects and prompts the whole team uses. Some become small internal apps. Some become real agents that run on their own with approvals, like order entry or inbox triage. To keep this from becoming shadow IT, give people one approved place to publish what they build, behind your company login. Our team has built that kind of internal platform for a company, so people had a sanctioned place to share what they built.
What about Claude Code for engineering teams?
Treat it as a change to how software is delivered, not as autocomplete. The teams that get the most from coding agents write clearer tickets, have good tests, review agent-written pull requests like any other, and let background agents take well-defined work. Our team has led an engineering organization's move to AI-native work, where background coding agents now open a real share of the pull requests.
How do you measure the rollout?
Weekly active users by department, the number of shared projects and workflows in use, and two or three before-and-after numbers from the workflows you rebuilt: time per report, time to first reply, tickets handled per person. Usage without workflow change is a warning sign. See how to measure AI ROI.
Common questions
Is our data used to train Claude?
Anthropic's business plans don't train models on your data by default. Confirm the current terms for your plan with Anthropic.
Should we choose Claude, ChatGPT or Copilot?
Pick by where your work lives and which tools your teams prefer after a short trial. Many companies end up with more than one. The policy, connections and training matter more than the choice.
How long does a rollout take?
Access and policy take days. Department training and the first rebuilt workflows take a few weeks. Getting to real workflow change is a quarter-long effort.
Can you help us roll it out?
Yes. We run rollouts as an advisory engagement, and build the agents that come out of it as fixed-scope projects. Book a consultation.
Sources
- Claude plans and pricing, Anthropic.
- Model Context Protocol: the open standard behind Claude's connectors.