Team and Enterprise: admin controls, policy and rollout

0:00 Rolling Claude out to a team

Giving a team AI licences is easy. Getting a team to use AI well, safely and measurably is a management skill. In this lecture you will learn what Claude's Team and Enterprise plans add, how to write a one page policy people actually follow, how to configure the admin console, how to roll out in phases, and how to prove value with honest metrics.

0:28 Why rollout needs a plan

Why does rollout need a plan at all? Because licences do not equal adoption. Give sixty people access with no guidance, and you usually get three patterns. A few power users who get huge value, a large group who try it twice and stop, and a handful who paste things they should not. Think of it like opening a new office. You would not just hand out keys. You would explain how things work, who to ask, and what is off limits. That is what a rollout plan does.

1:07 What business plans add

Team and Enterprise plans add organisational control. Identity features like single sign on, automatic user provisioning and, on Enterprise, custom roles with admin permissions. Model entitlements, in beta, to control which models people can use. Toggles for web search, memory, connectors, skills, plugins, Claude Code and sharing, with connector access manageable through roles. Audit logs, a Compliance API, custom retention on Enterprise, and a self serve HIPAA ready configuration. Usage analytics, an Analytics API and Smart reports in beta. And commercial terms with no training on customer content by default. Check exact availability per plan with Anthropic.

1:49 The one-page policy

Before buying seats, write a one page policy. Which tools are approved. Four data classes, with restricted data never entering AI chats. Accountability, whoever sends or publishes AI assisted work owns its accuracy. Verification by stakes. Honesty rules, no fabricated reviews, quotes or statistics. Least privilege and human approval for connectors and agents. An approved list for skills and plugins. Respect for client contract clauses. Incident reporting within a set time. And a review date with a named owner.

2:23 Admin setup

Then configure the console to match the policy. Set up single sign on and provisioning, and decide who gets standard or premium seats. Start conservative with models and features. Approve a short connector list, read only where possible. Publish approved skills, like house styles and report templates, and shared Projects per team or client. Configure retention and audit logging, and connect the Compliance API to your tools if required. Finally, switch on analytics and decide who reviews them each month.

2:58 Five-phase rollout

Roll out in five phases. A pilot of four to six weeks with five to fifteen people and three to five defined use cases, with baselines measured before you start. Playbooks that turn what worked into shared Projects, skills and prompt libraries with owners. Training and scaling with short, role based sessions using real tasks, and champions in each team. Governance, with a monthly review of analytics, incidents and cost. And improvement, retiring weak use cases and investing in strong ones.

3:33 Honest metrics

Measure outcomes against a baseline. Time to first draft of a standard deliverable. Revision rounds. Turnaround on client requests. Error or rework rates found in quality checks. And weekly active use among the roles you targeted. Avoid vanity numbers like prompt counts or response length. They tell you people are typing, not that work is getting better.

3:58 Simple example

A simple example to make it concrete. A five person team picks one use case, meeting follow up emails. Before starting, each person times three follow ups, and the average is about twenty minutes, an illustrative figure. They use a shared Project with the team's tone and a template, and after two weeks the same task takes about six minutes, with fewer corrections from their manager. One use case, one metric, a clear result. That small, measured win is what earns budget for a wider rollout.

4:35 Worked example: 60 people, three offices

An agency with offices in London, Dubai and Lahore piloted Claude Team with twelve people across content, paid media and account management. They measured their own baseline first, hours to first draft of a monthly report and revision rounds. After six weeks with shared client Projects and a reporting skill, both improved in the pilot group. They scaled with custom roles, giving client facing staff Drive and Slack connectors and finance none, plus weekly champion office hours. One incident, a draft containing an unverified statistic, led to a checks needed rule in every Project. The monthly governance review then became routine. Analytics showed one office using Claude far less, so the champion there ran two short role based sessions using real client tasks, and usage rose. An incident about a shared link exposing a draft led to a change in sharing settings. Governance was not a brake on adoption. It was how adoption spread safely.

5:42 Client AI clauses

Finally, client contracts. More and more of them say whether and how suppliers may use AI. Look for bans on AI processing of client data, disclosure obligations, who owns AI assisted outputs, and data residency requirements. Record each client's rules in its Project instructions, so nobody has to remember them.

6:04 Try this now

Try this now. Copy the one page policy template from the lesson and fill in the placeholders for your team, the approved tools, the incident owner and the review date. Then choose one pilot use case that happens often and is easy to measure, such as meeting follow ups or weekly reports. This week, before anyone changes how they work, time three real examples and record the average and the number of revision rounds. That baseline is the most important number in your whole rollout, because without it, you can never prove what changed.

6:45 Watch me do it, part 1

Let me set up the rollout for the three office agency. I open the policy template and fill the placeholders. Approved tool, the Claude Team workspace. Incident owner, the operations director, within twenty four hours. Review date, March. Then the admin console. Single sign on and provisioning are already connected to our identity provider. I set model entitlements so most staff use the balanced models by default. I turn memory and web search on, but leave plugins off until we have an approved list. Connectors, Google Drive and Slack for client facing roles only, and none for finance, using custom roles.

7:29 Watch me do it, part 2

Next, shared assets. I publish three client Projects, each with a named owner, and add our monthly reporting skill to the approved list. Then the step most rollouts skip, the baseline. Before anyone in the pilot changes how they work, each of the twelve people times three monthly reports, and I record the average and the revision rounds in a sheet. Finally, I book a monthly governance review, forty five minutes, with the usage analytics, the incident log and costs on the agenda. Now when leadership asks whether it worked in six weeks, I can answer with numbers.

8:12 Recap and next step

Recap. Write the policy before buying seats. Configure the admin console to match it, starting conservative. Pilot with real baselines, turn wins into playbooks, scale with training and champions, and govern monthly. Your next step: adapt the policy template in the lesson for your team, choose one pilot use case, and record its baseline metric this week.