How do you roll Claude out to a 100-person business?

Start with twelve people, one page of rules and one connected system. Give the whole company access on the first morning and you get a spike of curiosity, a week of "what do I use this for?", and a licence bill for people who logged in twice.

For most Australian businesses of around 100 people, the first 30 days look like this: pick the plan (usually Team, unless you need SCIM, audit logs or custom retention), name one person accountable, write a one-page policy, give seats to a pilot group chosen for the work they do, connect HubSpot with read-first use cases, train each team on its own tasks, and measure weekly use before you buy another seat.

At day 30 you decide whether to expand, using numbers from your own business. That is the whole plan. The rest of this post is the detail, week by week, with the checklist we use at the end.

Team or Enterprise: which plan does a 100-person business need?

Anthropic sells two business plans. Here is what separates them, taken from the Claude pricing page on 25 September 2026. Team prices are what the page shows an Australian visitor, in Australian dollars with GST included. Enterprise is quoted in US dollars on the same page. Anthropic can change either, so check before you budget.

The deciding question is simple. Does your IT team provision every app through Entra ID or Okta, and does someone (an auditor, a regulator, a big customer's security team) expect to see logs of who did what? If yes, you want Enterprise. If you are a 100-person business where the ops lead adds new starters by hand, Team covers you, and SSO is on both plans.

One thing people miss with Team: you can mix seat types. Your two heaviest users (usually a RevOps admin and whoever writes the most proposals) get premium seats, and everyone else starts on standard.

Before week one: three decisions that take an afternoon

Australia's National AI Centre published its Guidance for AI Adoption in October 2025. It boils responsible AI use down to six practices, and the first one is "decide who is accountable". That is the right place to start, because a rollout without an owner turns into twelve people doing twelve different things.

1. Name the owner. One person, with the time to do it. In a business this size it is usually the ops lead, the RevOps lead or the head of sales. They own the policy, the seats, the connectors and the day 30 decision.
2. Write the one-page policy. What people can use Claude for, what they must never paste in, who approves a new connector, and who to tell when something goes wrong. We wrote up a one-page AI policy you can copy, so you are not starting from nothing.
3. Decide what data goes in. The OAIC's guidance on commercially available AI products says Privacy Act obligations apply to personal information you put into an AI system and to anything it generates about a person. Know where the data goes before anyone connects a CRM. Our piece on where your data goes when you use AI, the Australian answer covers the questions to ask.

Week 1: who gets a seat first?

Pick the pilot group by the work they do. Enthusiasm is a bad filter, because the keenest people are often the ones with the least repetitive work.

A good pilot group for a 100-person business is about twelve people:

• Two or three salespeople who write a lot of follow-ups and proposals
• Two people from service or support who answer the same questions all day
• Your RevOps or HubSpot admin
• One person from finance and one from operations
• One marketer
• Two managers, so the people approving the work can see what it looks like

Ask each of them to bring one task they do every week that they would happily never do again. Write the list down. It becomes your training plan in week three and your measure in week four.

The admin work for week one:

• Turn on SSO, so nobody signs up with a personal email
• Switch off every connector you have not approved yet
• Start an AI register. The National AI Centre publishes a template, and it only needs a line per tool: what it is, who uses it, what data it touches
• Set up a shared Project for each team, holding the documents they reach for most (price list, product sheets, the style guide)

Six minutes on which jobs to hand to AI and which ones still need a person, which is the conversation to have with your pilot group before they start.

Week 2: connect one system, and start by reading

If you run on HubSpot, that is the first connector worth switching on. The HubSpot connector for Claude works with every HubSpot plan and any paid Claude plan. A Super Admin approves it first, then everyone else connects their own account.

Start with jobs where Claude reads and a person acts:

• "Summarise every open deal over $50k with no activity in 14 days."
• "Before my 2pm call, tell me everything we know about this company."
• "Which tickets this week mention the same problem?"

Leave Claude writing to HubSpot (creating tasks, updating properties, logging notes) until week three or four, once people trust the answers. The connector respects each person's HubSpot permissions, so a salesperson cannot see through Claude what they could not see in HubSpot. It is still worth deciding who is allowed to let it write. We went through that in detail in what we locked down when we connected HubSpot to Claude.

Week 3: train people on their own work

A generic "intro to AI" session gets polite nods and changes nothing. What works is 45 minutes per team, built around the tasks they wrote down in week one.

Run each session the same way:

1. Take one task from the list and do it live, with the person who brought it driving.
2. Get it wrong on purpose once. Show what happens when the prompt is vague, then fix it.
3. Save the prompt that worked into the team's Project, so the next person starts from it.
4. Agree what a person must check before anything goes to a customer.

Then open one channel in Slack or Teams where people post what worked. By the end of the week you have the start of a prompt library written by your own people, about your own work.

Week 4: what should you measure before you buy more seats?

Keep it to four numbers and one list. Anything more and nobody fills it in.

Treat the time-saved number as what it is, a self-reported estimate. It is useful for deciding where to go next and useless as a business case on its own.

The list is the "not working" list: tasks people tried and gave up on. Those tell you as much as the wins. Usually the reason is missing context (Claude did not have the price list), which a Project fixes, or a task that needs a system you have not connected yet.

Then make the day 30 call. If most of the pilot group is still using it in week four, expand to the next 30 or 40 people, starting with the teams whose tasks looked most like the ones that worked. If usage fell away after week two, find out why before you spend more. A second wave on top of a stalled first one only multiplies the problem.

The 30-day Claude rollout checklist

Before week one

• Plan chosen: Team, or Enterprise if you need SCIM, audit logs or custom retention
• One accountable owner named, with time set aside
• One-page AI policy written and shared
• Decision made on what data can and cannot go in

Week 1

• Pilot group of about twelve chosen by the work they do
• Each person has named one weekly task to move over
• SSO on, unapproved connectors off
• AI register started
• One shared Project per team with its key documents

Week 2

• HubSpot connector approved by a Super Admin
• Read-only use cases running
• Decision recorded on who can let Claude write to HubSpot

Week 3

• 45-minute session per team, using their own tasks
• Working prompts saved to each team's Project
• Checks agreed before anything reaches a customer
• Sharing channel open

Week 4

• Weekly active users, tasks moved, time estimates and the issues log reviewed
• "Not working" list reviewed and the causes named
• Expand, hold or fix decision made and written down

If you want a sense of where you sit against everyone else, our post on what the numbers say about AI adoption in the Australian mid-market has the ABS figures.

Frequently asked questions

Should a 100-person business choose Claude Team or Enterprise? Most start on Team, which includes SSO, admin controls for connectors, and Claude Code and Cowork. Choose Enterprise if you need SCIM provisioning, audit logs, custom data retention or fine grained role-based permissions, which Team does not include.

Can we put customer data into Claude? The OAIC says Privacy Act obligations apply to any personal information you put into an AI system and to personal information it generates. Decide in your AI policy what can go in, check how your plan handles data, and keep sensitive information out unless you have a clear reason and the right controls.

Do we need an AI policy before we roll out Claude? Yes, and one page is enough to start. It should say what people can use Claude for, what they must never paste in, who approves new connectors and who to tell when something goes wrong.

How long before everyone in the business has access? Plan on a 30-day pilot with about twelve people, then expand in waves of 30 to 40 if weekly use holds up. At that pace a 100-person business has everyone on in about three months, and still knows who is using what.

Sources

• Claude pricing, Anthropic, for plan features and seat prices as listed on 25 September 2026.
• Guidance for AI adoption: foundations, National AI Centre, for the six essential practices and the AI register.
• Guidance on privacy and the use of commercially available AI products, OAIC, for how the Privacy Act applies to what goes into and comes out of AI tools.

---

We build on Claude and connect it to HubSpot for Australian businesses. If you want a hand planning your first 30 days, get in touch.