Get more out of your AI tools
You already pay for Claude Team, Microsoft Copilot or Google Workspace with Gemini. We set it up properly: who can see what, which connectors are on, the limits on what the AI may do, and your first custom agents in the tool. Then your people use it for real work, within the permissions and limits you set.
This is for you if
- You pay for AI licences, and most people use them as a chat window.
- Nobody has checked who can open what in SharePoint, OneDrive or Google Drive, and the AI can find anything a person can open.
- Nobody in the company owns the setup, and nobody has time to learn the admin settings.
What we set up in each tool
The parts are the same in every tool: the admin role, who sees what, connectors, limits and custom agents. Where they are set differs.
| What we set up | |
|---|---|
| Claude Team | Anthropic's plan for companies. We set up the Owner roles, sign-in with your company's domain, and which connectors the company allows; each person then connects with their own account and sees only what they may see. We check the feedback setting, which can send a rated conversation to Anthropic. Where your people should ask questions of your own data, we build an MCP server and add it as a custom connector. |
| Microsoft Copilot | Microsoft's assistant in Word, Excel, Outlook and Teams. We find the SharePoint and OneDrive files shared too widely before Copilot is switched on, choose which agents are allowed, and decide with you whether Anthropic's models may be used in Copilot. Your first agents are built in Copilot Studio, Microsoft's tool for building agents. |
| Google Workspace with Gemini | Google's AI in Gmail, Docs, Sheets and Drive. We go through the same parts: who holds the admin role, what is shared too widely in Google Drive, which connectors are switched on, and your first custom agents. |
What you leave with
- The result
- An AI tool set up for your company's work. The admin role stays with a named person in your company, the settings are checked, and your people know the rules for what may go in.
- Who sees what
- Files shared too widely are found and fixed before the tool is switched on, because the AI shows each person everything they can already open.
- Connectors and limits
- Only the connectors people need, reading only to begin with. Sending, creating or changing comes on when you have decided which steps the AI may take.
- Your first custom agents
- Custom agents in the tool for your team's own tasks, and where you want it, an MCP server so your people can ask questions of your own data.
What we do, what you do
| We | You |
|---|---|
| Go through the admin roles, licences and sign-in, and keep the top role in your company. | Name the person who owns the setup, and a backup. Give us our own admin account for the work, which you remove afterwards. |
| Find the files shared too widely, and check the settings for training on your data, feedback, history and sign-in. | Decide who should have access to finance, salary and HR files. |
| Turn on the connectors people need, set the limits on what the AI may do, and build your first custom agents. | Choose the tasks, and try the agents on your own work. |
| Write one page of rules for what may go in, and train the people who use the tool. | Take part in the training, and update the rules when the tool or the way you use it changes. |
Worth knowing
An AI tool helps each person with their own work, as it comes. When a task repeats across systems and must follow your rules, with a person approving the result, that is work for your AI agents, built on your own data. Setting up the tools you already pay for is often where that starts.
Your questions
- Which tool should we choose?
- Choose by where your documents, mail and numbers live. If they live in Microsoft 365, Microsoft Copilot. If they live in Google Workspace, Slack or systems you connect yourself, or the work is analysis across several sources, Claude Team. If your numbers are spread across spreadsheets nobody can trace, neither yet: put your data in order first.
- Can our people ask questions of our own data?
- Yes, through an MCP server: a program that gives your AI tool a defined way to read one of your systems. Each person signs in as themselves and sees only what they may see. At DuffBeauty, employees ask their questions in Claude Team through MCP servers and get the figures with the records behind them.
- May the AI send, create or change things?
- Not to begin with. We start with connectors that only read. Writing comes on one tool at a time, when there is a job that needs it, and a good first one makes drafts a person approves. Money leaving the company, anything going outside the company and a commitment always go to a person.
- Who in our company needs to take part?
- The person who owns the setup, and a backup; the people who will use the tool, in the training; and for a few steps, such as single sign-on, someone with admin rights for your domain or your Microsoft tenant.
- How is it priced?
- A fixed price for the setup, agreed before we start, never billed by the hour. If we keep the setup up to date for you afterwards, that is a fixed monthly fee.