Good Claude training leaves a team with shared projects set up for their own recurring work, real tasks finished and checked during the session, a habit of verifying figures and sources, and a one-page policy. A tour of features and a list of generic prompts is not enough.
When a firm buys Claude for its staff, the next question is usually about training. There are plenty of providers, and most sessions look similar from the outside: a morning or a day, a presenter, some slides and a demo.
The difference shows up three months later. In some firms the whole team is using Claude on real work. In others, a few enthusiasts are and everyone else has gone back to doing things the old way.
This post is a buyer's guide. It sets out what a Claude training day should leave your team with, the warning signs in a proposal, and the questions to ask before you book. It applies whether you use us or anyone else.
What the team should have at the end of the day
Working shared projects
A Claude project holds instructions and reference documents for a piece of recurring work. On Team and Enterprise plans, projects can be shared with colleagues, so the whole team uses the same set-up.
By the end of a good session, your team should have two or three of these working for their own tasks: a monthly report, a standard client letter, a tender response, a case summary. Each one should have plain instructions saying who the work is for, what good looks like, which documents to use and what to check.
If people leave with nothing set up, they go back to starting every chat from a blank box, and the habits fade.
Real tasks finished
People should bring their own work: a document they need to review, an email thread they need to answer, a report that is due. They should leave with it done, or close to done, and checked.
This is the single best test of a training session. A finished piece of real work proves the method fits the team's job, and it gives people something to repeat on Monday.
A checking habit
Claude writes fluently whether it is right or wrong. Training should spend real time on where it needs watching for your kind of work:
- figures that must be traced back to the source
- citations and references that must be checked against the original
- drafts that read well but miss a condition or a caveat
- anything going to a client, which a person reads in full first
People should practise catching these in the session, on their own material. A warning on a slide does not build the habit.
A one-page policy
The team should know which account to use, what may and may not go into Claude, who approves new connectors and who checks consequential work. That fits on a page. A good provider will either help you write it or work from the one you have. Our AI governance advice covers the wider policy if you need it.
A follow-up
A short follow-up a few weeks later, to look at what people are actually using and fix the projects that are not working, is often worth more than the second half of the training day.
Red flags in a proposal
- Generic prompt lists. "50 prompts for productivity" handouts are easy to produce and rarely used. Good instructions are specific to the task, the reader and the source documents.
- A tool tour. An hour of clicking through every menu teaches people where buttons are, not how to get a client letter right.
- Demo data only. If every exercise uses a made-up company, people have to translate it to their own work afterwards, and most will not.
- No mention of checking. If the agenda has no time for verifying outputs, the session is teaching speed without care.
- No setup work. Training that does not touch your Claude organisation, your projects or your connector set-up leaves the hard part to you.
- No follow-up. A single session with nothing after it tends to produce a burst of interest and then a slow return to old habits.
- Vague outcomes. "Increased AI confidence" is not something you can check. "Three working projects and a finished tender section" is.
Questions to ask a provider
- Will my team work on their own documents during the session, and how do you handle confidential material?
- What will exist in our Claude account at the end of the day that did not exist at the start?
- How much of the time is hands-on, and how much is presentation?
- How do you teach people to check outputs for our kind of work?
- Will you help us set up shared projects and the Microsoft 365 connector, or is that our job?
- What does the follow-up look like, and when does it happen?
- Can you show an agenda from a session for a firm like ours?
- Is the trainer independent of the vendors, or do they also sell licences?
The answers tell you quickly whether the session is built around your work or around the tool.
Before the day
A few things make any training session more useful:
- Buy the business plan first. Training on personal accounts teaches habits you will have to undo. Licence costs are set out in our Claude pricing guide.
- Get your IT provider to approve the Microsoft 365 connector for the training group if you want to use it on the day.
- Ask each person to bring one or two real tasks they do every week or month.
- Agree who owns each shared project after the session.
The post on rolling out Claude company-wide sets out the order to do these in.
How we run it
Our Claude training is built around these outcomes. Each person brings two or three real tasks and works through them with Claude. The team agrees what information belongs in the tool and where a person must review, and leaves with written actions and an owner for each. For firms that already have Claude Team or Enterprise, we review the setup with your Claude owner first, build the sessions around shared projects for recurring work, and look at the usage reports a few weeks later. Acuity has no commercial tie to Anthropic or Microsoft. The course formats and workshop agenda are on that page.
If you want to see whether it fits your team, tell us about your team and we will suggest a format.