Most AI rollouts do not fail because the technology is wrong. They fail because licences get handed out with no plan, no setup and no training, and usage quietly fades after a fortnight. This is the approach we use with small and medium businesses to make Claude stick.
1. Start with the work, not the tool
Before anyone logs in, list the jobs that eat your team's time. Proposals, inbox triage, reports, meeting notes, research, quotes. Pick two or three that are frequent, repetitive and easy to check. These are your first use cases.
Starting with real work gives people an immediate reason to use Claude and gives you something to measure.
2. Use a business plan, not personal accounts
Move everyone onto a Claude business plan from the start. You get a shared workspace, admin control and commercial data terms, and you avoid a scatter of personal accounts that nobody can manage. Our guide to Claude Team and Enterprise covers which to choose.1
3. Write a one-page AI use policy
Keep it short enough that people actually read it. It should answer three questions:4
- What is Claude approved for in our business?
- What should never go into it, such as passwords, card details, health information or other sensitive personal data?
- What must a person check before anything Claude drafts is sent, published or acted on?
4. Set up Projects for each team
A Project gives Claude the knowledge and instructions a team needs: your services, pricing rules, tone of voice, templates and examples of good work. This is the single biggest difference between generic answers and useful ones. Start with one Project per team and improve it over time.2
5. Connect the tools that matter
Connectors let Claude work with the tools your team already uses, such as email, files, calendars or your CRM, within each person's access. Connect only what supports your first use cases. You can add more once the basics are working.3
6. Train people on their real tasks
Generic AI training rarely changes behaviour. Run short, hands-on sessions where each person uses Claude on their own work: drafting their proposals, summarising their documents, answering their customers. Give each role a one-page playbook to keep.
7. Name a champion in each team
Every team has someone who picks this up quickly. Give them a little extra time and training, and make them the first port of call for questions. Champions spread good habits faster than any document.
8. Measure at 30, 60 and 90 days
Agree two or three simple measures at the start: hours saved on a specific task, turnaround time for proposals, or how many people use Claude each week. Review them at 30, 60 and 90 days, then double down on what works and drop what does not.
Common mistakes to avoid
- Buying licences for everyone before proving value with one or two teams
- Skipping the policy, then worrying about what has been shared
- Leaving Claude unconfigured, so everyone gets generic answers
- One-off training with no follow-up
- No measures, so nobody can say whether it is working
Sources
Links checked
- 1Anthropic Privacy Center. Is my data used for model training?privacy.claude.com ↗
- 2Claude Help Center. What are projects?support.claude.com ↗
- 3Claude Help Center. Use connectors to extend Claude's capabilitiessupport.claude.com ↗
- 4Information Commissioner's Office. Guidance on AI and data protectionico.org.uk ↗
Want help with this?
Claude implementation, done properly.
We roll Claude out across your business: Claude Team or Enterprise setup, workspaces, access, data settings and connectors to your tools. UK and US.
Claude is a trademark of Anthropic, PBC. Plans and features change over time, so check Anthropic's current documentation for the latest details. This guide is general information, not legal advice.