01 the business
Accounting practice
65 staff, Microsoft Copilot licences bought
The firm bought AI licences, but only a handful of people use them well. Others avoid them, and some paste client data into free tools. Generic online courses do not show a bookkeeper how AI fits their own day.
02 before
The problem
Many teams already pay for AI tools, but only a few people use them well. Others avoid them, or paste client data into them without thinking. Generic online courses do not show a bookkeeper or a dispatcher how AI fits their own day.
- Licences are paid for but barely used.
- Client data goes into unapproved tools.
- No guidance on what is allowed.
03 the build
What the AI does
- Step 1: Interviews each team about their week and picks the tasks where AI saves real effort, such as drafting, summarising and checking.
- Step 2: Runs short, role-based workshops using your own documents and tools, such as Microsoft Copilot, ChatGPT Enterprise or Gemini for Google Workspace.
- Step 3: Builds a shared prompt library for those tasks and stores it where people already work, such as Notion, SharePoint or Confluence.
- Step 4: Writes a one-page AI use policy that says which data can go into which tool.
- Step 5: Provides a practice assistant in Slack or Teams that answers 'how do I do this with AI?' questions and points to the right prompt.
- Step 6: Reviews usage with team leads after a few weeks and adjusts the library and training where people are stuck.
paused: waiting for a person to approve
How it was put together
- Interviewed each team about their week and chose tasks where AI saves real effort.
- Ran short, role-based workshops using the firm's own documents and tools.
- Built a shared prompt library and a one-page AI use policy.
- Provided a practice assistant in Teams and reviewed usage with team leads after a few weeks.
04 human in the loop
Where a person stays in charge
Your leadership team approves the AI use policy and the list of approved tools before training starts. Team leads review the prompts for their area before they are published to staff.
- Leadership approves the AI policy and approved tools before training.
- Team leads review prompts before they are published.
- The policy says clearly which data may go into which tool.
05 rollout
From kickoff to live in 6–8 weeks
- Weeks 1–2Interviews, policy draft and tool approval.
- Weeks 3–4Workshops by team and the prompt library.
- Weeks 6–8Usage review and a second round where people are stuck.
06 after
What changes
AI use spreads beyond a few enthusiasts to people in every team, within rules everyone understands. Managers can see which tasks people now handle with AI and where more support is needed.
Assume 65 staff each saving 1 hour a week on drafting and summarising after training.
65 × 1 hour = 65 hours a week, or over 3,000 hours a year, against the cost of licences already bought.
What to measure
- Weekly active users of approved tools.
- Prompts used from the library.
- Time saved on target tasks, as reported by teams.
- Use of unapproved tools.
07 evidence
What the research says
In a survey of 31,000 people in 31 markets, 75% of knowledge workers said they use AI at work, and 78% of AI users bring their own AI tools.
In a study of 453 professionals doing writing tasks from their own jobs, access to ChatGPT cut the time taken by 40% and raised quality, as rated by evaluators, by 18%.
In a field experiment with 758 BCG consultants, those using GPT-4 on tasks within its abilities finished 12.2% more tasks, 25.1% faster and at over 40% higher quality, but did worse on a task outside its abilities.
08 lessons
Mistakes to avoid
- Generic training. Use people's real tasks.
- No policy. People will use whatever is easiest.
- One-off workshops. Habits need follow-up.
09 questions
Common questions
- How much difference can training make?
- In a study of 453 professionals, ChatGPT cut writing time by 40% and raised quality by 18%. Studies also show AI can make people worse on tasks outside its abilities, which training must cover.
- Are staff already using AI?
- Probably. Microsoft and LinkedIn found 78% of AI users bring their own tools to work, which is why a clear policy matters.