How a UK accountancy firm approached Copilot adoption without overwhelming the team

The firm already used Microsoft 365 heavily across Outlook, Teams, Word, Excel and SharePoint. Leaders could see that Copilot might help with repeated client communication, meeting follow-up, document preparation and internal knowledge retrieval.

The concern was not whether AI looked interesting. The concern was whether staff would use it well during normal accountancy work, especially when deadlines, client confidentiality and review standards mattered.

The team had mixed confidence. Some people were curious and experimenting. Others were unsure what Copilot was allowed to touch, whether the output could be trusted and how it fitted around existing review processes.

The firm needed to avoid a launch that felt exciting for a week and then faded. It also needed to avoid pushing junior staff towards shortcuts that looked faster but created review risk later.

The adoption plan deliberately stayed away from broad promises about AI transformation. It focused on the situations where Copilot could help a person prepare, draft, summarise or organise work while keeping professional judgement with the team.

• Generic prompt training with no link to accountancy workflows.
• Unreviewed use of Copilot output in client-facing work.
• Using sensitive client data before permissions and acceptable use were understood.
• Rolling out too much change during already busy delivery periods.
• Treating automation or agents as the first answer before the workflow was clear.

The rollout was reframed around a small number of recognisable tasks. Instead of asking staff to adopt Copilot in general, the firm introduced practical patterns for client emails, meeting notes, internal guidance, report commentary and routine admin preparation.

Training also made the review layer explicit. Staff were shown how to check source context, mark output as a first draft and decide when Copilot should not be used for the work in front of them.

The training worked best in smaller groups, with examples matched to tax, accounts and admin roles. The language stayed plain: what Copilot can help with, what it cannot decide and what a good prompt needs to include.

The firm also needed a follow-up rhythm. That meant giving people somewhere to bring real questions after the first session, because most useful Copilot habits appear when staff try it against live work and hit the first awkward edge case.

• Role-based examples for client-facing teams.
• Prompt patterns for drafting, summarising, rewriting and preparing follow-up.
• Safe-use guidance around client data, permissions and review.
• Partner-visible examples so leaders could see the quality bar.
• A simple feedback route for questions after training.

The next sensible move was not to build everything at once. The firm could strengthen the first use cases, clean up the source material people rely on and decide which repeated questions or intake steps would justify an agent later.

That staged route kept adoption practical. Copilot could support everyday work first, while agents and automation stayed connected to workflows that had a clear owner and a real reason to exist.