Most companies start their AI journey by buying Claude, ChatGPT or Gemini subscriptions for the team. Two months later, a handful of people use them for general questions and everyone else has gone back to the old way. The problem isn’t the tool. It’s the starting point.
Start with the list of repetitive tasks
Ask each department to list the tasks that repeat weekly and take more than an hour: reconciling statements, summarising contracts, preparing performance reports, answering similar supplier queries, building presentations. That list is your real roadmap.
Rank by return and ease
Not every repetitive task is the right first step. Start with work that meets three conditions: it repeats often, its inputs are clear, and its output is easy to check. Document summaries, correspondence drafting and spreadsheet analysis are good places to begin. Sensitive financial decisions are not.
Set rules before the tools spread
Which data must never be entered? Who reviews output before it reaches a client? A one-page usage policy protects the company more than any technical setting.
Train on real work
Generic “how to write prompts” training is forgotten within a week. Training that takes an accountant’s actual task and turns it into a new workflow in front of them sticks. That’s why we build our programmes around each department’s own tasks, and finish with a template library teams use the very next day.