What changes
AI capability in-house means the practice understands AI well enough to choose tools, use them every day and explain them to clients. It isn’t bought in as a black box from a software supplier. In an accountancy practice this matters twice over: you need it to run your own work, and your clients will ask you about AI in their businesses long before they ask anyone else.
Capability has two parts. The first is skill: every person in the practice, from partners to trainees, knows what the tools do well, where they get things wrong and how to check the output. The second is rules: a short written policy on which tools may see client data, how outputs are reviewed and who signs off. Your professional body’s ethical code and your anti-money laundering (AML) obligations still apply when a task is done with AI.
Why it matters now
Every later move depends on this one. You can’t hand over the legwork, build a tax checker or advise a client on AI until your people are confident with the tools. Starting now means the practice learns on its own work, where mistakes are cheap, before it sells AI-based services to clients.
Checklist
- Enrol the whole team on the AI training for accountants course. Partners should start with Module 1: AI Fundamentals for Accountants; bookkeeping staff with Module 2: Data Processing & Bookkeeping and Module 3: Bank Reconciliation & Automation.
- Write a two-page AI policy: approved tools, what client data may go into them, how every output is checked, and who is accountable. Review your engagement letters and privacy notice so clients know how AI is used on their work.
- Name an AI lead for the practice with two hours a week set aside, and a monthly team session to share what’s working.
- Choose three internal tasks to practise on for a month, such as drafting client emails, summarising HMRC correspondence and reviewing a trial balance for anomalies. Keep a simple log of time taken and errors found.
Illustrative example
A practice with a partner and four staff sets aside 30 minutes a day each for six weeks: 5 people times half an hour times 5 days times 6 weeks is 75 hours in total. They work through the course modules and practise on their own tasks. By the end, drafting a standard client letter takes 10 minutes instead of 30, and the practice has a written policy, an approved tool list and a log of where AI went wrong and how it was caught. The partner uses that log to decide which processing work moves first in Move 3.
Common mistakes
- Training the partners and leaving the team to pick it up. The people doing the processing are the ones whose work changes first, so they need the training most.
- Letting staff put client data into free consumer tools with no policy. Agree the approved tools and the data rules before anyone uses AI on client work.
- Treating training as a one-off. The tools change month by month, so set a regular slot to review what’s new and update the policy.