What changes
Most of the hours in a small practice go on processing: coding transactions, reconciling bank accounts, chasing missing records, preparing working papers and drafting returns and letters. AI can now do a large share of that first pass. In the new model it does, and the accountant’s job moves to setting up the process, reviewing what the AI produced, dealing with the exceptions and talking to the client.
The word that matters is checked. Every workflow you hand over needs a defined review step, a named reviewer and a record of what was checked. That record is what lets you stand behind the numbers, and it’s the same discipline you’ll later sell to clients as assurance of their own AI-produced figures.
Why it matters now
Quarterly digital reporting under Making Tax Digital for Income Tax means more submissions per client, more often. Doing that by hand multiplies the processing load. Doing it with AI and a review step keeps the load flat and gives you up-to-date figures on every client, which is the raw material for the foresight services in Move 4.
Checklist
- Pick the two highest-volume processing tasks from your Move 1 mapping, usually transaction coding and bank reconciliation, and redesign each as a written workflow: what the AI does, what a person reviews, and what gets escalated.
- Run each new workflow alongside the old one for one month on a sample of 10 clients, and compare accuracy and time before you switch fully.
- Set up AI-drafted first versions of routine client correspondence, such as records requests, deadline reminders and payments on account reminders, with a person approving every one before it’s sent.
- Record review evidence in the client file: who checked, what was changed, and why. Make sure it would stand up to a quality review by your professional body.
- Book the time you free up straight into client diaries as short forward-look calls, so it doesn’t disappear into more processing.
Illustrative example
A practice prepares 80 sole-trader returns a year, and records preparation takes about 4 hours per client, or 320 hours in total. With AI categorisation and first-draft working papers, plus a person reviewing every file, it drops to about 2.5 hours per client, or 200 hours. That frees 120 hours. The practice spends 40 of them on a 30-minute call with each of the 80 clients about the year ahead, and keeps the other 80 hours for taking on new clients.
Common mistakes
- Switching on AI features in your software and assuming the output is right. Without a review step and a record of it, you can’t stand behind the numbers.
- Automating a messy process as it stands. Tidy the workflow first, or the AI will reproduce the same errors faster.
- Letting the saved hours fill up with more of the same work. Decide in advance where they go, and put them in the diary.
Coaching your clients
Your clients’ businesses have the same legwork: invoicing, credit control, stock, rotas and admin. As you hand over your own, tell clients what you did and what it saved. A 20-minute conversation at the year-end meeting about which of their tasks AI could take on is the start of the readiness review in Move 4, and it’s usually the first time anyone has offered them practical help with it.