AI for accountants: what helps, what to check, and how to start
Where AI helps a small UK accountancy practice, where it must not work alone, what ICAEW and ACCA say, and a safe way to start.
Most small practice owners know the pain. Clients send records late, in the wrong format, or not at all. Your team spends hours chasing, typing and checking before the real work starts.
AI for accountants can take some of that load, and some of the claims are fair. This guide is for the owner of a small UK practice. It covers where AI helps today, where it must not work alone, what the professional bodies have published, and a safe way to start. It is general guidance, not legal or tax advice.
Where AI helps today
These are drafting and sorting jobs. In each one, a person reads the result before it goes anywhere. If the term is new to you, what an AI agent is explains the basics.
- Chasing client records. Draft a polite reminder that lists exactly what is missing, ready for you to send.
- Emails and engagement letters. Produce a first draft from your own template, then edit it to fit the client.
- Summarising HMRC guidance for a client. Paste in the gov.uk page and ask for a plain English summary. Then check it against the page yourself.
- First pass bookkeeping checks. Flag duplicates, odd amounts and missing receipts for a person to review.
- Meeting notes. Turn a call transcript into a short note and a list of actions.
One caution on engagement letters. AI can draft the wording, but it does not know your terms, your insurer's requirements or your regulator's rules. Use your own approved template as the base, and have a partner sign off the final text.
The same goes for client emails about tax. A draft that sounds warm and clear can still say something you would never advise. Read every line as if you wrote it, because your name is on it.
Where AI must not be trusted alone
AI writes fluent text whether or not it is right. That is the danger in three areas.
- Tax advice. A confident answer is not a correct answer. It may not fit your client's facts or the current rules.
- Figures. Numbers should come from your accounting software, not from a chat reply. Check any figure an AI gives you against the source record.
- Filing. Nothing goes to HMRC or Companies House until a qualified person has checked it.
The risk is real in tax work. ACCA's guidance on AI in tax warns that generative AI can invent content such as case law. It points to the tribunal case Harber v HMRC as an example.
What ICAEW and ACCA have published
In January 2026, the seven bodies that jointly prepare Professional Conduct in Relation to Taxation, including ICAEW, issued advice to members on using AI in tax work. ACCA's version is dated 19 January 2026. Its main points are plain:
- You are ultimately responsible for any work you produce, even if AI helped.
- Treat AI output with the same scepticism you would apply to a junior colleague's work.
- Putting client data into a public AI tool is likely to breach confidentiality unless the client has consented.
- Check output against current legislation and guidance, and consider keeping copies of your sources and prompts.
- Train anyone who uses the tool.
ICAEW's generative AI do's and don'ts say much the same. It advises small, non-critical pilots, people reviewing the work, and disclosing when AI has contributed. It also says to keep client and confidential internal data off public AI tools.
If you belong to a different body, read its own guidance too.
Client data and the ICO
Client records are full of personal data, and UK GDPR still applies when a tool processes it. The ICO's guidance on AI and data protection covers accountability, transparency, lawfulness, accuracy and fairness.
That page was last updated on 15 March 2023 and is under review because of the Data (Use and Access) Act. So check the ICO site for the current position before you rely on it.
In practice, ask three things of any tool. Where does the data go? Can the provider use it for training? Have clients been told? Put the answers in your engagement terms and privacy notice.
Making Tax Digital: why your clients will need help
At the time of writing (November 2026), gov.uk sets out the dates for Making Tax Digital for Income Tax by qualifying income:
| Qualifying income | Tax year it is based on | Start date |
|---|---|---|
| Over £50,000 | 2024 to 2025 | 6 April 2026 |
| Over £30,000 | 2025 to 2026 | 6 April 2027 |
| Over £20,000 | 2026 to 2027 | 6 April 2028 |
Some people are exempt, and partnerships will follow on a timeline still to be set. Check gov.uk for the latest, as dates can change.
The effect for you is more clients who must keep digital records. That means more chasing and more checking. AI can help with the chasing and a first look at the records. Your software and your people still carry the compliance.
Questions to ask before you pay for a tool
Salespeople will tell you it is safe. Ask them to show you in writing.
- Where is client data stored, and who at the provider can see it?
- Is our data used to train the provider's models?
- Can we delete our data, and can we export it if we leave?
- Does the tool keep a log of what was asked and answered?
- What happens if the provider has a breach?
If the answers are vague, treat that as an answer. Record what you were told and when.
A safe way to start
- Write a one page rule on what staff may and may not put into AI tools. Our AI policy template for UK businesses is a starting point.
- Pick one low risk job, such as reminder emails or meeting notes.
- Read the tool's terms on data first. While you test, leave client names out.
- Have a person read everything before it is sent. Keep a short note of what you asked and what you checked.
- After a month, decide whether to keep it, change it or drop it.
- Only then add a second job. Go slowly, because a small mistake repeated across many clients becomes a large one.
When AI is the wrong tool
Skip it where judgement carries the weight, such as tax planning for a complex client. Skip it where you cannot check the answer quickly. If checking takes as long as doing the job, do the job.
Sometimes a template or a rule in your accounting software does the job better and cheaper than AI.
Where Agentic Scale fits
We build AI employees for small practices. They chase records, draft client emails and prepare first-pass checks, working from an AI Brain: your engagement terms, checklists and tone of voice saved as plain files in the practice's own Drive.
The figures and the advice stay with a qualified person, who approves each item before it is sent.
Begin with the AI Audit. The cost is £197, and you get the full amount credited if you build within 30 days. Our page on AI Agents for accountants shows how it looks in a working practice.
Sources
Can AI do bookkeeping for a small business?
Is it safe to put client data into ChatGPT or Claude?
Do I have to tell clients that I use AI?
Will AI replace accountants?
An AI receptionist can answer calls and book appointments, but it needs rules, testing and a person watching. Here is what UK firms should check.
5 min read · 12 November 2026A plain guide to the HMO licence in England: when you need one, what it costs, the penalties, and how to keep renewals on track.
5 min read · 12 November 2026Agentic AI plans and carries out jobs, not just answers questions. Here is what that means, where it helps and where to be careful.
5 min read · 12 November 2026Not sure where to start?Book an AI Audit.
We look at how you work today and tell you honestly what AI should and shouldn't do for you. £197, credited in full if you build within 30 days.