AI for accountants: practical, safe uses in a small practice
Accountants can use AI and automation to chase clients for records, capture receipts and bills, draft routine emails and summaries, and write up meeting notes. Keep client data in approved business tools, follow your professional body's confidentiality rules, and have a person review anything that goes to HMRC or a client.
I'm Managing Partner of a two-office law firm with around 30 people, so client confidentiality and hard deadlines are my working life too. This guide covers what I'd put on systems in a small practice, and what I'd keep with a person.
Sound familiar?
The work comes in waves, and the waves are getting closer together. January was always hard. Now Making Tax Digital adds four quarterly deadlines a year for some of your clients.
- Half my week goes on chasing clients for bank statements and receipts.
- I'm doing admin in the evenings instead of the work clients pay for.
- The team would love to use ChatGPT, but I'm not putting client data into it.
- Every new client means the same forms, the same ID checks and the same emails, typed again.
How can accountants use AI in a small practice?
Start with the jobs that repeat and don't need judgement: chasing records, capturing documents, drafting routine emails and writing up meetings. Automation handles the reminders and the paperwork flow. AI helps with first drafts and summaries. The judgement stays with you: the numbers, the advice and anything that goes to HMRC or a client.
Automation and AI do different jobs, and people often lump them together. Automation follows rules you set: when a client uploads a file, tick it off and stop the reminders. AI handles the fuzzy parts, such as reading a document or drafting a reply. The most useful set-ups use both, with a person checking the output.
| Job | What changes | Tools that often do it | Who checks |
|---|---|---|---|
| Client onboarding | Forms, ID documents and the engagement letter sent and tracked in one flow | Your practice software, online forms, e-signature | You decide on due diligence and sign off the engagement |
| Record chasing | Polite reminders on a schedule, stopped when the files arrive | Power Automate, Zapier or your practice software's own reminders | A person picks up anyone who doesn't respond |
| Receipts and bills | Key details read from a photo or PDF and prepared for posting | Hubdoc in Xero, Dext | A person reviews each entry before it's published |
| Routine client emails | A first draft written from your notes | Microsoft Copilot, or ChatGPT on a business plan | You read and edit before sending |
| Meeting notes | A summary and actions written up after the call | Teams intelligent recap or a similar notes tool | You check the notes before they go on file |
| Workload view | Deadlines and missing records on one screen | A dashboard in Excel or Power BI | You set the priorities |
Drowning in record chasing before every deadline? Let's look at where the hours go in your practice.
Book a free 30-minute callHow do you automate client onboarding and record chasing?
Map the steps once, then let software run them. A new client gets a welcome email, a secure upload link and the engagement letter to sign. Reminders go out on a schedule and stop the moment the files arrive. Anyone who hasn't responded after the last reminder lands on a person's task list.
An example record-chasing sequence (the timings are illustrations, not rules)
- Day 1. The request goes out with a checklist of exactly what's needed and one upload link.
- Day 7. A friendly reminder that lists only what's still missing.
- Day 14. A firmer reminder, with the deadline stated plainly.
- Day 21. No more emails. The client appears on a named person's call list.
- Any day. When the last file lands, the reminders stop and the job moves to "ready to prepare".
In the systems I've designed at my firm, an internal request is chased after two days, and a review request gets one reminder and then stops. Clients need longer than colleagues, which is why the sequence above leaves a week between reminders and ends with a phone call, not a third email.
If your practice runs on Microsoft 365, my Power Automate examples show the kind of flows involved. Chasing your own unpaid fees works the same way: see chasing unpaid invoices automatically.
Onboarding also includes anti-money laundering checks. Automation can collect ID documents and flag gaps, but the customer due diligence decision stays with a person. HMRC's guidance for accountancy service providers says due diligence with little or no face-to-face contact must reflect the extra risk.
Can AI read receipts and bills for you?
Yes, for the typing part. Document capture tools read a photo or PDF, pull out details such as the supplier, date and total, and prepare an entry for someone to review. Xero's Hubdoc works this way, and Dext does a similar job. The review step matters: a person should check entries before they're published.
Xero says Hubdoc extracts details such as the contact, date, total and invoice number, ready to code and publish with the original attached. Dext describes its app as capturing and categorising receipts, invoices and expenses, then syncing them to your accounting software. I won't repeat anyone's accuracy figures. Test any tool on a batch of your own clients' documents and count what needs correcting.
For a practice, the bigger win is often the collection, not the reading. If clients photograph receipts into one place all year, nobody is sorting a carrier bag of paper in January.
Is ChatGPT safe for accountants to use with client work?
Not with client data on a free or personal account. ICAEW has warned that putting client data into tools such as ChatGPT is a potential breach of confidentiality. Use a business plan your practice controls, keep identifying details out where you can, and review every draft as carefully as you'd review a trainee's work.
The trainee comparison is ICAEW's own, from its regulatory news warning on AI tools. The safest first uses need little or no client data:
- Drafting the standard emails you send every week, such as year-end reminders and requests for records.
- Turning your own notes into a plain-English summary for a client, which you then check line by line.
- Explaining a rule change in simpler words for your newsletter. Check it against GOV.UK, because ICAEW notes many tools can't draw on the latest legislative developments.
- Tidying an internal procedure or checklist.
If your practice runs on Microsoft 365, Microsoft says Copilot respects each person's existing access rights and doesn't use prompts to train its models. Because it surfaces any file a person could already open, tidy your folder permissions first. Whatever you choose, write the rules down: my one-page AI policy template adapts well to a practice.
Can AI take notes in client meetings?
Yes, and it can save a lot of typing afterwards. Tools such as Teams' intelligent recap write up notes and suggested tasks after a call. Tell clients at the start that notes are being taken this way, make sure your privacy notice covers it, and check the notes before they go on the client file.
In Microsoft Teams, intelligent recap needs a Teams Premium or Microsoft Copilot licence, and transcription has to be switched on. Check what your current licences include before buying anything new. AI notes can miss nuance or get a figure wrong, so treat them as a first draft.
What do ICAEW and ACCA say about AI and confidentiality?
Both say your existing ethical duties apply in full. ICAEW's Code of Ethics covers confidentiality in subsection 114, and ICAEW expects due diligence, contractual protections and internal approval before confidential client information goes into any system. ACCA's research with CISI advises against unapproved or public AI tools without authorisation.
| What they say | What it means in a small practice |
|---|---|
| ICAEW: know where data is stored, who can access it and whether it may be used to train models | Read the vendor's data page before you approve a tool, and keep a note of what you found |
| ICAEW: due diligence, contractual protections and internal approval before confidential information goes in | One named person approves each AI tool, on a business plan with proper terms |
| ICAEW: review AI output with professional scepticism | Nothing AI-drafted leaves the practice unchecked |
| ACCA and CISI: safeguard client data and avoid unapproved or public AI tools without authorisation | An approved list, and personal accounts off-limits for client work |
| ACCA and CISI case study: a supplier sharing anonymised client data raises questions under the engagement letter | Check your software terms against your engagement letters and privacy notice |
ICAEW's July 2026 article on client confidentiality is worth ten minutes of any partner's time. It reminds firms that not every member of staff needs access to every client's information. That matters the day you switch on a tool that can search your files.
The ACCA Code of Ethics and Conduct binds members and partners in ACCA firms. ACCA's AI Monitor report, published with CISI in October 2025, says accountants using AI must still maintain confidentiality and exercise professional judgement. Its case study on a supplier sharing anonymised client data is a useful prompt to reread your own engagement letters.
Please note: this is general information from ICAEW, ACCA, HMRC and GOV.UK pages, not legal, tax or regulatory advice. Check your own professional body's current guidance, and take advice for your practice where you need it.
Does Making Tax Digital change what's worth automating?
Yes, for practices with sole trader and landlord clients. Since 6 April 2026, those with qualifying income over £50,000 must keep digital records and send quarterly updates. The threshold drops to £30,000 from April 2027. That turns an annual chase into a quarterly one, which is exactly the kind of repeat work automation handles well.
| From | Qualifying income over | Based on the tax return for |
|---|---|---|
| 6 April 2026 | £50,000 | 2024 to 2025 |
| 6 April 2027 | £30,000 | 2025 to 2026 |
| 6 April 2028 | £20,000 | 2026 to 2027 |
Source: GOV.UK, check if you're eligible for Making Tax Digital for Income Tax, checked 30 September 2026. GOV.UK notes the government is introducing legislation to confirm the 2028 start.
Quarterly updates are due by 7 August, 7 November, 7 February and 7 May. The Self Assessment return for 2025 to 2026 is still due online by 31 January 2027. A reminder sequence tied to each date, stopping as each client's records arrive, means your team only chases the clients who are behind.
One thing doesn't change: a person checks every figure before it goes to HMRC. The HMRC Standard for Agents expects agents to take reasonable steps to make sure third-party inputs, such as software, give accurate results. The software can prepare the update. You're still the one submitting it.
Will AI replace accountants?
Not the judgement, and that's what clients pay for. ICAEW's own guidance points out that AI tools are not qualified accountants. They can draft, sort and summarise, but they can't take responsibility for a set of accounts or a tax return. What can change is how much of your week goes on admin.
Around half of businesses using AI told the ONS it had made no difference to their headcount. The ONS also found that businesses in professional, scientific and technical activities are more likely to get AI through software they buy than through free tools. That's the industry group accountancy sits in. So for most practices, AI is arriving inside the software they already use.
In practice: what I see
In my own firm, I've built two things that map onto a practice's pressures. A live dashboard takes its numbers straight from our case management system, so nobody pieces reports together by hand. Work handed between colleagues follows a fixed route with one-click sign-off, rather than disappearing into someone's inbox.
Both keep a person in charge of the decision. The software moves the information; people sign it off. That's the model I'd use for record chasing and document capture too.
January is when this matters most. With the Self Assessment deadline on 31 January, every hour spent chasing a client for statements is an hour not spent on the return itself.
Common pitfalls
- Automating a messy process. If nobody agrees what "complete records" means, a reminder sequence sends confusion faster. Write the checklist first.
- Client data in personal accounts. ICAEW's dos and don'ts of AI say to keep client and confidential data off public AI tools. That includes a junior's personal ChatGPT on their phone.
- Publishing captured data unchecked. Extracted entries still need a person's review before they reach the ledger.
- Forgetting the paperwork. If a tool handles client data in a new way, check your engagement letters and privacy notice still describe what happens. Privacy information must name the recipients, or kinds of recipients, of personal data.
- Chasing people who've already sent it. Every reminder should stop the moment the file lands. Nothing irritates a client faster.
What to do next
How I can help
I set up business automation for small firms using software they already pay for, such as Microsoft 365 and Xero, with a person checking anything that matters. If you'd rather see where the hours go before anything is built, the AI audit maps your week first. And if your own clients ask you about AI, you're welcome to point them my way.
Questions people ask
Which AI tools do accountants use?
Many practices will meet AI first inside software they already use: capture tools such as Hubdoc and Dext, Microsoft Copilot in Outlook and Word, and meeting recaps in Teams. General assistants such as ChatGPT and Claude help with drafting, on a business plan. Start with what you already pay for before adding anything new.
Should we tell clients we use AI?
ICAEW stresses honesty about AI use, especially where it has supported research or analysis delivered to a client. Your privacy notice should also cover the kinds of organisations that receive client personal data, which can include AI providers. A short line in your engagement letter is a simple way to be open.
Can software submit returns to HMRC for us?
Making Tax Digital relies on compatible software to send updates and returns, but you remain responsible for what's sent. The HMRC Standard for Agents expects reasonable steps to make sure software gives accurate results. So a person should always review before anything is submitted.
How much time will automation save my practice?
It depends on your clients and your processes, so I won't promise a number. The honest way to find out is to measure one job before and after, such as how long record chasing takes in the month before a deadline.
Can you help practices in Sussex and Kent?
Yes. I'm based in Eastbourne and cover East Sussex, Brighton, West Sussex and the Kent border, in person or by video. I'm a solicitor, so I understand confidentiality in a regulated practice, but this consultancy doesn't give legal advice.
Related guides
Chasing unpaid invoices: late payment interest and automatic reminders
Chasing unpaid invoices from other businesses? How statutory late payment interest works, a worked example and a polite reminder routine for Xero or QuickBooks.
Read the guideAutomation · 14 min read12 Power Automate examples for small businesses on Microsoft 365
12 Power Automate examples for a small business on Microsoft 365, each set out as trigger and actions in plain English, with the licence each one needs.
Read the guideBusiness automation
Business automation consultant in Sussex. Daniel Andrews connects the software you already use so enquiries, quotes, invoices and reports run themselves.
Business automation