AI for accountants: what it does in a UK practice, and how to get it working.
For an accounting practice, the useful version of AI is not a robot accountant. It is the hours it takes out of the work that surrounds the accounting: the same client questions answered again and again, figures retyped from PDFs, letters written from scratch, and knowledge stuck in someone's inbox. The catch is that buying AI software is not the same as getting it working in your practice. I do that second part, as one senior operator, and you own what gets built.
In short
- AI pays off in a practice in a few specific places: client queries, document and data extraction, first drafts of letters and reports, onboarding, and finding things in your own files.
- Buying a tool does not get you those gains. Getting AI working inside your practice, on your data and your systems, does.
- Client confidentiality comes first. ICAEW and ACCA guidance both point the same way: a written policy, care with what goes into AI tools, and a human checking the output.
- You can find out what to build first for free, and a costed plan is £1,495, credited against the first build.
Where AI pays off in an accounting practice
Ignore the headlines about AI replacing accountants. In a real UK practice, the value sits in a handful of unglamorous places:
- Client queries answered from your own information. "When is my return due?", "What do you need from me?", "Did you get my receipts?" An assistant that answers from your engagement letters, deadlines and FAQs, and passes anything it cannot answer to a person.
- Documents turned into data. Bank statements, invoices and receipts read and pulled into a structured format for review, instead of being retyped. Your bookkeeping software may already do part of this; the gap is usually everything that arrives outside it.
- First drafts of letters, reports and emails. Engagement letters, management report commentary, chaser emails and client updates started from a good draft in your house style. A qualified person still reviews and signs off every one.
- Onboarding that runs itself. Collecting ID, documents and information from new clients, chasing what is missing, and getting everything into the right place before anyone in the practice touches it.
- Your practice's knowledge, searchable. Past advice, procedures and how-we-do-it notes that live in shared drives and inboxes, made findable in seconds by the whole team.
- The routine chasing. Reminders for missing records and approaching deadlines, drafted and sent to a schedule, so nobody spends Monday morning writing them.
None of this replaces professional judgement. It removes the repetitive work around it, which is where most of a small practice's lost hours go.
Tools versus implementation: why buying software is not the same as getting AI working
Most practices already have AI somewhere: in their accounting platform, in a document tool, or in a ChatGPT account someone signed up for. That is a fine start. What it rarely gives you is AI that works across your whole practice, on your own information, under your control.
| Off-the-shelf AI features | An AI system built around your practice | |
|---|---|---|
| What it works on | One app's data | Your documents, inboxes, engagement letters and systems together |
| Fit | The vendor's idea of your workflow | Your actual process, step by step |
| Who owns it | The vendor; it stops when you stop paying | You, on your own accounts, with no lock-in |
| Client data | Goes where that vendor sends it | Kept where you decide, with access you control |
| Cost | Per user, every month, indefinitely | A fixed price to build, then the genuine running costs |
| Best for | Getting started, single tasks | Removing whole chunks of practice admin |
Both have a place. The question is which one moves your numbers. Usually it is a small amount of the right build, wired into what you already use. That is the work I do, and you own it, which is why I call it AI you actually own.
Client data, confidentiality and professional guidance
For an accounting practice, this is the section that matters most, and the professional bodies have been clear about the direction.
ICAEW has published a Generative AI Guide. Its ethics section applies the fundamental principles, including confidentiality, to AI use, and points out that you need to understand who information is shared with, how it is secured and how long it is kept. Its section on using generative AI says accountants should be transparent by labelling or acknowledging AI-generated output, and that easy access to free tools means organisations need their own guidance on use.
ACCA's Accounting and Business magazine makes the same practical points: set clear rules on how AI is used, keep confidential information out of public tools, and do not rely on AI without human oversight, because once data goes into a generative AI system, control over where it is stored and shared can be lost.
The ICO sets out how UK GDPR applies to AI in its guidance on AI and data protection, covering principles such as data minimisation, security and transparency.
In practice, that means three things for any AI I build for a practice:
- Client data stays where you decide. Systems run on your accounts. By default I use business or API tiers, or private deployments, that don't train on your data, and we agree the set-up per project. Nothing is pasted into public chat tools.
- A person checks anything a client or HMRC will see. AI drafts; your qualified staff review and sign off.
- It is written down. You get documentation of what each system does, what data it touches and who has access, which helps with your own policy and your data protection records.
This is not legal or regulatory advice, and it is no substitute for your own professional body's guidance or your data protection obligations. Check the current versions of the pages above, and speak to your professional body if you are unsure.
What it costs to implement AI in a practice
Software subscriptions are not the cost that matters. Getting AI working around your practice is. Here is how I price it, all fixed and agreed before any work starts:
- First AI System Recommendation: free. The single most valuable thing to build in your practice, in writing.
- AI Readiness Assessment: £1,495, fixed, and credited in full against your first build. A costed, prioritised plan.
- AI Growth Sprints: £1,800 a month. One AI system built, integrated and owned each month.
- AI Business Transformation: £15,000 to £30,000. A core process reworked around AI end to end.
- Custom AI Product Build: £20,000 to £40,000. A bespoke tool built around how your practice works.
- AI Implementation Days: £2,500 to £3,500 a day, for a defined problem.
Full detail is on the services page, and I explain where the money goes in what good AI implementation actually costs.
What to build first
The tool is never the first problem. The first problem is knowing which one thing will give you back the most time. The tool is never the first problem explains why I always start there.
An illustrative example (not a real client)
A six-person practice handles around 300 personal tax and small-company clients. Every January the team loses days to the same three questions by email and phone, and to chasing missing records. The first system worth building is not a new accounting tool. It is an assistant that answers those questions from the practice's own deadlines and engagement letters, plus automated, personalised chasers for missing records, with anything unusual passed straight to a person. It touches no figures and no advice, so the risk is low, and the hours come back in the busiest month of the year. The practice owns it, and the second system builds on it.
Your practice will be different. That is the point of the free recommendation: it names the one thing worth building first in your practice, not a generic one.
Common questions
Will AI replace accountants?
No. AI is good at repetitive, rule-based work around the accounting: reading documents, drafting, answering routine questions and chasing. Judgement, advice, sign-off and the client relationship stay with qualified people, and the professional bodies' guidance assumes a human checking the output. The practices that gain are the ones that use AI to take the repetitive work off their team.
Can AI do bookkeeping?
Partly. AI can read and categorise transactions and pull data out of documents, and most bookkeeping platforms now include some of this. It still needs a person to review and correct it, and it works best when it is connected to the rest of your process rather than used on its own.
Is ChatGPT safe for client data?
Not in the way most people use it. Pasting client information into a public chat tool means losing control of where that data goes, which is exactly what ICAEW and ACCA guidance warns about. Business and enterprise versions, or systems run on your own accounts, can be set up so your data is not used for training and access is controlled. Check the terms of any tool before client data goes near it.
What do ICAEW and ACCA say about using AI?
Both point in the same direction: have clear rules for how AI is used, protect confidential information, be transparent about AI-generated output, and keep a qualified person reviewing it. See the ICAEW Generative AI Guide and ACCA's guidance linked above for the detail.
How much does AI cost for a small practice?
Off-the-shelf tools are usually a monthly subscription per user. Getting AI working across your practice is a fixed-price build: the AI Readiness Assessment is £1,495, credited against the first build, and ongoing monthly build is £1,800 with everything owned. The first recommendation is free.
Do I need an AI consultant, or just software?
Software is enough for trying things out. You need implementation if you want AI built into how the practice runs, on your own data, owned by you, without hiring a developer. See what an AI implementation consultant does.