Loading...
Sector · Accountancy and bookkeeping
AI for accountants, with a qualified person still signing off
A practical guide to AI for accountants in UK practice: what to automate now, what a qualified person must still review, and how we help you build it.
In focus →
Live routingHover to pause
Incoming
Checking…
AI handles0
A person decides0
MTD for Income Tax qualifying income threshold from April 2026£0Falls to £30,000 from April 2027 and £20,000 from April 2028.
Source: HM Revenue & Customs (GOV.UK) ↗UK firms (10+ staff) using AI0%Up from around 12% in late 2023.
Source: Office for National Statistics ↗Where the line sits
Most of the work in a practice is volume, and volume is what AI is good at
Bank lines, receipts, supplier invoices and client emails arrive faster than most teams can process them.
AI can now read, code and match much of that volume with useful accuracy.
It cannot take responsibility for the numbers, and your professional body is clear that you still do.
This guide sets out where AI for accountants earns its keep in a practice, where a qualified person stays in the loop, and how escalation should work.
We design, build and train teams on those workflows, so the AI does the grind and your people keep the judgement.
What AI handles in the practice
- Bank reconciliation suggestions and matching
- Data extraction from receipts, invoices and statements
- Transaction coding against your chart of accounts
- First-draft management accounts commentary
- Client query triage and routing
- Chasing records ahead of quarterly MTD updates
What a qualified person keeps
- Reviewing and approving every return and set of accounts
- Judgement on treatment, estimates and adjustments
- Tax advice and anything a client will rely on
- Professional scepticism on anomalies the AI flags
- Client relationships, fees and engagement terms
- Deciding which client data can go into which tool
What to automate
AI for accountants starts with the repetitive, rule-bound work
ICAEW’s generative AI guide suggests starting with tasks that need little judgement and are low risk, time consuming and repetitive.
In a practice, that points straight at bank reconciliation, transaction coding and data extraction from receipts and invoices.
The AI suggests matches and codes, and flags anything below a confidence threshold for a person to look at.
The same pattern works for first-draft management accounts commentary, where the AI explains movements and a manager rewrites what matters.
Client query triage is another quick win, with the AI tagging each email and routing it to the right person with a draft reply.
Making Tax Digital for Income Tax brings quarterly updates for more sole traders and landlords, so record chasing and completeness checks are worth automating.
The HMRC timetable starts at £50,000 of qualifying income from April 2026, then £30,000 from April 2027 and £20,000 from April 2028.
85%
AI acts alone
130
65% of 200 cases
Goes to a person
70
35% of 200 cases
Illustrative batch with a typical confidence spread. Your real split comes from a pilot on your own work. Some decisions go to a person whatever the score.
Sign-off
What a qualified person must still review and sign
The first rule of AI for accountants is that output is a draft, however polished it looks.
The PCRT guidance on AI for tax work says output should be reviewed as if prepared by a less experienced junior colleague.
Every return, set of accounts and piece of advice still goes out under a named person’s judgement.
That person approves VAT returns, year-end adjustments, estimates and any treatment that needs interpretation.
They also own the client conversation when the AI finds something odd.
ACCA reminds members they cannot abdicate or outsource professional scepticism and judgement to technology.
We build the review step into the workflow, so nothing is filed or sent until the right person has approved it.
AI drafts
Check
Person decides
Learns
01 / 04
Professional scepticism
Scepticism and ethics when the software sounds confident
ICAEW’s Code of Ethics lists automation bias among the biases accountants should be aware of.
In practice, scepticism means checking the source document behind each AI suggestion, not just the suggestion.
It means sampling auto-matched items rather than trusting the match rate.
It means treating a tidy commentary paragraph as a hypothesis until the numbers support it.
Confidentiality matters as much, and ICAEW advises against loading confidential client information into public generative AI tools.
Our AI policy template gives you a starting point for which tools can see which data.
ICAEW’s Code of Ethics lists automation bias among the biases accountants should be aware of.
How we help
Build it once, with escalation designed in
We start by mapping where hours go across bookkeeping, compliance and advisory work.
Then we pick two or three workflows where AI for accountants can take volume without touching judgement.
Each workflow gets a confidence threshold, a named reviewer and a clear route for exceptions.
Anything unusual, such as a large unmatched receipt or a query that sounds like tax advice, goes straight to a person.
We train your team to review AI output properly, because that is where most of the risk sits.
If you want the wider picture first, our AI readiness assessment is a sensible place to start.
We start by mapping where hours go across bookkeeping, compliance and advisory work.
How it works
Four steps, one owner
Each system ships with a named person accountable for what it does.
01
Step 1
Map the work
We look at where time goes across your client portfolio and pick the tasks with the most volume and the least judgement.
02
Step 2
Set the rules
We agree confidence thresholds, named reviewers, data boundaries and what always escalates to a qualified person.
03
Step 3
Build and test
We build the workflow against your real data in a safe environment and measure accuracy before anything goes live.
04
Step 4
Train and hand over
Your team learns to review AI output with scepticism, and you own the process, the policy and the audit trail.
Questions
Asked and answered
Further reading
From the source
ICAEWGetting started with generative AI ↗Start with low-risk, repetitive tasks that need little judgement.ICAEWGenerative AI and ethics ↗How the fundamental principles apply, including confidentiality.ICAEWThe role and mindset expected of professional accountants ↗The Code of Ethics on automation bias and an inquiring mind.ACCAKey areas of risk when adopting AI ↗Governance, confidentiality, client communication and training.Professional Conduct in Relation to Taxation bodies (hosted by ACCA)PCRT topical guidance: the ethical use of AI for tax work ↗Members remain accountable for work produced with AI.HM Revenue & Customs (GOV.UK)Find out if and when you need to use Making Tax Digital for Income Tax ↗Start dates and qualifying income thresholds from 2026 to 2028.
Keep reading
Related pages
SectorsAI for financial advisers →AI for financial advisers, where the adviser still owns the adviceGovernanceAI readiness assessment →An AI readiness assessment that names the human checkpointGovernanceAI policy template →An AI policy template your team will actually followTrainingAI training for business →AI training for business, with people who know when to step in
Next step
Tell us the work. We’ll show you where the line goes.
A 30-minute call. We map one workflow, what the AI could take, and where a person must stay in the loop.