---
title: "AI for UK accountants in 2026: I tested what 12 firms actually use (not what Sage sells you)"
description: "A practitioner walk-through of which AI workflows actually stuck in 12 UK accounting firms and which were quietly switched off. Based on hands-on work across tax prep, bookkeeping, audit, payroll, and advisory. Includes a comparison of Sage AI, Xero JAX, Dext, Iplicit, AccountsIQ AI, and ChatGPT-on-private-data, plus an honest take on the 60% data-entry stat the vendors love. The post finishes with a four-step audit framework any UK firm of 5 to 50 staff can run this quarter to find the one workflow worth redesigning, and a frank take on advisory work in the wake of the SRA's Garfield AI authorisation for legal services."
canonical: https://richardbatt.com/blog/ai-for-accountants-uk-2026-12-firms-tested
date: 2026-05-05
author: Richard Batt
tags: [AI Implementation, AI for Accountants, SMB AI, UK Business]
type: blog_post
---

# AI for UK accountants in 2026: I tested what 12 firms actually use (not what Sage sells you)

_A practitioner walk-through of which AI workflows actually stuck in 12 UK accounting firms and which were quietly switched off. Based on hands-on work across tax prep, bookkeeping, audit, payroll, and advisory. Includes a comparison of Sage AI, Xero JAX, Dext, Iplicit, AccountsIQ AI, and ChatGPT-on-private-data, plus an honest take on the 60% data-entry stat the vendors love. The post finishes with a four-step audit framework any UK firm of 5 to 50 staff can run this quarter to find the one workflow worth redesigning, and a frank take on advisory work in the wake of the SRA's Garfield AI authorisation for legal services._

**Richard Batt** — AI implementation specialist. 120+ projects across 15+ industries, serving SMBs (5-200 employees) worldwide from Middlesbrough, UK (working globally). Contact: richard@richardbatt.com · https://richardbatt.com

The Google AI Overview for "ai for accountants uk" cites Sage, Xero, AI Accounts, Wolters Kluwer, Fusion Accountants, and CloudBook. None of them's a practitioner. Each one's selling something. So the honest answer to the question every UK accountant's asking right now, which is which AI workflows actually save real hours and which ones get switched off after three weeks, never quite arrives in those Overviews.

This piece is what I've got in my notes from the last quarter of work with 12 UK accounting firms. They range from a four-person high-street practice in Stockton to a 38-partner firm in Manchester. The composite descriptions below pool patterns from those engagements and from the wider book of 120+ AI projects across 15 industries I've shipped over the past few years. Names have been removed, ratios have been rounded, and a couple of the firm sketches blend two or three real practices to keep client confidence. But the numbers, the workflow choices, and the failure modes are real.

**The short version**

- The Sage 60% data-entry reduction stat is real for a narrow workflow (invoice-line capture) and almost nowhere else. Treat it as a ceiling for one task, not a firm-wide promise.
- Across 12 firms, three workflows reliably stuck: bookkeeping data capture, first-draft client emails, and audit working-paper review. Two reliably failed: tax narrative drafting and full-fat advisory generation.
- The tool that mattered least was the AI feature inside the practice management system. The tool that mattered most was a private-data ChatGPT instance with the firm's own playbooks in it.
- Advisory work is changing fast. After the SRA's Garfield AI authorisation in April 2026, accounting clients are starting to ask the same "are you charging me for AI work" questions that solicitors are facing.
- The four-step audit at the end of this piece is the same one I run on a £2,000 AI Roadmap audit. You can run it yourself in a Friday afternoon.

## What the vendors are actually selling

Before the firm-by-firm walk-through, it helps to put the major UK tools in one table. There are six that come up in nearly every conversation I have with UK accountants. The promises are similar. The fit isn't.

| Tool | What it does well | What it doesn't | UK pricing baseline |
| --- | --- | --- | --- |
| Sage AI (Sage Intacct, Sage 50 add-ons) | Invoice-line capture, supplier matching, anomaly flags on transactions | Doesn't touch advisory or narrative work; weak on UK SME chart-of-accounts variety | Bundled with Sage subscriptions; effective AI add-on £15 to £30 per user per month |
| Xero JAX (the AI assistant launched in 2024) | Reconciliation suggestions, drafting queries, basic bookkeeping Q&A | Hallucinates on UK-specific tax rules; not yet trustworthy for client-facing output | Included with Xero Premium tiers from £55 per month |
| Dext (formerly Receipt Bank) | Best-in-market receipt and invoice capture; clean integration with Xero, QuickBooks, Sage | Not really an AI assistant; document capture only | From £20 per user per month |
| Iplicit | Mid-market cloud accounting with AI-assisted automation; strong reporting | Limited adoption outside multi-entity firms; smaller ecosystem | From £150 per user per month, scaling with modules |
| AccountsIQ AI | Multi-entity consolidation with AI-assisted variance analysis | Designed for groups and PE-backed mid-market, not high-street firms | £250+ per user per month |
| ChatGPT (private-data instance, OpenAI Team or Enterprise) | Drafting and summarising client correspondence; Q&A across loaded training docs | Not a bookkeeping engine; needs scaffolding to keep client data inside the firm's tenancy | £20 to £50 per user per month, plus setup |

If you take one thing from that table, take this. Three of those six aren't really AI products at all. They're accounting products with AI features bolted on. And the capabilities that actually moved hours in the firms I worked with came from a small number of well-scoped tools, not from a sprawling stack.

## The 12 firms, sketched

I'm going to walk through them in three groups: the ones where AI saved real hours, the ones where it saved partial hours, and the ones where it cost more than it saved.

### Group A: real hours saved (4 firms)

**Firm 1. A 4-partner audit and accounts firm in Leeds, 22 staff, mostly owner-managed business clients.** They built a private-data ChatGPT workspace with their own audit-file templates, ISA 240 working papers, and the last three years of internal training memos loaded in. Audit juniors now ask the workspace for first-draft sample selections, materiality narratives, and prior-year comparisons. The seniors review and edit. Time saved on a typical small-company audit file: 4 to 6 hours over a 60-hour engagement. The investment was a £900 one-off configuration and £25 per seat per month. Payback was inside two months.

**Firm 2. A high-street firm in Newcastle, 11 staff, heavy on personal tax.** They use Dext for receipt capture (saving roughly 6 hours a week of input across the team) plus a private-data ChatGPT for client query triage. When a client emails about a CGT calculation, the AI drafts a structured reply that pulls from the firm's own past explanations. The senior partner reviews and sends. The reply rate has gone from 36 hours to 5 hours during the busy season.

**Firm 3. A 6-person bookkeeping bureau in Hull.** They use Xero JAX for inside-Xero work, plus Dext for capture, plus a custom workflow that auto-flags VAT codes that look wrong before the client review. On a typical 100-transaction month, manual touches dropped from 38 to 11. The bookkeeper now handles 40% more clients per FTE.

**Firm 4. A 9-partner mid-tier firm in Manchester, 38 staff.** They invested in a properly scoped AI documentation reviewer that reads client-supplied schedules, flags inconsistencies against the prior year, and writes a draft variance commentary. Manager review time on year-end accounts dropped by about 20%. The build cost £18,000, came in three weeks, and was paying for itself by month four.

### Group B: partial gains, ongoing tuning (5 firms)

**Firm 5. A 3-person tax practice in Bristol.** Tried using AI for SA100 narrative drafts. The output looked confident but contained two HMRC errors in the first ten cases. They now use AI only for client-explanation paragraphs, not for technical tax positions. So the time saved is real but smaller than expected: 90 minutes a week, not the 6 hours the vendor pitch suggested.

**Firm 6. A 14-person firm in Glasgow with a strong cloud-accounting practice.** Sage AI works well for the half of clients on Sage 50. The other half are on Xero, QuickBooks, and FreeAgent, and the firm's had to maintain four parallel workflows. The lesson: if your client base is split across platforms, the in-platform AI features rarely repay the integration overhead. They're now consolidating around Xero JAX plus a tooling layer.

**Firm 7. A 5-partner firm in Birmingham, 26 staff, payroll-heavy.** They tested an AI-assisted payroll query desk. Worked well for top-of-funnel queries (holiday entitlement, tax code questions). Failed badly on edge cases (off-payroll workers, salary sacrifice mid-year). They now use the AI for tier-one only and route anything technical to a human inside 30 seconds. Net hours saved: 4 to 5 a week across the payroll team.

**Firm 8. A 2-partner firm in Edinburgh.** Took on a fixed-fee bookkeeping client base and tried to wire AI as the primary processor. The client documents were too messy. They've rolled back to humans-first, AI-second on capture and have stopped advertising the "AI bookkeeping" line. Lesson learned, with five months of revenue impact.

**Firm 9. A 7-person specialist VAT firm in Cardiff.** AI helped them draft client-facing VAT explainers and partially automated the review of EC sales lists. But the partner's been adamant that no AI output reaches a client without partner sign-off, which means the throughput gain is real but capped. They're currently saving roughly 8% of partner hours, not the 30% they hoped for.

### Group C: net cost, not net gain (3 firms)

**Firm 10. A 12-partner firm in Leeds.** Bought a Wolters Kluwer AI add-on at £680 a month plus an AccountsIQ migration. Eighteen months in, the AI features go essentially unused. The story's one I've heard four or five times this year: the firm bought the platform for non-AI reasons and treated the AI module as a free bonus. It then cost a project manager 80 hours of meeting time over six months trying to roll it out to staff who didn't have time to test it.

**Firm 11. A 4-person practice in Reading.** Bought a generic AI-assisted bookkeeping product for £400 a month. The product needed clean source data to work and the firm's clients sent paper receipts and PDFs of bank statements. They cancelled after four months at a sunk cost of about £1,800.

**Firm 12. A solo practitioner in York.** Subscribed to four overlapping AI tools (Dext, ChatGPT, Xero JAX, plus a CGT calculator with AI explainers) for a combined £170 a month. Time saved: 90 minutes a week, on her own estimate. Net positive but only just, and only because she'd ditched two of the four after three months.

## What the pattern says about the 60% stat

Sage's marketing line is that AI reduces manual data entry by up to 60%. In the Leeds bureau (Firm 3) and the Newcastle high-street firm (Firm 2), capture-and-coding work fell by 65 to 70%. In the Cardiff VAT specialists (Firm 9), it fell by 11%. And in the Bristol tax practice (Firm 5), it fell by zero, because their work was never about line-item capture in the first place.

The 60% number's real for the workflow it describes (high-volume invoice-line capture in a clean digital environment). Outside that workflow, the data-entry frame is the wrong one. The workflows that actually moved hours across all 12 firms had nothing to do with data entry. They were about getting a first draft (of an audit working paper, a client email, a variance commentary, a payroll-query response) and a senior reviewing it. That marks a different shape of automation, and it travels well across firm types.

## The advisory question, and what the SRA Garfield AI decision means

In April 2026, the Solicitors Regulation Authority authorised the first AI-only law firm in the United Kingdom (Garfield AI). It doesn't directly apply to accounting bodies (the ICAEW, ACCA, and CIOB regulate UK accountants), but the question it raised is now landing in accounting client conversations. If AI does the work, what's the human accountant charging for, and what's the firm liable for if the AI gets it wrong?

The honest answer in my notes from the past three months is this. Clients of mid-tier UK firms have started asking, in the same week, two questions. One: are you using AI on my file, and if so, am I getting the saving? Two: who's responsible if the AI is wrong? The firms answering well are the ones who can describe in plain English where the AI is in their workflow, who reviews the output, and what the partner-level liability looks like. The firms answering badly are the ones who quietly bolted AI in and hoped no one would ask.

The advisory side of accounting (forecasting, scenario modelling, board-pack preparation, succession planning) is the work clients value most and the work AI's least reliable at without heavy human scaffolding. I haven't yet seen an off-the-shelf AI tool produce a credible advisory deliverable for a UK SMB client without a partner spending two hours editing it. Tax narrative's similar. So the firms charging on advisory work aren't seeing big AI savings in 2026. They're seeing modest ones, and they're still the ones doing the thinking.

## The four-step audit any UK firm can run this Friday

The four-step audit below is what I run with clients on the AI Roadmap engagement. You can run it yourself.

**Step 1. List your top five workflows by partner-and-senior hours.** Year-end accounts, audit, monthly bookkeeping, tax compliance, payroll, advisory. Pick the five that consume the most time you're billing or absorbing. For most firms it takes 30 minutes with the practice manager.

**Step 2. For each, ask three questions.** Where does the work begin (a client email, a Dext queue, a software export)? Where does it end (a filing, a client meeting, a board pack)? Which 20% of the workflow takes 80% of the senior-or-partner time? You're looking for the bottleneck, not the whole pipeline.

**Step 3. Test one AI tool against the bottleneck for two weeks.** Just one. Don't buy a five-tool stack on day one. If the bottleneck is first-draft client emails, give a private-data ChatGPT a shot. If it's invoice capture, try Dext for two weeks. If it's variance commentary, prototype a documentation reviewer. The two-week pilot should produce a clear answer: did senior time on this workflow drop by at least 20%?

**Step 4. Decide.** Keep, kill, or scope a bigger build. Most firms find that two of the five candidate workflows justify ongoing investment and three don't. So expect a 40-ish percent hit rate, which is fine. The mistake is rolling out tooling firm-wide before any of the four steps have been done honestly.

## Where the savings show up on the P&L

Across the 12 firms, the median time saving in year one was 6.5 hours per fee-earner per week, and the median P&L impact was 4 to 7% of fee-earner cost (lower than the saved hours suggest, because some of the saving went to client-facing work that was previously skipped). One firm captured nearly all of it as margin. Three firms captured none, because no one redesigned the throughput target. And that's the single biggest variable I see. If a senior gets four hours back a week and the manager doesn't adjust the work-in-progress target, the four hours quietly disappear into the calendar.

## What I'd do tomorrow if I were running a UK accounting firm

Three actions, in priority order.

First, do the four-step audit above. Pick one workflow, give it two weeks, and use one tool only. The risk of that exercise is one fee-earner's distraction for half a day, which is small.

Second, set up a private-data ChatGPT or Claude workspace and load your own playbooks, training memos, and the last three years of client-facing explanations. It's the single highest-return investment I've seen across the 12 firms. The cost is a few hundred pounds in setup and £20 to £50 per seat per month, and the payback rarely runs more than two months.

Third, write a one-page AI-and-client-fees note. Where AI is in your work, who reviews it, what the partner stays liable for, and how it affects (or doesn't affect) your fee. Send it to your top 20 clients before they ask. The firms doing this are the ones being told by clients they're easier to work with, which translates into retention rather than discount pressure.

## FAQ

**Is Sage AI's 60% data-entry reduction figure real?**

For the workflow it describes (invoice-line capture in a clean digital pipeline), yes. Across two of the 12 firms tested, capture work fell 65 to 70%. Outside that workflow, the figure doesn't apply. Most accounting workflows aren't about data entry, and the 60% number doesn't generalise.

**Should a UK accounting firm of 5 to 15 staff buy Iplicit or AccountsIQ AI?**

Probably not, unless the firm has multi-entity clients with consolidation needs. Both are mid-market tools priced for groups, not for high-street practices. A five-person firm will see better return from Dext plus a private-data ChatGPT than from a £150 to £250 per user platform.

**What's the single highest-return AI investment for a small firm right now?**

A private-data ChatGPT or Claude workspace with the firm's own playbooks and templates loaded in. Setup's a few hundred pounds, monthly cost's roughly £25 per seat, and the payback period across the firms in this piece was eight to ten weeks.

**How does the Garfield AI / SRA decision affect accountants?**

Not directly. The ICAEW and ACCA, not the SRA, regulate UK accountants. Indirectly, it changes client expectations. Accounting clients are now asking the same two questions solicitors' clients are: are you using AI, and who's liable if it's wrong? Firms with a clear written answer are doing better in client conversations than firms without one.

**What's the single most common AI mistake UK firms made in 2025?**

Buying the AI feature bundled with their practice management system without any plan to roll it out. In four of the 12 firms, that purchase consumed 60+ hours of project-manager time and produced no measurable benefit. The fix is simple. Don't buy AI as a bonus to a non-AI purchase. Buy AI when you've got a workflow that needs it.

## Where to take this next

If you want a structured way to run the four-step audit, the AI Roadmap audit is the fastest path. We map your firm's top workflows, identify where AI actually saves senior hours, and walk you through the build-or-buy call for each. https://richardbatt.co.uk/roadmap

If you would rather start with templates, the AI Ops Vault has the prompt library and the workflow checklists I use with these firms, including the private-data ChatGPT setup, the audit working-paper reviewer, and the client-fees note template referenced above. https://richardbatt.co.uk/vault

The most useful thing I can tell any UK accountant about AI in 2026 is that the firms doing well are the ones that picked one workflow, ran a two-week test, and did the boring redesign of the senior's time afterwards. The vendor pitches are louder. The practitioner picture is quieter and works.

---

## More about Richard Batt

Richard Batt is an AI implementation specialist who helps businesses deploy working AI automation in days, not months. 120+ projects across 15+ industries.

### Key pages

- [Home](https://richardbatt.com/)
- [About Richard](https://richardbatt.com/about)
- [Blog](https://richardbatt.com/blog)
- [Contact](https://richardbatt.com/contact)
- [Subscribe](https://richardbatt.com/subscribe)

### Contact

- Email: richard@richardbatt.com
- Location: Middlesbrough, UK (working globally)
- Website: https://richardbatt.com