Referrer briefing · Insolvency Direct · Insolnet

AI and the small general practice accountant, November 2026

What has actually changed in bookkeeping, accounts, tax and client work, and what it is doing to fees, people and the value of a practice.

Executive summary

Almost every small UK practice has now tried AI. Very few have rebuilt a workflow around it. ICAEW's August 2026 survey of firms with fewer than eleven principals found that only 32% of sole practitioners and 43% of small firms consider themselves early adopters, and AccountingWEB's 2026 member survey found just 7% reporting a transformational effect. The vendors say otherwise, but the vendors are selling. The measurable change is narrower and more interesting than the headlines: the review step has moved rather than disappeared, the real workload shock of 2026 was Making Tax Digital rather than AI, and the client who arrives with a ChatGPT answer is now the single most reported day-to-day consequence.

1. The adoption picture: tried by nearly everyone, transforming almost nobody

Two sets of numbers describe the same profession and appear to contradict each other. Xero's commissioned survey of 515 UK practices in August 2025 found that 98% use AI "in some way". ICAEW's own research among its smallest member firms, with fieldwork in August 2026, found responses "weighted toward the late majority and laggards". Both are true. One measures whether anybody in the office has opened a chatbot. The other measures whether the firm thinks of itself as having changed how it works.

32%sole practitionersself-described AI adopters, ICAEW Aug 2026
7%"transformational"AccountingWEB / Sage 2026, 600+ members
29%micro businessesusing any AI, ONS BICS Oct 2026
44%UK accountantsuse AI in their role, ACCA 2026

The honest reading for a small practice is this. Roughly seven in ten practitioners use ChatGPT, Gemini or Copilot in some form. Two-thirds believe their core accounting software already contains AI, and they are right. Around a third would call themselves adopters. Structured deployment, meaning an approved tool list, a written policy and some training, is rare. AccountingWEB's analysis of this year's Accounting Excellence entries put AI mentions at 72% of large-firm entries and 38% of small-firm entries, with small firms describing "ad hoc, task-specific" use focused on immediate time savings: drafting engagement letters, summarising meeting notes, producing a first draft of advice.

The official statistics say the same thing about the clients. The ONS Business Insights survey for late September 2026 found 30% of all UK businesses using at least one AI technology, up from 9% three years earlier, but only 29% of businesses with fewer than ten employees. The Department for Science, Innovation and Technology's 2026 adoption survey found that 47% of businesses spent nothing at all on AI in 2025, and that limited skills were the top barrier for 67% of those who had adopted it. ACCA's 2026 talent survey adds a telling pair of figures: UK accountants receiving AI training from their employer nearly doubled in a year, from 23% to 41%, and the proportion concerned about AI's effect on their role rose from 30% to 49% over the same period. Adoption and anxiety are rising together.

Almost everyone has tried a chatbot. Roughly a third call themselves adopters. Almost nobody has rebuilt a workflow.

2. Bookkeeping: the review step has moved, not disappeared

This is where embedded AI has gone furthest, and where the vendor claims are loudest. Xero reports more than 100 million transactions auto-reconciled since launch and says practices using its Auto Bank Reconciliation save about half their monthly reconciliation time. Dext claims extraction accuracy above 99% on receipts and invoices. Accountancy Age reported in September 2026 that practices are cutting routine bookkeeping preparation time by 30% to 40%, and quoted an unnamed 45-staff regional firm that took quarterly VAT preparation from about two days to about three hours per file after connecting bank feeds to receipt matching.

None of those figures is independently measured, and no UK survey has yet found what share of a typical client's transactions are coded without a human touching them. Blog-level estimates run at 95% to 99% for familiar counterparties and 80% to 90% for new or ambiguous ones. The practical point is that a 97% match rate on a client doing 300 transactions a month still leaves nine items a month that need a decision, and the items that need a decision are the ones that matter: a first-time supplier, a receipt that is half personal, a VAT treatment under partial exemption, a round-sum transfer to the director. The vendors know this. Dext publishes nothing to the ledger without review, and Xero routes exceptions to a human. The review burden has shifted from keying to exception-checking. It has not gone.

The one AI bookkeeping category that has genuinely moved is the market itself. Botkeeper, the best-funded US AI bookkeeping start-up, closed in February 2026 after eleven years. Its function has been absorbed by the ledgers and capture tools firms already pay for: Dext AI Assist (March 2026), FreeAgent's "Jack" bookkeeper built with Jenesys, Xero's Smart Document Capture (July 2026) and Sage's AutoEntry. A small firm no longer needs to buy AI bookkeeping. It arrives with the subscription, and the subscription is going up (see section 6).

3. Accounts preparation: assistive checking, not AI-drafted accounts

For a practice on TaxCalc, Capium or IRIS, "AI in accounts production" in November 2026 means anomaly flags and document extraction, not AI-authored FRS 102 or FRS 105 disclosure notes. TaxCalc's summer 2026 release contains no AI feature at all; it is rules-based validation and iXBRL checks, which is what most small firms want from it. Capium describes its AI as "limited". Silverfin offers AI mapping of ledger codes to its standard chart at a claimed 95% to 98% first-pass accuracy, plus a continuous checking assistant, but its templates still produce the notes.

The most concrete UK number is IRIS's: its AI tax anomaly detection, which first shipped in November 2025, has been used by more than 8,000 practices across more than 60,000 returns, flagging HMRC rejection risks and likely future queries. It is free "for now", a phrase worth noting. An AI virtual assistant for IRIS Elements is on the roadmap for April 2027.

The regulator has moved ahead of the software. ICAEW's Code of Ethics 2026, in force from 1 July 2026, adds "automation bias" (the tendency to favour output generated by automated systems) to its examples of bias in professional judgement, and amends the objectivity principle to cover undue reliance on technology. The practical meaning is that a set of accounts signed off from an AI-assisted draft without a sceptical review now carries an explicit ethics hook. No practitioner survey has yet measured the accuracy of AI-drafted accounts or disclosure notes at small firms. That gap is itself a finding.

4. Tax: MTD was the workload story, research was the AI story, and HMRC is the biggest adopter of all

Making Tax Digital for Income Tax

The single largest change to a small practice's tax workflow in 2026 was not generative AI. It was MTD for Income Tax going live on 6 April 2026 for sole traders and landlords with qualifying income above £50,000. HMRC estimated 864,000 taxpayers in scope. By 12 August 2026 more than 570,000 had signed up and more than 436,000 had filed a first quarterly update, with agents responsible for over two-thirds of sign-ups. From September 2026 HMRC began signing up the remainder itself, in batches of up to 294,000, without notifying the agent.

The first deadline on 7 August 2026 was, in AccountingWEB's words, reached "in chaos". A five-week filing window, clients signed up for the wrong year, and successful submission receipts that HMRC later treated as incomplete because trade and property income must be filed as separate quarterly updates. One Nottingham firm, Page Kirk, found seven clients with "outstanding" first-quarter updates despite holding success receipts for all of them. No penalty points apply in 2026/27. The threshold drops to £30,000 in April 2027 and £20,000 in April 2028.

Every major vendor has positioned an "MTD agent" as the answer: Sage announced the UK's first in September 2025, Xero, IRIS and Dext followed. They chase records by email and WhatsApp, draft quarterly updates for review and flag anomalies. There is no independent evidence yet of their effect on per-client time, and the first two quarters' problems were HMRC data-structure problems that a chasing agent does not solve. What MTD has unquestionably done is force a quarterly cadence on the smallest clients, which is the real driver behind the shift to monthly retainers discussed in section 7.

Tax research is the one paid AI category with hard adoption numbers

Tolley's February 2026 survey of more than 400 UK tax professionals found 84% using AI for tax research, with tax-specific tools (53%) now ahead of general chatbots (31%). Three-quarters were concerned about fabricated output, and 71% said they were more confident with AI grounded in authoritative tax sources. That concern has shaped the products. Tolley+ with Protégé gives citation-backed answers from Simon's and De Voil; Croner Intelligence, launched in June 2026, is grounded in Croner's editorial content and is explicitly positioned to set out the rules and decisive facts so the accountant reaches the conclusion, not to recommend a filing position. At the smaller end, Regulus, a constrained model for UK tax queries built by a Worcester sole practitioner, and FD Intelligence's TIA assistant trained on HMRC manuals, both appeared in 2025 and 2026.

HMRC published its own guidelines for generative AI in tax software on 28 January 2026. Tools should draw on current legislation and HMRC guidance rather than forums, should "support, not replace, human judgment", should flag nuanced areas and point the user to a qualified adviser, and should remind users that "it is their responsibility to make sure their tax returns are accurate". HMRC endorses no product.

HMRC's own AI

HMRC is now the largest AI deployment any small practice deals with. Its July 2026 roadmap update reports more than 28,000 staff using Microsoft Copilot daily, scaling to 50,000 during 2026; 6.3 million interactions handled by the "Ask HMRC" digital assistant in 2025/26; 78% of customer interactions already digital or automated, with a target of at least 90% by 2030; and AI-flagged returns feeding caseworkers who "make the final decisions". In September 2026 HMRC signed a 15-year contract with Salesforce worth up to £2bn for an AI-enabled CRM, alongside a £175m deal with Quantexa for real-time fraud risk scoring. It is recruiting 5,500 additional compliance officers by 2029/30, and tax adviser registration became mandatory in May 2026 with new powers against advisers who facilitate non-compliance from 1 April 2026.

Two consequences follow for small firms. First, the Connect data-matching system now starts "the vast majority" of enquiries, so the random HMRC enquiry is effectively a thing of the past; letters arrive because something in the data did not match, and under MTD that something can now be a quarterly figure rather than an annual one. Second, taxpayers have an official-looking AI source for wrong answers in addition to ChatGPT: GOV.UK Chat went live on 15 May 2026 and within hours Dan Neidle had it telling a £99,000 earner that a £1,000 pay rise would not affect their tax-free childcare, omitting the £100,000 cliff edge entirely.

5. Client communications: the client with a ChatGPT answer

Drafting client correspondence is the most-cited AI use in UK practices, and meeting transcription is the most widely adopted third-party AI tool category. Neither has caused much trouble. The phenomenon practitioners actually talk about is the other direction of travel: clients using AI themselves and bringing the results to their accountant.

Dext's December 2025 survey of 500 UK accountants and bookkeepers is the only quantified UK evidence, and it is vendor-commissioned, but the pattern it describes is corroborated by named practitioners in AAT Comment, Accountancy Age and AccountingWEB. Ravical's separate survey of 500 UK businesses found two-thirds consult ChatGPT before their accountant, and seven in ten have acted on AI-generated financial advice before speaking to a professional.

What UK accountants reported seeing in 2025Share
More clients using public AI tools for financial, tax or bookkeeping advice77%
Clients using AI output to question or challenge professional advice72%
Clients suggesting AI could replace professional accounting services68%
Encountering client mistakes caused by AI advice at least weekly31% (7% daily)
Know of a business that suffered a direct financial loss from AI advice50%
Most common error: misinterpreting business expenses46%
Second most common: incorrectly claiming or charging VAT41%
Want public AI tools regulated or restricted for tax advice92%

Source: Dext / Censuswide, 500 UK accountants and bookkeepers, fieldwork first two weeks of December 2025.

The anecdotes are specific. A client of Streets in Chorley used Claude to convert PDF bank statements into an Excel cash book; numerous transactions were missing. A HaysMac client connected an AI tool directly to Xero and received a profit and loss account with items misclassified and commentary that was, in the partner's words, invented. Azets has declined work where re-engineering a client's AI output made the job uneconomic. A sole trader asked a chatbot about the VAT threshold, was told not to register, and received a backdated bill with a failure-to-notify penalty. One AccountingWEB reader observed that the questions have changed: "It's no longer 'can I claim for petrol' but 'if I do a share for share exchange and insert a holding company can I move assets up'."

The time cost is real but modest: 93% of those encountering AI errors spend up to ten hours a month fixing them. The commercial problem is that most of it is unbilled. ICB and Dext found 64% of bookkeepers already give at least an hour of unpaid guidance a month, and the ICB's chief executive put the cause plainly: bookkeepers are not charging for advice "because they don't have the confidence". Mark Lee's counterpoint is also worth having: the loudest complaints come from larger firms whose fees are high enough to make clients look for ways round them. His advice to sole practitioners is to tell clients that checking an AI answer often takes longer than researching the question from scratch, and to consider whether a client who runs every piece of advice through a chatbot is one worth keeping.

What clients want, when asked directly, is less dramatic than practitioners fear. AAT Comment's February 2026 round-up quoted Switchfoot Accounting: "Clients haven't requested we use AI more." Plus Accounting: clients want better outcomes, not more AI. The US client surveys that do exist (there is no UK equivalent) find that around nine in ten want to be told where AI is used in their work, that speed of response matters far more than speed of completion, and that willingness to pay more for faster, more advisory service has risen sharply since 2024. The consistent ask is speed, transparency and a human who puts their name to the numbers.

6. Software as a service: AI bundled "free", prices up, and which apps are actually gaining traction

The incumbents' playbook

Every major platform has followed the same pattern in 2025 and 2026: bundle an AI assistant or "agents" into the existing subscription at no separately stated price, raise the list price, and signal that usage-based AI charging comes next. Xero's chief executive told analysts in May 2026 that AI monetisation, through bundling, à la carte options and usage-linked pricing, would "start rolling out in FY27". Intuit says of its UK agents that "additional terms, conditions, and fees may apply" and that its Accountant Suite, launched free to UK firms in February 2026, will be priced "as determined by market". IRIS's AI features are "currently free and uncapped". Dext's AI Assist had introductory pricing to 31 October 2026. Nobody has said the word "AI" in a price-rise notice; Xero cites "a wave of new capabilities". UK accountants at Xerocon London in July 2026 drew the connection themselves.

UK list price, ex VAT per month2024/25Sep/Nov 2025Sep/Nov 2026Two-year change
Xero Grow£33£37£39+18%
Xero Comprehensive£47£50£55+17%
Xero Ultimate£59£65£70+19%
QuickBooks Essentials£33£38£40+21%
QuickBooks Plus£47£56£60+28%

Sources: Xero UK pricing updates page (1 September 2025 and 1 September 2026, multi-organisation discount withdrawn from the latter); QuickBooks UK list prices compiled from reseller and independent tester reports (13 November 2025 and 2 November 2026). Sage publishes no UK price-change table; its own forum carries user reports of rises between 20% and 66% in 2026.

Two further moves matter for the practice's cost base. Xero replaced its 15% app-store revenue share with tiered API pricing from March 2026, a cost that app partners feeding the small-firm stack will pass on. And the independent hands-on tests that exist show the shipped AI is narrower than the marketing: Xero's Claude connector exposes five read-only summary functions for one organisation at a time, and QuickBooks' AI chat is tier-gated with an unpublished monthly cap. Roughly 20,000 of Xero's five million subscribers connected the Claude integration in its first 60 days.

ICAEW's small-firm research puts the practitioner's side of this crisply. Twenty-nine per cent of small firms now use five or more software providers. The top enabler of greater AI use is better integration with existing tools, and the second is "clearer or more flexible licence fee structures". Firms are not short of AI features. They are short of features that talk to each other and of pricing they can predict.

Which AI apps are gaining traction

No UK survey publishes AI tool market share. The best available proxies are set out below, with the evidence graded.

CategoryToolsEvidence of small-firm traction
General chatbotsChatGPT, Microsoft Copilot, Gemini, ClaudeStrong. Around seven in ten practitioners use one; ChatGPT and Copilot together account for 55% of reported AI use among UK finance professionals. Mostly individual accounts, not firm-wide licences.
Meeting transcriptionTeams, Otter, Fireflies, FathomStrong. The most widely adopted third-party category in AccountingWEB's awards data. Some clients dislike being recorded.
Tax researchTolley+ with Protégé, Croner Intelligence, TIA, RegulusStrong among tax specialists: 53% use a tax-specific AI tool. The only paid AI-native category with hard numbers.
Embedded ledger AIXero JAX, Sage Copilot, Intuit agents, FreeAgent JackPassive but universal. Arrives with the subscription; 2.6m Xero customers used an AI feature in the year to March 2026.
Capture and anomaly detectionDext AI Assist, IRIS Elements AI, Silverfin AssistantGrowing. IRIS: 8,000+ practices, 60,000+ returns. Dext line-item extraction use more than doubled in a year.
Practice managementKarbon Kai, Ignition AI proposals, Intuit Accountant Suite, Xero Partner HubEarly. Karbon prices in US dollars at $59 to $99 per user; the incumbents' free bundles are squeezing the paid tools.
Agent buildersXero Force, Copilot Studio, Zapier, Make, n8nNiche. Used by a "builder" minority or via outside consultants.
AI receptionists, AI bookkeeping start-upsVariousWeak. Vendor anecdotes only; practitioner scepticism on the forums; Botkeeper closed.
Audit AICaseware, MindBridge, Inflo, CircitMid-tier and above. Irrelevant below ten principals except through Caseware's existing footprint. FRC guidance issued 30 March 2026.

7. Fees: pressure without measured deflation

No UK dataset shows average compliance fees falling in 2025 or 2026. The measured direction is still up. TaxCalc's autumn 2025 survey of 215 practitioners found 47% expecting to raise prices for MTD clients and 48% saying lack of capacity was already holding back growth. Xero's May 2026 survey found the highest-margin firms more than twice as likely to plan a price rise above 20%. ICAEW's mid-tier research found "almost all" firms growing fees in their latest year. Published fixed-fee packages for a simple limited company have not moved visibly: the volume online firms charge £40 to £135 a month plus VAT, and a VAT-registered one-director company lands at around £120 to £150 a month. In real terms, after four years of inflation, a static nominal fee is a cut. Nobody has measured that either.

What has deteriorated is the expectation. Ravical's February 2026 survey of 500 UK firm leaders found 63% expecting compliance work to be "only marginally profitable or worse" within five years, around 40% saying technology is already reducing billable hours and 37% citing pricing pressure. The mechanism practitioners describe is not clients demanding discounts. It is unpaid work expanding: longer AI-written client emails to read, "run this past you" questions, and AI-generated bookkeeping that doubles clean-up time on a VAT return. For an hourly-billing firm that is billable, if it dares. For a fixed-fee firm it is scope creep absorbed as margin.

AI has not cut the fee. It has cut the time, and left the fee for the firm to defend.

The pricing-model debate has been loud and the data thin. Every consultant and vendor on the Accountex stage in May 2026 said the same thing: AI "removes the link between time and value", so time-based billing is finished. The one broad 2026 survey, from BILL, found 60% of firms still relying heavily on hourly billing, 68% of leaders ruling out or undecided on new models, and AI the direct catalyst for only 20% of those who had switched. One North West mid-tier firm cut direct processing hours on bookkeeping engagements by 40% and saw revenue on those engagements fall, because the fees were still hourly. The lesson is to reprice before automating, not after.

The quieter truth is that MTD, not AI, is doing the repricing. A quarterly filing cycle does not fit an annual fixed fee. TaxCalc, Sage and the practice consultants all now recommend converting annual fees to monthly retainers, separately priced quarterly submissions, and tiered packages. That shift will survive whatever the AI vendors do next.

On advisory, the vendors disagree with each other. Xero says three in five UK practices direct AI time savings to advisory work; Ravical says only 6% would use freed time to generate new advisory revenue, and half say the barrier is a skills mismatch rather than time. The two are compatible if most firms are using the hours for capacity relief and MTD absorption rather than selling new engagements. Measured advisory revenue attributable to AI at small firms is, as of today, undocumented.

8. Inside the firm: how practices are using AI, and the rulebook that arrived in 2026

Ad hoc use, not built systems

The AccountingWEB and Sage 2026 survey gives the shape of day-to-day use: automating repetitive tasks and data entry (44%), document OCR (39%), transcribing calls and meetings (32%), automated insights and reporting (31%), client chatbots (31%), generating or summarising documents with a language model (29%), and none at all (21%). Small firms, in the awards data, prioritise engagement letters, meeting summaries, first-draft advice and fee proposals. Mid-tier and large firms have governance frameworks, technology working groups and gated pilots. AccountingWEB's conclusion was that "governance, not enthusiasm, is becoming the real point of difference".

Named small firms that have actually built something are scarce, and the accounts are vendor-authored. Hartley Accounts, a six-person practice in Yorkshire with around 140 clients, had an automation consultancy build three agents connected to its email and Xero: email triage with drafted replies, monthly management report drafting, and WhatsApp onboarding with 48-hour chasers. The reported results after 60 days were email triage down from 14 hours a week to three, monthly report preparation from 46 hours to six, onboarding from 12 days to five, for a total cost of £3,200 and £42 a month in running costs. Treat it as an upper bound until a firm publishes its own figures. Consultant pricing for a five to twenty staff firm runs £3,000 to £15,000 per project. PKF Littlejohn built Copilot Studio agents in-house after a four-week engagement and made Copilot its only approved AI tool. Clever Accounts spent eighteen months building an in-house AI platform and won an award for it in September 2026, without publishing a single metric.

The hiring consequence is beginning to show. Xero's 2026 survey found 63% of firms changing their hiring criteria, with technology fluency and relationship skills now ranking above traditional accounting skills, and the highest-margin firms twice as likely to hire data analysts and tax technologists. For a typical small practice the "technologist" is a part-time process owner or an outside consultancy, not a hire. Average AI spend per practice, at £1,746 a year on Xero's figures, is tiny against the 5% to 10% of turnover that most firms now spend on technology overall.

The rulebook

Three things happened in 2026 that every small firm should know about.

On 19 January 2026 the seven PCRT bodies (AAT, ACCA, ATT, CIOT, ICAEW, ICAS and STEP) issued joint guidance on the ethical use of AI tools in tax work. Its key lines are worth quoting. "The output from an AI tool should also be regarded as if it were prepared by a less experienced junior colleague." "Members remain ultimately accountable for any work produced, regardless of whether AI has been involved in producing the work." And: "The input of client data into publicly available AI tools is likely to constitute a breach of client confidentiality" unless the client has consented. It recommends an engagement-letter statement that AI-enabled software may be used, anonymised data in any public model, and an audit trail of prompts. On 9 November 2026 ICAEW's regulatory team went further, saying that entering client data into a public AI tool "can be akin to posting it publicly", and that firms are expected to have written, regularly reviewed AI policies and to train staff on them.

On 8 January 2026 the First-tier Tribunal in Elden v HMRC [2026] UKFTT 41 (TC) dealt with a skeleton argument whose case summaries had been produced with AI and not checked: one case name fabricated, three real cases with cited passages that bore no relation to the judgments. The judge said whether AI was used was "not directly relevant"; the failure was the lack of verification, which the tribunal characterised as professionally incompetent. Future submissions in that matter must state whether AI was used.

Professional indemnity has not yet caught up. UK broker commentary in 2026 says standard PI wordings do not exclude AI-assisted work as a class, but underwriters now ask at renewal which tools a firm uses, in which workflows, and what oversight applies. Where exclusions are appearing, they target advice "generated by or based primarily on AI output without human professional review". The minimum defensible package for a sole practitioner is therefore short: a written AI policy with a tool list, no client-identifiable data in consumer tools, an AI clause in the engagement letter, a record of prompts and outputs, and a documented human review. Globally, only 21% of firms have an AI policy at all.

9. The official numbers in one place

SourceDateWhat it says
ICAEW Sole Practitioners and Small Firms in Focus6 Nov 2026 (fieldwork Aug 2026)32% of sole practitioners and 43% of small firms self-describe as AI early adopters or ahead; internal knowledge management the most common use; integration, training and data security the enablers. 53% of sole practitioners considering or pursuing exit.
ICAEW mid-tier research (35 firms)22 May 202686% have a technology strategy that includes AI; 68% expect AI to reduce early-career roles; 17% confident in assessing workforce impact.
ACCA Global Talent Trends (UK cut)Sep 202644% use AI in their role; 41% get AI training (23% in 2025); 49% concerned about AI's effect on their role (30% in 2025).
PCRT AI topical guidance, seven bodies19 Jan 2026Treat output as a junior's work; member remains accountable; public tools and client data likely a confidentiality breach.
ICAEW Code of Ethics 20261 Jul 2026Automation bias and reliance on technology added to objectivity and professional judgement provisions.
FRC Key Facts and Trends 20265 Oct 2026 (2025 data)UK and ROI members 418,986 (+2.5%); students 156,367 (+0.9%); registered audit firms 3,421 (down 9% in a year).
FRC AI in audit guidance26 Jun 2025; 30 Mar 2026First guidance and thematic review of the six largest firms, then generative and agentic AI guidance. Firms "continue to hold regulatory accountability". Nothing scaled to small firms.
HMRC Transformation Roadmap update2 Jul 202628,000+ staff on Copilot, 50,000 target; 6.3m Ask HMRC interactions; 78% digital, 90% by 2030; 5,500 extra compliance officers by 2029/30.
HMRC generative AI software guidelines28 Jan 2026Support not replace human judgment; reliable sources only; user responsible for accuracy; HMRC endorses nothing.
HMRC MTD for Income Tax12 Aug 2026864,000 in scope; 570,000+ signed up; 436,000+ first quarterly updates filed; over two-thirds of sign-ups by agents.
ONS BICS wave 1658 Oct 202630% of businesses use AI; 29% of those under ten employees; 51% of those with 250+; up from 9% in Sep 2023.
DSIT AI Adoption Survey 202629 Oct 202632% adoption, double 2025; 67% of adopters cite limited skills; 47% spent nothing on AI in 2025.
Companies House transition plan23 Oct 2026; 18 Nov 202688% of companies compliant with identity verification. Filer verification and mandatory ACSP registration "no earlier than November 2027". Software-only accounts filing, loss of abridged accounts and small-company P&L filing from April 2028.
ICO agentic AI8 Oct to 20 Nov 2026Call for evidence; guidance due spring 2027; organisations "remain responsible for data protection compliance" for AI they deploy.

10. People: trainees, the newly qualified and the owners

The disappearing first rung is a Big Four story

The numbers that made the headlines are real: the Big Four's combined UK graduate and apprentice intake fell from 6,500 in 2023 to 5,400 in 2025, with KPMG down from around 1,400 to around 1,000, and AI cited alongside the economy and offshoring. Those intakes have since stabilised; EY and Deloitte each took around 1,500 to 1,600 in 2026. Below the Big Four the direction is reversed. Grant Thornton planned a 30% rise to 340 trainees for 2026, Azets took 384 in 2025 (up 15%) and Cooper Parry 140-plus (up 40%). Graduate competition is nonetheless ferocious: Azets received more than 10,500 applications for just over 200 graduate roles, and the Institute of Student Employers recorded 140 applications per graduate vacancy across all sectors, the highest in thirty years.

The profession's pipeline is not shrinking. FRC data for 2025 shows UK and Irish student numbers up 0.9% to 156,367, with ICAEW up 2.5%, ACCA up 1.8% and AAT up 17%, and the number of students becoming members up 4.9%. ICAEW's intake of 10,750 new ACA students in 2025 is about 10% below its 2022 peak, but it is not collapsing. The mid-tier expectation, from ICAEW's May 2026 research, is that AI reduces the need for some early-career roles (68% of firms) while total headcount does not fall (83%), and that the mix shifts from graduates to school leavers.

At the level of the small practice, there is no survey evidence of firms hiring fewer trainees because of AI. The evidence runs the other way. Advancetrack's 2026 survey of around 500 practice leaders found 73% turning away work because of staff shortages, and 45% saying the shortage is worse than three years ago. The bigger shock to small-firm training economics is a policy decision: from 1 January 2026 the Department for Education funds new Level 7 apprenticeships only for starters aged 16 to 21. The Accountancy or Taxation Professional standard accounted for 9,200 starts in 2023/24, 39% of all Level 7 starts, and only around 11% of Level 7 starters were under 22. The pull-forward was spectacular: Level 7 starts in December 2025 were up 845% on the prior year, and January 2026 starts were down 93%. A small practice can no longer levy-fund an ACA or ACCA graduate aged 22 or over, which pushes it towards the AAT school-leaver route, an offshore team, or AI tooling for the work that trainee used to do. No UK data yet shows which of the three small firms are choosing.

+0.9%UK/ROI studentsFRC, 2025 vs 2024
5,400Big Four intake2025, from 6,500 in 2023
73%firms turning work awaystaff shortage, Advancetrack 2026
-93%Level 7 startsJan 2026 vs Jan 2025

The newly qualified

The squeeze is at entry level, not at qualification. Recruiter guides for 2026 put a newly qualified ACA in regional practice at £35,000 to £51,000, typically around £43,500, and £48,000 to £70,000 in commerce or London. Robert Half and Hays report employers paying a premium for data and AI capability, assessed as "appetite to learn, adaptability and comfort with technology", though no UK percentage has been published. Trainees quoted by ICAEW in June 2026 describe AI removing data entry and moving them to analysis sooner; the risk ICAEW flags is juniors being asked to supervise AI output before they have done the underlying work themselves. ICAEW's small-firm survey found 39% expecting the profession to become less attractive because AI is seen as replacing accountancy. That is a recruitment problem small firms will feel before the Big Four do.

Owners: retiring on schedule, for the usual reasons

ICAEW's November 2026 research is the best data the UK has on exits. Fifty-three per cent of sole practitioners and 33% of small firms are considering or actively pursuing a sale or other exit. Among those exiting, 85% cite retirement and 44% cite the complexity and volume of regulation; a third cite both. AI is not cited as a driver. The population of UK accountancy enterprises fell 8.6% between 2019 and 2025, from 43,720 to 39,975, with each surviving firm now serving 68 businesses against 62 before. The average UK accountant is 46.

Two things did pull sales forward in 2025 and 2026, and neither was AI. Business Asset Disposal Relief rose from 10% to 14% in April 2025 and to 18% from April 2026, and the private equity bid for practices peaked in 2024 and 2025. The evidence on the other side of the ledger is that owners are not rushing to sell into consolidation: 82% of firms with five or more principals have been approached by an investor, yet only 5% of sole practitioners and small firms agree that private equity benefits the sector and 53% disagree. The more plausible reading is that a sole practitioner who can run a smaller team with better tooling has one more reason to stay independent, and one fewer reason to find a successor. No UK survey has yet asked whether AI has brought retirement forward or pushed it back.

11. Practice values: GRF multiples hold at the bottom, AI bites at the top

The small-practice market still prices in gross recurring fees. UK brokers and ACCA's exit guidance in 2026 quote an anchor of about 0.8x GRF for sub-£300,000 fee banks, with a range of roughly 0.6x to 1.2x set by client mix, concentration, systems and owner dependency. Converted to earnings, the same deals sit around 4x to 7x EBITDA for small regional practices against 10x to 14x or more for prime assets. The standard structure is 50% on completion and 50% after 24 months for a small practice, or three equal instalments over 36 months for a larger one, with deferred consideration tied to client retention. An open-ended clawback window remains the red flag.

What moves a small-practice multiple in 2026UpDown
SystemsCloud ledgers, documented processesDesktop software, undocumented manual work
BillingMonthly direct debit, fixed feesAnnual billing in arrears
ClientsRecurring, spread, growing, youngerConcentrated, winding down, banking with the owner personally
OwnerLow dependency, team stays, owner stays onRelationships tied to the retiring principal
ComplianceEngagement letters, AML, working papers in orderLooming losses, regulatory gaps

Notice what that checklist is. Cloud, documented process, recurring monthly billing, low owner dependency: these are exactly the attributes an AI-ready practice has and an owner-dependent, desktop-based sole practice lacks. No transaction dataset shows an "AI discount" or "AI premium" on sub-£1m GRF multiples, and no UK broker has published a 2025 or 2026 multiples report. The apparent drift in broker anchors from around 1.0x to around 0.8x is suggestive of softening, but it rests on marketing pages, not completions. MarktoMarket's cross-sector data shows deals under £2.5m enterprise value averaging 3.9x EBITDA against 10x for mid-cap, a 61% discount for being small, which is where most of the pricing gap actually comes from.

Where AI has visibly moved value is at the top. In May 2026 Penta Capital paused the planned £1bn sale of Sumer, the twelfth-largest UK firm, with a source quoted as saying buyers "don't know what cashflows in three years are going to look like because you don't know the impact AI is going to have". Exponent shelved its £1bn-plus sale of Xeinadin, 122 firms and 80,000 SME clients, after bids missed an ask of 15x to 17x EBITDA. The consolidators have started consolidating each other: Affinia and UHY Hacker Young combined in a £100m merger in February 2026. The UK accounts for more than 40% of European private equity accountancy deals, and around 20 to 25 of the top 60 firms are now PE-backed, but the arbitrage that funded aggressive bolt-on pricing, buying at 4x to 7x and hoping to exit at 15x, has narrowed.

For the owner of a £1m to £5m practice, the consolidators are still buying: Xeinadin, DJH and Sumer all completed bolt-ons of that size through 2026, and Draper Hinks expects "AI-enhanced practices" to feature more among buyers in the next twelve months, with PE capital funding automation, cloud, offshoring and centralised back office after the deal. AI appears in the thesis as a post-acquisition margin tool, not as a reason to pay more. Below £1m the market is what it always was, local trade buyers at a GRF multiple, and Draper Hinks notes it has been shrinking by about 3% a year. The first explicit AI-arbitrage buyer of retiring owners' practices, Numica, has raised $21m, bought seven practices, automated about 75% of their workflow in year one and lifted margin at its first acquisition from 37% to 52%. It is in New York. It is not yet here. It will be.

12. What this means for referrers, and for us

Insolvency Direct works with small accountancy practices every week, and three threads in this briefing bear directly on that relationship.

First, the client who acts on AI advice before calling anyone is now the norm, not the exception. Two-thirds of UK businesses consult a chatbot before their accountant. The errors that reach the practice cluster around expenses, VAT and dividends, which is also where we see the overdrawn director's loan account and the dividend that was never covered by profit when a company gets into difficulty. An accountant who is already correcting a client's AI-generated books is looking at the same bank data a liquidator would read. The earlier that conversation happens, the more options the director has.

Second, HMRC's data-matching now starts most enquiries, MTD gives it quarterly visibility of the smallest traders, and it is recruiting 5,500 more compliance officers. Tax arrears that used to surface at year end will surface within months, and the nudge letter will arrive with the data behind it. Practices should expect more clients forwarding HMRC letters they do not understand, and should have a referral route ready for the ones where the arrears are not going to be paid.

Third, the standard the profession has set for itself, "the output of an AI tool should be regarded as if it were prepared by a less experienced junior colleague", is the right one and it is the one we apply to our own tooling. Our bank analysis module uses machine categorisation and AI to reconstruct a director's loan account and trace dividends across two years of statements in minutes. Every finding is then reviewed by a licensed insolvency practitioner before it reaches a director or a creditor. The model finds; the practitioner decides. That is the only version of AI in professional services that survives a tribunal, an insurer or a regulator, and it is the version small practices should insist on from every vendor who tells them otherwise.

Watch list for the next twelve months

  1. AI charging. Xero has said AI monetisation starts in its FY27; Intuit and IRIS have reserved the right. Budget for the "free" features to acquire a price.
  2. MTD threshold to £30,000 in April 2027. A second, larger cohort of sole traders and landlords, with the first penalty points live from 6 April 2027.
  3. Companies House. Mandatory ACSP registration and filer identity verification no earlier than November 2027; software-only accounts, the end of abridged accounts and small-company profit and loss filing from April 2028.
  4. Insurer questions at renewal. Expect a specific AI section on the proposal form. Have the policy, the tool list and the review process written down before you are asked.
  5. ICAEW member survey results. The 2026 survey closes on 27 November and should give the first ICAEW-only usage figure.
  6. DfE full-year apprenticeship data. The 2025/26 release will show whether the Level 7 funding cut has structurally reduced accountancy starts or merely moved them.
  7. ICO agentic AI guidance, spring 2027. The first data-protection rulebook written for tools that act rather than draft.
  8. Consolidator appetite. If PE bolt-on programmes slow as the brokers predict, the £1m to £5m practice loses its premium buyer and the GRF anchor matters again.

Frequently asked questions

How many small UK accountancy firms are actually using AI in 2026?

Almost all have tried it, but structured use is rare. ICAEW's survey of firms with fewer than eleven principals (fieldwork August 2026, published 6 November 2026) found only 32% of sole practitioners and 43% of small firms describing themselves as AI early adopters or ahead. AccountingWEB's 2026 survey of more than 600 members found 7% reporting a transformational effect and 53% minimal or moderate benefit. Vendor surveys reporting 98% adoption measure any use at all, including a single chatbot query.

Is AI reducing accountancy fees for small businesses in the UK?

No UK dataset shows compliance fees falling in 2025 or 2026. TaxCalc's autumn 2025 survey found 47% of firms planning to raise prices for Making Tax Digital clients, and published fixed-fee packages for a simple limited company remain in the £40 to £135 a month range. What has changed is the expectation: 63% of UK firm leaders surveyed by Ravical in February 2026 expect compliance work to be only marginally profitable within five years, and practitioners report growing unpaid work correcting clients' AI-generated bookkeeping and answering questions clients have already put to ChatGPT.

Has Making Tax Digital for Income Tax increased accountants' workload?

Yes. MTD for Income Tax went live on 6 April 2026 for sole traders and landlords with qualifying income above £50,000, with around 864,000 taxpayers in scope. By 12 August 2026 more than 570,000 had signed up and 436,000 had filed a first quarterly update, with agents responsible for over two-thirds of sign-ups. The first deadline on 7 August 2026 was widely reported as chaotic, with a five-week filing window and successful submission receipts later treated by HMRC as incomplete where trade and property income needed separate updates. The threshold falls to £30,000 in April 2027 and £20,000 in April 2028.

What do the professional bodies require when an accountant uses AI?

The seven PCRT bodies (AAT, ACCA, ATT, CIOT, ICAEW, ICAS and STEP) issued joint guidance on 19 January 2026. It says the output of an AI tool should be regarded as if it were prepared by a less experienced junior colleague, that members remain ultimately accountable for any work produced regardless of whether AI was involved, and that entering client data into publicly available AI tools is likely to constitute a breach of client confidentiality unless the client has consented. It recommends an engagement-letter statement that AI-enabled software may be used and an audit trail of prompts. ICAEW's Code of Ethics 2026, in force from 1 July 2026, adds automation bias and reliance on technology to its objectivity provisions.

Are accountancy firms hiring fewer trainees because of AI?

The cuts are concentrated at the Big Four, whose combined UK graduate and apprentice intake fell from 6,500 in 2023 to 5,400 in 2025 before stabilising in 2026. Across the profession the FRC reports UK and Irish student numbers up 0.9% in 2025, with ICAEW up 2.5%, ACCA up 1.8% and AAT up 17%. Small firms report a shortage rather than a surplus: 73% told Advancetrack in 2026 they were turning work away for lack of staff. The larger change to small-firm training economics is the withdrawal of Level 7 apprenticeship funding for starters aged 22 and over from 1 January 2026.

What is a small UK accountancy practice worth in 2026?

UK brokers and ACCA's exit guidance quote an anchor of about 0.8 times gross recurring fees for sub-£300,000 fee banks, within a range of roughly 0.6 to 1.2 times, equivalent to around 4 to 7 times EBITDA for small regional practices. Typical structures pay 50% on completion and 50% after 24 months, or three instalments over 36 months, with deferred consideration tied to client retention. Buyers reward cloud systems, monthly direct-debit billing, documented processes, a spread of recurring clients and low owner dependency. No transaction data yet shows an AI discount at this size; at the top of the market, Sumer and Xeinadin both failed to find buyers at £1 billion in 2026 with AI uncertainty cited.

What should a business do if it has acted on ChatGPT tax advice that turns out to be wrong?

Speak to a qualified adviser before HMRC does. Accountants surveyed by Dext in December 2025 reported that the most common AI-driven client errors were misinterpreted business expenses (46%) and incorrectly claimed or charged VAT (41%), and half knew of a business that had suffered a direct financial loss. HMRC's data-matching now starts most enquiries, and under MTD it sees quarterly figures, so errors surface faster than they used to. Where the resulting arrears cannot be paid, an early conversation with a licensed insolvency practitioner preserves more options for the director than waiting for a petition.