Artificial intelligence (AI) for accountants: how to chase records, hit deadlines and see the numbers, process by process
See how AI can be applied to the real processes of a UK accountancy practice: client questions, deadlines and Making Tax Digital (MTD), onboarding and anti-money-laundering (AML), bookkeeping, VAT and accounts, and the practice's own numbers. A record 11.48 million Self Assessment returns were filed by the 31 January 2026 deadline, and Making Tax Digital for Income Tax went live in April 2026, so the quarterly rhythm only tightens from here. Each process comes with examples and an honest view of the technology.
Client requests: the recurring questions answered correctly, the moment they are asked
Most of what a practice does for a client is production work, but a surprising share of the day goes on the same handful of questions, asked over and over by different people. Which records do you need from me. When is the deadline. How much will I owe. The questions are routine; the rules behind them are not, and 2026 has moved several of them at once.
Making Tax Digital already governs VAT for every VAT-registered business, Making Tax Digital for Income Tax went live on 6 April 2026 for around 780,000 sole traders and landlords with qualifying income over £50,000, and mandatory e-invoicing has been announced for 2029. Each change arrives as a wave of what-do-I-do-now messages from every client it touches, and they tend to land in the same week.
They land, too, on the channels people actually use. A client photographs a letter and sends it on WhatsApp at nine in the evening, or emails a one-line question over the weekend, long after the office has closed. Answer late and the client worries, and the worry is not misplaced: around a million taxpayers missed the 31 January 2026 Self Assessment deadline.
An AI assistant built for your practice can take the recurring ones off your desk. It answers from official HMRC sources and your own firm's notes, gives a dated reply with a link the client can check, and hands anything that needs professional judgement to you. It never issues binding advice in the practice's name; it clears the routine so your time goes where it is actually needed.
The recurring questions answered from official sources, at the moment they are asked
The assistant is connected to your practice's knowledge base, the client's own record and the current HMRC guidance. When a client asks which records are needed, when a deadline falls or roughly how a rule applies to them, it replies in plain English, cites the official source and dates the answer, because the rules change through the year. Where a question turns on judgement, it drafts a reply for you to approve rather than sending its own.
A landlord in Leeds messages on a Sunday to ask whether Making Tax Digital affects the flat they let. The assistant explains the income threshold, links the GOV.UK guidance and notes that a firm position on their figures will come from the accountant. On Monday you approve or adjust, instead of typing the same explanation for the fifth time this month.
The questions that used to interrupt production work all week get a correct, sourced first answer straight away. Your hours go to the judgement calls, not to repeating the same explanation to a dozen clients.
The Making Tax Digital wave handled client by client
As Making Tax Digital for Income Tax reaches each affected client, the assistant produces a short, tailored explanation of what actually changes for them, drawn from their record and the statutory calendar, dated and sourced. It sets out the new quarterly rhythm in general terms and books nothing binding, leaving the personal position for the accountant to confirm.
A self-employed client over the threshold asks what they need to do now that the rules have started. The assistant explains, in general terms, that quarterly updates replace the single annual return, links the policy paper and flags the item for the accountant to tailor to their figures. The client feels answered; nobody on your team has drafted the explanation from scratch again.
The surge of what-do-I-do-now messages that a rule change sets off gets a consistent, accurate first response across the whole client base, instead of a scramble of half-answers written between other jobs.
A correct first response on informal, out-of-hours channels
Clients send short questions and documents on WhatsApp and email, in the evening and at the weekend, not through a formal system. The assistant gives a correct, dated first response on those channels around the clock, recognises which client and which matter it is dealing with, and escalates anything that needs a person, with the full thread attached, so the accountant is never guessing at context.
A client emails at eleven on a Saturday night asking whether their return was filed on time. The assistant confirms from the record that it was submitted and acknowledged, dates the reply and closes the worry there and then, rather than leaving it to fester until Monday morning.
The pressure to reply instantly, at any hour, lifts off the accountant. Clients get a solid answer when the question is on their mind, and the cases that genuinely need judgement arrive with their history already gathered.
Judgement escalated, advice never issued in the firm's name
The assistant is built to know its own edge. Informative, sourced answers from official guidance it will give; a binding view on a client's specific position it will not, routing that to the qualified accountant instead. Every answer carries its source and its date, and anything that could bear on a client's decision is handed over rather than resolved in the chat.
A client asks whether a particular expense is deductible against their rental income. The assistant explains the general rule with a link, then says plainly that the accountant will confirm how it applies to their case, and creates the task. The client is informed; the professional judgement stays where responsibility for it sits.
You get the throughput of an always-on assistant without the risk of an unqualified answer going out under the practice's name. The line between information and advice is designed in from day one, not patched on later.
The routine-question workload is exactly the kind of work an assistant can take on now, and the profession is already moving this way.
- Answering the recurring questions from official sources and your own notes is a buildable system today, not a promise. Making Tax Digital for Income Tax is live from 6 April 2026 for around 780,000 sole traders and landlords over £50,000, the first quarterly update falls due on 7 August 2026, and every affected client generates the same small set of questions. An assistant that replies from the GOV.UK guidance and your practice's standard answers, dated and sourced, can be connected to your channels now.GOV.UK
- The demand behind those questions is measurable. A record 11.48 million Self Assessment returns were filed by the 31 January 2026 deadline, yet around a million taxpayers still missed it, which is precisely the anxiety that produces the did-my-return-go-in questions. Giving each of those a correct, immediate, sourced first answer is well within reach today.GOV.UK / HMRC
- One note of realism on the adoption figures. ACCA reports that 52% of its respondents regularly use AI tools and 82% feel confident learning them, and ICAEW found 83% of its youngest chartered accountants use AI at least weekly. Those come from professional-body member surveys and describe the profession at large, not your practice, so read them as a pointer to where the profession is heading, not a promised result. The numbers that matter are your own: how many recurring questions the assistant answers first time, and how much production time that hands back.ACCA
The assistant answers clients in the practice's name, so these three rules are not optional.
- There is no UK AI Act and no blanket duty to disclose AI use, but the rules that do exist are technology-neutral: the ASA applies its codes however a reply is generated, and a misleading answer is misleading whoever wrote it. So the assistant introduces itself as an AI assistant and never poses as the accountant.ASA
- Informative answers from official sources are one thing; binding tax advice is another. Professional judgement stays with the qualified accountant, who carries personal responsibility under their body's rules, and because the guidance changes often, every answer is sourced and dated.GOV.UK
- Client questions, chat transcripts and the documents attached to them are personal data under the UK GDPR, as amended by the Data (Use and Access) Act 2025. An external chat or AI provider processes them on the practice's behalf, so a processor contract, data minimisation and a retention schedule are part of the implementation.legislation.gov.uk
Onboarding and AML: a clean start that maps the obligations and orders the due diligence
Onboarding is the riskiest moment in the client relationship, precisely because nobody on the team has yet internalised the new client's obligations. Taking on a company means mapping quickly whether it is VAT registered and on which cycle, whether Making Tax Digital for VAT and now for Income Tax apply, what PAYE and payroll run, and when corporation tax and accounts fall due. Miss one in month one and the first deadline is already gone.
It is also a regulated moment. An accountancy practice is a supervised business under the Money Laundering Regulations 2017: before it acts, it must identify the client and any beneficial owner and carry out customer due diligence proportionate to risk, whether simplified, standard or enhanced. The UK National Risk Assessment 2025 rates accountancy high-risk, and a firm is supervised either by its professional body or, by default, by HMRC.
The duty does not end at sign-up. The regulations require a documented risk assessment, ongoing monitoring, record-keeping of the checks and a suspicious-activity report to the National Crime Agency where warranted. And the record-keeping is enforced: HMRC issued 134 penalties to accountancy service providers totalling £513,930 in just six months of 2025.
Most of that is structured, repeatable work, which is where an AI assistant earns its place. It can build the obligations checklist from the client's data and the statutory calendar, order and pre-fill the identity and beneficial-owner capture, and keep a clean trail of every check carried out. What it never does is make the call. The risk judgement and any suspicious-activity report are professional acts of the accountant, and they stay that way.
An obligations map built on day one
From the client's data and the statutory calendar, the assistant generates an obligations checklist: VAT registration status and cycle, whether Making Tax Digital for VAT and for Income Tax apply, PAYE and payroll, and the corporation tax and accounts dates. The accountant validates it, so the first deadlines are mapped before the client's history has been fully absorbed by the team.
A Manchester company moves its accounts across. Before the first week is out, the assistant has laid out the VAT quarters, the corporation tax date and the accounts filing date from the company's record, and flagged that Making Tax Digital for VAT already applies. The accountant checks the map rather than assembling it under time pressure.
The month-one gap, where a deadline slips because the obligations are not yet internalised, is closed. Onboarding starts from a validated calendar instead of a blank page and a good memory.
Customer due diligence, ordered and pre-filled
Under the Money Laundering Regulations 2017 the practice must identify the client and any beneficial owner before acting. The assistant runs a structured capture of the names, dates of birth and addresses, requests identity and proof-of-address documents through a secure checklist, and pre-fills what it can. The risk assessment, whether simplified, standard or enhanced, and the decision to proceed stay with the accountant.
For the incoming company with two director-shareholders, the assistant collects ID and proof of address for both, confirms the shareholdings and orders the file into the practice's standard format. The accountant opens a complete, consistent pack and makes the risk call, rather than chasing scattered documents first.
The repetitive, error-prone part of due diligence is handled consistently for every client, in the same structure each time. The accountant's time goes to the judgement the regulations reserve for a person, not to collecting and retyping details.
A clean, retained trail of every check
AML is not only onboarding: the regulations require a documented risk assessment, ongoing monitoring and record-keeping of the checks carried out. The assistant keeps that trail as the work happens, logging what was collected, verified and when, so the evidence exists in order rather than being reconstructed later. The suspicious-activity report to the National Crime Agency, where warranted, remains a human act.
Twelve months on, a supervisory review asks what checks were done at onboarding. The trail is already ordered by client and by date, so the practice produces it directly instead of piecing it together from emails and folders.
With HMRC issuing 134 penalties to accountancy service providers totalling £513,930 in six months of 2025, a retained, ordered trail has real value. The record-keeping is enforced, and the assistant keeps it current without adding admin to the accountant's day.
The data transfer handled properly from the first message
Taking on a client transfers personal data, not only the directors but staff and the client's own customers. From the first onboarding message the assistant works under a clear lawful basis and a written processor contract, holds documents with defined retention, and tells the client plainly when an AI tool is part of the conversation, so the compliance floor is respected before any work begins.
A client asks whether their information is safe with an AI in the loop. The assistant confirms that documents are held under a data-processing agreement with set retention, in line with UK data-protection law, and that only the practice team sees the file, then continues the onboarding.
The data-protection groundwork is laid at the start rather than retrofitted, and the client hears it stated openly. Trust is built in the exact moment, handing over identity documents, where it matters most.
Onboarding and anti-money-laundering work is a strong fit for automation now, because so much of it is structured and repeatable.
- Generating an obligations checklist from the client's data and the statutory calendar, so the first VAT, Making Tax Digital, PAYE and corporation tax dates are mapped before month one, is buildable today. Onboarding is where a gap is most likely, because the client's history is not yet internalised by the team, and a checklist the accountant validates is exactly what closes it.GOV.UK
- The due-diligence flow is equally suited to it. Under the Money Laundering Regulations 2017 a practice must identify the client and any beneficial owner and carry out customer due diligence proportionate to risk. Collecting and ordering names, dates of birth and addresses is structured work an assisted flow can pre-fill, leaving the accountant the risk assessment and the decision to proceed.GOV.UK
- The record-keeping this supports is enforced, not nominal. HMRC issued 134 penalties to accountancy service providers totalling £513,930 in six months of 2025, so a clean, retained trail of the checks carried out has real value. An assistant can keep that trail while the risk judgement and any report to the National Crime Agency remain human acts.GOV.UK
- Set expectations against your own onboarding, not a headline. Whatever time-saving figure a tool quotes will have come from another firm's process, so treat it as a rough heading and prove it here: measure how long onboarding takes end to end, how complete the due-diligence pack is first time, and how few first deadlines slip, before and after. Those are the numbers that decide it.
Onboarding carries the strictest limits on this page, and they come first.
- Customer due diligence, the risk assessment and any suspicious-activity report are personal obligations of the practice under the Money Laundering Regulations 2017. AI can collect and order the data, but the judgement and the report stay human, and the checks must be retained.GOV.UK
- Identity documents and beneficial-owner details are personal data, and often special category. An external service that processes them is a processor that needs a written contract and a defined retention period under the UK GDPR.legislation.gov.uk
- Taking on a client also transfers other people's data: directors, staff and the client's own customers. The lawful basis and the processor contract are settled from the outset, and the client is told plainly if an AI tool is used in the onboarding conversation.ICO
Deadline management: a computed calendar that chases the records, not a wall you meet by surprise
The UK compliance year is unusually predictable, which is exactly what makes it automatable. Self Assessment is filed by 31 January, with payments on account due on 31 January and 31 July. VAT returns and payment fall roughly one month and seven days after each quarter end. Corporation tax and company accounts run on each company's own cycle. None of these dates is remembered; every one of them can be computed from the client's own records.
From 6 April 2026 a second clock sits on top of the annual one. Making Tax Digital for Income Tax adds quarterly updates due on 7 August, 7 November, 7 February and 7 May, plus a final declaration by 31 January. For every sole trader and landlord over £50,000 that is four extra filing moments a year, each needing the records in on time, layered onto clients who are used to a single annual scramble.
And the scramble is real. A record 11.48 million returns were filed by the 31 January 2026 deadline, around a million were still missed, and 475,722 went in on the day itself, the busiest hour being between five and six in the evening. The bottleneck is never the software. It is the last-mile chase to get every client's records in before the wall.
An AI assistant can hold that whole calendar for you, client by client, and work the chase. It computes who is due what and when, flags early what is ready and what is still waiting on documents, drafts the reminders and shows you a live who-is-due list. The filing, the sign-off and the responsibility stay with the practice and the taxpayer; the assistant simply makes sure nobody reaches the wall by surprise.
A deadline board computed client by client
The assistant reads each client's record and builds their statutory calendar: the 31 January Self Assessment filing, the payments on account on 31 January and 31 July, each VAT quarter, the corporation tax and accounts dates on the company's own cycle, and now the Making Tax Digital quarterly updates. Because the dates are fixed in law and in the record, the board is calculated rather than kept by hand, and it flags early what is ready against what is still outstanding.
In a Bristol practice the board shows, weeks ahead, that a client's VAT return is ready to file while three others are still missing bank statements. The accountant works from a single computed view instead of a spreadsheet somebody has to remember to update.
Nothing falls between the annual, quarterly and monthly clocks, because they all resolve into one calendar. The practice sees the month ahead early enough to act, not on the day a deadline lands.
The new quarterly rhythm chased before it swamps the practice
For every client in scope for Making Tax Digital for Income Tax, the assistant computes which quarterly update is next, on 7 August, 7 November, 7 February or 7 May, and starts the chase early rather than in the final week. It knows who is over the £50,000 threshold, tracks whose records are in and drafts the reminders for the rest, so the extra filing moments are managed on a schedule instead of absorbed in a panic.
In late June the assistant lists the nine clients with a 7 August update due and drafts a first reminder to each. The chase starts six weeks out, so by early August the records are mostly in, rather than arriving in a rush the day before.
Four new filing moments a year per client stop being four new emergencies. The quarterly rhythm becomes routine, and the practice absorbs Making Tax Digital without adding a scramble for every quarter.
The 31 January last mile worked from a live outstanding list
As the January peak approaches, the assistant maintains a live list of who is still outstanding and what each of them is missing, drafts the reminders and shows the accountant a who-is-due view that updates as records arrive. The people, not the software, remain the constraint, and the assistant puts the practice's attention exactly where the records are still absent.
On 20 January the list names eleven clients still to file: four who have sent nothing and seven missing one item each. The assistant has a tailored reminder ready for each, so the team spends the last stretch on the four hard cases rather than working out who to chase.
A frantic January becomes a managed one. No client is forgotten in the crush, and the practice reaches the deadline having chased deliberately rather than firefighting on the day.
The VAT cycle read straight from compatible software
Because Making Tax Digital for VAT already requires digital records and filing through compatible software with digital links, the VAT calendar is machine-readable end to end. The assistant reads the return cycle from the client's own system rather than tracking dates by hand, and folds it into the same computed calendar as the Self Assessment and Income Tax clocks, for the accountant to validate.
A quarterly VAT return period closes in the client's software; the assistant registers the one-month-and-seven-day filing date automatically and places it on the board next to that client's other obligations, with no date typed in by anyone.
The three clocks, quarterly VAT, annual Self Assessment and the new Income Tax updates, are managed from one computed calendar. Hand-tracking of VAT dates disappears, and the accountant validates a calendar rather than assembling one.
The deadline work sits on firm, buildable ground, because the dates are fixed in law and in each client's record.
- Because Self Assessment is filed by 31 January, with payments on account on 31 January and 31 July, and VAT falls about one month and seven days after each quarter end, a deadline board is exactly computable rather than remembered. An assistant that holds that calendar per client and flags what is ready against what is still waiting can be built now.GOV.UK
- The new quarterly rhythm is just as fixed and just as buildable. Making Tax Digital for Income Tax updates fall due on 7 August, 7 November, 7 February and 7 May, with the final declaration by 31 January, for every sole trader and landlord over £50,000. Computing who is due which update and starting the chase early is what keeps the extra clock from swamping a practice used to a single annual return.GOV.UK
- The pressure point it relieves is documented. A record 11.48 million returns were filed by the 31 January 2026 deadline, around a million missed, 475,722 on the day itself, the busiest hour between five and six in the evening. A live outstanding list and drafted reminders turn that last mile into a managed one, and do the same for each Making Tax Digital quarter.GOV.UK / HMRC
- Keep your own scoreboard rather than a vendor's. No published percentage will tell you what a deadline assistant is worth in your office, and any productivity claim attached to a tool comes from another firm and from the party selling it. The figures worth watching are the ones you can measure here: how many clients file on time, how early the records arrive, and how much of January stops being a scramble, each before and after.
A deadline assistant is only trusted if the filing stays with the practice, so keep these limits in from the start.
- Deadlines and filing formats are set by HMRC. The assistant can hold the calendar and prepare the data, but the filing, the sign-off and the tax responsibility stay with the practice and the taxpayer.GOV.UK
- A computed due date is validated against the official HMRC source before it reaches a client. A wrongly calculated deadline would undermine the very trust the automation is built to earn, so the source travels with every reminder.GOV.UK
- Reminder lists that slice clients by who owes what are personal data processing under the UK GDPR. Purpose limitation applies, and any external tool that holds them acts as a processor under a written contract.legislation.gov.uk
VAT returns, Self Assessment and MTD updates, prepared and checked before you sign
Return and accounts preparation concentrates the year into predictable peaks. VAT returns every quarter, the Self Assessment wall on 31 January, corporation tax and statutory accounts on each company's own cycle, and now four MTD Income Tax quarterly updates plus a final declaration per client from 2026/27. This is wave work, with the dates known well in advance and the pressure arriving all at once.
The scale of that 31 January wall is not abstract. A record 11.48 million Self Assessment returns were filed by the deadline for the 2024 to 2025 tax year, yet around 1 million taxpayers still missed it and 475,722 filed on the day itself. For a practice the risk is rarely the software, it is the last-mile scramble to get every client's records in before the wall, and MTD for Income Tax now layers three more quarterly walls on top of that annual one.
The preparation itself is well suited to an assistant. MTD for VAT already requires returns to be filed through functional compatible software with digital links, so the return is assembled from the digital records rather than transcribed, and a coherence check before filing, client details, boxes, rates, year-on-year carry-forwards, is a natural second pair of eyes. Catching an out-of-range value before submission is what heads off the HMRC query and the penalty later, and that matters more as the quarterly cadence multiplies the number of filing moments per client.
An AI assistant built for your practice gathers and orders the data per client, flags what is ready and what still awaits documents, and runs that coherence check before anything goes out. What it does not do is file. The preparation is the assistant's; the sign-off and the submission stay with the practice and the taxpayer, so every filing is signed by a person.
Gather and order the data per client for each deadline
For every client with a filing due, the assistant assembles the records in one place and marks plainly what is ready and what is still outstanding, a missing bank statement, a dividend voucher, a piece of paperwork not yet supplied. Instead of the accountant reconstructing where each client stands from memory, the readiness of the whole list is visible, and the chase for what is missing goes out in good time.
A practice preparing a run of returns opens the assistant's list: for one client everything is in, for another only a single interest certificate is outstanding, for a third three items are still awaited. The accountant works the ready files first and chases the rest early, rather than discovering the gaps in the last week of January.
The practice sees the true state of every client ahead of the peak, not on the eve of it. Work flows to the returns that can be finished now, and the missing-records scramble is spread out instead of stacked against the deadline.
A coherence check before every submission
Before a return is signed, the assistant checks it against itself and its history: do the client details agree across the pages, are the boxes internally consistent, do the VAT figures fit the applicable rate, do the year-on-year carry-forwards line up, is any value sitting outside a sensible range. Each flag comes with the reason it was raised, so the accountant reviews a short list of concrete points rather than re-reading the whole return blind.
On a VAT return the assistant flags that the output tax does not reconcile to the recorded sales at the standard rate, and that a carried-forward figure differs from last period without an obvious cause. Both are checked and resolved before submission, so what would have become an HMRC query is dealt with while it is still a two-minute fix.
The errors that generate queries and penalties are caught before filing, not after. As the quarterly cadence multiplies the filing moments, a reliable pre-submission check keeps that growing number of returns from becoming a growing number of mistakes.
The quarterly rhythm kept on the rails
MTD for Income Tax replaces one annual return with four quarterly updates and a final declaration per client. The assistant tracks where each client sits in that cycle, prompts for the records each quarter needs and prepares the update from the reconciled ledger, so the rhythm is managed rather than remembered. The figures are still reviewed and signed by the accountant every time.
Across a book of sole-trader and landlord clients now in MTD for Income Tax, the assistant keeps each one's quarterly position current and readies the update as its date approaches. The first final declaration for 2026/27 is due by 31 January 2028, and the assistant carries the quarter-then-finalise trail into it rather than leaving a year to reconstruct.
The three extra quarterly walls become a managed routine instead of a recurring panic. Nothing depends on one person holding every client's cycle in their head, and each update starts from a prepared, reconciled position.
See the whole client base before the 31 January wall
In the run-up to the Self Assessment deadline the assistant gives the practice a single readiness view: which returns are done, which are ready to sign, which are waiting on the client and which have not been started. The list re-sorts as records arrive, so the principal always knows where the real risk sits without opening twenty files.
In mid-January the view shows the practice is clear on most clients but waiting on six, two of whom have gone quiet. The reminders go out at once and attention concentrates on the genuine stragglers, rather than everyone finding out on the 30th which files are still open.
The busiest window of the accountancy year is steered on facts instead of adrenaline. Fewer clients slip toward deadline day, and the practice decides where to push while there is still time to push.
Return preparation is ready to build around today, so long as one number stays in perspective.
- Assembling the data per client, flagging what is ready against what is outstanding and running a coherence check before filing can be built now. MTD for VAT already requires returns to be filed through functional compatible software with digital links, so the return is assembled from the same structured records HMRC expects, and a check on the boxes, the rates and the carry-forwards is a natural second pair of eyes before a person signs.GOV.UK
- The load this relieves is real and dated. A record 11.48 million Self Assessment returns were filed by the 31 January 2026 deadline for 2024 to 2025, around 1 million taxpayers still missed it and 475,722 filed on the day itself, and MTD for Income Tax now adds three more quarterly walls per client on top of that annual one. A readiness view across the whole client base, prepared and kept current, is exactly the workflow that is realistic to build today against that pressure.GOV.UK / HMRC
- One thing to keep in perspective: the time-saved and error-reduction percentages quoted for return-prep tooling come from the vendors that sell it, and often from other markets. Treat them as a direction of travel rather than a guarantee for your practice, and build the case on your own figures instead: preparation time per return, HMRC queries raised per filing, and the share of clients filed well before deadline day, each measured before and after.
The hard limits to design in before a single return is prepared this way.
- Deadlines and the content of returns are HMRC's, not the assistant's. AI can prepare and check the data, but the submission, the sign-off and the tax responsibility remain with the practice and the taxpayer. Every filing is signed by a person who has reviewed it; the assistant readies the return, it never files.GOV.UK
- AI-prepared figures still need human sign-off before filing. Automation does not transfer responsibility for the accuracy of the return, the VAT position or the accounts, so the pre-submission check is a second pair of eyes for the accountant, not a substitute for one.GOV.UK
- Clients' tax data is confidential personal data under the UK GDPR, so processing it through an external AI service requires a processor contract with defined retention. And if a solely automated step ever produced a significant decision, the Article 22A to 22D safeguards, in force since 5 February 2026 via the DUAA 2025, would apply, which is a further reason the sign-off on every filing stays human.legislation.gov.uk
Source documents into clean books, without the manual re-keying
Bookkeeping begins with a pile of source documents. Every month a practice works through invoices, receipts and bank statements, then transcribes the supplier, the amount, the VAT and the date, line by line, before a single entry is posted. It is volume work rather than judgement, and in a small or sole-practitioner firm it lands on the same person who also holds the client relationship and carries the year-end.
Making Tax Digital has quietly turned that volume into a compliance question. MTD for VAT requires digital records and digital links between systems, and MTD for Income Tax has required digital record-keeping since it went live on 6 April 2026. Re-typing figures from one program into another breaks the digital-links rule, so extracting document data straight into the bookkeeping system is now the compliant route as well as the faster one.
The shape of the input is changing too. Mandatory e-invoicing for all VAT invoices has been announced for 2029, which means a growing share of documents will land as structured data rather than images to scan. There is nothing to buy yet, but the practices that let structured records flow now will meet the mandate as a formality instead of a project.
An AI assistant built for your practice works precisely this step. It reads the receipts and statements as they are, proposes the field extraction and a suggested posting, and hands you the classification decision and the final check. The account code, the VAT treatment and the deductibility stay where they belong, with the accountant, and nothing reaches the records until you approve it.
Read receipts, invoices and bank statements as they arrive
The assistant takes the documents in whatever form the client sends them, a scanned batch, phone photos, a bank statement PDF, and extracts the fields that matter: supplier, date, net, VAT and gross. It proposes a posting for each and parks anything unreadable or contradictory in a query list with a plain note on what is wrong. Nothing is posted until your bookkeeper approves it, so the classification decision stays with a person.
A client of a two-partner practice sends a month of receipts and a bank statement at the end of the quarter. The assistant extracts the lot, then flags one receipt where the VAT shown does not reconcile to the net at any standard rate, and another that is too creased to read. Both go into the query list rather than into the books, and the rest arrive already prepared.
The transcription that used to eat an afternoon per client is done in the background, while the account code, the VAT treatment and the deductibility stay human decisions. Because the data flows digitally instead of being typed twice, the process lines up with what Making Tax Digital already expects of your records.
Suggested postings, with the judgement left to you
From the patterns in each client's own history the assistant proposes the account code and the VAT treatment, but it presents them as suggestions for the accountant to confirm, never as a done posting. It learns the house rules over time, so recurring items settle into a steady proposal while the genuinely ambiguous ones are pushed forward for a decision rather than guessed at.
A monthly software subscription is proposed to the same code it has carried all year and waved through in seconds. A one-off payment that could be a capital purchase or a repair is held back with both options set out, so the accountant makes the call that actually needs judgement.
Routine coding is prepared and the judgement calls are surfaced instead of buried in a hundred identical-looking lines. The work that needs an accountant reaches an accountant, and the work that does not stops taking their evening.
Chase the missing paperwork before the month can close
The assistant knows which clients still owe documents for the period, drafts the polite reminder and tracks what has come back in. Instead of the principal holding the whole list in their head, the outstanding items are visible, and the chase goes out on time rather than in the last-minute scramble before a deadline.
For an 80-client practice the assistant lists the eleven clients missing a bank statement or a run of receipts for the quarter, drafts a short reminder to each and re-checks as they trickle in. The bookkeeper starts the close with the gaps already narrowed, not with a fortnight of chasing still to do.
The last-mile scramble to get every client's records in before the wall shrinks. Documents arrive earlier and more completely, which is exactly what the quarterly MTD rhythm now depends on.
Answer questions straight from the documents
Because the assistant holds the documents and their extracted data, it answers plain questions at the record: what is still in the query list, what a given supplier charged last quarter, whether a client's paperwork for the period is complete. Every answer links back to the underlying document, so it can be checked in one click rather than hunted for across folders and inboxes.
A bookkeeper asks which receipts are still unresolved for a client before running the VAT figures. The assistant lists the three open items with the reason each was parked, and the query is cleared in minutes instead of at the next month end.
The overview of the paperwork stops depending on the one person who knows where everything lives. Time goes on decisions rather than searching, and nothing quietly falls out of the pile.
The document step sits on solid, buildable ground today, with one place to keep a level head.
- A flow that reads source documents, proposes the extraction and prepares a posting for approval can be built now, and in the UK it works with the grain of the law. Making Tax Digital for VAT requires digital records and digital links between systems, and MTD for Income Tax has required digital record-keeping since 6 April 2026. Re-typing figures between programs breaks the digital-links rule, so structured, machine-read document data is the compliant path, not a nice-to-have.GOV.UK
- The input is only going to get more structured. Mandatory e-invoicing for all VAT invoices, both B2B and B2G, has been announced for 2029, so an increasing share of documents will arrive as data rather than as images to scan, and AI extraction gets easier the more structured the input becomes. There is nothing to buy for it yet, but structured digital records already pay off under Making Tax Digital, so preparing now costs nothing extra.GOV.UK
- This is not speculative inside the profession. ACCA reports that 52% of respondents regularly use AI tools and 82% feel confident learning them, while ICAEW found 83% of its youngest chartered accountants, aged 18 to 24, use AI at least weekly. The demand and the caution already sit side by side in the profession, which tells you the ecosystem has moved, not that there is a finished product to plug in.ACCA
- Where to stay realistic: the extraction-accuracy and time-saved percentages quoted for document tooling come from the companies selling it, and from cleaner sample data than a real client's shoebox of receipts. Read them as an indication from other firms' books, not a promise, and judge the build on your own numbers: minutes per document, the share of receipts extracted correctly first time, and the errors caught before they reached the books.
Nothing the assistant reads should reach the books until these three points are settled.
- The posting the AI suggests is checked by a person. The account classification, the VAT treatment and the deductibility remain professional decisions of the accountant, not the system, and a wrong extraction is no excuse for an incorrect set of records. The working rule is simple: the assistant reads, extracts and proposes; the entry reaches the books only after a human has signed it off.GOV.UK
- There is nothing to purchase for 2029 e-invoicing yet. The technical standards and rollout detail arrive with the government's roadmap due at Budget 2026, and anyone selling certainty before that is selling ahead of the facts. What is worth doing now is structured digital records, which already earn their keep under Making Tax Digital today.GOV.UK
- Invoices, receipts and bank statements carry personal and commercial data, so an external extraction service acts as a processor. That means a written contract under the UK GDPR, data minimisation and a retention schedule that is actually followed, with what is no longer needed leaving the system rather than sitting in an inbox forever.legislation.gov.uk
Bank reconciliation matched and flagged, with the judgement calls left to you
Reconciliation is rule-based comparison work, repeated for every client and every period. You match the lines on a bank statement to the movements already recorded, find the receipts and payments that pair up, and isolate the open items and the differences. It is the same shape of task month after month, which is precisely why it is such a natural fit for an assisted flow that proposes the answer in draft.
Where the records already live in MTD-compatible software, the statement and the ledger are machine-readable, so matching starts from structured fields rather than re-keyed figures. That turns a manual tick-and-bash into a process where the assistant proposes the matches and flags the anomalies, a payment with no invoice, a duplicate, a VAT figure that does not fit the rate, and a person confirms before anything is locked. The cleaner the source data Making Tax Digital forces, the higher the share the assistant can match on its own.
Reconciliation is also a control point, and the cadence has tightened. Quarterly VAT returns and, since April 2026, quarterly MTD Income Tax updates are only ever as good as the reconciled ledger behind them. A clean reconciliation each period now matters more than it did under a single annual cycle, because there are simply more filing moments to be ready for.
An AI assistant built for your practice keeps that reconciliation current between deadlines rather than in a rush the week one falls due. It does the comparison, proposes the matches and queues the doubtful items for your decision. The treatment of an unmatched item, a write-off or a disputed difference stays a professional call, and nothing is locked until you confirm it.
Match the statement to the ledger, in draft
The assistant lines up each statement entry against the movements already recorded, pairs the receipts and payments that clearly correspond and isolates whatever is left over. Confident matches are proposed together; the doubtful ones are set aside for a person. It never posts the match itself, it prepares it, so the accountant reviews a short list of decisions instead of ticking through every line.
In the week after the VAT quarter closes, a practice reconciles the period for a retail client. The assistant matches the great majority of the statement to the ledger straight away and presents fourteen paired items for a quick glance, leaving a handful of genuine unknowns clearly separated. What used to be an afternoon of manual matching becomes a ten-minute review.
The repetitive comparison is done in the background, and the accountant's attention goes to the items that actually need a view. The reconciliation is prepared to a consistent standard whoever picks it up.
Flag the anomalies a tired eye misses
As it matches, the assistant watches for the tell-tale problems: a payment with no matching invoice, a duplicate that would double-count, a VAT figure that does not fit the applicable rate, a receipt that never arrived. Each is raised with the reason it was flagged, so the exception is explained rather than just highlighted, and a person decides what to do about it.
During a reconciliation the assistant flags a supplier payment that appears twice on the same day and a sales receipt whose VAT does not reconcile to the net. Both are surfaced before the period is closed, so a duplicate is not carried into the return and a misposted rate is caught while it is still cheap to fix.
The errors that would otherwise surface as an HMRC query, or not at all, are caught at the point of reconciliation. The ledger behind each return starts from a cleaner position without anyone reading every line twice.
Keep the reconciliation current between deadlines
Rather than reconciling a whole quarter in the week it is due, the assistant keeps the comparison up to date as statements and records come in, so the position is always close to current. When a quarterly update or a VAT return approaches, the reconciliation is already most of the way there and only the recent items need a look.
With quarterly MTD Income Tax updates now in the calendar, the assistant keeps a landlord client's reconciliation ticking over through the quarter. When the update is due, the accountant confirms a short tail of recent items instead of starting the whole quarter cold against the clock.
Each quarterly update begins from a trustworthy ledger rather than a scramble. The more frequent rhythm stops meaning more late nights, because the work is spread rather than stacked against the deadline.
The unmatched items queued for your decision
Everything the assistant cannot match with confidence is collected in one place with its context: the amount, the date, why it could not be paired and any near-matches it considered. The write-off, the disputed difference and the genuinely unexplained item are presented for the accountant to resolve, and the reconciliation is only locked once a person has signed it off.
At the end of a period three items remain: a small unexplained credit, a payment the client disputes and a difference that looks like a timing gap into the next month. The assistant lays out all three with its reasoning, and the accountant decides on each rather than the system quietly forcing a match.
The awkward last mile of a reconciliation, the part that actually needs judgement, is handed over cleanly instead of hidden inside an auto-matched total. Control of the records stays firmly with the practice.
Reconciliation is one of the clearest wins to build today, provided you keep sight of one honest limit.
- An assisted flow that proposes matches and flags anomalies can be built now, because the ingredients are already in place. Where records live in MTD-compatible software the statement and the ledger are machine-readable, so matching starts from structured fields rather than re-keyed figures, and the cleaner the source data Making Tax Digital forces, the higher the share the assistant can match on its own before a person confirms it.GOV.UK
- The value has grown with the cadence. Since MTD for Income Tax went live on 6 April 2026, quarterly updates sit alongside the quarterly VAT return, and each is only as good as the reconciled ledger behind it. Keeping reconciliation current between deadlines, rather than reconciling a whole quarter the week it is due, is a workflow within reach today and matters more with every filing moment the quarterly rhythm adds.GOV.UK
- One caution on the numbers: the auto-match rates quoted for reconciliation tooling come from the firms selling it, and usually from tidier books than a new client's first quarter. Treat any headline percentage as a direction of travel, not a guarantee for your ledgers, and prove it on your own figures instead: your real auto-match rate, the exceptions raised per period, and the hours to a clean reconciliation before and after.
Keep these limits in the build so the control point stays a control point.
- Matches and differences suggested by the assistant are verified by a person. The treatment of an unmatched item, a write-off or a disputed difference remains a professional decision of the accountant, and nothing is locked without that human sign-off. The assistant proposes and explains; it does not force a match to make a total balance.GOV.UK
- The accuracy and reality of the records remain the accountant's responsibility under their professional body's rules, and an automatic match does not transfer that. The ICO's guidance on AI and data protection expects meaningful human review of an automated output before it is relied on, which is exactly the confirm-before-lock step designed into this flow.ICO
- Bank statements and ledgers are personal and commercial data, so an external reconciliation service acts as a processor. A written contract and a defined retention schedule under the UK GDPR are part of the build, not an afterthought bolted on once the flow is live.legislation.gov.uk
Pricing and proposals: charge where the work has genuinely grown, on evidence not gut feel
A fixed fee is a promise made once, against work that no longer stands still. Making Tax Digital for Income Tax turns what used to be a single annual return into four quarterly updates plus a final declaration for every client over the threshold, and that sits on top of MTD for VAT and the practice's anti-money-laundering duties. The work grows even where the fee does not, and the margin erodes quietly, one unbilled hour at a time.
Most small practices cannot see that happening. The profession is large and fragmented, over 408,000 members across the UK and Republic of Ireland bodies, and in the many sole-practitioner and small firms the real profitability of each client is sensed rather than measured. A proposal built on gut feel is hard to defend and harder still to raise when a client's compliance footprint has clearly grown.
The proposal and the engagement letter are also where honesty about AI belongs. There is no UK statute forcing you to disclose that you use AI, but ICAEW and ACCA guidance pushes firms towards transparency with clients, and the engagement letter is the natural place to set out, in plain terms, where AI supports the work and where the accountant's judgement stays in charge.
An AI assistant built for your practice works exactly that seam. It prepares draft proposals, documents the real complexity of each client, the hours, the obligations, the risks, and gives you the evidence to price where the work has genuinely increased. It drafts, you decide. The fee, the wording and the sign-off never leave the principal's desk.
Draft proposals built on documented workload, not gut feel
The assistant assembles a proposal from each client's obligations and the statutory calendar: whether they are VAT registered and on which cycle, whether they fall under MTD for Income Tax, PAYE and payroll, corporation-tax and accounts dates. It sets out the estimated hours and the recurring filing moments so the fee has a visible basis. It drafts; the principal sets the number.
A practice takes on a sole trader who also lets two flats and is VAT registered. The assistant lays out the year: four quarterly VAT returns, four MTD quarterly updates plus the final declaration, and the rental record-keeping, with an hours estimate against each. The principal reads it in a few minutes and sets a fee that reflects the actual work rather than last year's going rate.
The proposal has a spine. When a client asks why the fee is what it is, the answer is a documented list of obligations and hours, not a shrug, and the number is far easier to stand behind.
Evidence to re-price when a client's compliance footprint grows
MTD for Income Tax adds roughly four filing moments a year per client in scope, and the qualifying-income threshold falls from £50,000 now to £30,000 in April 2027 and £20,000 in April 2028, pulling more clients in each year. The assistant flags the clients whose obligations have grown since the fee was last set and prepares the before-and-after picture the principal needs for a fee conversation.
A sole-trader client who was one annual return last year now carries quarterly updates and more record-keeping. The assistant shows the change on a single page: what the engagement used to involve, what it involves now. The principal opens the renewal with the evidence in hand rather than an apology.
Fee reviews land where the work has actually moved, on evidence, instead of an awkward across-the-board rise or a silent loss of margin that nobody chose.
Engagement-letter wording, including honest AI-use transparency
The assistant drafts the engagement letter and proposal wording, including a plain sentence on where AI supports the work and where the accountant keeps the judgement, in line with ICAEW and ACCA guidance. The content, the pricing and the sign-off stay with the principal, who remains responsible for the work whatever tools are used.
A practice adds a short, clear paragraph to its standard engagement letter explaining that AI helps prepare and organise the work, and that a qualified accountant checks and signs off everything that matters. Clients read candour rather than concealment, and the point is settled once, in the right document.
Transparency about AI is handled cleanly and in advance, in the document built for it, rather than improvised later when a client asks.
Fixed fee against actual hours, checked not assumed
Where the practice records time, the assistant compares the fixed fee on each client against the hours actually spent, and surfaces the engagements where the two have drifted apart. It presents this as decision support for the principal, never as an automatic verdict on a client relationship.
The assistant flags three fixed-fee clients whose recorded hours have crept well past what the fee assumed, two of them newly inside MTD for Income Tax. The principal decides which to raise, which to hold and which to restructure, working from one checkable list.
The gap between fixed fees and actual hours stops being a year-end surprise and becomes a visible, manageable number the principal reviews on their own terms.
The ground here is firm, and the pressure that makes it worth doing is already in the statute book.
- The growth in workload is real and computable, not a guess. MTD for Income Tax replaces one annual return with four quarterly updates plus a final declaration for every client in scope, and the qualifying-income threshold falls from £50,000 now to £30,000 in April 2027 and £20,000 in April 2028, so the client base in scope roughly triples over three years. An assistant that reads each client's obligations from that calendar and documents the hours behind a proposal is buildable today.GOV.UK
- Pricing on evidence rather than instinct suits the shape of the profession. With over 408,000 members across the UK and Republic of Ireland bodies and thousands of small and sole-practitioner firms where the principal does production, client relationships and admin at once, the margin on each client is usually sensed, not measured. Turning recorded obligations and hours into a defensible proposal is a realistic first build, not a research project.FRC
- The engagement letter is the right home for AI transparency. There is no UK law compelling disclosure, but ICAEW and ACCA guidance steers firms towards being open with clients about where AI supports the work, and an assistant can draft that wording for the principal to approve.ICAEW
- Keep a firm grip on the vendor numbers. The time-saving and efficiency percentages quoted around AI proposal and pricing tools tend to come from other markets and from the firms selling them, so treat them as a rough guide rather than a promise for your practice. The figures that settle it are your own: hours per client against the fee, measured before and after.
The fee and the numbers behind it belong to the practice, which sets the limits below.
- The assistant prepares, the principal prices. Draft proposals and workload estimates are useful, but setting the fee is a professional decision, and the internal figures behind it, the hours, the margins, the rates, are confidential to the practice. They belong in the practice's own systems, never in dev tools, tickets or client messages.
- Where the engagement letter describes AI use, it must be clear and honest, in line with ICAEW and ACCA guidance, and the accountant remains responsible for the work whatever tools are used. An assistant can draft the wording; it cannot carry the responsibility.ICAEW
- Profitability and workload estimates are decision support, not automatic verdicts. Which clients to re-price, hold or let go stays a human judgement made on the numbers, not an output the system applies on its own.
Team and workload: plan the peaks you can already see coming
Ask any principal when the pressure comes and they can name the dates before you finish the question. The 31 January Self Assessment wall, the quarterly VAT cycle, the corporation-tax and accounts deadlines that fall on each company's own clock, and now the MTD for Income Tax quarterly updates due on 7 August, 7 November, 7 February and 7 May. The peaks are set in law, which means they are calculable in advance rather than something you brace for and hope.
That is what makes them plannable. An assistant can turn the deadline calendar into a picture of load, per person and per client, so the work is spread across the month and the year instead of banking up in the days before each due date. The difference matters more now that the critical moments arrive every quarter, not once a winter.
The organising job is not only about efficiency. Under the Money Laundering Regulations 2017 a practice needs clear roles, who carries out the client due diligence, who reviews it, who keeps the records, proportionate to its size, and adopting AI itself needs the same governance, which is where the generational skills gap inside the profession bites. Structure is partly a compliance matter, not a nicety.
And the sector is built on small practices where the principal does production, client relationships and admin at once, the classic single point of failure. Distributing work with real visibility of deadlines and of progress per client reduces the dependence on one person at exactly the moments it is most dangerous. The assistant lays out who is due what and who is at risk of slipping; the assignment and the responsibility stay with the practice.
The statutory calendar turned into a load map
The assistant reads the fixed dates from law and from each client's own cycle, 31 January, the quarterly VAT clock, corporation tax and accounts, and the MTD quarterly-update dates, and builds a view of load per person and per client for the weeks ahead. You see where the work banks up before it does.
Three weeks out from the January peak, a practice looks at the month ahead. The assistant shows one senior carrying a disproportionate share of the limited-company accounts while two colleagues have room, and proposes a redistribution for the principal to approve.
Capacity is planned against dates that are known in advance, so the bottleneck is smoothed before it forms rather than survived after it has.
The January crunch managed, not endured
In the run-up to 31 January the assistant keeps a live who-is-outstanding list: which clients still owe records, which returns are ready to file, which are at risk of missing the wall. It drafts the chase messages; the principal decides what goes out and when.
On the Monday before the deadline the assistant flags eleven clients whose records have still not arrived and ranks them by how much work remains once they do. The team works the list in order rather than firefighting whoever shouts loudest.
A frantic January becomes a managed one. A record 11.48 million returns were filed by the 2026 deadline yet around a million people still missed it, and the practice's job is to keep its own clients well off that list.
The new quarterly rhythm spread across the team
MTD for Income Tax adds four filing moments a year for every sole-trader and landlord client in scope. The assistant computes who is due which quarterly update, on 7 August, 7 November, 7 February and 7 May, and starts the record-gathering early so the new cadence does not swamp the practice.
In July the assistant lists the clients due their first MTD quarterly update by 7 August and schedules the record-gathering across the fortnight before, instead of leaving it to the final days.
The quarterly wall is worked ahead of time, so a rhythm capable of overwhelming a once-a-year practice becomes just another planned cycle.
AML roles and checks tracked, not misplaced
The assistant keeps track of who did the client due diligence, who reviewed it and who holds the records for each client, proportionate to the practice's size. It orders the evidence and keeps a trail of the checks carried out; the risk judgement and any report to the authorities stay with the accountant.
During onboarding a new client, the assistant records which team member verified the identity and beneficial-owner details and flags the review still outstanding. Nothing sits in a limbo where everyone assumes a colleague did it.
The organisational side of AML, who checks, who verifies, who retains, is visible and assignable, which is exactly where a supervised practice needs it to be.
Planning the peaks is a solved problem in principle; the value is in doing it for every client, every cycle.
- Because the deadlines are fixed in law they are computable, not remembered. The 31 January filing date, the quarterly VAT clock and the MTD for Income Tax quarterly updates on 7 August, 7 November, 7 February and 7 May can all be held client by client and turned into a load map weeks ahead. That calendar layer is a realistic build today.GOV.UK
- The January wall is measurable and unforgiving, which is exactly why planning against it pays. A record 11.48 million Self Assessment returns were filed by the 2026 deadline, yet around a million people still missed it and 475,722 filed on the day itself, the busiest hour being 17:00 to 17:59. An assistant that shows who is still outstanding turns that last-mile scramble into a worked list.GOV.UK / HMRC
- The first build suits the shape of the trade. With over 408,000 members and thousands of small firms where the principal is also the admin department, distributing work with visibility of deadlines and progress per client is the sensible first step, and it directly eases the single-point-of-failure risk these practices carry.FRC
- Be sceptical of the capacity-saving headlines. Efficiency percentages attached to workload and practice-management AI usually come from other markets and from the companies selling the tools, so read them as a direction rather than a guarantee. Judge this on your own numbers: how many clients slip past a deadline, and how evenly the load actually falls across the team.
Two of these are law, one is common sense; plan all three in.
- Workload data links hours, people and clients, so it is personal data about both staff and clients. Data minimisation applies under the UK GDPR, and any external tool that holds it acts as a processor under a written contract.UK GDPR (Regulation (EU) 2016/679 as retained in UK law)
- AI used to monitor staff activity is a higher-risk use in its own right. The ICO names performance and productivity monitoring as higher-risk, so it needs a lawful basis, transparency and usually a data protection impact assessment, and workers must be told. Workload planning should help the team, not grade it in secret.ICO
- Capacity and load estimates are decision support. Who is assigned what, and how the practice is staffed, stays a human decision, not an output the system applies on its own.
Reporting and margins: ask your own numbers which clients actually pay
In most small practices, profitability per client is a feeling rather than a figure. The team's time on each client, and the margin it leaves, are inferred from a general sense of who is a nuisance and who is easy, not read off anything. The profession is large and fragmented, and that instinct is what pricing and portfolio decisions quietly rest on.
An assistant can put a number where the feeling is. It summarises, in plain language, the hours and activities spent on each client and flags the ones consuming far more than the fee they pay, so the principal can decide what to do about them. This is the first realistic office win, an honest picture of margin, not another dashboard nobody opens.
The pressure behind it is the wave of digital reporting. MTD for VAT, MTD for Income Tax from 2026, the AML obligations and e-invoicing, announced for 2029, all add volume and complexity per client without automatically adding to the fee. Margin falls silently unless someone can see it moving, and clear internal reporting is what gives the principal the arguments, the hours, the complexity, the risk, to renegotiate where the work has genuinely grown.
The realistic first step is not a new reporting suite. It is being able to ask plain-language questions of your own data: which clients exceed their budgeted hours, how much margin is lost each month, which service line weighs most. Appetite for this already exists inside the profession, and the assistant answers from your own numbers, which stay the basis for human decisions rather than replacing them.
Plain-English questions of your own numbers
Between any fixed reports, the assistant stays available to answer questions as they occur to you: which clients are over their budgeted hours, how this month compares with last, which service line takes the most time. It reads the answer from the practice's own records, changes nothing, and names the figures it relied on so you can check them.
On a Thursday evening a principal asks from the phone which clients have run over their budgeted hours this quarter. The answer arrives in under a minute, in a few sentences, with the underlying numbers, instead of waiting for a month-end pack that never quite gets built.
Decisions are made on current numbers rather than last quarter's impression, and the reporting job stops being a dreaded evening chore and becomes a short conversation.
Which clients cost more hours than their fixed fee
The assistant joins recorded time to the fixed fee on each client and surfaces the engagements where the hours have outgrown the money. It presents this as a ranked, checkable list, decision support for the principal, never an automatic verdict on a relationship.
The assistant flags four fixed-fee clients whose hours this year sit well above what the fee assumed, two of them newly inside MTD for Income Tax. The principal sees which to raise, which to restructure and which to keep as they are.
The gap between fixed fees and actual hours becomes a visible monthly number instead of a nasty surprise at the year end.
Early warning as compliance adds weight
As each new obligation lands, the MTD quarterly updates, the digital-records duty, e-invoicing ahead, the assistant shows which clients have taken on more work since their fee was last set, and assembles the before-and-after evidence for a fee conversation.
A client who was a single annual return last year now carries quarterly updates and more record-keeping. The assistant lays the change on one page so the renewal opens on evidence rather than on apology.
Margin erosion is caught while it is happening, and fee reviews land where the work has actually moved rather than in an across-the-board rise.
Where the hours really go across the book
The assistant summarises time by service line and by client segment, so the principal can see which work makes money, which only makes turnover, and where the hours disappear. It shows its working on request and rests only on the practice's own history.
The monthly summary shows VAT and the new MTD quarterly work absorbing more of the team's time than their share of the fees, while the annual-accounts work carries the margin. The decision to buy in extra help, or to adjust fees, is planned around that picture.
The practice steers by where the time and the margin actually are, not by which clients merely feel busiest.
This is buildable now, and it is the office win most practices feel first.
- The thing missing in a small practice is rarely the data, it is the reading of it. Profitability per client is sensed, not seen, across a large and fragmented profession, and an assistant that summarises the hours per client in plain language and flags the ones consuming more than their fee is a realistic first build rather than a distant ambition.FRC
- The pressure is structural and rising. MTD for VAT, MTD for Income Tax live since April 2026 and e-invoicing, announced for 2029, each add volume and complexity per client without lifting the fee, so margin erodes quietly. Internal reporting that shows the hours, the complexity and the risk is what gives a principal the grounds to renegotiate where the work has genuinely grown.GOV.UK
- The appetite is already there. ACCA reports 52% of respondents regularly use AI tools and ICAEW found 83% of its youngest chartered accountants use AI at least weekly, so querying the practice's own numbers in plain language, without hand-building a single report, is the most accessible way to give the principal an honest picture of margin.ACCA
- Treat the vendor uplift figures with care. The margin and efficiency gains quoted by reporting-tool sellers were earned in other markets on their own products, so they point a direction rather than promising an outcome. The only benchmark that counts is your own book: budgeted hours against fee, client by client, before and after.
Know the boundaries, one legal and two of judgement.
- The moment a report links hours to a named client or a named member of staff, it is processing personal data under the UK GDPR. Data minimisation applies, and any external analytics or AI tool that holds those figures acts as a processor under a written contract.UK GDPR (Regulation (EU) 2016/679 as retained in UK law)
- Profitability and workload estimates are decision support for the principal, not automatic verdicts. Which clients to re-price and how to staff the work stay human decisions, made on the numbers rather than applied by the system.
- Internal figures, the hours, the margins and the fees, are confidential to the practice. They must not be placed in dev or ticketing tools or in client messages; they belong in the practice's own internal reporting and nowhere else.
We build them, on Claude
These AI flows do not stay on paper. Svennis Cloud Solutions builds and integrates them into your systems, with a team of certified Claude architects, on Anthropic technology, from the first WhatsApp message to the finished invoice.
See what this looks like in your practice
Describe your practice in one sentence and Claude will show you live where AI would make the biggest difference, using your own numbers.
Sources
- 1. GOV.UK - Making Tax Digital for Income Tax (policy paper)
- 2. GOV.UK / HMRC - 11.48 million beat the Self Assessment deadline
- 3. ACCA - AI is reshaping the work of accountants; ICAEW AI in accountancy research
- 4. ASA - Disclosure of AI in advertising: striking the balance between creativity and responsibility
- 5. legislation.gov.uk - UK GDPR (Regulation (EU) 2016/679 as retained in UK law)
- 6. GOV.UK - Money laundering supervision for accountancy service providers (MLR 2017)
- 7. ICO - Guidance on AI and data protection
- 8. GOV.UK - Self Assessment tax returns: deadlines
- 9. GOV.UK - VAT record keeping / Making Tax Digital for VAT
- 10. legislation.gov.uk - Data (Use and Access) Act 2025, section 80 (UK GDPR Articles 22A to 22D)
- 11. GOV.UK - Promoting e-invoicing across UK businesses and the public sector (consultation response)
- 12. FRC - Key Facts and Trends in the Accountancy Profession 2025
- 13. ICAEW - AI and accountants: the rules and guidance you need to follow
- 14. ICO - Monitoring workers