A practice can use AI to chase quarterly Making Tax Digital records by comparing each client's expected income and expense evidence with what has arrived, drafting a precise request for the gaps, and sorting replies into a reviewer queue. Keep digital records in compatible software, check cumulative totals there, and have a practitioner approve the update.
The first stage applies from 6 April 2026 to qualifying income above GBP 50,000 on the specified earlier return; later stages use above GBP 30,000 and above GBP 20,000. The chase below starts once a client's scope and software route have been confirmed. A practice still deciding who belongs in the programme should first use the client-list triage workflow.
Define the evidence you expect for each client
A small practice loses time when the first reminder says only “send your quarterly records”. Turn the client engagement into a concrete evidence checklist. For a sole trader, that might include sales invoices, payment-platform statements, bank-feed exceptions, purchase invoices, mileage or travel records if relevant, and a note of cash takings. For a landlord, it might include rent statements, managing-agent deductions, repair invoices and receipts for other property expenses. These examples are illustrative; the actual request should match the client's activities and the income and expense categories used in their software.
Make the checklist a small table or portal form with four columns: item, expected source, last date received and status. Use “not applicable” only after a person has confirmed it. Missing is different from a true zero. A client who had no cash takings should confirm “none”; an empty cash-takings row should trigger a question. The official guidance says digital records need amount, date and category, while ordinary source documents such as invoices and statements must still be kept. AI may label and compare files, but a client or practice reviewer owns the accuracy of those records. The same guidance warns that a bank feed may need extra detail, such as the Self Assessment category, and that some transactions may not appear in full in the feed, so bank-feed exceptions deserve their own line on the checklist.
Illustrative filled checklist: “C-208, sole trader, period ending 5 October: sales export received through 5 October; bank feed received through 1 October; purchase invoices missing for September; cash takings confirmed none; reviewer assigned.” The useful chase is for the missing four bank-feed days and September purchase invoices, not every document on the original list. If the client's records are maintained by a bookkeeper, ask that person for the gap rather than sending two conflicting requests to the client.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowWork backwards from the submission date
Official guidance gives the same update deadlines for standard and calendar update periods: 7 August, 7 November, 7 February and 7 May after the fourth period. Standard periods end on 5 July, 5 October, 5 January and 5 April; calendar periods end on the last day of June, September, December and March. Confirm which basis the client chose in software before writing the request schedule. The choice cannot simply be changed after a first update has been sent in the tax year. An update can go any time after the period ends, up to the deadline, or up to 10 days before the period ends if the client expects no further transactions, which helps with a client who will be away at quarter end.
Set internal dates earlier than those deadlines, allowing for missing records, reconciliation, client questions and practitioner review. These are practice targets, not statutory dates. An illustrative schedule for a 7 November deadline might be: first tailored request on 10 October; reply target on 20 October; missing-item follow-up on 24 October; reviewer pack by 31 October; client approval, if required by the engagement, by 3 November; final submission check before 7 November. If a client regularly delivers a week late, the practice should change its request date or service arrangement rather than letting AI write increasingly urgent but ineffective emails.
Keep one status per client and business: awaiting first request, waiting for client, partial evidence, ready for review, clarification needed, approved, submitted or escalated. Do not use “done” for both approved and submitted. The quarterly update is a summary generated from digital records, not a separate return typed into an AI assistant. It covers the tax year to date, so the reviewer checks cumulative totals against the previous quarter as well as this period's new records.
Build a chase table that can explain every reminder
| Column | Illustrative entry | Why it matters |
|---|---|---|
| Client and business ID | C-208 / trade A | Avoids sending the wrong business request |
| Period and deadline | Standard, 5 October / 7 November | Controls timing |
| Expected item | September purchase invoices | Makes the request specific |
| Evidence received | Bank feed to 1 October | Prevents duplicate requests |
| Owner and next action | Reviewer B / ask client about four days | Supports follow-through |
| Approval state | Not yet approved | Blocks premature submission |
Use the table as the source of truth for automation. If an AI tool reads a mailbox or portal, have it propose updates to this tracker, then require a person to accept them. A statement called “purchase documents” may cover only one supplier; a payment-platform export may stop before the period end. The owner checks the coverage dates and contents before marking an item received. Log the evidence location rather than copying entire documents into the tracker.
Another illustrative failure: two team members maintain separate spreadsheets, one marking a bank feed complete and the other emailing the client for it. A single tracker with an update timestamp and named owner stops duplicated chasing. If the firm cannot provide that shared view, start with a simple controlled list before adding AI. The broader missing-records tutorial covers general request mechanics; the quarterly version must also account for update periods and cumulative totals.
Make the tracker usable by the person who covers an absent colleague. A status reading “chased” says little; log the item requested, channel, date, response due and exact evidence still outstanding. If the client replies to a different colleague, the cover reviewer can see whether that answer closes the original gap. An illustrative entry might read “6 October, portal request for card statement covering 1 July to 30 September; client replied 9 October with July only; August and September outstanding; follow-up owned by manager B.” That history is longer than a green cell, but it prevents an update being marked ready on the strength of a partial attachment.
Draft one question per missing item with AI
Feed AI only the minimum approved fields: client ID, preferred salutation token, period, missing items, last request date and your approved wording for the deadline. Do not paste tax references, full bank statements or private notes into an unapproved tool. Ask for a draft with a clear action, a reply route and no unsupported claim about penalties. A person must inspect every message, especially where the client has disputed an earlier amount. For guidance on handling sensitive practice data, read the accountant data-safety tutorial.
Draft a polite client email for a quarterly records request.
Use only these reviewed fields: period [period], missing [items],
internal reply target [date], secure upload route [route].
Ask one clear question for each missing item. Do not invent amounts,
penalties, tax conclusions or a claim that everything else is complete.
Use [first name] and [your name]. Return subject and body only.
Illustrative output: “Subject: Two records needed for your October review. Hello [first name], please upload your September purchase invoices and confirm whether the bank feed includes 2 to 5 October by 20 October. Use the secure portal link already in your account. If there were no purchases in September, please tell us. [your name]”. A reviewer should fix the subject if “October review” could imply the legal deadline; “quarterly records request” may be clearer. The reviewer should also ensure the dates and upload route correspond to that client.
A second sample is a landlord with a managing-agent statement that bundles rent and maintenance deductions. The request should say, “Please upload the statement showing rent received and the deduction detail for July to September,” not “send all rental paperwork”. If an AI classifier marks the statement as complete because the filename contains “quarterly”, a reviewer must open it and check that the period and deductions are actually present. Filename classification is an aid, never evidence of completeness.
Sort incoming replies into work, not into tax conclusions
Give replies labels such as complete file, partial file, question, no activity declared, wrong period, duplicate or needs specialist review. AI can propose a label and extract a plain-language summary; it should not silently alter the ledger. For each proposed complete item, require a source link and a coverage date. If a client writes “nothing new”, ask whether that means no income, no expenses or no additional documents. Official guidance says an update may still be needed when there was no activity in the period.
Illustrative before and after: a raw reply says, “Attached the shop statements. The card one isn't updated and I paid for tools on my personal card.” A poor AI summary is “all statements received”. A useful reviewer note is “bank statement attached; card statement pending; personal-card tool purchase needs date, amount, receipt and business-purpose check”. The next client request can then name the missing card statement and receipt. The purchase category and allowable amount remain for the qualified reviewer, not the model, to decide.
Watch for email threads where a client answers only the oldest request. The reply parser should compare the date and item IDs with the latest tracker, then flag any mismatch. If a client sends a scanned image with unreadable figures, mark the item “received, unreadable”, not complete. An AI extraction may produce convincing numbers from a blurred document. A second human or the client should confirm the source before the digital record changes.
Decide how to handle a file that arrives after the reviewer pack was prepared. Freeze a dated copy of the pack, identify which categories the new file affects, and send it to the reviewer as a change rather than quietly replacing an attachment. For example, a late credit note of GBP 420 may alter the sales total or may already be reflected in the accounting software. Ask the reviewer to compare the source, ledger entry and prior period. The AI can draft the change note, but it cannot infer whether the transaction was recorded twice. This late-evidence path matters more to accuracy than another generic reminder.
Follow a 24-client practice through one quarter
In an illustrative small practice, 24 clients need a quarterly chase. Six maintain their own compatible records and require only a confirmation and review. Twelve send most evidence through a portal. Six rely on mixed email and paper documents and need closer help. The practice tracks 96 expected evidence items across these clients. On the first internal review, 70 are complete, 16 are missing, six cover the wrong dates and four are marked “received but unclear”. These invented counts show the workload arithmetic, not a typical performance benchmark.
The owner groups the 26 unresolved items into 18 tailored client requests because some clients are missing two related items. AI drafts those requests from the tracker. A practitioner reviews every draft, spending an illustrative four minutes each, or 72 minutes for 18 messages. The practice logs the time it would otherwise spend composing them manually during this quarter, but also logs extra correction time. Speed without fewer unresolved items or cleaner updates is no win. If three drafts invent a penalty claim, that is a control failure and the prompt must be fixed before another batch.
After a week, suppose 12 clients answer. Seven provide complete evidence, three provide only part of it and two ask what a requested document means. The team does not send the same first reminder again. It sends a short clarification to the three partial responders and calls the two confused clients to explain the record in plain language. Four non-responders receive a follow-up with the outstanding item and the practice's internal deadline. Two clients need manager escalation because no complete record can be prepared by the review date.
At review, imagine the software shows cumulative sales of GBP 84,000 where the prior quarterly update showed GBP 63,000. The new period's sales are therefore GBP 21,000 if no earlier corrections exist. If the incoming sales export totals GBP 18,500, the reviewer investigates the GBP 2,500 difference rather than asking AI to make the numbers fit. It may be an earlier-period correction, a duplicate record or a missing export. Only the source documents and compatible software ledger can establish the cause.
Give the practice an escalation rule before the last week. At the illustrative 31 October internal review date, any client lacking a full bank feed goes to the manager; any client who has supplied records but has an unresolved categorisation question goes to the practitioner; and any client who has not replied at all gets a direct call. These are internal service choices, not official deadlines. The owner records the outcome and whether the client was offered a realistic alternative to the requested format, such as exporting a statement from their banking portal. Repeated failure to get a particular file is a process problem to solve with the client, not simply a cue for more AI-generated messages.
Put a human gate between records and submission
The reviewer checks client identity, business, chosen period basis, source coverage, digital record categories, duplicate transactions and the year-to-date movement. They then confirm what their engagement says about client approval. Send the quarterly update through compatible software, not a generated email. Keep evidence of the submitted status and receipt in the practice system. If a client has more than one self-employment business, the software and tracker may need distinct updates for each; do not roll separate records into one reminder merely because they share a person.
For a jointly let property, the permitted quarterly treatment can differ depending on whether the update includes expenses or income only. Do not build an automatic request for every expense category until a practitioner has checked the client's treatment. For a client using more than one software product, confirm the digital link rather than moving figures manually between products. These are setup decisions that a generic email-chase model cannot make. The client-books error-check tutorial is useful when the missing evidence reveals a ledger problem beyond the reminder itself.
In the first year of the scheme, official guidance says late quarterly-update penalty points will not be applied for the 2026 to 2027 tax year, while the updates still have to be sent before the return can be submitted. Do not use the temporary penalty treatment as a reason to delay the records chase. Confirm the current rule before mentioning penalties to any client. Ask a qualified tax adviser about a disputed obligation, exception or correction; the AI draft should never answer it as if it were final advice.
Measure whether the chase improved the quarter
After submission, compare the tracker with the previous quarter: percentage of evidence items complete by the internal review date, days between first request and usable response, number of duplicate chases, number of drafts corrected for factual errors, unresolved clients at deadline, and updates sent only after approval. Record the actual reviewer minutes as well as drafting time. A workflow that saves composition time but doubles verification time needs redesign.
Ask one client who struggled with the request to explain what was confusing. If “bank statement” meant a different file to them than to the practice, revise the checklist. If clients repeatedly upload everything at once near the deadline, give a monthly micro-request or show them how to maintain records throughout the period. The goal is a steady flow of accurate digital records and a reviewable quarterly submission, with AI used only for the matching, drafting and sorting work it can support.
Quarterly chase questions from clients
Do quarterly updates contain every receipt and invoice?
The update sent through compatible software contains totals for income and expense categories, rather than individual receipts and invoices. The underlying digital records and supporting documents still matter because those totals must be accurate and the client must retain the normal evidence used for a tax return. Ask for the missing source material when a category cannot be checked.
What if a client had no transactions in the period?
Do not remove the client from the chase solely because the period was quiet. Official guidance says an update is still needed when no income was received and no expenses were incurred. Confirm that the statement is complete for every relevant business, review the software totals, and record who approved the nil position before submission.
Can the practice send an update from an AI chat?
No. The AI assistant can draft questions, classify replies and prepare a review list. The submission itself goes through compatible Making Tax Digital for Income Tax software. A responsible person checks the digital records, cumulative totals and correct client account before authorising the update.
Further reads
- How Accountants Use AI to Explain Tax to Clients in Plain English — Explain a confusing request without jargon.
- Making Tax Digital for Landlords: Keep Rental Income Records With AI — See the property-record side of the workflow.
- Tax Services AI for a Small Practice: Intake, Chasing and Review Notes — Adapt the chase to other tax intake work.
- AI Engagement Letters for Accountants: Faster Drafts, Clear Scope — State who collects and submits each record.
- How Small Tax Practices Use AI Through the Busy Season — Plan capacity around repeated deadlines.
- AI Bookkeeping in the UK: What Works With MTD and VAT — Four MTD and VAT tests for any AI bookkeeping tool, applied to Xero, QuickBooks, Sage and Dext, plus a test pack that catches silent coding errors.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: official tax authority guidance, Send quarterly updates; Create digital records; Find out if and when you need to use Making Tax Digital for Income Tax; Find software that works with Making Tax Digital for Income Tax (read 4 November 2026).