How accounting practices can automate client chasing and MTD quarterly prep

Direct answer

Practices can automate the routine parts of client chasing and MTD quarterly preparation: reminders by client and deadline, an intake queue that checks which documents have arrived, and draft summaries for a person to review. Professional judgement, sign-off and advice stay with the accountant. With MTD for Income Tax reaching sole traders and landlords with qualifying income over £30,000 from 6 April 2027, the number of quarterly updates a practice handles is about to rise.

The split that works is simple. Rules handle anything with a date or a checklist: who owes what, by when, and whether it has arrived. A bounded AI step can help with reading what came in, such as sorting a pile of uploads into receipts, statements and letters, but it suggests rather than decides. A person reviews every figure before it goes to HMRC, and nothing is submitted without them.

We build the record-keeping and the automation; HMRC's guidance, your professional body and your own procedures confirm what you must do for each client. This guide is not tax or legal advice; HMRC's MTD for Income Tax guidance is the source for the rules. AutomateFlow hasn't published a case study with an accounting practice, so the workflow below is illustrative.

Client chasing by deadline and client type

Chasing works when it's driven by the calendar and the client list, not by whoever remembers. The quarterly dates are fixed:

Clients whose accounting period runs from 1 April to 31 March can choose calendar update periods with the same deadlines, but HMRC says the choice has to be made before the first update and can't be changed for the rest of the tax year. That is one more field to hold per client.

On penalties, HMRC says it won't apply penalty points for late quarterly updates in the 2026 to 2027 tax year. After that, each missed update earns a point, and four points bring a £200 penalty. The grace year is a good time to get the chasing right.

A chasing setup that holds up usually has:

• One client list with the fields that drive the dates: client type (sole trader, landlord, both, limited company), MTD start date, update periods, VAT quarter if any, year end, and who at the practice owns the client.
• A reminder sequence per deadline, for example three weeks out, ten days out and three days out, each with a message written for that client type. A landlord gets asked for rent statements and letting agent reports; a sole trader for sales records and receipts.
• A stop rule. The sequence stops the moment the expected records arrive, so nobody gets chased for something they've sent.
• An escalation to a person. After the last automatic reminder, the client goes on a named person's list with the history attached, for a call rather than another email.
• A log of every reminder sent and every reply, so the practice can show what it asked for and when.

This is the same pattern as staged reminders in any business with fixed steps. In the PaxBespoke case we built reminders at each stage of a bespoke wardrobe sale, from consultation to signature to deposit, with the team keeping control of prices and exceptions.

Document intake: what arrived, what's missing, what needs a person

Most of a quarter's chasing time goes on working out what a client has actually sent. An intake queue answers three questions for every client and period:

1. What arrived? Uploads to the portal, email attachments and bank feeds land in one queue per client. An AI step reads each document and suggests what it is (a bank statement, a receipt, a letting agent statement, an HMRC letter) and which period it covers. A rule files it only if the suggestion matches a document type the client is expected to send.
2. What's missing? The queue compares what arrived with the client's expected list for the period. The gap becomes the next reminder, written as a specific list ("the March statement for the second account") instead of "please send your records".
3. What needs a person? Anything the AI isn't confident about, anything that doesn't match the expected list, and anything from HMRC goes to a named person. So do duplicates, documents for the wrong period and anything that looks like a new source of income.

The rule of thumb: AI can label and summarise, rules can file and check, and a person handles everything that would change the advice or the numbers.

Quarterly update prep with a person signing off

Once the records are in, the preparation has a fixed shape. An illustrative version:

1. Pull the period. Transactions for the update period come from the client's software or bank feed into the practice's working file.
2. Suggest categories. Rules categorise anything that matches a known pattern for that client. A bounded AI step suggests categories for the rest, with its reason shown. Nothing is posted until a person accepts it.
3. Flag the odd ones. Rules compare the quarter with the same quarter last year and the previous quarter, and flag large swings, new payees, missing regular income and round-sum transfers.
4. Draft the summary. A short draft for the reviewer lists totals, the flags and what was assumed. The reviewer edits it; it is never sent to the client unreviewed.
5. Review and approve. The accountant reviews the figures and the flags, and approves the update in the practice's MTD-compatible software. The approval, the reviewer and the time are logged.
6. Submit and confirm. The update goes to HMRC from the software, the confirmation is filed against the client, and the client gets a short note.

Automation shortens steps one to four and makes step five faster to do well. It doesn't remove step five.

An AI policy for the practice (one page)

If anyone at the practice uses AI tools on client work, write down the rules before you automate anything. One page is enough:

• Approved tools. Which AI tools may be used, on which accounts (practice accounts only, never personal ones).
• Data allowed. What client data may go into each tool, and what never does: for example, no full bank statements in a tool whose terms allow training on your data.
• What AI may do. Label, summarise, draft and suggest. Never submit, never send to a client unreviewed, never give advice.
• Review. Who checks AI output, and how that check is recorded.
• Client information. How you tell clients that AI tools are used in preparing their work, if your engagement letters need to say so.
• Logging. Where prompts and outputs that affect a client's figures are kept, and for how long.
• Ownership. Who keeps the policy up to date, and when it's next reviewed.

Check the result against your data protection obligations and your professional body's guidance.

What stays with the accountant

Some work should never move to a rule or a model, however good the tools get:

• Deciding how a transaction should be treated when it's genuinely unclear.
• Anything that changes a client's tax position, or the advice you give them.
• Signing off a quarterly update or a return.
• Conversations about late records, disputes, or a client's circumstances changing.
• Taking on a new client, and the checks that go with it.

The point of the automation is to get the accountant to those decisions sooner, with the records already in order.

If your clients keep asking to buy a method your practice has built, such as a bookkeeping routine or a reporting pack, we also build software products with firms on that basis. Fees come first and any shared upside is agreed in writing. Talk about a partnership, or read more on how partnerships work.

Limitations

MTD dates, deadlines and penalty rules were checked on GOV.UK on 24 October 2026. HMRC updates its guidance often, and later threshold changes are not covered here. The workflow is illustrative, not a client project.

Related AutomateFlow pages

• Integrations between the tools you already use
• Records apps and reporting built around your systems
• PaxBespoke operations automation
• Employment Rights Act holiday and absence records
• What AI implementation costs in the UK
• Book a call