MTD for Income Tax: Triage Your Client List and Send Reminders With AI

To triage an accounting practice's MTD for Income Tax client list, combine each client's verified gross self-employment and property income with the relevant return year, then assign a start date, software status and named reviewer. Use AI to sort incomplete records and draft reminders, but have a practitioner confirm scope and approve every client message.

The threshold test is based on qualifying income before expenses, not profit or the most recent bank balance. Current official guidance stages clients above GBP 50,000 from 6 April 2026, above GBP 30,000 from 6 April 2027 and above GBP 20,000 from 6 April 2028, each measured against the specified earlier return. Check that guidance again before a bulk send.

Follow me on Instagram@sagnikteaches

Build one row per client before asking AI to sort anything

Start with your practice management list, not a prompt asking a model to guess who is affected. For each sole trader or landlord, create a restricted working row containing the internal client ID, responsible manager, income sources, filed return year, gross self-employment amount, gross property amount, date of source check, known correspondence, software product, digital-record status, contact preference and next action. Keep the actual return in your normal system; the AI export should carry only the fields needed for sorting. A row without an evidence date is not ready for an automated category.

Connect on LinkedInSagnik Bhattacharya

Separate a client who does not appear to meet a threshold from one whose data is absent. They require different messages. A blank income cell must never be interpreted as zero, and a note saying “turnover 40k” is insufficient if it is unclear whether that excludes rent or refers to profit. If your list has several businesses under one person, add the relevant amounts to the person's qualifying-income review while keeping distinct bookkeeping workstreams visible. Refer disputed or unusual income treatment to the tax practitioner rather than allowing the sorting rule to settle it.

Subscribe on YouTube@sagnikteaches

That filled row is illustrative, not a client story. The trading and property amounts total GBP 52,600, which is above the first published threshold, but the row still needs the practitioner to confirm the return, any corrections and applicable exceptions. The “spreadsheet” note identifies a software question; it does not mean the client must replace the spreadsheet. Official guidance says spreadsheets can be used with linking or bridging software, provided the complete software route does the required work.

Turn the official stages into a reviewable triage rule

Use a deterministic spreadsheet formula or practice-system rule for the simple comparison. Have AI explain exceptions and propose tasks around that rule. The published stages compare the 2024 to 2025 return with GBP 50,000 for a 2026 start; the 2025 to 2026 return with GBP 30,000 for a 2027 start; and the 2026 to 2027 return with GBP 20,000 for a 2028 start. These are “more than” thresholds. Keep the source year beside the number so a future team member cannot sort a 2026 amount against the 2024 threshold. Sum only the qualifying streams: official guidance leaves employment pay, a partner's profit share, dividends and pensions out of qualifying income, so a rule that totals every income box on the return will pull in clients who are not affected. For a jointly owned let property, only the client's share counts.

A useful queue has four states: confirmed in scope now, likely to enter at the next stage, review needed and currently outside the published stages. “Review needed” covers missing returns, mixed income whose source is uncertain, amended returns, potential exemption, conflicting correspondence and a software route that has not been checked. “Outside” means the evidence supports that classification as of a stated date; it does not mean the client can be ignored forever. A yearly return can change the position, so add a recheck event after filing. Because the first stage has already begun, the guidance now says those clients should have started from 6 April 2026 and can still sign up; a confirmed first-stage client with no sign-up is this week's task, not a planning item. Sign-up needs Self Assessment registration and a return submitted in the last two years, so check both before promising a client a start date.

For an illustrative edge case, imagine a client with GBP 29,950 of gross trading income on the relevant return and GBP 350 of gross rental income. A practice list filtered only on trading income would miss the combined GBP 30,300. The correction is to store each qualifying stream separately, calculate the total once, and retain the component values for the reviewer. For another client, a gross figure of GBP 50,000 exactly must not be silently rounded upward. Ask the reviewer to confirm the exact return amount and the official “more than” wording.

Run the current official eligibility checker where a case is unclear. Record the result or reference in your practice system. AI is useful for forming a question such as “Which income components are missing from this row?” It should never be asked “Is this person legally required to join?” without a human checking the underlying return and guidance. If the matter turns on an exemption or an amended return, ask a qualified tax adviser within the practice to decide.

Make software readiness a separate traffic light

A client may be in scope but operationally ready; another may be out of scope for now yet have records that need attention. Do not collapse these into one red or green status. Software readiness should ask whether the client has chosen a product, whether it supports their mix of income, whether the practice has access where needed, whether digital records are being created, and whether the route can send quarterly updates and the year-end return. The official software finder lists recognised options, but selection still depends on the client's income sources and agent arrangement. Agents see the full list with filters rather than the personalised questions a client answers, and the guidance allows only one product for each separate submission: a landlord cannot split property quarterly updates across two products.

For example, mark “amber” when a landlord says “I use an app” but cannot say whether it supports property income or whether the accountant can access the underlying digital records. Ask for the product name, the module or service used, and a test of the hand-off. Do not ask for passwords in an email. A separate case might use a spreadsheet plus bridging software; mark it ready only after a team member checks the link and the intended submission workflow, not simply because the file is electronic.

Record a small set of evidence codes: S0 for no product named, S1 for product named but fit unchecked, S2 for fit checked but onboarding incomplete, and S3 for an end-to-end route demonstrated. These codes are a practice convention, not a tax authority classification. Give each code an owner and a next action. “Choose software” is vague; “send the two products you are considering by Friday so we can check property support and access” is actionable.

Use AI to find missing fields, then review its queue

Export only the minimum data needed for this sorting task into an approved AI environment. Replace names with client IDs and remove addresses, account numbers, tax references and free-text notes that might contain sensitive details. If your firm has not approved a tool for client data, run the prompt on invented or thoroughly de-identified rows and apply the resulting rule in the practice system. The related tutorial on accountants and client data in AI tools covers the broader control decision.

Task: Review this de-identified client readiness table.
Return one row per client ID with: missing field, likely triage queue,
software question, suggested next human action and reason.
Use only the values supplied. Do not infer a blank as zero.
Do not decide exemptions or final tax scope.
Flag any row where income components or source return year conflict.
Published stage rules supplied by the practice: [insert checked rules].

Illustrative sample output: “C-184: likely current-stage review; gross components total GBP 52,600; source return present; bridge unselected; ask manager to confirm the return and schedule software-choice contact.” That is helpful because it separates arithmetic, uncertainty and action. A bad output would say “C-184 is enrolled” merely because the total exceeds a threshold. Edit it to “possible scope, practitioner confirmation pending”, then check the status in the official process.

Inspect a random sample from every queue and all rows containing blanks or exceptions. Compare the AI grouping with a manual spreadsheet filter. Save the reviewed list, prompt version and reviewer date. If the model confuses source years in even one row, stop bulk use and revise the task instructions or return to a formula. Your objective is a dependable worklist, not an elegant narrative from the model.

Send a specific first reminder for each readiness gap

One generic “MTD is coming” mail rarely moves a client to the next step. Draft messages from the verified queue: income confirmation, software decision, digital-record setup, agent access or missing correspondence. Use a template with approved factual text and a single request. Do not put an uncertain threshold determination in the subject line. A human sender should check the recipient, status, dates, attachments and wording before dispatch. For a more general document-chase process, see chasing missing client records with AI; here the point is readiness for this tax change.

Illustrative message for a software gap: “Subject: Your digital record-keeping setup. Hello [first name], we have reviewed the information currently on file and need to confirm how you intend to keep and submit your income records. Please reply with the name of your current bookkeeping product, whether it holds your trading and property income, and whether you want our practice to prepare the quarterly updates. We will check the route with you before confirming the next step. [your name]”. The message asks for evidence without presenting an unreviewed legal conclusion.

Illustrative message for a missing figure is different: “Hello [first name], we cannot finish our readiness review because the gross property-income figure on your last return is not in our working list. Please send the filed return or authorise us to check the copy already held in your portal. We will then confirm what action is required. [your name]”. If the practice already holds the filed return, the internal owner should first retrieve it rather than burden the client. The message should be generated only after that internal check.

For an apparently ready client, ask for a short confirmation of responsibilities: “We have recorded [product] as your software and [role] as the person keeping records. Please confirm who will review quarterly totals and who will approve submission.” Do not assume that an existing agent authorisation automatically gives a particular staff member access to every data source. Place the answer back into the readiness row and advance the status only after a reviewer checks it.

One more illustrative failure appears when the practice sends a “please select software” message to a client who already purchased a product after the list was exported. The client replies with a complaint, and the team discovers its mail merge used a two-week-old status field. Put the last client contact date beside the evidence date; suppress a draft when a newer email or portal note exists. A manager can then ask the narrower question that remains unanswered, such as whether the new product can handle both the client's income streams. This check protects the relationship while keeping the queue useful.

Work through a 60-client practice queue

Consider an illustrative two-manager practice with 60 sole-trader and landlord clients. An initial export has 14 clients above the first-stage threshold on the matching return, 11 with income near a future threshold, 25 apparently below the published stages and 10 with incomplete or conflicting figures. These counts are invented for planning; they are not a claim about typical practices. The team first verifies the 14 and the 10 exceptions, then checks the 11 future-stage cases against the correct return year. The 25 lower cases get a recheck date after their next return.

Within the 14 current-stage records, suppose four have a complete software route, six use a product whose property support or practice access is unconfirmed, and four still send paper receipts. The practice creates three separate tasks rather than blasting 14 identical emails. A manager spends an illustrative 12 minutes checking each high-priority return and software record: 14 × 12 minutes equals 168 minutes. The 10 exception records may need longer, so the team reserves review time rather than claiming AI has removed it. AI drafts 18 messages for review: six software questions, four record-setup notes and eight income or evidence requests. The sender logs delivery and reply status in the same list.

Suppose nine clients answer within the first reminder period, five provide enough information to move from amber to green, two expose a mismatch in the source year, and two still need a call. That result is more meaningful than counting emails generated. The source-year mismatches trigger a correction across the entire list: filter for every row with a threshold but no return-year field. The team can then see whether the triage rule remains trustworthy.

Another practical case is a client with two distinct self-employment activities. The person-level qualifying-income review can add the relevant gross amounts, while the record-keeping plan needs to preserve separate business records and updates where the rules require them. A spreadsheet with one “all income” column hides that operational issue. Flag it for a practitioner and verify the structure against official guidance before onboarding the software.

Schedule reminders from status changes, not from a static calendar

Attach a next-action date to each row, but trigger the content from what has changed. If a client supplies a product name, the next message should ask about fit and access; it should not repeat “choose a product”. If they upload the missing return, re-run the scope review before any reminder about joining. Use a three-contact sequence as a practice choice: initial request, a short follow-up that repeats the outstanding item, and a personal call or manager review for non-response. Do not claim this sequence is mandated by the tax authority.

Set a send gate: every generated draft must display client ID, verified category, evidence date, requested action and reviewer initials. A mail merge should reject rows missing any of these. If the stage or deadline has changed, update the approved source text once and re-review affected drafts. The practice can measure useful progress through the percentage of current-stage clients with confirmed software and a named record keeper, the number of unresolved exceptions, and reply rates by request type. Email opens alone do not tell you whether the client can submit an update.

After onboarding, hand the client to a separate quarterly collection workflow. The quarterly records chase tutorial deals with obtaining evidence before each update; this triage list should remain focused on who needs preparation, what is missing and who owns the follow-up. If a landlord wants to understand their own ledger, point them to the landlord record-keeping tutorial rather than sending an internal practice checklist.

Audit the queue before each bulk send

Choose a reviewer who did not create the export to check five things: a sample of income totals against filed returns, the threshold and source-year pairing, all exception decisions, every software status promoted to ready, and the actual rendered email for each contact group. Check that sample messages have no other client's information. Do one test send to an internal address and inspect subject, salutation and links on a phone. Store approval with the queue version so the practice can reconstruct what was sent.

Revisit the official guidance each time a new stage starts and whenever a client challenges the classification. The tax authority's letter can help, but absence of a letter does not remove the client's responsibility to check. If the list contains an uncertain exemption, unusual property arrangement or an amended return, have a qualified adviser decide it. AI can keep the questions and reminders moving; accountability for tax scope and submission stays with the practice and client.

Client-list decisions that need a second look

Should I mark a client as outside scope when their income is exactly the threshold?

The published stages use income that is more than the stated threshold. Record the exact amount and source year, then confirm the result against current official guidance before changing a client's status. An amount exactly on the line is a review item, especially if the source return has been amended or an income stream was omitted.

Can AI decide whether a client is exempt?

No. AI can flag a possible exemption and prepare the facts a reviewer needs, but the practice should check the official criteria and any correspondence before recording an exemption. Keep the evidence, decision date and reviewer in the client record. Do not let a model turn a vague note about computer use into a final tax conclusion.

What if a client already uses a spreadsheet?

A spreadsheet can remain part of the record-keeping process if the client has compatible software that links to it for the required submissions. Ask which product will connect the records, which income sources it supports and who will send updates. A file labelled digital does not prove the entire route works.

Further reads

• Tax Services AI for a Small Practice: Intake, Chasing and Review Notes — Extend the chase to intake and review notes.
• How Accountants Use AI to Explain Tax to Clients in Plain English — Explain obligations clearly to clients.
• AI Engagement Letters for Accountants: Faster Drafts, Clear Scope — Clarify the practice's role in client paperwork.
• How Small Tax Practices Use AI Through the Busy Season — Plan review capacity around recurring deadlines.
• How Accountants Use AI to Spot Errors in Client Books — Check the books after records arrive.
• AI Powered Bookkeeping Platform: How a Practice Chooses One — A test-file method for picking an AI bookkeeping platform: three real client months each, scored on accuracy, review time and cost per client.
• 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, Find out if and when you need to use Making Tax Digital for Income Tax; Using Making Tax Digital for Income Tax for the first time; Work out your qualifying income for Making Tax Digital for Income Tax; Find software that works with Making Tax Digital for Income Tax; Choose the right software for Making Tax Digital for Income Tax; Send quarterly updates (read 4 November 2026).