Client Tax Organiser: Build One With AI and Send It by Portal

Build a client tax organiser by giving AI your approved intake checklist, asking for plain-language questions per client type, and having a tax professional verify every branch. Send it through a secure client portal, request documents only where answers call for them, and chase from a reviewed missing-items list. A submission is complete only when its evidence is checked.

One long generic checklist tends to produce blank boxes and irrelevant uploads. The useful unit is a question linked to a reason, a document request, and a follow-up condition. A client who says “no” to rental activity should not see a rental schedule. A client who says “unsure” should be routed to a person, rather than silently treated as “no”.

Follow me on Instagram@sagnikteaches

Start with the firm's approved questions

Collect the practice's existing engagement checklist, prior-year organiser, document naming convention and portal instructions. Ask the responsible tax professional which questions are mandatory for everyone and which depend on the client's type or changed circumstances. Do not ask AI to invent legal requirements from a vague prompt. Rules and forms vary and change; the professional should check current authoritative guidance for the work being prepared. Keep a version date and reviewer name on the organiser.

Connect on LinkedInSagnik Bhattacharya

Group questions into a small core and conditional modules. The core might cover contact details, changes in household or business arrangements, income sources, deductions the client wants considered, payments already made and documents expected from third parties. Conditional modules might cover employment, self-employment, property, investments or a company. Keep the wording within the scope of the firm's engagement; do not promise that every possible tax issue is captured by one questionnaire.

Subscribe on YouTube@sagnikteaches

Write each question so a client knows what to do. “Any investment events?” is too broad. A clearer illustrative question is: “Did you sell or transfer any shares or other investments during the period? If yes, upload the transaction statement showing dates, amounts and fees. If unsure, tell us what account you used.” The wording gives a yes/no/unsure path and describes the evidence without deciding the tax treatment. The professional may add or remove detail for the client's actual circumstances.

Another example: “Did you start or stop renting out a property during the period?” If yes, ask for the start and end dates, income summary, expense records and a note about any major repairs or improvements. Do not combine all costs into one unlabeled upload box, because the distinction may matter later. If the client did not rent property, hide the whole module. This kind of conditional design is separate from automated missing-document chasing, which starts after the organiser has been issued.

Let AI draft branches, then audit them

Give AI a sanitised version of the approved checklist and a table of client types. Ask it to rewrite questions for clarity, propose conditional branches and identify where it needs a professional decision. Do not include real names, identity numbers, return figures or documents while designing the template. Review every proposed question for ambiguity, scope creep and false certainty. A branch that asks “Were you self-employed?” may be unclear to a client who did occasional freelance work; the wording should describe activity, not rely on a label.

Using only the approved question list below, draft a client organiser.
For each question give: plain-language wording, answer choices,
document requested if yes, when to show the question, and a flag
for professional review. Include an UNSURE option. Do not add tax
rules, thresholds or form names that are not in the approved list.
Return a branch table and three sample client journeys using
fictional, generic circumstances. Do not request identity numbers.

An illustrative output might put “Did you receive any income from a side business?” under the self-employment module and request a sales summary and expense records if yes. The reviewer notices that “side business” could exclude a client who regards substantial freelance work as casual. The final wording becomes: “Did you receive payment for work you did outside employment, including freelance or occasional work?” The document request stays conditional. That single wording change can prevent a missing income stream.

Check the negative path too. Suppose AI drafts “If no employer document is uploaded, mark employment income absent.” That is unsafe: a client may have worked but not yet received the document. The branch should ask whether employment occurred, then record the document as expected, unavailable or not applicable. The portal should permit a client to submit the questionnaire while the document is pending, but the preparation status must remain incomplete. Silence and absence are different data states.

Have a second reviewer test at least three journeys: a simple repeat client, a client with a new income source, and a client whose facts are uncertain. Read the exact screen each would see. If a “no” answer hides a required follow-up, fix the branch. If the questionnaire requests the same document twice, consolidate it. If a client can reach the final submit screen without answering a critical question, set a validation rule or a manual review flag.

A complete organiser for an illustrative practice

Consider an illustrative practice with 120 annual clients: 70 mainly employment-based, 30 self-employed and 20 with investment or property activity. The practice used to send the same 45-question PDF to everyone. In a pilot design, the core has 12 questions, with 10 employment, 14 self-employment and 12 property or investment questions displayed only when relevant. These numbers are examples of questionnaire design, not a claim that any set is legally complete. The professional checks the content against current requirements before release.

One illustrative client has employment income, sold an investment and did occasional paid design work. The core asks about changed income sources, and the answers open the employment, investment and self-employment modules. The portal requests the employment statement, investment transaction history, sales records and relevant expenses. It does not ask for property documents. The client uploads four files, marks one employment document “not yet available”, and leaves a note about an investment transfer. The organiser is submitted but not complete; the preparer has two specific issues to resolve.

The practice's reviewer can quantify the intake without confusing it with accuracy: 12 core answers received, three relevant modules opened, four files uploaded, one expected file missing, one transfer query open. A dashboard may show “80% of requested items present”, but the return cannot proceed where the missing item is material. Use an item-level status: received, readable, complete for purpose, requested again, or not applicable. An uploaded photograph of half a statement counts as received but may not be complete for purpose.

In the same illustration, a second client with only employment income answers “no” to other activity. The portal shows the core and employment module, not the extra 26 questions. The reviewer still checks whether last year's known property activity was addressed. If last year's file shows rent but this year's organiser says “no property”, ask whether the property was sold, vacant or overlooked. AI can flag a difference between last year's category list and this year's answers, but a professional must phrase the follow-up and assess the response.

A third illustrative client selects “unsure” for a one-off payment from a former customer. The portal should create a review question rather than assigning that payment to a tax category. The preparer asks for the payment date, amount, reason and supporting correspondence. The answer may change which module is required. This illustrates why a good organiser has an uncertainty route; forcing clients to choose a category they do not understand creates false certainty.

Make the portal request easy to complete

Build the final organiser in a client portal approved by the practice. Use a clear invitation with the period, deadline, estimated effort if measured by your own pilot, and the instruction to answer “unsure” where appropriate. Show a progress indicator if the portal supports one, but do not rely on it as proof of completeness. Make the questionnaire usable on a phone, with short labels, readable instructions and a way to save and return. Offer an accessible alternative through a staff member for clients who cannot use the portal.

Label upload requests by purpose rather than “other files”. An illustrative request might say: “Investment activity: upload the statement covering the sale and purchase details for the period. If the account provider offers a full transaction export, upload that too.” Tell clients which formats the portal accepts, how to photograph multi-page documents and how to check an upload succeeded. Avoid telling them to place sensitive documents in ordinary email replies. Keep notification emails free of private figures.

Before launch, test the portal with dummy client accounts. Follow every branch, upload a deliberately unreadable file, close the browser halfway, and check that a returning user can finish without starting over. Confirm that staff see the answer and its attachment together. If the portal separates files from questions, assign an internal naming rule and a review queue. A document with a filename like “scan3” is difficult to match to an answer unless the portal preserves the association.

The invitation can be brief: “Hello [first name], your organiser for the current period is ready in our client portal. Please answer the questions and upload the requested records by [date]. Select ‘unsure’ wherever you need us to check something. We will review the submission and send one list of any remaining items.” Review the text against your actual service process. Do not promise a first-time complete return merely because the form has conditional logic. For a wider intake workflow, see automating accounting practice onboarding.

Use a staged reminder sequence

Reminders should reflect the client's status, not repeat the original invitation. A reasonable illustrative sequence is an invitation, a gentle reminder several days before the agreed deadline, and a specific follow-up after a partial submission. Set timing based on the engagement, not an invented universal rule. Pause automated messages when the client has asked a question or the portal has an access problem. Let staff override the sequence for sensitive circumstances.

A client who has not opened the organiser might receive: “Hello [first name], your organiser is waiting in the portal. Please start with the core questions; you can save and return. If access is difficult, tell us and we will help.” A client who completed the answers but missed two attachments should receive a different message naming the two missing categories, with a secure portal link. Do not mention the details of income or identity documents in the email subject line.

In an illustrative batch of 20 pilot clients, eight submit everything requested, seven submit with one or two open items, three have not started, and two report portal access trouble. These are fictional counts for planning. AI can segment the reminder queue into four groups, but the reviewer should verify the status before sending. For the seven partial submissions, generate individual missing-item lists from the approved checklist. A blanket “you have not completed your organiser” message would frustrate clients who did most of the work.

Keep a simple record of invitation date, reminder date, client response and staff action. If a client says a document will arrive later, set a new follow-up date and mark the item pending rather than sending the same reminder daily. If a client uploads a document after the cut-off, notify the preparer and review any work already done that depended on the missing information. A dedicated missing-records workflow can handle that queue without turning the organiser into a barrage of messages.

Protect return data and record consent decisions

Tax records can contain identity, financial and household information. Keep template design separate from client data: draft generic wording with AI first, then use only tools and data flows the practice has approved for real submissions. Document who can see portal responses, how access is removed, how long files are retained and what happens if the wrong person receives an invitation. A written information security plan, often called a WISP, should address these handling steps. One official WISP template for tax and accounting practices, prepared by a government and industry security partnership, includes sample attachments for record-retention policies, security breach procedures and a list of the staff authorised to access client personal information. Its requirements section also expects a firm to select service providers that can maintain appropriate safeguards, with the contract requiring it, and to use multi-factor authentication for anyone accessing its information systems. The portal provider and any AI service that touches return data belong on that provider list, with their contracts on file. A qualified adviser should check any applicable consent or disclosure duties before client data is sent to a third-party AI service.

An illustrative failure is a shared family email address used for two clients. Sending a portal invitation that exposes one client's private questionnaire to the other would be a confidentiality problem, even if the form itself is accurate. Verify recipient identity and portal access before sending. Another failure is pasting a completed return into a consumer AI chat to “improve” questions for next year. Use a sanitised pattern instead and follow the practice's approved data controls. See questions to ask before using AI with client data.

Consent is not a single checkbox that solves every use. Tell clients what the portal does, what information you need and how the practice handles it. If a separate authorisation is required for AI processing under the applicable professional rules or engagement terms, obtain it in the correct form and record it. Do not ask AI to decide whether a consent is legally sufficient. The professional who owns the engagement should approve the wording and the processor list.

Accept the organiser only after a human check

When the client presses submit, compare answers with the expected document list and prior known facts. Check each uploaded file opens, covers the correct period, contains all pages and matches the question it was attached to. Mark unclear answers for a client query. Do not infer “none” from an empty field or treat an irrelevant upload as fulfilment. The reviewer records accepted, partial or awaiting clarification, with a dated note.

For the illustrative client with investment activity, the file named “statement.pdf” opens but covers only the final quarter. The reviewer asks for the full-period transaction history and explains why the current file is insufficient. The organiser remains partial even though the upload box is ticked. For another client, a self-employment expense total of $12,400 has no underlying detail; the reviewer asks for a category breakdown and records, rather than simply copying the number into a return. These are precisely the gaps a portal progress bar cannot detect.

Give the reviewer a short acceptance record for each requested item: question ID, answer, file name, period covered, page count if relevant, readable yes or no, further query and reviewer initials. In the illustrative investment case, the item might read “INV-04; sale reported; statement.pdf; final quarter only; readable; ask for earlier transactions; pending”. This record is more useful than a single “documents complete” toggle because it explains what the preparer may rely on. If a client replaces the file, keep the earlier version in the controlled record and note which version was accepted.

Check the request list against the engagement scope before marking the intake ready. An organiser may collect enough evidence for one return while a related business account remains outside the engagement. The reviewer should state which file the accepted documents belong to and which unanswered questions require a separate conversation. For example, an illustrative client uploads a spreadsheet of business receipts and asks for bookkeeping corrections in a free-text box. The organiser records the request and alerts the engagement owner; it should not quietly expand the preparation task or let AI classify the spreadsheet as reviewed accounts.

Use a small release test before sending the next version to all clients. Give three staff members dummy profiles with different answers, then compare the portal's requested-document list with the approved question map. If one dummy client reports an investment sale but the portal asks only for year-end holdings, correct the conditional rule and repeat the test. Record the organiser version used in the test and the date it was approved. That simple trace helps explain why two clients received different questions when the template changed mid-season.

Review the pilot after the first group of clients. Count unclear questions, duplicate uploads, “unsure” selections, missing evidence after the first submission and staff time spent on follow-up. Rewrite questions that repeatedly confuse clients, then have the tax professional approve the new version. Keep a change log so a later reviewer knows which organiser version each client answered. The measure of success is a clearer, better evidenced intake and fewer avoidable exchanges, not a claim that AI made the tax decision.

Questions about client document intake

Can AI decide which tax forms a client must supply?

AI can draft a candidate checklist from a template, but a qualified tax professional must approve it against the client's circumstances and current requirements. Use conditional questions so the organiser asks for evidence only when relevant. Keep an option for clients to say that they are unsure and ask for help.

Should clients email their records if the portal is awkward?

Use the firm's approved secure transfer route for sensitive documents and give clients clear instructions for using it. A reminder email can point them to the portal without attaching private records or inviting a reply with identity documents. Offer an accessible alternative through the practice when a client cannot use the portal.

What if a client submits an organiser with blank answers?

Treat it as partially complete. Review mandatory fields and expected attachments, then send a short list of specific gaps through the agreed channel. Do not infer a negative answer from silence. Record the client's confirmation and the reviewer who accepted it before beginning the dependent preparation work.

Further reads

• AI Tax Return Prep for Small Practices: Extract, Review, Sign Off — Carry accepted records into a reviewed preparation workflow.
• ChatGPT Prompts for Tax Preparers: Client Emails and Checklists — Draft related client emails and checklists.
• Tax Services AI for a Small Practice: Intake, Chasing and Review Notes — Connect organiser responses to intake status.
• ChatGPT for Tax Preparers: How to Redact a Return Before You Paste — Reduce exposure when testing AI on return data.
• How Accountants Use AI to Explain Tax to Clients in Plain English — Explain later findings in client-friendly language.
• Companies House ID Verification: A Client Chaser Workflow for Accountants — A tracking sheet, a five-status chase sequence and email templates for getting every director and PSC verified, with AI kept away from ID documents.
• Form 8867 Due Diligence With AI: Interview Notes and Checklists — For practice owners supervising Form 8867 files: structure interview notes with AI, spot gaps before filing and keep records a reviewer can follow.
• Year-End Tax Planning Letters With AI: Personalise Without Errors — A tax practice's method for personalised year-end letters: data merged from your records, AI used for wording only, every figure checked before sending.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Publication 5708, Creating a Written Information Security Plan for your Tax and Accounting Practice (Rev. 8-2024), read 4 November 2026; the sample organiser and figures are illustrative, not tax advice.