AI for Small CPA Firms: Chasing Missing Client Documents in Tax Season

A small CPA firm can chase missing documents with AI by turning each client's completed organiser into a list of outstanding items, then drafting a short reminder sequence from that list using placeholders for names and amounts, so that no return data enters the AI tool. Check consent under Section 7216 and your written security plan first.

Chasing is a good first AI job for a practice because it needs the least sensitive data. A reminder says "we still need your mortgage interest statement", not what the statement says. Keeping the data out is the design, and the sections below build the process around that principle.

Follow me on Instagram@sagnikteaches

Why the chasing job is the right place to start

During tax season, the bottleneck is rarely preparation. It is waiting. A preparer cannot start a return that is missing a brokerage statement, so the file sits and the same three clients are chased by hand in the same fortnight. Typical chasing costs a partner or admin ten to fifteen minutes per client per round once you count finding the organiser, working out what is missing, writing the email, and logging it. With 150 individual returns and an average of two rounds, that is 50 to 75 hours of senior time in a busy period, and the cost is higher still when the clients who answer fastest crowd out the ones who need the most nudging.

Connect on LinkedInSagnik Bhattacharya

AI helps with the middle of that job: comparing a checklist with what arrived, and writing the message in a tone that suits the client. It does not need a return, a taxpayer identification number or an amount. That puts it at the low end of the risk range, and it is why this task suits a firm that is still deciding how far to go. For other first jobs and the review gate each needs, see AI agents for CPA firms: starting jobs and a review gate for each.

Subscribe on YouTube@sagnikteaches

The data rule that makes the whole process safe

Section 7216 of the tax code makes it an offence for a paid preparer to knowingly or recklessly disclose information supplied for preparing a return, or use it for anything other than preparing the return, unless an exception or valid consent applies. The text provides for fines and up to a year's imprisonment on conviction. Whether sending a client's data to an AI vendor counts as a disclosure depends on the facts, the vendor's terms and the regulations, and that is a question for your professional adviser. Two firm rules reduce the question before it arises:

• The AI sees a checklist, not a return. Use client codes (C-0142) or first names only. No identification numbers, no dates of birth, no income figures, no account numbers, no addresses where you can avoid them.
• Anything that does involve return data stops at a line you define. Where a task genuinely needs it, use a tool under a vendor agreement you have reviewed, obtain the client's consent in the form your adviser confirms is valid, and note it in the file.

The second leg of the safeguard is your written information security plan, often called a WISP: a written document describing how your firm protects client data. For a tax practice it is a legal requirement rather than good practice: the Safeguards Rule made under the Gramm-Leach-Bliley Act counts tax preparation firms as financial institutions, whatever their size, and requires them to keep that written plan. Some of the rule's provisions are relaxed for firms holding data on fewer than 5,000 consumers, but the written plan itself is not one of them, and the rule also expects you to oversee your service providers, which now includes your AI vendor. Publication 5708, a free sample WISP written for small tax and accounting practices, is a sensible template to adapt. A WISP should say which tools are approved for client data, who has access, how data is deleted and what happens in an incident. Add AI to it explicitly: the approved tool, the plan, the prohibited data types, and a named owner. The AI readiness checklist for professional firms covers that groundwork.

Choose the tool for the job

You have three realistic routes, and none needs a large budget.

1. Your practice management software. Platforms such as TaxDome and Canopy, named here only as examples of the category, include client portals, organiser or document request lists and automatic reminders. If your system already tracks what is outstanding per client, the cheapest step is to switch on its reminder automation and use AI only to write better wording. Check what AI features your plan includes and what they send to which vendor; pricing and plans change, so look at the vendor's page.
2. A business plan of a general assistant. ChatGPT Business and Claude Team each need at least two seats, and, according to the vendors' pages as checked in November 2026, business plans do not use content for model training by default. Claude Team lists a standard seat at $20 per user a month on annual billing or $25 monthly. ChatGPT Business lists the same $20 annual and $25 monthly figures for a standard seat. Check the current figures before buying.
3. A spreadsheet plus the assistant. For a firm without a portal, a shared spreadsheet with one row per client and one column per expected document works, and the assistant drafts the emails from a pasted row with names replaced by codes.

Whichever you pick, keep a note of why, in your WISP. If you are comparing software options, the tutorial on practice software automation features sets out what to look for.

From organiser to missing-documents list

The organiser is the best starting point because it already states what the client claims to have. A good process has three stages.

Stage 1: a standard expected-documents list

Create a master list of documents by client type: employee, self-employed, landlord, investor, retiree, owner of a small company. For each, list the documents normally needed. An illustrative extract for an employee with investments: wage statement, interest statement, dividend statement, brokerage consolidated statement, mortgage interest statement, property tax bill, charitable receipts above your threshold, health coverage forms, prior-year return if new to the firm, and identification, which you collect only through the secure portal.

Stage 2: a per-client "expected" list, from the organiser answers

When the client returns the organiser, a staff member ticks which items the answers imply. "Yes, I sold shares" means the brokerage statement is expected. "Yes, I paid childcare" means the provider's details. This is a short checklist per client, and it contains no amounts. You can ask AI to produce a first draft from the organiser's yes/no answers if you strip the identifying and financial fields first, but I would begin with staff ticking it by hand for two seasons, because the cost of a wrong inference is a missing document discovered in April.

Stage 3: expected minus received, kept in one place

The comparison is a spreadsheet or portal feature, not an AI job. A formula or filter that shows items marked expected and not received is exact and free. Use AI downstream, where words matter: turning that list into a message that reads like a person wrote it, which client by client is what a hurried partner cannot do well at 9 pm.

The reminder sequence

An illustrative four-step sequence for a firm with a portal:

The prompt below drafts step 2. Replace the bracketed fields from your own list; do not paste any figures.

Write a friendly, brief reminder email from a small accounting firm to
a client. Use only the details I give. Client first name: [first name].
Outstanding items: [mortgage interest statement; brokerage consolidated
statement]. Upload link: [portal link]. Deadline for us to start your
return: [date]. State why we need each item in a few words, offer to
help find it, and mention they can reply to this email with questions
but should not email documents. Under 120 words. No threats and no
guilt. Do not include any numbers or personal details beyond those given.

An illustrative draft: "Hi [first name], a quick reminder that we still need two items to start your return: your mortgage interest statement (it supports your deduction) and your brokerage consolidated statement (it covers your share sales). If you can't find either, tell me which lender or broker and I'll explain where it usually sits in their online account. Please upload through your portal at [link] rather than email, since the portal is secure. We start returns for this round on [date], so the sooner the better. [your name]".

The phrase "should not email documents" is deliberate. If your clients habitually email PDFs, the reminder is also your chance to move them to the portal.

A worked example with numbers

An illustration. A firm of three partners and two staff handles 160 individual returns, with 55 clients regularly slow to send documents. In the first round, the firm uses its portal's expected-documents list and the AI-written sequence. Of the 160 organisers, 105 arrive complete within a week. The other 55 enter the sequence. After step 1, 22 complete; after step 2, a further 15; step 3 brings in 8; leaving 10 who get a phone call. Eight of those 10 resolve on the call; two need a second visit and a client meeting.

Time: building the master lists and the four message templates takes about a day in January. Each chase round thereafter takes about three minutes per client to check the list and approve the draft, against 10 to 15 minutes by hand. For 55 clients and two rounds, that is about five and a half hours against 18 to 27 hours. The saving of roughly 13 to 21 hours of senior time is the sort of figure a firm should measure for itself: record time spent on chasing for two weeks before you change anything. These numbers are an illustration, not a forecast.

Examples of messages for specific situations

A client who has gone quiet

For a client who ignored the first two reminders, ask the AI to draft a shorter, warmer message with a way out: "Hi [first name], I've sent a couple of notes about your documents and I don't want to be a nuisance. Is there something making this tricky, such as a lost password, a missing form or a busy month? If it helps, I can do a ten-minute call and we can work through it together. [your name]". The line offering a call matters; it changes the tone from nagging to help.

A client who sent the wrong thing

Staff note the problem; the AI writes the message. "Thank you for the upload. The file we received is the summary page of your mortgage statement; we need the annual interest statement (usually titled with the tax year and sent in January). If your lender's site is confusing, tell me the lender and I'll point you to the right page." You check the sentence about where the document sits is accurate for that lender, because the model does not know.

A deadline that is close

Use specifics, not alarm: "To finish by the filing date, we need your documents by [date]. If we can't have them by then, we can discuss an extension, which would give you more time to file but not more time to pay any tax due." Check that sentence with your adviser; the point is that your firm's real policy decides the wording, not the model.

A before and after: the same client, two seasons

An illustrative contrast. Before: the partner opens the client's file on a Friday evening, reads the organiser, remembers that the client sold shares last year, checks whether a brokerage statement has arrived, finds it has not, and types: "Hi, can you send the rest of your stuff?" The client replies "what stuff?", and a week is lost.

After: the staff tick-list shows two expected items, the portal lists them by name, and the reminder says which two and why, with the link. The client uploads one item the same day and replies about the other, saying the broker has changed and the statement is on another site. The firm's answer, written once and reused: "If you changed broker during the year, you may need statements from both; please upload whichever you have and tell me the dates of the change." That sentence now lives in the template library because the first season showed it came up often.

Keep a template library of such answers, tagged by situation: changed lender, new job mid-year, joint accounts, a late-arriving consolidated statement, a deceased relative's affairs. Each is a short paragraph you have checked once; the AI personalises it, you approve. After a season, the library holds 20 to 30 answers and most chase messages are assembled rather than written.

Wording for your security plan and engagement letter

These are starting points to discuss with your adviser, not finished policy text. For the WISP, a short AI section might say: the approved assistant is [tool and plan]; it may be used only for drafting communications and generic checklists; client names are replaced by codes; tax identifiers, account numbers, dates of birth and income figures must never be entered; use on personal accounts is prohibited; the practice manager owns the list of approved tools and reviews it every quarter. For the engagement letter, a sentence saying that the firm may use technology tools, including AI-assisted drafting, to prepare client communications, and that return information is not entered into such tools without the client's consent, sets expectations honestly. The tutorial on AI engagement letters for accountants goes into scope wording more fully.

Reviewing the season afterwards

In May, spend an hour on the numbers. How many clients entered the sequence, how many needed the call, and which documents were chased most? Compare the average days from organiser to complete file with last year, if you recorded it. Ask the AI to cluster your log's free-text notes into causes (portal login problems, confusion about statement names, clients waiting on third parties) after you strip names, and check its clusters against a sample you read yourself. The output feeds next year's organiser: reword the confusing questions, move the commonly missed items to the first page, and send a pre-season email in January listing the documents by name so fewer rounds are needed at all.

Review gate and quality checks

• A person reads the list before anything is sent. The list is the facts, and a wrong "still missing" item is the most embarrassing error because the client may have already sent it.
• Stop rules. The sequence stops automatically when an item is uploaded. If your tool cannot do this, a weekly sweep of the list removes finished clients before the next send.
• Tone check. Read two drafts per batch aloud. If they would annoy you to receive, change the prompt.
• No attachments or figures in messages. Messages point to the portal and nothing else.
• Weekly log. Track organisers received, items outstanding, reminders sent, replies, calls made. Watch which document types are missed most and add a line about them to the organiser itself next year.

What goes wrong in practice

An illustrative failure: a staff member pastes a client's full organiser PDF into a free account of a chat tool to "save time", including a tax identifier and bank details. The fix is policy and technical: a rule in the WISP naming the approved tool, a short training note, and ideally a block on personal accounts on firm devices. Another failure: a reminder claims a document is missing that the client uploaded to the wrong folder. The fix is a weekly check of unfiled uploads. A third: the sequence sends at 11 pm and clients reply to the sender address, which is a shared mailbox nobody watches. The fix is a reply-to address that a person checks and a send window in office hours.

For wider guidance on trust, the tutorial on AI mistakes that put an accounting firm's client trust at risk lists the patterns to avoid, and how small tax practices use AI through the busy season shows where else it fits once chasing works. If you want help setting it up on the tools you already own, that is what my AI implementation consultation covers.

Questions small CPA firms ask about AI and client data

Does Section 7216 stop me using AI at all?

No, but it makes you careful about what leaves your control. The law restricts disclosing or using return information for anything other than preparing the return. A reminder built from a document list with no return figures is low risk. Anything involving return data needs a hard look at consent and the vendor's terms, ideally with your adviser.

Can I paste a client's prior-year return into ChatGPT?

Treat that as a no unless you have settled consent, a suitable business agreement with the vendor, and a written policy that allows it. For chasing documents you do not need the return, only a list of what is outstanding. Use placeholders for names and amounts.

Which reminder channel works best?

Email is the default, a text message for a short nudge, and a phone call for the clients who ignore both. Ask clients in the engagement letter which channel they prefer, and never put sensitive data in the message body. Link them to your secure portal instead.

How many reminders are too many?

Three or four reminders spread over three weeks, ending with a personal call, is a workable pattern. More than that irritates people and trains them to ignore you. Stop the sequence the moment the client uploads the missing item.

Further reads

• Is AI Worth It for a One-Person Bookkeeping Business? — Whether AI pays off for a very small practice.
• Can AI Do My Business Taxes? What It Can Safely Prepare — What AI can safely prepare for business taxes.
• How Accountants Use AI to Spot Errors in Client Books — Catch errors in client books once the documents arrive.
• How Accountants Use AI to Explain Tax to Clients in Plain English — Explain tax points in plain English to clients.
• AI Receipt Capture and Bank Categorisation: A Bookkeeper's Setup — Cut the document handling that follows the chase.
• Rolling Out AI in a Bookkeeping Practice Without Losing Control — A staged rollout pattern from a bookkeeping practice.
• AI Receptionist for Accounting Firms: Tax-Season Calls and Cost — Monthly cost ranges from vendor pages, a peak-versus-quiet sum, and word-for-word tax-season scripts for an accounting firm's phone.
• AI Software for Tax Professionals: Run a Pilot and Measure Cost per Return — A four-week pilot plan for a small tax practice: one task, pass or fail rules, a consent and WISP check, and a cost-per-return formula.
• AI Tax Prep Software for CPAs: What to Test Before You Switch — A pre-switch test for CPA firms: ten filed returns, a scoring sheet, planted errors and the checks that separate Canopy Smart Prep partners, Juno and others.
• AI Tax Organiser: Build One That Chases Missing Documents — Turn a prior-year return into a client document list, track uploads row by row, and send reminders that name exactly what is still missing.
• AI for Tax Audit Support: Document Requests, Timelines and Draft Replies — Build a request tracker and reply templates for tax audits and enquiries, with AI drafting and the preparer reviewing every word before it goes out.
• Tax Services AI for a Small Practice: Intake, Chasing and Review Notes — A three-part workflow for a small tax practice: intake checklist, document-chasing emails and reviewer notes, with a named person owning each step.
• Tax Workpaper Prep With AI: Tie the Return Back to Source Documents — How AI can draft a source-to-return tie-out schedule for a tax return, and the reviewer checklist that proves each figure before sign-off.
• TaxDome Automations: What to Automate First in a Small Practice — Which TaxDome automations to set up first in a small practice: onboarding, document requests and reminders, in order, with what to leave manual.
• Claude (Formerly Cowork) for Accounting Firms: Safe Folder Tasks — Safe folder tasks for an accounting firm, the client files that stay out, how Claude's permission modes work, and a worked bank-reconciliation example.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Section 7216 statute text (Cornell Legal Information Institute), Safeguards Rule business guidance and Publication 5708 on WISPs, all checked 4 November 2026; ChatGPT and Claude pricing and privacy pages, TaxDome and Canopy product pages (4 November 2026).