AI for Tax Prep Offices: Seasonal Intake and Document Chasing
A seasonal intake agent builds a checklist for each client, chases missing documents and flags deadlines, while preparers review and sign every return.

An AI agent for a tax preparation office can run the seasonal intake work that fills the first weeks of a filing period: opening a file for each client, listing the documents that client needs, chasing what is missing, and flagging deadlines to the preparer who owns the return. The agent prepares the file and the follow-up. The responsible preparer reviews every figure, decides every treatment question and signs every return, and the agent has no authority to file anything on a client's behalf.
Key takeaways
- Intake is a strong first place for an agent because the work is repetitive, deadline-driven and easy to check against a finished file.
- A per-client checklist should start from the prior-year file the preparer already approved, then add the client's answers about what changed.
- Document chasing works best with a fixed cadence, one agreed channel per client, and a clear handoff to a person after a set number of attempts.
- Deadlines should be flagged from the office's own calendar for the preparer to confirm, never treated as the agent's own judgment.
- Preparers sign every return. The agent's permissions should stop at drafting, reminding, filing documents as received and routing questions.
What a seasonal intake agent actually handles
Intake is the stretch between a client saying yes and a preparer having enough material to start. It involves opening the client record, sending the first request, collecting uploads from email or a portal, sorting what arrives, noticing what does not, and keeping the preparer informed about which files are ready. In many offices this work is done by whoever has spare hands in the busiest weeks, and quality drops exactly when volume rises.
A seasonal intake agent takes over the repeatable parts of that stretch. It sends the checklist, files incoming documents under the right client and category, reminds clients about missing items on the schedule the office has approved, and writes a short status note for each file. It does not interpret tax law, estimate liability, suggest positions or contact tax authorities. Those tasks stay with the people who hold the credentials and carry the responsibility.
The boundary matters more than the feature list. Before any tool is connected, the office should write down what the agent may read, what it may write, what it may send and what it must never do. A short written boundary is easier to review in the middle of a busy season than a general instruction to be helpful.
How to build a per-client document checklist
A generic list of "tax documents" is not enough, because two clients with similar incomes can need very different files. A useful checklist is built for one client, from the evidence the office already has. The agent can draft it, but the preparer should approve the structure before it is sent.
- Pull the prior-year file the preparer approved, and note the income sources, deductions, entities and accounts it contains.
- Ask the client a short set of change questions: new employment, sold or purchased property, new dependents, new accounts, business activity, or life events the office has agreed to ask about.
- Map each answer to the documents that usually support it, and mark every item as required, conditional on an answer, or optional.
- Attach the checklist to the client record with a version note and the preparer's name as owner, so a later change is traceable.
- Send the list once, grouped by source (employer, bank, property, brokerage, personal), so the client can answer in one pass.
- Keep the checklist open until the preparer marks it complete, not until the client says they have finished.
The last step prevents a common failure. A client who says "that's everything" may have sent the wrong year or a partial statement. The preparer's review, not the client's confidence, closes the item.
How the agent chases documents without becoming a nuisance
Chasing is where clients either feel supported or feel pestered. The difference usually comes from three choices: how often reminders go out, how specific they are, and when a person takes over. An agent should send reminders that name the exact missing item, not a general request to "send your paperwork."
Illustrative scenario: A client with a small rental property sends a bank statement for the year but not the expense receipts for repairs. The agent files the statement as received, marks the repair records as missing, and sends one message naming only those records, with the office's upload link. Five days later, with no response, it sends a second and shorter reminder in the same thread. After that, it does not send a third message. It writes a summary for the preparer: which items are missing, what was requested and when, and whether the client has replied at all. The preparer decides whether to call.
This pattern keeps the agent useful without making it the office's voice in a tense conversation. Preparers stay the people clients speak to when something is unclear, when a number looks wrong, or when a client is frustrated.
When to flag a deadline and when to route a question
Deadlines are where a poorly scoped agent can cause real harm. The agent should work from the dates the office has entered and confirmed, and it should flag items early enough for a person to act. It should never present a calculated date as official without the preparer's confirmation, and it should never decide that a deadline has passed for a client's circumstances.
The following decision guide is a practical starting point. The office should adapt it to its own rules and review it with the responsible preparer before the season begins.
- If a checklist item is due within the window the preparer set, flag the file to that preparer the same working day.
- If a client has not answered two requests, route the file to the preparer with a summary, rather than sending another reminder.
- If an upload arrives with an unclear name, the wrong year or a partial page, file it as received pending review and say so in the status note.
- If a client asks how something is treated for tax purposes, do not answer. Route the question to the preparer with the client's exact wording.
- If a client asks about an extension, an amendment or a notice from a tax authority, route it immediately and do not send a substantive reply.
Each rule has a clear owner, which makes the agent's behavior auditable. When a deadline is missed in any office, the question afterward is always who was responsible for seeing it. A written decision guide answers that question before it is asked.
Keeping the preparer in control of every return
The preparer's signature is the control point. The agent can assemble the file, identify what is complete and what is not, and draft a summary of the client's answers. It should not sign, submit, transmit or modify a return, and it should not present its summary as a finished review. Every figure that reaches a return must pass through the preparer, and the office's records should show who approved what.
The NIST AI Risk Management Framework is a useful reference for this kind of governance. It frames AI risk as something to govern, map, measure and manage over time, rather than a one-time setup decision. For a tax office, that translates into named owners, a review schedule for the agent's behavior, and a record of incidents when the agent sends the wrong message or files a document under the wrong client.
Client documents also carry a security risk that is easy to overlook. A PDF uploaded by a client can contain text that looks like instructions, for example a note telling the agent to ignore its rules. The OWASP guidance on prompt injection recommends treating retrieved or uploaded content as data to be analyzed, not as commands to be followed, and keeping sensitive actions behind explicit checks. The intake agent should be built on that assumption.
Memory, permissions and client data
An intake agent needs memory, but the memory must be narrow and sourced. The useful memory is the checklist, the client's answers with dates and sources, the list of documents received, and the open questions routed to the preparer. Anything the client said in conversation should be labeled as the client's statement, not as a verified fact.
Permissions should follow the principle of least privilege. If the agent reads from a shared mailbox or document folder, it should use the narrowest scope the tool allows, and read-only access is preferable wherever writing is not needed. Access grants should be reviewed at the start of the season and again when staff change. The OAuth 2.0 authorization model is one common way to grant an agent limited access to a service without sharing a password, and the office should confirm that each grant matches the written boundary.
Retention is a separate decision. The office should define how long intake material is kept, who can see it, and how it is deleted after the engagement ends. Offices serving Brazilian clients should also review data protection obligations with the ANPD guidance in mind, and offices in the European Union should check the EU Artificial Intelligence Act overview for any obligations that apply to their tools. This is a conversation for the office's compliance owner and counsel, not something an agent can settle by configuration.
How to start before the busy weeks
The safest first engagement is narrow. Choose one role, such as intake for individual returns, and one set of documents. Define the memory the role needs, the tools it may touch, the approvals a person must give, and one observable success criterion, such as whether preparers find fewer missing items during their first review of a file. Run the agent against files from the prior season, compare its checklists with what the preparer actually needed, and fix the gaps before any client sees a message.
Be Human designs, implements and operates this kind of personalized AI infrastructure for companies. In a managed engagement, the first deliverable defines the single role, its memory and tools, its approval boundaries and its success criterion. Ongoing operation can include monitoring of the agent's messages, recurring evaluations against approved files, permission reviews, maintenance of integrations with the office's document tools, incident response when something goes wrong, and controlled changes that the preparer signs off on. The service is remote and available in English and Portuguese.
If you run a tax office and want to find the first task worth handing to an agent, take the free Be Human AI diagnostic.
Frequently asked questions
Can an AI agent sign or file a tax return for my office? No, and it should not be set up that way. In a responsible design, the agent drafts checklists, files incoming documents, sends reminders and writes status notes. The responsible preparer reviews the figures, makes every treatment decision and signs or submits the return.
How does an intake agent decide which documents a client needs? It starts from the prior-year file the preparer already approved, then uses the client's answers about changes to add or mark conditional items. The preparer approves the checklist structure before it is sent, and the preparer can edit any item on a specific client's file. The agent does not decide tax treatment on its own.
Will clients feel that an agent is chasing them too aggressively? They can, if reminders are frequent, vague or endless. A good design uses a fixed cadence, names the exact missing item in each message, stops after a small number of attempts, and routes the file to a person with a summary. Clients then hear from the agent for the routine requests and from a preparer when the situation needs a conversation.
How should my office start with an AI agent before the busy season? Start with one role and one document type, run the agent against files from the prior season, and compare its output with what the preparer needed. Write down the boundaries, review who has access to what, and define a success criterion the office can observe. A short diagnostic can help identify which task is the best first candidate.
Your next step
See where an AI agent fits in your company first.
Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.
Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.


