Skip to content
be-human.ai
← Back to the journal

AI by industry

AI for Tax Preparers: Document Intake and Extension Follow-Up

AI agents for tax preparers can run intake, missing-document chasing and extension reminders under approval rules, while filing decisions stay with staff.

Be Human typographic cover with the title: AI for Tax Preparers: Document Intake and Extension Follow-Up
Conceptual illustration created with AI · Be Human

AI agents for tax preparers work best on the steps around the return: collecting client documents, chasing what is missing, and sending extension reminders that follow the firm's own calendar. A well-scoped agent can run those tasks through approved templates and logged actions, while preparers keep every judgment about deductions, positions and filing. The first project should be narrow enough that a staff member can review every outbound message and every status change.

Key takeaways

  • Start with intake and follow-up, the work that repeats every season and rarely needs professional judgment.
  • Write down what the agent may read, send and change, and what it must route to a preparer before connecting any tool.
  • Track each client document as a status with an owner and a timestamp, so the agent can tell what is missing and what was already sent.
  • Tie extension reminders to a deadline calendar that a licensed person has confirmed, and stop the agent when that calendar is unclear.
  • Keep filing decisions, positions, signatures and taxing authority contact outside the agent's scope in the first engagement and in every later one.

Where an intake agent starts and where it stops

In many tax offices, the bottleneck is not the return itself. It is the document that arrives as a blurry photo, the form the client forgot, and the email that asks the same question three times. This work is repetitive, rule-based and easy to reverse, which makes it a sensible place to start with an AI agent.

The same inbox also holds moments that need a licensed person. A client may describe a life event that could change how the return is prepared, a missing form may point to a larger issue, or a message may ask the firm to take a position. The agent's job is to notice those moments and route them with the relevant facts attached, not to resolve them.

Write the boundary down before any connection is made. A one-page scope should name the portal or inbox the agent can read, the templates it may send, the status fields it may update, and the categories it must escalate. Anything that changes a return, accepts a document as final for filing, or contacts a tax authority belongs outside that scope.

A step-by-step plan for document intake

A reliable intake flow is built before the agent is switched on. The steps below keep the firm in control while the agent takes over the repetitive parts.

  • Map the current path from first contact to a complete file: which channels clients use, which documents each return type usually needs, and who checks completeness today.
  • Write a checklist for each client type, using plain names a client will recognize, such as last year's return, a year-end brokerage statement or a property tax bill.
  • Give every item one of five statuses: requested, received, unreadable, incomplete or accepted by a preparer. The agent may move items between requested and received, but only a preparer may mark an item accepted.
  • Define escalation triggers in advance: a document dated for a different tax year, a name that does not match the client record, a message that mentions a change in household circumstances, or any question about how something should be reported.
  • Run the agent in shadow mode for a short period. It drafts each message it would send, and staff compare those drafts with what they would have written.
  • Turn on automatic sending only for the templates that passed that review, and hold a weekly review of escalations and corrections.

Firms that handle other document-heavy intake can compare this approach with how a family law intake team sets document checkpoints, where the same question of who accepts a document comes up early.

How to chase missing documents without damaging trust

Chasing is where clients feel the process most. Messages should be specific, polite and short: name the missing item, explain in one line why it is needed, and give one clear way to respond. Reminders should follow a cadence the firm agrees on in advance, not whatever the person with spare time decides that week.

Illustrative scenario: a household client uploads a year-end statement for one account but not for a second. The agent marks the first item as received, marks the second as requested and sends one message that names only the missing statement and the upload link. The client replies that the second account was closed. The agent does not decide what that means for the return. It routes the thread to the assigned preparer with the account name, the date of the message and the client's exact words attached.

The agent should never turn a client's situation into a tax conclusion. Its summaries should quote the client and state what is missing, leaving interpretation to the preparer. Limit what the agent can see to what the task requires, so a reminder about one form does not expose the full return. Documents belong in the firm's storage under its retention rules, and sensitive identifiers should stay out of chat logs and message drafts.

Extension follow-up that respects the calendar

Extension follow-up helps when it keeps clients informed about what the firm needs and by when. It becomes risky when the agent guesses at a deadline. Filing dates, extension rules and penalties depend on jurisdiction, return type and the client's circumstances. The firm should maintain its own deadline calendar, reviewed by a licensed person, and let the agent read from that calendar rather than from its general knowledge.

Use a simple decision guide for every reminder:

  • If a preparer has confirmed the calendar entry, the agent may send a reminder that quotes that entry exactly.
  • If the entry is missing, unclear or recently changed, the agent stops and flags the client record to staff.
  • If a client asks whether an extension has been filed, the agent answers only with the status recorded in the firm's system and routes anything it cannot confirm.
  • If a client asks whether they owe tax or how much, the agent does not answer and escalates the question at once.

Each reminder should store the template used, the calendar entry it cited, the recipient and the time it was sent. That record lets the office answer later what a client was told and when, without reconstructing it from memory.

Approval rules and an audit trail staff can read

Approval rules are what make an agent useful in a tax office instead of risky. Sort actions into three groups: actions the agent may take alone, actions it may take only after a staff member approves a draft, and actions it may never take. A standard document request and a status change from requested to received can run automatically. A message that mentions a deadline change needs approval. Changing a figure, accepting a document as complete for filing, or contacting a taxing authority is never in scope.

Permissions should follow the principle of least privilege. The agent's access to the document store should cover the intake folders, not the whole archive, and its credentials should be revocable without disrupting staff work. Guidance on authorization design, such as the OWASP authorization cheat sheet, is a useful reference before the first connection.

Client messages can also contain text that tries to instruct an assistant, for example a pasted paragraph asking it to ignore its rules or send a file elsewhere. The agent should treat anything a client sends as data to be read, never as a command to be followed. The OWASP guidance on prompt injection describes this class of risk in more detail.

Finally, every action should be logged with what triggered it, which record it touched and what changed. Staff should be able to read that log without special tools. A weekly review of escalations, edited drafts and rejected messages is often the fastest way to find a template that needs rewriting or a rule that is too loose.

How Be Human runs this and prepares the office for more capable assistants

Be Human designs and operates this kind of agent as managed infrastructure. A first engagement defines one useful role for the office, such as intake and follow-up, along with the memory and tools that role needs, the approval boundaries, and one observable success criterion, such as whether intake files reach a preparer complete on the first handoff. Managed operations then cover monitoring, recurring evaluations of the agent's drafts, maintenance of the connections to the portal and inbox, periodic permission review, incident response and controlled changes to templates and rules.

The preparation for more capable assistants is the same whichever system comes next: clean status records, written boundaries, logged actions and a named owner for every decision. An office with those in place can give a stronger system more intake work later without handing it the filing decisions. If you want to find the first task worth handing to an agent in your office, take the free Be Human AI diagnostic.

Frequently asked questions

Can an AI agent prepare or file a tax return for a client? No, and the first engagement should not be designed as if it could. The agent handles intake, status tracking, reminders and routing, while a licensed preparer makes every judgment about the return and any filing action.

Do clients need to know that a reminder came from an automated process? Clients should be able to tell that a message is automated, and the firm should decide the disclosure wording with its own compliance advisor. Every message should still be signed or attributed to a named staff member or the office, so the client knows who to contact.

What should a tax office automate first? Start with the intake checklist, the status tracking for each document and the standard requests for items that are clearly missing. These tasks repeat every season, have clear rules and can be reversed if a template turns out to be wrong.

How is this different from a generic chatbot on the firm's website? A generic chatbot answers questions in one conversation and usually has no durable record of each client's documents. An operational agent keeps the client's document status, follows written permissions, escalates defined cases to a preparer and logs every action so the office can review what happened.

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.

Less busywork. More room for you.

Bring us the work you wish you had time for.

Book a call to try a personalized AI agent free for 7 days ↗Or take the free AI diagnostic →