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

AI by industry

AI for Tax Preparation Offices: Document Intake and Status Updates

Intake, document checklists and client status updates suit an AI agent today. Preparers keep every filing decision behind explicit approval gates.

Be Human typographic cover with the title: AI for Tax Preparation Offices: Document Intake and Status Updates
Conceptual illustration created with AI · Be Human

An AI agent for a tax preparation office can take over the repeated front half of each engagement: receiving client documents, checking them against a written checklist, and sending status updates that answer the question every client asks, which is what is still missing. It should not decide what belongs on a return, which figures are final, or when a client is ready to sign. Those judgments stay with the preparer, and the agent's job is to make sure they arrive with a complete, organized file.

Key takeaways - Start with intake: receipt, sorting and missing-item follow-up are bounded tasks an agent can own. - Have preparers write the checklist for each engagement type, so the agent applies their rules instead of guessing at them. - Give the agent read access to the status tracker and write access only to client messages and draft notes. - Place filing, signature, figure changes and payment instructions behind named approval gates owned by a person. - Measure success by how many files reach the preparer complete and correctly labeled, not by how many messages were sent.

Why document intake is the right first job Intake absorbs a lot of attention in a tax office on work that is repetitive, rule-driven and easy to check. A client uploads a photo, a staff member asks whether it is the right form, and a week later someone is still chasing the one missing page. None of this requires tax judgment at the moment a document arrives, but all of it determines how quickly and cleanly the preparer can begin.

That is why intake suits an agent. The work has clear inputs, such as a document, a checklist item and a client record, and clear outputs, such as a document marked received, a request sent or a gap flagged. When something goes wrong, such as a misfiled year or a duplicate upload, the error shows up at the point of entry instead of surfacing later inside a return.

The same pattern appears in other document-heavy businesses. AI Agents for Import Export Companies covers a similar split between document handling and status communication. Keep the first task narrow on purpose: one client type, one checklist, one tracker. A whole office handed to an agent at once produces a vague role that nobody can evaluate.

A step-by-step plan for the intake workflow Build from the artifacts the office already uses, such as its current client questionnaire and its upload instructions. The plan below fits one client type, and you repeat it for each engagement type you want to cover.

  • Write the checklist with the preparers who own it, listing each document, the reason it is needed and the accepted formats.
  • Define the states a document can be in, such as requested, received, unreadable, needs clarification and ready for preparer review.
  • Give the agent access to the client upload channel and the status tracker, and nothing else in the practice systems.
  • Agree on a reminder rhythm with the preparer, including how many follow-ups are allowed before a person calls the client.
  • List the conditions that force escalation: a tax year that does not match the file, a name or identifier mismatch, a suspected duplicate, or any message that mentions a notice, an audit or a legal deadline.
  • Run a shadow period in which preparers compare the agent's sorting and gap flags with their own, and record each disagreement with its reason.
  • Only after the disagreements are rare and explained, let the agent mark files complete for handoff.

The shadow period matters more than the launch date. It shows whether the checklist is written clearly enough for someone other than its author to apply it.

What a useful status update contains Clients rarely object to waiting. They object to not knowing what is happening. A good status update says four things: what the office has received, what is still missing, what happens next and who owns the next step. It does not promise a filing date unless the preparer has set one, and it never interprets a document or estimates a result.

Illustrative scenario: a small business client uploads a bank statement that covers only part of the year. The agent acknowledges receipt, compares the upload with the checklist, and sends a message naming the missing months and where to upload them. Two days later it sends one reminder, the number of reminders the preparer approved. If the client replies that the other statements are with a bookkeeper, the agent records that, tells the client who will follow up, and flags the file for the preparer rather than asking again.

Keep messages in the office's voice and language, and identify the sender as an AI agent working for the office. A clear identity protects the relationship if a client later asks who wrote to them.

Where the approval gates belong The useful question is not whether the agent can write to clients, but which messages and actions carry consequences that a preparer must own. The guide below is a starting point for that conversation.

  • The agent may acknowledge receipt, sort uploads and send requests that match the approved checklist word for word.
  • The agent may draft an explanation of why an item is needed, but a preparer approves the draft before it is sent.
  • The agent may flag a figure that looks inconsistent with prior-year records, but it may not change the figure or suggest a tax treatment.
  • A licensed preparer must approve any message about tax treatment, penalties, legally binding deadlines or a request for a client signature.
  • Filing submission, payment instructions, extension requests and decisions about a client's identity stay with named people and never run automatically.

Name the owner of each gate as a person, not a team, and write the rules where the agent can read them. Every gate also needs a fallback. If the owner is unavailable, the work waits and the client receives a truthful holding message. The [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) is a practical reference for mapping who is accountable for each of these decisions and how the risks are reviewed over time.

Access, permissions and document safety Tax documents carry identifiers, income details and sometimes information about dependents, so the agent should hold the smallest set of access that lets it do the intake job. Give it scoped, revocable credentials to the upload channel and the tracker, not a general login to the practice. Review those permissions on a schedule and after any change to the tools it uses. The [OWASP authorization guidance](https://cheatsheetseries.owasp.org/cheatsheets/Authorization_Cheat_Sheet.html) describes the principle of checking each access against an explicit policy instead of assuming that a connected system is safe to use.

Uploaded documents are also an input channel. A scanned PDF or a note inside a spreadsheet can contain text that tries to redirect an assistant, such as an instruction to skip verification. Treat document content as data to be read, never as instructions to follow. The [OWASP prompt injection prevention guidance](https://cheatsheetseries.owasp.org/cheatsheets/LLM_Prompt_Injection_Prevention_Cheat_Sheet.html) lays out the basic defenses. In practice, the agent should have no ability to change checklists, approve its own escalations or alter a client record because a document says so.

Finally, collect only what the checklist requires and agree on retention with the office before go-live, not afterward.

Running it after launch Checklists drift with each season, and clients find new ways to send files. Keep a log of each request sent, each state change and each escalation, so a preparer can reconstruct why a file was marked complete. Run recurring evaluations on a sample of permitted past files, including deliberately messy ones such as cropped photos and mixed-year bundles. Track a few plain measures: how often files reach the preparer complete, how often a wrong-year or duplicate document is caught at intake, and how often an escalation was needed but not raised. For firms that handle evidence and deadlines in other settings, [AI for Property Tax Appeal Firms](/en/blog/ai-for-property-tax-appeal-firms) shows how the same discipline applies to assessment records.

Be Human designs and operates this kind of agent as managed infrastructure. A first engagement defines the one role, the checklists and memory it uses, the approval boundaries and a success criterion the office can observe. Managed operations then cover monitoring, recurring evaluations, permission review, integration maintenance and controlled changes, so a checklist revision made by a preparer reaches the agent through a tested change rather than a quiet edit. The service is delivered remotely in English and Portuguese.

Choosing the first task Most offices already know which intake task produces the most follow-up messages and the most rework, and that is usually the right place to begin. Count those messages for a month before choosing any tool, and let the preparers name the checklist they trust most. Start where the rules are clearest and the cost of a missed item is easiest to see. If you are not sure which intake task deserves an agent first, take the free Be Human AI diagnostic to find the first task worth handing to an agent.

Frequently asked questions Can an AI agent prepare or file tax returns? No, and this plan does not ask it to. The agent handles intake, document checklists and status messages, while preparers decide tax treatment, approve any figure change and submit filings. Filing and signature steps stay behind named human approval gates.

Should clients know they are talking to an AI? Yes. Messages should identify the agent as working for the office, and clients should always have a clear way to reach a person. Disclosure is simpler to defend than a message that later turns out to have come from a machine presented as staff.

What happens when a client sends the wrong year's documents? The agent should detect the mismatch against the checklist and the file's tax year, hold the document in a flagged state, and tell the client only what is needed to resend. It should not reclassify the document on its own. Preparers see the flag in the same tracker they already use.

How do we start without rebuilding the whole intake process? Begin with one engagement type and one checklist that preparers already trust. Run a shadow period before the agent sends anything on its own. Expand only after the first checklist produces files that arrive complete and are rarely corrected.

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 →