AI for Tax Resolution Firms: Intake, Notice Deadlines and Client Updates
How a tax resolution firm can use an AI agent for intake, notice deadlines and client updates, with licensed staff owning every representation decision.

An AI agent can carry most of the administrative load at a tax resolution firm: collecting intake information, logging incoming notices, tracking the response windows printed on each one, and drafting client status updates for staff to approve. It should not decide how to respond to a tax authority, which positions to take, or whether anything is signed or filed for a client. Those decisions belong to the licensed professionals who hold representation authority, and the agent's job is to make sure the right person sees the right file at the right time.
Key takeaways
- Intake, notice logging and deadline reminders are the strongest first tasks because they are structured, repeatable and easy to audit.
- Every deadline the agent tracks should be copied from the notice itself and confirmed by a licensed person before it drives any work.
- The agent drafts client updates in plain language, and staff review and release them. Updates never predict outcomes.
- Positions, settlement options, responses to the tax authority and signatures stay with enrolled agents, CPAs or attorneys.
- Permissions should be narrow: read access to client documents, write access to the case log, and no access to filing or payment systems.
What a tax resolution firm needs from intake
Intake is where most matters go wrong before anyone has opened a return. A prospective client arrives with a stack of letters, a partial set of tax years, a spouse or former business partner who may or may not be on the account, and a belief about what the notices mean. Someone at the firm has to establish who the client is, what authority the firm holds, which years and notice types are involved, and whether a conflict exists. Those questions are answered through a process the firm already runs, usually with a questionnaire, a document request and a short call with a staff member.
An agent can make that process faster and less dependent on one person remembering each step. It can send the approved questionnaire, collect answers, build a document checklist for the matter, and chase missing items on a schedule the firm defines. What it cannot do is decide whether the firm should take the matter. Acceptance, fee discussions and conflict decisions are professional judgments, and the agent should only route them to the person who makes them.
Which intake tasks an agent can own
The useful question is not whether an agent can "do intake." It is which parts of intake are mechanical enough to hand over and well enough documented that a reviewer can check the result. In most firms, the following tasks fit that description:
- Send the intake questionnaire in the client's language, using wording a licensed staff member has already approved
- Turn the answers into a document checklist for the matter and follow up on items still missing
- Check whether a signed authorization form is on file before a case is opened for staff review
- Route the matter to the right licensed person according to rules the firm writes down in advance
- Flag incomplete answers, unusual entity structures, mismatched names and possible conflicts for a human to decide
Each of these tasks produces a record. That record matters more than the speed. When a supervising professional asks why a file was routed a certain way or why a document was requested, the answer should be visible in the case log without anyone reconstructing it from memory.
Notice deadlines: from the envelope to a confirmed date
Notice tracking is the most valuable and the most sensitive part of this work. A missed response window can narrow options that were open a week earlier, so the process has to be boring and exact. The agent should never calculate a deadline from a general rule of thumb. It should copy the date the notice states, record which tax years and notice type it covers, and leave the professional judgment about what to do to the assigned person.
A workable sequence looks like this:
1. Log the notice when it arrives: date received, issuing office, notice type, tax years, and the contact information printed on it. 2. Copy the response deadline exactly as printed, including any page where it appears, and attach the scanned page to the entry. 3. Ask the assigned licensed person to confirm the deadline in the file before any reminder schedule is treated as final. 4. Set reminders at intervals the firm chooses, with escalation to the supervising professional if the item is not acknowledged. 5. Record what was sent, when, and by whom, and close the item only when a licensed person closes it.
Illustrative scenario: a notice arrives for one client covering two tax years, and the response window is printed on the second page. The agent logs each year as a separate line, attaches the page with the date, and sends the reminder to the enrolled agent who owns the file. The agent does not decide whether a response is needed or what it should say.
Mail that arrives on paper usually needs scanning first. The firm should decide who scans, where scans are stored, and how long they are kept, and the agent should only read files that fall inside that policy. Clients sometimes send documents with text that looks like an instruction, such as a note telling the reader to ignore earlier requests. The agent should treat that text as content to be reviewed, never as a command.
Client updates that stay accurate and calm
Clients in a tax matter are often anxious, and a slow or vague update can create more calls than the matter itself. An agent can help by drafting updates from fields the case log already contains: what has been received, what is waiting on the client, who owns the next step, and when the next step is expected. The draft gives the staff member a clean starting point and keeps the wording consistent across matters.
The review step is not optional. A licensed person should release every update, because an update can easily slide into a prediction about penalties, amounts or outcomes, and the firm should never make those promises in an automated message. The firm should approve the templates in advance, including the language it uses for English and Portuguese clients if it serves both. When the case log is missing a field, the draft should say so plainly instead of filling the gap with a guess.
What stays with licensed staff
A clear boundary protects the client, the firm and the agent's usefulness. The following decisions stay with enrolled agents, CPAs, attorneys or other professionals the firm has authorized, and the agent should never be configured to make them:
- Whether a response is needed, and what it says
- Which collection alternatives, payment arrangements or settlement paths are considered
- How ambiguous notices are interpreted and what they imply for the client
- Anything signed, submitted or communicated to the tax authority on the client's behalf
- Advice about penalties, liability, confidentiality or professional obligations
- Any exception to the firm's intake or conflict rules
The agent can support each of these decisions by putting the relevant facts in front of the right person. It should not be the person making them. This is also the simplest way to keep the system inside the firm's professional responsibilities, since a reviewer can always point to the human who owned each choice.
How permissions and records should be set
The access an agent holds should match the work it does and nothing more. Read access to the documents needed for intake and notice logging is reasonable. Write access to the case log and the reminder queue is reasonable. Access to filing systems, payment tools or any channel that reaches the tax authority is not part of this role. The access model should follow the principle that each action is authorized explicitly for a specific purpose, which is the same discipline described in the [OWASP authorization guidance](https://cheatsheetseries.owasp.org/cheatsheets/Authorization_Cheat_Sheet.html).
Every action the agent takes should leave a trace: which document it read, which field it wrote, which draft it produced, and which person approved it. Reviewers need that trail to check the work, and the firm needs it when something looks wrong. Risk expectations for systems like this are discussed in the [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework), which is a useful reference when a firm writes its own internal policy. For a broader view of how agents are designed to stay bounded, the [Anthropic guide to building effective agents](https://www.anthropic.com/engineering/building-effective-agents) describes the value of simple, controllable patterns over open-ended autonomy.
Firms that already run similar workflows can borrow from neighboring practices. The property tax appeal firm approach to assessments, evidence and deadlines follows the same separation between tracking dates and deciding positions, and the family law intake and court date guide shows how intake checklists can be structured before any agent touches them.
How Be Human runs this in practice
Be Human designs the role first: the single job the agent does, the memory and tools it needs, the approval points where licensed staff take over, and a success measure the firm can observe, such as whether every notice has a logged deadline confirmed by a professional. After that, Be Human implements the integrations and operates them through monitoring, recurring evaluations of the agent's drafts and logs, permission reviews, integration maintenance and a defined response when something goes wrong. The service is remote and available in English and Portuguese, and the firm keeps control of the agent's identity, memory, channels and permissions throughout. If you want to find the first task in your firm worth handing to an agent, take the free Be Human AI diagnostic.
Frequently asked questions
Can an AI agent respond to a tax notice for a client? No, and a well-designed agent should not be set up to try. The agent can log the notice, copy the response window exactly as printed and route the file to the enrolled agent, CPA or attorney who owns the matter. Any response, position or communication with the tax authority is a professional decision made and signed by a licensed person.
How should a firm handle a deadline the agent cannot read clearly? The agent should flag the notice as unclear instead of guessing. It attaches the scanned page, records what it could and could not read, and asks the assigned professional to confirm the date. Until a person confirms it, the deadline should not drive any automated reminder or client message.
Do client updates need to be reviewed before they are sent? Yes. The agent can draft them from the case log, but a licensed staff member should approve each update before release. Reviewers check for accuracy, for any implied prediction about outcomes or penalties, and for wording that matches the firm's approved templates in English or Portuguese.
Where should a firm start if it wants to try this? Start with intake and notice logging, because those tasks are structured, produce a clear record and keep the agent away from representation decisions. Limit the agent to read access on documents and write access on the case log, review its work weekly for the first period, and expand the role only after the firm has seen the logs and agreed on the boundaries.
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.


