AI for Property Tax Appeal Firms: Assessments, Evidence, Deadlines

For a property tax appeal company, the useful form of AI is a records agent. It reads each assessment notice when it arrives, keeps every parcel's evidence file current, and flags the appeal window before it closes. It can draft packets and correspondence, but it never submits anything to an assessor or board without a named person approving the exact text and attachments.

Key takeaways

• Start with the notice, not the appeal: a parcel record that links every notice, deadline and owner is the foundation the rest depends on.
• Keep each evidence item in a per-parcel file with its source, retrieval date and review status.
• Let the agent extract, draft and flag, but require a named approver before any filing, signature, payment or client-facing commitment.
• Treat every deadline the agent produces as a claim to verify against the official source, not as a fact.
• Measure success by confirmed deadlines with linked sources and time to a reviewable packet, not by the number of drafts produced.

Why assessment notices are the right place to start

A property tax appeal firm runs on two clocks. The first is the assessment cycle, which determines when notices arrive and what value each parcel carries. The second is the appeal window, which each jurisdiction sets and which can close quickly after a notice is received. Most operational failures in this work are not failures of argument. They are failures of records: a notice saved in the wrong folder, a date copied from an outdated email, or an evidence file built on a value that has already been superseded.

An agent is useful here for a narrow reason. It can read every notice the moment it lands in the shared inbox or upload folder, extract the fields that matter, and attach them to the correct parcel. That habit alone removes much of the risk, and it does so without asking the agent to judge the merits of any appeal. Valuation judgment stays with the appraisers, attorneys or agents of record who are responsible for the case.

What the agent should track for each parcel

Design the record before you design the agent. Each parcel should have one file that the agent maintains and that a person can open and understand in under a minute. At minimum, it should hold:

• The parcel identifier, the owner of record, and the taxing jurisdiction that governs the appeal
• Every assessment notice, with the date received, the assessed value as stated, and the appeal deadline as printed
• The official source for that deadline, with the date it was checked and the name of the person who checked it
• Evidence items such as comparable sales, income statements, condition reports or photographs, each with its source, retrieval date and version
• Engagement and authorization status, so that no work starts on a parcel where the firm's authority to act is unclear
• Open questions for the appraiser, the attorney of record or the client, each with a named owner

Write this template down before any tool is connected. If the team cannot agree on what a complete file looks like, an agent will only produce inconsistent files faster. The file also serves as the agent's working memory, and it should stay factual. When the agent summarizes evidence, the summary should point to the underlying document, so a reviewer can trace every sentence back to its source.

Where the approval boundary belongs

The approval boundary is the most important design decision in this kind of engagement, and it should be written down before the first agent run. A workable split looks like this:

• Read and extract: the agent may read notices, move fields into the parcel record, and flag anything it cannot parse with confidence
• Draft: the agent may prepare evidence summaries, cover letters and appeal packets, saved as drafts with their sources attached
• Propose: the agent may place a deadline on the shared calendar, labeled as unverified until a person confirms it against the official source
• Submit: the agent may not file, sign, pay, send anything to a board or assessor, or email a client; each of these requires a named approver to sign off on the exact version and attachments
• Change: any edit to a confirmed deadline, evidence source or valuation figure requires a person to re-check it and record who made the decision

Approval should mean that a named person reviewed the final text and the attachments that will be sent. A general click that confirms everything is not approval in any useful sense. Keep the credentials the agent uses limited to reading and drafting, and leave the filing step as a separate action that a person performs in the portal or system of record. The OWASP authorization guidance is a practical reference for writing these scopes down, and the NIST AI Risk Management Framework offers a shared vocabulary for recording who is accountable for each risk.

How to keep deadlines honest

Deadlines are where a records agent earns its place, and also where a confident wrong answer does the most damage. Apply three rules. First, extract the date from the notice and the rule from the official source, and store both with their links. Second, require a second check by a person before a date is treated as confirmed on the calendar. Third, send reminders to the approver on a schedule that leaves room for the work the window requires, not a single reminder on the last day.

Also decide who owns changes to jurisdiction rules. Holiday schedules, weekend rules and procedural amendments can shift a window, so a recurring review of the rules the firm relies on belongs on the calendar too. A firm that handles deadlines in other regulated areas will recognize the pattern, which is also described in AI Agents for Law Firms: Setup Guide for Intake and Deadlines.

Illustrative scenario: a notice for a commercial parcel arrives late on a Friday. The agent creates the parcel record, extracts the printed appeal date, and places it on the shared calendar with an unverified label. On Monday, an analyst compares the date with the county's published calendar, confirms it, and removes the label. The approver receives a reminder well before the window closes, along with the status of the draft packet and the list of missing evidence. At no point does the agent file anything.

Keeping evidence traceable and safe

Evidence files become messy because they mix firm documents with material gathered from outside sources, such as listing pages, public records and letters from owners. Each source should keep its own retrieval date and link, and each version should be identifiable. Retrieval-based designs suit this work because the agent builds its answers from stored documents that a reviewer can open, rather than from what it recalls. The same discipline appears in Superintelligent Agents for Internal Audit: Workpapers and Follow-Up, where every finding has to point back to a workpaper.

Treat outside material as data, never as instructions. A web page or a forwarded email can contain text written to steer an assistant, such as a line telling it to skip a check or to mark a deadline as confirmed. The OWASP prompt injection guidance explains why content from untrusted sources should be kept separate from the agent's instructions and why the agent's actions should be limited by the permissions it holds. In practice, the agent can read an owner's letter, but the letter cannot change the agent's approval rules.

How Be Human approaches this in a managed engagement

Be Human begins with one useful role, in this case a deadline and evidence records agent for the appeal team. The first engagement defines the memory and tools that role needs, the approval boundaries described above, and one observable success criterion, such as every confirmed deadline carrying a linked official source. Managed operations then cover monitoring of the agent's activity, recurring evaluations against a set of past notices, permission review as staff and systems change, incident response if a document or credential is misused, and controlled improvements that are tested before the team relies on them. The service is delivered remotely, in English and Portuguese.

Where to start

Choose the first task by looking at the record your team re-checks most often, which is usually the deadline calendar or the evidence index. Give the agent one narrow job, keep the approval step with a person, and review the results against your success criterion before widening its role. Once that works, the same parcel record can support the next task without rebuilding the foundation. If you are not sure which task is the right first one, take the free Be Human AI diagnostic to find the first task worth handing to an agent.

Frequently asked questions

Can an AI agent file a property tax appeal on its own? No, and it should not be set up to do so. The agent can read notices, organize evidence, draft packets and propose calendar entries, but the filing and signature steps should belong to a named person who reviews the exact version being submitted. Keeping that boundary in the permissions, not only in a policy document, makes it enforceable.

What should an appeal firm set up first? Start with the parcel record rather than the agent. Define the fields each file must contain, including the official source for each deadline and the retrieval date for each piece of evidence, and get the team to agree on that template. Only then connect the tools the agent will use.

How does an agent avoid missing a deadline? It cannot guarantee that on its own, which is why deadlines are treated as claims to verify. The agent extracts the date and the rule, a person confirms the date against the official source, and reminders go to the approver on a schedule that leaves time for the required work. Recurring checks of jurisdiction rules cover changes that a single notice would not reveal.

Does this replace appraisers or attorneys? No. The agent handles records, retrieval and drafting so that the professionals spend their time on valuation and legal judgment. Those decisions remain with the appraisers, attorneys or agents of record, and the agent's outputs are only as useful as the review they receive.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.