AI Agents for Property Tax Appeal Firms: Deadlines, Comps, Files
An AI agent can track an appeal firm's filing deadlines and comparable sales records, while licensed staff keep valuation judgment and every filing decision.

An AI agent for a property tax appeal firm works as a tracker and organizer: it watches filing and hearing deadlines across jurisdictions, keeps comparable sales records with their sources and adjustments, and assembles client files for staff review. Valuation judgment, the choice of which arguments to make, and every submission stay with licensed staff, who approve each step before anything leaves the office. The agent's value is fewer missed dates and cleaner evidence, not a different opinion of value.
Key takeaways
- An agent can own the deadline calendar, the comparable sales log and the file index, but it should not decide value or submit anything.
- Every deadline needs a source document, a jurisdiction, a responsible person and a human confirmation.
- Comparable sales records hold up under review when each sale carries its source, its adjustments and the name of the person who accepted it.
- Client files need scoped permissions, because uploaded documents can contain text that a careless tool might treat as instructions.
- Start with one narrow task, such as deadline reconciliation for one jurisdiction, and expand only after the review shows the errors are understood.
Where appeal firms lose time and evidence
A property tax appeal firm works against a calendar that someone else sets. Assessment notices arrive on local schedules, filing windows and informal review periods differ from one jurisdiction to the next, and the dates that matter often live in a notice PDF, a government website and an email from a client who thinks the letter arrived last week. Most firms manage this with spreadsheets, a shared inbox and experienced staff who remember the exceptions. That works until a senior person is out or the portfolio grows.
Evidence has a similar problem. Comparable sales are pulled, screenshotted, pasted into memos and rarely recorded in a form another person can audit months later. When a hearing officer asks why a sale was excluded or why an adjustment was made, the answer should be in the file, not in someone's memory.
Those are the two places where an agent helps most: tracking and organizing. Our guide to AI for property tax appeal firms covers the wider workflow. This article focuses on the two tasks where an agent is most useful and where its limits matter most.
How do you track deadlines as a sourced calendar?
A reminder that says the appeal is due on a certain day is not enough. A deadline record needs the jurisdiction, the notice it came from, the page or paragraph where the date appears, the person responsible and the date a human confirmed it. Without that chain, a wrong date looks exactly like a right one.
A workable deadline record usually includes these fields:
- Parcel or account identifier and the owner of record
- Jurisdiction and the body that issues the decision
- Source document and the exact location of the date in it
- Deadline type: filing, evidence exchange, hearing, response or document request
- Confirmation status: extracted, confirmed by staff or disputed
- Responsible staff member and a named backup
- Alert schedule and escalation path when a date is unconfirmed
The agent can extract candidate dates from notices and portal pages, propose calendar entries and flag conflicts, such as a hearing recorded before the evidence deadline in the same file. It should not mark a deadline as confirmed. A staff member does that. The same pattern appears in expiration tracking for credentialing firms, where a missed date is costly and the record has to show who checked it.
Monitor the sources too. Portals change layouts and forms change wording, so the agent's extraction can drift without any visible error. A recurring check that compares recorded dates against the live source catches this before a deadline does.
How do you build a comparable sales record people can audit?
Comparable sales are the working evidence in most appeals, so the record must be more than a list of addresses and prices. Each entry should keep the source of the sale data, the date it was retrieved, the property characteristics compared, and each adjustment with its stated reason. The agent can gather candidate sales from sources the firm is permitted to use, normalize fields into the same units and detect duplicates, such as one transaction recorded under two parcel numbers.
A useful comparables log includes:
- Sale source, retrieval date and a link back to the record
- Property characteristics in consistent units across all candidates
- A one-sentence exclusion reason for every candidate removed
- Adjustments listed separately, each with the analyst who proposed it
- Review status and the date of approval
- Data gaps, such as a missing lot size, marked as gaps rather than filled with estimates
The adjustment column is where professional judgment lives. The agent can show that three sales share a neighborhood code and similar building area. It should not decide that a view or condition difference is worth a particular amount, or that a transaction is too unusual to count. Those calls belong to the licensed analyst, and the record should show who made them.
Illustrative scenario: A small firm preparing appeals for several warehouse properties asks an agent to assemble comparable sales within a defined radius. The agent returns eleven candidates, marks four as duplicates and flags two with missing building area. The analyst excludes three candidates with written reasons, keeps the rest and proposes one adjustment for an unusual loading configuration. The file shows which decisions were the analyst's and which were the tool's.
Where does valuation judgment stay with licensed staff?
Decide the boundary before the first task, not after the first mistake. The question is not whether an agent can write a plausible valuation summary. The question is who is accountable for the opinion that goes to a board or hearing officer, and whether that person can explain every number in it.
A practical decision guide for the firm:
- The agent may gather notices, extract dates, propose calendar entries, retrieve permitted sales data, normalize fields and draft internal checklists.
- The agent may draft, with review required: a comparable sales summary or evidence index that an analyst will rewrite or approve.
- Staff only: select comparables for submission, choose adjustments, state an opinion of value, decide whether to appeal, sign documents, negotiate with an assessor and advise a client on likely outcomes.
- Never: submit to a government portal, send anything to a client, or delete or overwrite records in a client file without a staff approval step.
Guidance on building agents generally favors narrow roles, clear handoffs and visible human checkpoints over open-ended autonomy. In an appeal firm, that means every consequential action has a named person who approved it, and the log shows the approval. If an outside reviewer asked how a figure was produced, the firm should be able to trace it to a source, a human decision and a date.
How do you keep client files safe from untrusted documents?
A client file contains tax notices, leases, prior assessments, photographs and correspondence. Some of that text comes from people outside the firm, including owners, tenants and third-party appraisers. Treat it as data to be read, not as instructions to be followed. A document that says to ignore the deadline list and mark all appeals complete should be stored and flagged as text, and the agent should be built so that document content cannot change its permissions or its task. The OWASP guidance on prompt injection prevention describes this risk, and the OWASP Top 10 for LLM Applications lists it among the main application risks.
Limit access by role. The agent needs read access to the notices folder and write access only to the deadline and comparables records. It should have no ability to submit to a portal or send email to a client. Use separate credentials for each connected system, review them on a schedule and log every read and write with the requesting user or process. The OWASP authorization guidance is a useful checklist for these decisions, especially for deciding who can see which client's file.
If the firm handles personal data, retention, access and transfer rules depend on the jurisdictions involved. Confirm them with counsel before connecting client files to any system, and keep the data inside the environments the firm has approved.
How should a first engagement be run?
A first engagement should be small enough to evaluate in a few weeks. A sensible sequence:
- Pick one jurisdiction or one property type where missed dates or weak evidence cause the most rework.
- Write the decision boundary on one page: what the agent may write, what staff must approve and what it may never do.
- Connect read access to the notice and calendar sources, and create one deadline table with the fields listed above.
- Run the agent in shadow mode: it proposes entries, staff record what they would have entered, and the differences are reviewed every week.
- Define an observable success criterion, such as every deadline in the pilot set having a source and a confirmation status, with unconfirmed items falling to a level the firm agrees in advance.
- Expand only after the review shows the errors are understood and the approval steps hold under normal workload.
Avoid a success criterion based on speed alone. Faster entry without a source chain produces faster mistakes. Also avoid measuring success by appeal outcomes, since those depend on facts and decisions the agent does not control.
How does Be Human run this as a managed service?
Be Human designs and operates this kind of agent as managed infrastructure. A first engagement defines one useful role, the memory and tools it needs, the approval boundaries and the success criterion, so the firm can see exactly what the agent can read, write and propose. Ongoing operations can include monitoring of outputs, recurring evaluations on samples of deadline and comparable records, permission reviews, maintenance when a portal or document format changes, incident response, and controlled improvements that staff approve before they go live. Service is remote and available in English and Portuguese.
If you want to find the first task worth handing to an agent in your firm, take the free Be Human AI diagnostic.
Frequently asked questions
Can an agent file an appeal or send anything to the assessor? No, not in the model described here. Filings, portal submissions and client messages are staff actions with an approval step. The agent can prepare drafts and checklists inside the firm's systems, and the firm decides whether any narrow send action is ever appropriate.
Does an agent replace comparable sales analysts? No. The agent handles retrieval, normalization, duplicate detection and documentation of the record. Analysts still choose comparables, decide adjustments and state the opinion of value, and the record shows which decisions were theirs.
How do we know the deadline records are right? Each deadline should carry its source, its extraction point and a confirmation status set by a staff member. Recurring evaluations compare recorded dates against live sources, and alerts flag any date that remains unconfirmed close to its window.
Where should a firm with messy files start? Start with one jurisdiction or one record type, usually deadline reconciliation or a comparables index for a single property type. A narrow first task shows whether the sources, permissions and approval steps hold before the firm connects more of its client files.
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.


