Superintelligence for Property Tax Appeal Firms: Deadlines and Comparables
A practical plan for property tax appeal firms to track filing deadlines and comparable-sale records with AI agents, with human review on every appeal.

Superintelligence for a property tax appeal firm means AI agents that hold deadlines, comparable-sale records and filing history in one controlled memory, then prepare work that a named person checks before anything is filed. The useful version does not decide whether an appeal should be filed. It keeps every date visible, builds evidence packets from sourced sales, and routes each file to a reviewer who can approve or stop it.
Key takeaways
- Deadlines are the highest-risk field in an appeal practice, so they need one register that agents can read and staff can own.
- Comparable sales should be stored with their source, sale date and adjustments, not pasted into documents as plain text.
- Agents can draft evidence packets and flag inconsistencies, but a named person approves every appeal before it is submitted or used at a hearing.
- Permissions should separate reading records, drafting documents and submitting to an assessor or review board portal.
- Start with one role and one measurable check, such as no open file reaching its filing window without an assigned reviewer, before adding more agents.
Where deadlines break down in appeal work
Property tax appeals run on local calendars. Assessment notices arrive on different schedules, filing windows are set by each jurisdiction, and some deadlines run from the notice date rather than the date a client signed the engagement. Many firms track these in a spreadsheet maintained by one person, email threads that hold the real instructions, and a folder per parcel. The failure is rarely a single forgotten date. It is more often a date copied from a letter that was later superseded, a parcel whose owner changed while the file was open, or a rule that changed after the firm last checked it.
The fix starts with treating each deadline as a claim that needs a source and a verifier. Staff should confirm current rules with the relevant jurisdiction. An agent cannot replace that confirmation, but it can make sure no deadline lives only in someone's memory or inbox.
How do you build a deadline register an agent can read?
The register is the first artifact to build, and it should be boring on purpose. Each open file needs a row that records the same fields every time:
- The parcel identifier, the owner of record and the assessment year under review
- The notice date, the filing window and the source document that set the window
- The hearing date when it is known, and the status of the evidence packet
- The named reviewer responsible for the file
- The date someone last verified the deadline against current jurisdiction rules, and who did it
Once the register exists, the working sequence is straightforward. Export every open file into one table. Attach the actual notice to each row instead of a typed summary. Have a staff member verify each deadline once and record the verifier. Give the agent read access to the register and write access only to draft notes inside each file. Then schedule a weekly run that lists files lacking a reviewer, an evidence packet or a verified deadline, and sends that list to the person who owns the practice.
The permission split matters. Least privilege is the core idea in the [OWASP Authorization guidance](https://cheatsheetseries.owasp.org/cheatsheets/Authorization_Cheat_Sheet.html): an agent should hold only the access its job requires. A deadline coordinator that can read the register and draft notes does not also need the ability to change a filing date or submit a document.
The same pattern appears in other regulated record work. AI for a Credentialing Company describes how application status and expirations become the core records an agent must track. The logic transfers well to appeals, where a missed expiration and a missed filing window carry similar costs.
Comparable-sale records that hold up at review
Comparables carry most of the weight in many appeals, and their weakness is usually provenance. A sale price copied from a listing site without a deed reference, a transaction that may not reflect a typical arm's length sale, or a comparable that differs in size, age, condition or zoning and was never adjusted: each of these gives a reviewer a reason to discount the entire packet.
Every comparable record should store the parcel or address, the sale date, the recorded price, the source document, the property characteristics at the time of sale, and each adjustment with a short written reason. An agent can gather candidate sales from records the firm already holds, compute preliminary differences and flag any record missing a source. The analyst then decides which sales are relevant and how the adjustments will be argued. Keeping that split clear makes the agent's output checkable, because every number traces back to a document a person can open.
For the wider picture of assessments, evidence and deadlines in this sector, see AI for Property Tax Appeal Firms.
An illustrative scenario: one appeal from notice to hearing
The following scenario is illustrative and does not describe a real firm. A small appeals practice receives a new assessment notice for a commercial parcel. The intake step logs the notice, pulls the filing window from the register and opens a file, confirming that the owner's authorization is on record before any work begins.
A research agent then proposes a short list of candidate sales drawn from the firm's existing records. Three of the candidates lack a deed reference, so the agent marks them as incomplete instead of leaving them in the packet. The analyst removes two candidates that were not comparable in use and sets the adjustments for the remainder. The draft evidence packet goes to the appeals lead, who rejects one adjustment and asks for a dated site photo log before the packet can move forward.
Only after the lead approves the packet does the filing step run. It uses a portal account that the agent can access solely for the documents approved in that file, and every step leaves a record of who approved it and when. The deadline was never in the agent's hands. The agent kept the date visible, the evidence traceable and the approval path clear.
Keeping a human on every appeal
Human review is not a formality in this work, so it should be defined before any agent runs. A practical boundary set looks like this:
- Agents may read the register, the firm's notes and public records the firm already holds
- Agents may draft letters, evidence packets and internal summaries for a named reviewer
- Agents may not submit a filing, accept a settlement or send a client a valuation opinion without a named approver
- Any change to a deadline requires a second person to verify it
- A file missing a source document is held with a visible reason, never silently dropped or silently completed
Document content needs its own rule. Notices, third-party reports and uploaded attachments can contain text that tries to steer an automated system, so the agent should treat document content as data to be summarized, not as instructions to follow. The [OWASP prompt injection prevention guidance](https://cheatsheetseries.owasp.org/cheatsheets/LLM_Prompt_Injection_Prevention_Cheat_Sheet.html) covers this risk in detail. For naming risks, owners and review points across the life of a system, the [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) is a useful reference that business owners can read without a technical background.
A parallel exists in other evidence-heavy work. AI for Process Servers shows why service logs must be complete enough to stand up when someone questions them years later, and appeal files face the same test at a hearing.
What a managed setup covers for an appeal firm
Be Human designs, implements and operates personalized AI infrastructure, with agents that keep their identity, memory, channels, tools, permissions and operating context under human control. A first engagement defines one useful role, the memory and tools that role needs, the approval boundaries and an observable success criterion. For an appeal practice, that role is often a deadline and evidence coordinator with read access to the register and draft access to the packet folder for each file.
Managed operations can then cover monitoring of deadline alerts, recurring evaluations of comparable-sale output against samples that staff have checked by hand, permission reviews whenever someone joins or leaves the team, incident response if a record is wrong, and controlled improvements to the agent's instructions. Each change is made deliberately and tested before it reaches live files.
Choosing the first task
The best first task usually has a clear owner, an output a person can check in minutes and a short feedback loop, such as the weekly list of files that lack a reviewer. That makes it a fair test of the whole setup before the firm trusts it with more. 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 a sound design does not allow it. The agent can prepare the file, draft the evidence packet and flag gaps, but submission should require a named approver and a portal permission limited to approved documents. The firm keeps responsibility for every filing it makes.
How should a firm handle a deadline that the agent and a staff member disagree on? Treat the disagreement as a hold, not a choice. The file should stay visible with the conflicting sources attached, and a second person should verify the deadline against the jurisdiction's current rules before anything proceeds. The verification and the verifier's name should be recorded in the register.
What makes a comparable sale strong enough to use? A strong comparable has a recorded source, a sale date close to the assessment date, physical characteristics that can be compared and adjustments explained in writing. Records that lack a deed reference or a clear source should be marked incomplete rather than quietly used. A human analyst still decides which sales are relevant to the argument.
Where should a firm start if it has no existing system? Start with the deadline register, because it is the smallest change with the clearest benefit. Populate it for open files, verify each deadline once, and run a weekly check for missing reviewers and packets. Once that routine is reliable, the firm can decide whether comparable-sale records or intake should be the next role.
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.


