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

AI by industry

AI for Public Adjusters: Tracking Claim Files and Carrier Deadlines

How public adjusters can use AI agents to track claim files, carrier deadlines and document requests, with settlement positions kept by the adjuster.

Be Human typographic cover with the title: AI for Public Adjusters: Tracking Claim Files and Carrier Deadlines
Conceptual illustration created with AI · Be Human

An AI agent for a public adjusting firm tracks each claim file, the carrier's response deadlines and the documents still owed, then tells the adjuster what needs attention and when. It can draft requests and reminders, but the adjuster keeps ownership of valuation, negotiation and every settlement position. The useful first deployment is narrow: one role, one set of files, and clear rules about what the agent may do without approval.

Key takeaways

  • Treat each claim file as a record with one owner, a current list of open requests and dated carrier obligations.
  • Let the agent track deadlines, file documents and draft requests, but keep valuation and settlement positions with the adjuster.
  • Start with read access and draft-only outbound permissions, and log every action the agent takes.
  • Begin with one carrier or one adjuster's caseload before widening the scope.

What a claim file tracker has to hold

A public adjusting file grows in layers. It includes the notice of loss, policy declarations, proof of loss forms, photographs, contractor estimates, the carrier's correspondence, inspection reports and the firm's own notes on what was promised and to whom. When these live in an inbox, a shared drive and someone's memory, the team spends time reconstructing the file before every call. An agent is only as useful as the record it can read, so the first build is a register that holds, for each claim:

  • The claim number, carrier, loss date and the policy language the firm relies on
  • The current stage, such as inspection, estimate exchange, proof of loss or negotiation
  • Every open request, with who owes it, when it was sent and when it is due
  • The adjuster responsible for the file and the approval rules that apply to it
  • A short decision log of positions the adjuster has taken, so the agent never restates a position as if it were new

Firms that already work with deadline-heavy files, such as property tax appeal firms, often recognize the same pattern: a deadline is easy to write down and hard to keep current. The register solves the writing down. The agent's job is keeping it current.

How do you model carrier deadlines without guessing?

Carrier deadlines come from several sources: the policy wording, written requests from the carrier, timelines set by state rules or contracts, and informal promises made on a call. If these are merged into one due date, the file hides which deadlines are binding. Separate them before the agent touches anything.

  • Step one: record each deadline with its source document and the exact sentence that creates it.
  • Step two: label each deadline as binding (written, from the policy or a rule) or working (mentioned on a call and awaiting written confirmation).
  • Step three: set two reminder points for each binding deadline, one early enough to gather documents and one for the adjuster's final check.
  • Step four: when a carrier stays silent past an agreed point, the agent raises a flag to the adjuster instead of sending a follow-up on its own.
  • Step five: when the carrier replies, the agent updates the source, the deadline and the decision log together.

The agent does not decide what counts as binding in a disputed file. The adjuster or the firm's compliance lead makes that call, and the register records it.

How do document requests stay accurate?

Illustrative scenario: an adjuster is handling a water damage claim. The carrier has asked for three items: the contractor's itemized estimate, moisture readings and prior repair invoices. Two weeks later, the estimate has arrived, but the moisture readings exist only as an attachment in a technician's email, and the invoices were requested again by a different person at the carrier. Without a register, the adjuster finds this out the afternoon before the call.

With an agent watching the register, the request list becomes one list per file, with each item's requester, the party it was requested from, the date received and where it is stored. When the estimate arrives, the agent marks it received, links it to the claim folder and notes its page count. The readings are flagged as received but not yet filed to the claim folder. The duplicate invoice request is flagged as possibly already answered. The agent then drafts a short note asking the insured's contact for the invoices, and the adjuster approves it before it leaves the firm.

What may the agent do, and what stays with the adjuster?

The split should be written down before launch, not negotiated file by file. A practical decision guide looks like this:

  • The agent may update the register, file received documents to the claim folder, summarize carrier correspondence and list open deadlines each morning.
  • The agent may draft requests, reminders and status notes, but none leaves the firm without adjuster approval.
  • The agent never proposes a settlement figure, accepts or rejects an offer, characterizes coverage in writing, changes an entry in the decision log or promises a timeline to a party outside the firm.

If an action is unclear, place it on the approval side. Other businesses that track claims and damage files face a similar split, which is why the approach used for equipment rental damage claims is a useful comparison for scoping permissions.

How do you control access and handle risky document content?

The agent reads carrier letters, contractor estimates and photo captions. Some of that text comes from parties with their own interests, and an email can contain sentences that look like instructions. The agent should treat document content as data to be summarized, never as commands to be carried out. The OWASP guidance on prompt injection prevention describes this risk and the controls that reduce it.

Access should follow the same discipline. Limit each agent to the register and the claim folders it needs, scope access by claim so that one file cannot expose another, and keep outbound messages behind approval. OWASP's authorization guidance is a useful reference for this: the question is not only who is asking, but what each action is allowed to touch. The NIST AI Risk Management Framework offers a practical structure for assigning owners to these risks before launch.

Keep a log of every read, write and draft. When an adjuster asks why a deadline moved, the answer should be one lookup away. Retention, consent and personal data handling should follow the obligations that apply to the firm and its clients, reviewed by whoever owns compliance. A mistake in a private claim file is expensive, so the log is not optional.

How should a firm roll this out?

Start small and measure what the agent gets wrong. A sound rollout has three steps:

  • Step one: choose one carrier or one adjuster's caseload, and write down how deadlines and document requests are actually handled today, including the informal parts.
  • Step two: build the register and the agent's read access, then run the agent beside the existing process for one full claim cycle. Compare its flags against what the adjuster already knew, and record every miss.
  • Step three: widen permissions only after the flags are consistently useful and the approval log shows no draft sent without review.

Be Human runs this as a managed engagement. The first phase defines one role, the memory and tools it needs (such as register access, read-only carrier correspondence and a draft queue), the approval boundaries and an observable success criterion, for example that every open binding deadline has a named owner and a scheduled reminder. After launch, managed operations cover monitoring of flags, recurring evaluations of whether drafts match how the firm writes, permission reviews, maintenance when a carrier portal or document system changes, and incident response when something goes wrong. The adjuster's judgment and the firm's settlement positions remain with the adjuster throughout.

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 AI agent negotiate a claim for a public adjuster? In the model described here, no. The agent tracks files, deadlines and documents and drafts communications, while valuation, negotiation and settlement positions stay with the licensed adjuster. Every outbound message to a carrier, insured or contractor passes through approval.

Does the agent replace our claim management software? It depends on what the firm already uses. Where a system of record exists, the agent should read from and write to it rather than becoming a second source of truth. If no such system exists, the register described above is the first thing to build.

How do we know the agent is missing deadlines? Set up a recurring evaluation: periodically compare the agent's deadline list against files that the adjuster has reviewed by hand, and record every miss with its cause. Monitoring should focus on misses and their causes, not on how many reminders were sent.

What should we check before letting an agent touch carrier correspondence? Confirm the authorization scope, how personal data is handled and retained, who approves outbound drafts and whether the compliance lead agrees with the rules. Start with read access and draft-only permissions, and widen them only after a documented review.

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 →