Your Claims Data Is Smarter Than Any Prompt

For Public Adjusters and Claims Professionals

A senior adjuster at a large public adjusting firm told us something recently that stopped the conversation cold.

"I can hold maybe fifty claims in my head at once. On a good day. But we've got four hundred active right now, and every one of them has a carrier who's done something in the last week that I should know about."

He wasn't complaining. He was describing the physics of the job. The human brain has limits. Claims don't care about those limits.

This is where AI enters the conversation, and where most of the conversation goes wrong.

The ChatGPT Trap

Every PA firm we talk to has at least one person who's tried using ChatGPT or Claude or some other general AI tool for claims work. They paste in a policy, ask it to identify coverage gaps. They drop an Xactimate export and ask for a comparison. They type out a carrier response and ask it to draft a rebuttal.

And it works. Sort of. For about fifteen minutes.

The problem isn't intelligence. These models are genuinely capable. The problem is amnesia. Every time you open a new chat window, you're talking to someone who has never met you, never seen your claim, and has no idea that this particular carrier revised their estimate three times last month and reduced HVAC by twelve percent in version four.

You're starting from zero. Every single time.

That's not a workflow. That's a parlor trick.

What "Purpose-Built" Actually Means

The phrase "purpose-built AI" gets thrown around in insurance technology, usually by companies that have bolted a chat interface onto an existing database and called it innovation. That's not what we're talking about.

Purpose-built means the system was designed from the ground up around a single question: what is the current truth of this claim?

Not "what is the general knowledge about claims." Not "what does the internet say about Xactimate." The specific, evolving, documented truth of claim number 2024-FL-08847 as it exists right now, across every email, every document, every estimate version, every carrier communication that has ever touched it.

That's a different kind of intelligence than a prompt-and-response tool. Here's why.

Context That Persists Gets Smarter

Consider what happens when a carrier sends revision four of their estimate on a large commercial loss.

A general AI tool sees a spreadsheet. It can compare numbers if you give it the right instructions and paste in the right data. It might catch that the total went down. Maybe.

A purpose-built claims system sees revision four in context. It already has revisions one, two, and three. It knows the PA's baseline estimate. It knows which trades were disputed in the rebuttal sent six weeks ago. It notices that the carrier agreed on electrical but quietly reduced HVAC by $47,000 -- the exact category that wasn't part of the dispute.

That's not a smarter prompt. That's a system that remembers.

Every document that enters the system adds a node to a living knowledge graph. Every email parsed is another data point about carrier behavior, response timing, negotiation patterns. Over time, the system develops something that looks a lot like the institutional knowledge that senior adjusters carry in their heads -- except it doesn't forget, it doesn't retire, and it doesn't top out at fifty claims.

The Reconstruction Tax

Here's something nobody in this industry talks about openly, because it's so universal that it's become invisible.

Every person on every side of every claim independently reconstructs the claim from email.

The PA opens Outlook in the morning and scrolls through overnight messages to figure out what happened. The carrier adjuster does the same thing on their end. The contractor pulling permits does it from their inbox. The insured checks their email to see if anyone responded.

Four parties. Same emails. Four independent reconstructions. Every single morning.

We call this the reconstruction tax, and it's the single biggest throughput killer in public adjusting. It's not the negotiation. It's not the field work. It's the twenty to forty minutes per claim, per day, that every person spends figuring out what the current state of reality is by parsing email threads.

A purpose-built system eliminates this tax entirely. Not by replacing email -- that will never happen and shouldn't -- but by structuring it as it arrives. When a carrier response comes in at 2 AM, the system has already classified it, linked it to the right claim, identified which documents were attached, flagged what changed, and updated the claim state before anyone opens their laptop.

The adjuster's morning shifts from "what happened overnight" to "here's what changed, here's what it means, here's what needs your judgment."

Constrained AI Is Better AI

There's a counterintuitive principle at work here that matters.

General AI tools are impressive precisely because they can do anything. Write poetry. Analyze financial statements. Explain quantum physics. Translate Mandarin. The breadth is the product.

For claims work, the breadth is the problem.

When you ask a general model to analyze a carrier estimate, it's drawing on everything it knows about everything. It might format the output beautifully. It might use the right terminology. But it has no guardrails specific to how PA firms actually work, no understanding of the difference between a trade summary comparison and a full line-item audit, no knowledge of why a $3,000 variance in a million-dollar claim matters less than a $3,000 variance in a $150,000 claim.

Purpose-built systems are constrained by design. They operate within the boundaries of claims data, claim workflows, and claim outcomes. That constraint isn't a limitation. It's the whole point.

A constrained system that knows your claims inside out will always outperform an unconstrained system that knows a little about everything. The world doesn't need another tool that can summarize a PDF. It needs a tool that knows this PDF is the third version of a carrier estimate that reduced concrete by nineteen percent relative to your baseline, and that the carrier's field adjuster has a pattern of undervaluing concrete on losses in this region.

Compounding Returns

The real argument for purpose-built AI in claims isn't about any single feature. It's about compounding.

Every claim processed makes the system smarter about claim patterns. Every carrier interaction teaches it something about how that carrier operates. Every estimate version builds a richer picture of how negotiations actually play out across different loss types, different geographies, different carrier desks.

A firm that processes two hundred claims through a purpose-built system in year one has a materially different tool in year two than a firm that just started. Not because the software was updated, but because the data underneath it is deeper.

This is the moat that generic AI tools can never cross. ChatGPT doesn't get smarter about your claims over time. It doesn't learn that Carrier X always low-balls HVAC on the first pass. It doesn't notice that your junior adjusters consistently miss depreciation recoveries on contents claims under $75,000. It can't tell you that your average cycle time on water damage claims has increased fourteen percent this quarter because carrier response times in that category have slowed.

Purpose-built systems can do all of this, because the data lives there. Permanently. Structured. Connected.

What This Means for Firms Today

If you're a PA firm evaluating AI tools -- and you should be, because this technology is no longer optional -- here's the honest filter:

Ask whether the tool remembers. If you have to re-explain your claim every time you use it, you're using a search engine, not an intelligence platform.

Ask where the data lives. If the AI is analyzing documents you paste in but doesn't retain the analysis or connect it to anything, the value evaporates the moment you close the tab.

Ask what it learns. Does the system get better at identifying issues specific to your practice, your carriers, your loss types? Or does it give the same generic output to every firm that uses it?

Ask about the graph. The most valuable thing in modern claims technology is the structured representation of a claim that connects documents to emails to estimates to parties to timelines. If the tool doesn't build this graph, it's a feature. If it does, it's infrastructure.

The firms that will lead this industry in five years won't be the ones with the most adjusters. They'll be the ones whose systems know the most about their claims. That knowledge compounds. And compounding always wins.

The firms that treat claims data as infrastructure -- structured, connected, persistent -- will outpace firms still pasting into chat windows. That's not a technology bet. It's an operational one.