The packet is dense
FNOLs, carrier letters, contractor estimates, recorded-statement transcripts — the signal is scattered across pages of boilerplate.
Claims intelligence
Read any claim in seconds.
The dense first read on every claim — the FNOL, the adjuster narrative, the estimate packet — is where the hours go. Paste it in and ClaimsReader returns a clean structured summary in seconds: claimant, peril, coverages, line items, and the gaps a careful reviewer would flag. Built on Astra AI.
For adjusters & claims teams · No claim text stored · A review aid, not a coverage call
Every claim starts with the same slow chore: reading it.
Pages of narrative, an estimate buried three documents deep, a date of loss that's mentioned once and never again. The facts are in there — you just have to dig them out before the real work begins.
The first read is the bottleneck
Before triage, before assignment, before a coverage call — someone has to read the whole thing and pull the key facts by hand. Every time.
The easy miss is expensive
A missing date of loss, no photos, a peril that might be excluded — the things that stall a file later are usually visible on page one, if anyone has time to notice.
ClaimsReader does that first pass for you — so the reading stops being the part that slows the whole desk down.
Try it now
Paste a claim. Get the facts back structured.
Drop in raw claim text — an FNOL, an adjuster's narrative, an estimate, or a packet — and ClaimsReader reads it the way a senior reviewer would, then hands back the structure. Nothing you paste is stored.
Output is a review aid for a licensed professional — it does not make coverage determinations.
How it works
Paste it. Read it. Move on.
Paste the claim text.
An FNOL, an adjuster's narrative, an estimate, a transcript — whatever you've got. No formatting, no template, no upload ceremony. Just the words.
ClaimsReader reads it like a senior reviewer.
It pulls the claimant, the numbers, the date of loss, the peril, the property, and the line items — and it never invents a fact that isn't in the text.
You get the structure — and the flags.
A clean summary plus the gaps a careful reviewer would raise: missing dates, absent documentation, a peril worth a second look. Copy it, drop it into intake, keep moving.
What it extracts
Everything you'd pull by hand — already pulled.
ClaimsReader reads a claim the way a careful examiner does: it finds what's there, structures it, and surfaces what's missing. It works only from the text you paste — it never fabricates a name, a number, or a coverage.
Claimant & claim IDs
Claimant name, claim number, and policy number — lifted straight out of the narrative, even when they're scattered across the page.
Date, peril & property
Date of loss, the cause (water, fire, wind, theft, liability), and the property or item insured — with a plain one-line summary of what actually happened.
Line items & estimate
Itemized damages and amounts pulled into a clean table, with an estimated total — so the dollars are visible at a glance, not buried in a paragraph.
Coverages referenced
Every coverage the claim touches, tagged by how it shows up — mentioned, applies, excluded, or unclear — so the policy questions are framed up front.
Coverage gaps & flags
The most valuable field: missing dates, absent photos or estimates, a peril that may be excluded, internal inconsistencies — the things a reviewer catches, surfaced first.
Structured, copyable output
A clean reviewer summary plus one-click JSON — ready to drop into intake, triage, a spreadsheet, or a downstream review like FRVerify.
A real read
From a wall of text to a clean file.
Here's an FNOL the way it usually arrives — and what ClaimsReader hands back. Same facts, no invention, in a form you can act on.
FNOL — Claim #WC-2025-33871. Insured: Robert Alvarez. Policy HO3-558210 (homeowners). Date of loss: not clearly stated, possibly last weekend. Loss location: 2271 Hollow Creek Rd, Unit B. Cause: washing machine supply hose burst on the second floor, water ran down through the ceiling into the kitchen and finished basement. Drywall, kitchen cabinets, and laminate flooring affected; insured also reports a ruined area rug and a TV. Mitigation company already on site, verbal estimate around $12,500 for water extraction and demo. No photos uploaded yet. Insured mentioned the unit is partially rented to a tenant. Possible mold starting per the mitigation tech.
Robert Alvarez
structured by ClaimsReaderClaim #WC-2025-33871
Policy #HO3-558210
Date of lossNot stated
PerilWater
Property2271 Hollow Creek Rd, Unit B
Coverages referenced
Dwellingapplies
Personal propertymentioned
Loss of use / rentalunclear
Moldexcluded?
Line items
| Water extraction & demo (verbal est.) | $12,500 |
| Drywall, cabinets, laminate flooring | — |
| Area rug & television | — |
Estimated total$12,500
Coverage gaps & flags
- No clear date of loss — "possibly last weekend" needs confirming.
- No photos or written estimate uploaded yet; estimate is verbal only.
- Possible mold — confirm policy form; mold is often sub-limited or excluded.
- Unit is partially rented — check loss-of-use vs. landlord/tenant interest.
Second-floor washing-machine supply hose failed, causing water damage down through the kitchen and finished basement. Verbal mitigation estimate ~$12,500; documentation and a firm date of loss are outstanding, and possible mold plus a tenant occupancy raise coverage questions to resolve before the file advances.
Illustrative example. Every analysis is generated live from the exact text you paste — nothing here is pre-filled into your results.
Who it's for
One first read. Four different desks.
Triage a file before you open it
Paste the FNOL and see the whole claim at a glance — peril, exposure, and the missing pieces — so you walk into the file already knowing what to chase.
Standardize the intake read
Every new claim gets the same structured first pass, regardless of who logs it — consistent facts and flags feeding triage and assignment.
Know the scope before the truck rolls
Turn a carrier's assignment or estimate into a clear scope and dollar figure in seconds — and spot the documentation a carrier will want before you start.
Catch the gaps on page one
Surface missing dates, absent documentation, and possible exclusions up front — so the easy misses don't become the costly ones downstream.
Built on Astra AI
Reading is the easy part to give away.
ClaimsReader is built by Astra AI on MIND — our knowledge-graph platform for software that reasons over messy, real-world documents. The judgment stays with your adjusters; the slow, mechanical first read is what the machine takes off their plate. It reads what you paste, in the moment, and hands the structure back.
Nothing stored
Analysis happens in the moment. Claim text isn't retained after your result is returned.
Never invents facts
It works only from the text you paste — no fabricated names, numbers, or coverages.
A review aid, by design
It frames and flags for a licensed professional. It does not make coverage determinations.
Patents pending
Built on Astra AI's MIND platform — the same engine behind our claims-review tools.
Questions
What teams ask before they trust it.
Does ClaimsReader make a coverage decision?
No. ClaimsReader is a review and triage aid. It structures what's in a claim and flags gaps a reviewer might raise — but it does not make coverage determinations or give legal advice. A licensed adjuster reviews every claim and owns the call.
What can I paste in?
Any claim text — an FNOL, an adjuster's narrative, a carrier assignment, an estimate, or a recorded-statement transcript. Paste the words and ClaimsReader reads them. There's no required format or template.
Will it ever make up details that aren't in the claim?
It's instructed not to. ClaimsReader works only from the text you paste — if a fact isn't there, the field comes back empty rather than guessed. The coverage-gaps field exists precisely to call out what's missing.
What happens to the claim text I paste?
Analysis happens in the moment to generate your result. The claim text isn't stored after the response is returned. Treat the tool as a working aid and follow your organization's own handling rules for sensitive claim data.
How accurate is the extraction?
It's strong on the facts that are clearly stated and conservative where the text is ambiguous — by design, it would rather flag "not stated" than invent an answer. It's a fast, consistent first pass, not a substitute for a reviewer's read of the source documents.
Can the output feed our existing workflow?
Yes. The result is structured and copyable as JSON, so it drops into intake, triage, a spreadsheet, or a downstream review such as FRVerify. ClaimsReader is meant to remove the slow first read, not replace your system of record.
The slow first read shouldn't set the pace of the whole desk.
Paste a claim and see the facts come back in seconds. No setup, no upload, nothing stored.