Automate claims processing
Unstract reads the first notice of loss, photos, police reports and estimates as they arrive, checks them against the policy, and opens the claim in your claims system.
Trusted by insurance teams around the globe
Claims arrive as attachments, not as data
The policy is already in your claims system. The first notice, the photos, the police report and the estimates arrive by email and upload, and an adjuster keys them in while the claimant waits for a call back.
A claimant on hold while an adjuster keys the claim form
A police report scanned crooked, with a stamp over the date
A loss date that differs between the form and the report
Photos and the estimate arriving in separate emails, days apart
A repeat claimant spotted only after the payment went out
Every one of them has to be read, checked against the policy and keyed into the claims system before an adjuster can decide anything.
Unstract reads the claim form, the photos, the police report and the estimate as they arrive, checks them against each other and the policy, and opens the claim. An adjuster sees only the claim files that don't match.
From first notice of loss to the claims system, without the keying
Six steps, run on their own for every claim. Parsing by LLMWhisperer, extraction by Document Studio, the whole run by Flow.
Pick up the first notice wherever it lands
First notice of loss comes in by email, web form or agent upload, with photos, a police or incident report, repair estimates and medical bills. Each document is picked up where it arrives and tied to the claim it belongs to, one record per claim.
Nobody saves attachments to a folder and forgets them. The run starts the moment the first notice exists.
Split the claim file into the documents inside it
A claim often arrives as one PDF, with the claim form, the photos, the police report and the estimate scanned together. The file is split into its documents and each one is classified by type, so the right fields come out of each.
A medical bill is read as a medical bill and an estimate as an estimate, in whatever order they were scanned.
Read every document, then pull the fields that matter
LLMWhisperer reads each document as text with its layout kept, so a crooked scan or a handwritten claim form comes out as cleanly as a PDF. Document Studio then extracts the fields you defined: claimant, policy number, date, place and cause of loss, damaged items, injuries, and the totals on every estimate and invoice.
The same definitions run on every claim, whatever the police department's report or the body shop's estimate looks like.
Check it against the policy before anyone touches it
The policy is checked as in force on the loss date, the loss type as covered, and the deductible and limits applied. Repeat claimants, dates that don't line up and invoices above the estimate are flagged. A claim that passes keeps moving. One that fails stops in a review screen with the field, the document and the mismatch side by side.
Set the human-in-the-loop conditions once (coverage in question, a fraud flag, an amount above authority) and only those claims wait.
Send each claim down the right path
Rules on the extracted fields set each claim's path. Simple claims are fast-tracked, complex or injury claims go to the adjuster with the right authority, and flagged ones go to the special investigations unit with the reason attached.
Claims triage happens the moment the file is read, not when someone reaches the queue.
Open the claim where your adjusters already work
The claim, its documents and the extracted data land in your claims system, Guidewire ClaimCenter, Duck Creek Claims or another, with the assignment in the adjuster's work queue. Each decision a person made along the way is recorded against the claim.
Nobody keys the claim twice, and the audit trail builds itself as the run goes.
See it run on a week of your own claims.
Bring the first notices, photos and estimates from one line of business. We run them live and show every field, every check and every exception.
Everything that arrives with a claim
Documents
- Loss and damage claims
- ACORD forms
- Automobile accident reports
- Police and incident reports
- Health insurance claim forms
- Itemized medical bills
- Employee reports of injury
- Disability benefit applications
- Travel insurance claim forms
- Pet insurance claims
- Warranty claim forms
- Damage photos, repair estimates and invoices
- Policy records and earlier claims from your claims system
- Witness statements
Fields captured
Our agents are extroverts.
Cloud storage like S3, Google Drive, Dropbox and Box. Databases and warehouses like Snowflake, BigQuery, Postgres and SQL Server. Accounting, ERP and CRM systems like QuickBooks, Xero, SAP, NetSuite, Salesforce and HubSpot. Project trackers like Jira, Asana, ClickUp and Linear. And anything with an API.
Live on one line of business first
No developers on your side, and no big-bang cutover. Claims processing is built on your own claim files and proven on one line of business before it goes further.
Book a consultation
We look at a week of your claim files together: which documents arrive, how, and what an adjuster checks before a claim is opened.
Built on your documents
The fields, the checks against your policy and claims data, and the connection to your claims system are set up on your claim forms, police reports and estimates, not samples.
Pilot on one line of business, then roll out
One line of business runs live within days. When the numbers hold, the other lines follow, and the exception rules carry over.
Built for the teams that run the claims desk
First notices opened in the claims system within minutes, and only the claims that need judgment on an adjuster's desk.
Medical claims with the bills, codes and provider details checked before anyone keys them, and disability claims read the same way.
Each client's claim forms read with that client's own coverage rules, checks and claims system.
Repeat claimants, dates that don't line up and invoices above the estimate flagged as the file is read, so fraud review starts at first notice.
Claims opened in minutes, exceptions caught at first notice
Safe to run your claim files through
Claim files carry policy numbers, driver's license and vehicle details, medical bills and injury details: the personal and health data a breach notice would have to list.
- Encrypted in transit and at rest.
- Claim files are never used to train models.
- Access by role, line of business and document type, with every review and correction recorded against the claim.
- Retention you set: processed claim files are deleted on the schedule you choose.
- Self-hosted deployment for carriers whose controls keep claim and health data inside their own environment.
What claims managers ask before they start
First notices of loss and ACORD loss notices, claim forms for auto, property, health, disability, travel, pet and warranty claims, police and incident reports, employee injury reports, damage photos, repair estimates, invoices and itemized medical bills. Policy records and earlier claims come from your claims system, and you can add anything else that arrives with a claim.
Your claims management software holds the policy and the claim once someone has opened it. The first notice, the photos, the police report and the estimate still arrive as attachments, and an adjuster reads them and keys the claim in. Automated claims processing takes that step over: Unstract reads them, checks them against the policy and opens the claim, so adjusters start from clean data.
Classify each document as it arrives, read the fields that decide the path, and apply your rules to them. Unstract splits the claim file, reads the loss type, injuries, amount and coverage, and routes each claim: simple ones to the fast track, injury or complex ones to an adjuster with the right authority, flagged ones to your special investigations unit.
Two kinds of tool are involved. Capture tools read documents into fields; a claims management system stores the claim. Unstract does the work between them: it reads every document in the claim file, checks it against the policy and the other documents, and opens the claim in your claims system, holding only the exceptions for a person.
Look for two things: a read that keeps the page layout, so a value comes from the right box on a crooked scan or a fax, and a confidence score on every value. LLMWhisperer reads with the layout kept, and any value below the threshold you set stops for a person with the page beside it instead of reaching the claims system as a guess.
Yes: the policy in force on the loss date, the loss type against the coverage, the deductible and limits, the loss date and place across the claim form, the police report and the photos, the estimate against the invoices, and earlier claims for the same claimant or vehicle. A failed check holds the claim with the reason attached.
It stops in a review screen that shows the field, the document it was read from and the value it was checked against, side by side. An adjuster confirms it, corrects it or sends it back, a fraud flag goes to your special investigations unit with the reason, the decision is recorded against the claim, and every other claim keeps moving.
To the systems your adjusters already use: the claim, its documents and the extracted data to your claims system, such as Guidewire ClaimCenter or Duck Creek Claims, the assignment to the right adjuster's work queue, and fraud flags to your special investigations unit. Flow posts it as part of the run, so nothing is downloaded and keyed again.
Unstract is SOC 2, ISO 27001, GDPR and HIPAA compliant. Claim files are encrypted in transit and at rest, never used to train models, and kept only as long as you choose. Self-hosted deployment keeps claim and health data inside your own environment; see the Trust Center.
You don't need developers. In a consultation we build the claims run around your own documents, your checks and your claims system, run it on one line of business, and most teams are live there within days. The other lines follow when the numbers hold.
Bring a week of claim files to the consultation
The first notices, photos and estimates from one line of business, the crooked scans included. You watch each claim open, and every held claim shows why it stopped.