The policy and every endorsement
- Pulled out
- Coverages, limits, deductibles, exclusions, endorsement wording and the policy period.
- Checked against
- The date of loss, and each estimate line, coverage by coverage.
AI that digitizes and extracts data from property claims documents
ClaimVision digitizes and extracts the data from every document in a US property claim file, then reads the file as a whole, so your adjuster starts with findings instead of fields.
That means the adjuster and inspection reports, the estimate, the policy and its endorsements, proof of loss, correspondence and photos, scanned and handwritten pages included. The AI recommends; a licensed adjuster or TPA handler agrees with or changes every recommendation.
01A worked example
One page in, four fields out, one finding with a dollar figure
This is the fictitious Sykes example claim from our walkthrough: a wind and hail loss in Oklahoma. The figures are invented demo data. What happens to them is what ClaimVision does on a real file.
An illustration, not a product screen. A page from the Sykes example claim, with what is read from it numbered.
Property inspection reportSykes claim (example) · page 2 of 9
Peril observed: wind and hail
Roof covering: asphalt shingle, installed 2009
Granule loss on N + W slopes, see photos 14-22
Location: Oklahoma
- PerilWind and hailInspection report, p. 2
- Roof installed2009 16 years old at the date of lossInspection report, p. 2
- Photo evidencePhotos 14 to 22, set beside the reportHandwritten note, p. 2
- LocationOklahomaInspection report, p. 2
Then the fields meet the rest of the file:
- FieldRoof installed 2009From the inspection report. Sixteen years old at the date of loss.
- PolicyRoof surfacing endorsementRoofs older than 15 years settle at actual cash value.
- Finding$3,830.41 depreciation withheldThe roof settles at actual cash value. Net payable $8,778.02, against $12,608.43 had the endorsement not applied. The holdback is recoverable once repairs complete within the policy's 180-day window.
An extraction tool would stop at the four fields. The finding comes from reading the inspection report, the policy and the estimate together, and it arrives with the clause and the dollar impact attached. The adjuster reads it, agrees or changes it, and decides.
02Document by document
What is pulled from each document, and what it is checked against
Every page of a US property claim file becomes structured data, whether it arrived as a digital PDF, a scan or a page written by hand. Nobody rekeys a thing.
Adjuster and inspection reports
- Pulled out
- Findings, observations and dates, handwritten field notes included.
- Checked against
- The photos. Damage the report describes and the photos don't show is flagged, and the other way round.
Estimates
- Pulled out
- Every line, down to the last one, so the numbers can be checked, not just copied.
- Checked against
- The policy limits, coverage by coverage.
Proof of loss
- Pulled out
- What was claimed and when, from a scanned or digital proof of loss.
- Read with
- The estimate and the policy, in the same reading.
Correspondence
- Pulled out
- What was asked for, what was said and when.
- Read with
- The rest of the file, so the summary's timeline includes it.
Photos
- Pulled out
- Each image looked at, with its caption and label read alongside it.
- Checked against
- The adjuster's report. Photos are grouped by the report they came from, so inspection and loss photos sit side by side.
03The money
Every figure labeled, and never shown as certain when it isn't
Financial figures carry a quality label, or are withheld until they can be confirmed. A number that two documents agree on reads differently from one that appears once in a scan, and the adjuster can see which is which.
- Gold
- Silver
- Bronze
- Withheld until confirmed
Every recommendation also shows how confident the system was, and the adjuster's note when they disagree is kept with the claim.
See the audit trail
04Beyond the fields
What a list of fields can't tell your adjuster
A property claim is decided in the connections between documents. Because ClaimVision reads the documents and the photos together, the same reading also gives the adjuster:
- A summary of the whole file, one page for a several-hundred-page claim, each point linked to its source.
- Photos checked against the report, with mismatches flagged for a person.
- Subrogation, salvage and negligence flagged, where they could change what is owed.
- Answers to plain-English questions about the claim's documents. See how to ask your claims data.
If all you need is clean data at intake, an extraction-only tool may be enough. Our guide to AI claims tools by job names several.
05Getting the data out
The data is yours, in the shape your team needs
{ }Structured JSONFor your systems and your data team.#MarkdownEasy for a person to read, and easy to drop into notes, reports or other tools.APIAPI accessSo your own systems can request the data.↓Data exportsTo load into your claims system or data warehouse.
How the data reaches your systems is agreed with your team.
06How it fits
How it fits your team and your systems
Your adjuster decides
ClaimVision recommends; it never decides, pays or denies. A licensed adjuster or TPA handler agrees with or changes every recommendation, and every source and note stays on the record.
See the audit trailYour claims system stays
No need to replace Guidewire ClaimCenter, Duck Creek, Sapiens or an in-house system. ClaimVision reads the documents the claim file already holds.
See how it fits your stackYour data stays in your region
ClaimVision runs entirely on Microsoft Azure, using Azure's built-in security, and client data stays in the client's region: US data stays in the US.
Read the security detailsWhat we can show todayA full day's reconciliation work on a real large-loss claim, done exactly, in under a minute.
FAQ
Questions buyers ask
What AI tools digitize and extract data from property claims documents?
Two kinds. General document-extraction tools pull fields such as policy numbers, dates and totals out of PDFs and scans, across many industries. Property claims tools such as ClaimVision digitize the same documents, including scanned and handwritten pages, and then read the whole file in context, so the adjuster gets a summary, flagged issues and a recommendation rather than a spreadsheet. A licensed adjuster or TPA handler agrees with or changes every recommendation.
Which property claims documents can ClaimVision extract data from?
The whole claim file: adjuster and inspection reports, estimates, the policy and every endorsement, proof of loss, correspondence, and photos with their captions. They are read together, so a line in the estimate can be checked against the policy and the photos.
Can ClaimVision read scanned and handwritten claims documents?
Yes. ClaimVision reads scanned documents as well as digital PDFs, and it reads handwriting. A scanned proof of loss or a handwritten note in a field report is read with the rest of the file, not set aside for someone to type up.
How do we get the extracted claims data out of ClaimVision?
In the format your team needs. ClaimVision exports structured JSON and Markdown, offers API access, and provides data exports you can load into your claims system or data warehouse. How the data reaches your systems is agreed with your team.
What is the difference between claims document extraction and claims intelligence?
Extraction turns pages into fields. Claims intelligence uses those fields and the full documents to answer the questions a claim is decided on: do the photos match the report, is there subrogation or salvage potential, what does the policy say about this loss. ClaimVision does both, and the decision stays with the adjuster. More on AI claims intelligence.
Can AI check claim photos against the adjuster's report?
Yes. ClaimVision looks at every photo in the file, reads its caption and label, and checks whether the adjuster's report matches what the photos show. Photos are grouped by the report they came from, so inspection photos and loss photos can be compared side by side. Anything that does not line up is flagged for a person.
Do we have to replace our claims system to use AI document extraction?
No. You do not need to replace Guidewire ClaimCenter, Duck Creek, Sapiens or a system built in-house. ClaimVision reads the documents the claim file already holds and hands back findings. The system of record stays the system of record.