Extract data from health insurance claim forms
Unstract reads CMS-1500 and UB-04 claim forms, scanned, photocopied or faxed, and returns checked, structured data to your claims or core admin system.
PINEWOOD HEALTH PLAN · CLAIMS · PO BOX 4410, SALEM, OR 97302
corrected claim — DX ptr fixed, see box 22
APPROVED OMB-0938-1197 FORM 1500 (02-12)
{ "form": "CMS-1500", "edition": "02/12", "claim_type": "corrected", "member_id": "***-****902-01", "patient": "Morales, Elena R", "diagnosis_a": "J02.9", "cpt_line_1": "99213", "total_charge": 177, "billing_npi": "1639201847", "confidence": 0.97}
Trusted by insurance teams around the globe
Paper claims rarely come in the way the form was designed
A 1500 printed on red dropout stock and fed straight to a scanner is the easy case. The mailroom sees this instead.
Photocopied in black, so the red form no longer drops out
Two editions of the 1500, and the UB-04 for facility claims
Service lines filled in by hand and corrected in pen
Claim, itemized bill, primary EOB and records in one PDF
Itemized statements and reimbursement forms in every layout
Each one is keyed by hand before it can be adjudicated, and every keying slip comes back as a denial or a corrected claim.
Unstract takes that on — it reads both editions of the 1500 and the UB-04, pulls the codes and charges you name, checks them against eligibility and the rest of the claim, and sends the record on to adjudication.
Parse, extract and automate claims on one platform
Three products, each useful by itself: read the claim, turn it into data, run the intake around it. Start with whichever layer your claims operation needs first.
A health insurance claim parser that keeps every box and service line
Get the full text of a 1500 or UB-04 in whatever form it reached you, scanned in the mailroom, photocopied without its red ink, or faxed from a practice, with the form's layout intact. Each service line stays on its row, so a charge is never paired with the wrong procedure code, and the diagnosis pointers stay beside the lines they point from.
Pass the text to your own LLM, or use parsing as the first step of a claims workflow.
Learn more about LLMWhispererTurn health insurance claim forms into structured, checked data
Describe the member, provider, diagnosis and service-line fields you need, and every claim comes back as the same JSON record, whether it was a professional 1500, a facility UB-04 or a member's reimbursement form. Values are checked as they come out: a member who wasn't eligible on the date of service, or a pointer to a diagnosis the claim doesn't list, arrives flagged with the reason.
Call it as an API from your intake system, run it as an ETL pipeline from a scan folder, or make it one step of an orchestrated claims process.
Learn more about Document StudioRun paper claims intake from mailroom to adjudication
Describe your paper claims intake in plain English and Flow's agents build the workflow: claims picked up from the scan queue and the fax line, each document in the file sorted, every claim read and checked for eligibility on the date of service, duplicates and code mismatches, then routed. Clean claims go straight on to adjudication in your claims or core admin system, with the images attached. You approve the workflow once, and it runs the same way on every claim.
Choose when a person steps in: coverage in question, a fraud flag, a billed amount above an examiner's authority, a read that isn't certain. Only those claims stop, and the examiner sees why.
Learn more about FlowStart with the extraction. Bring in the automation later.
Start for free covers parsing and extracting your claim forms now. Automating paper claims intake begins with a consultation.
Professional and facility claims, and what's sent with them
Documents
- CMS-1500 professional claim forms
- UB-04 institutional claim forms
- Corrected claims and claim adjustment requests
- ADA dental claim forms
- Claims for healthcare benefits
- Itemized bills and superbills
- Medical claim review and appeal requests
- Explanations of benefits from a primary payer
- Medical reimbursement requests
- Medical records sent with a claim
- Patient requests for medical payment
- FSA and HSA claim forms
- Provider inquiry forms
- A plan's own claim form, in its own layout
Fields captured
Processes these forms feed
For the teams that work paper claims
Paper and faxed claims into adjudication without a keying vendor in between, with only the unclear lines sent to an examiner.
Claims for many self-funded plans read the same way, whichever plan's form or layout each one arrived on.
Superbills, itemized statements and returned claims read back, so the missing field behind a denial is found before resubmission.
Codes, units and charges on every line, checked against the itemized bill and earlier claims before payment rather than after.
Less keying, faster to adjudication
Built to carry protected health information
A claim form carries a member's name, date of birth, plan ID, diagnoses and procedures. That is protected health information, and it is handled as such.
- Encrypted in transit and at rest, end to end.
- Claims you process are never used to train models.
- Access set by role, team and claim type, with every correction recorded against the claim it was made on.
- You decide retention: processed claims and images are deleted on your schedule.
- Self-hosted deployment for plans and TPAs that keep PHI inside their own environment. Ask us what yours needs.
What health plans and TPAs ask before they start
One that reads the form as a form. LLMWhisperer keeps every box and service line where it sits on the page, and Document Studio turns that text into the fields you defined, so a 1500, a UB-04 and a plan's own reimbursement form all come out as the same record.
Yes. Document Studio deploys an endpoint per document type: send a 1500 and get the claim back as JSON, service lines included. If you would rather drop scans in a folder or bucket, the same extraction runs as an ETL pipeline instead.
Yes. Traditional claim OCR depends on the red dropout ink, so a black and white copy defeats it. LLMWhisperer reads the form's layout rather than relying on the ink colour, and a value it can't read with certainty goes to an examiner with the image instead of into your system as a guess.
Yes: UB-04 facility claims with revenue codes and type of bill, dental claim forms, itemized statements, superbills and the primary payer's EOB on a secondary claim. Each is classified first, then read with the fields that belong to it.
The same one. FSA, HSA and medical reimbursement forms arrive with receipts and statements stapled on, so the claim is split by document, each part is read, and the amounts on the receipts are checked against the amount claimed.
Usually the keying, and not much else. Here the AI does the extraction: Document Studio uses LLMs to turn the parsed claim into your fields. The OCR underneath is not an LLM, it is a parser built for LLM pipelines. The checks, routing and handoff to adjudication are run by Flow, on rules you approve.
Leave adjudication where it is. Unstract sits in front of it: claims are read, checked and posted to your claims or core admin system through its API or a file drop, in the format it already accepts.
Run it as a pipeline rather than a document at a time. Scans land in a folder or bucket, every claim is read and checked as it arrives, and records go out in batches or one by one. Volume changes the queue length, not the steps.
Either. Use LLMWhisperer and Document Studio on their own to turn claims into data, or add Flow to run the intake around them: picking up scans, checking eligibility and duplicates, routing exceptions and opening the claim.
The claim stops in a review queue. The examiner sees the field, the box it came from and the reason it stopped, then approves or corrects it. The decision is recorded on the claim and everything else keeps moving.
Unstract is SOC 2, ISO 27001, GDPR and HIPAA compliant. Claims are encrypted in transit and at rest, never used for model training, and kept only as long as you choose. Plans and TPAs that need PHI to stay in their own environment can self-host; see the Trust Center.
In a consultation we set the intake up around your claim types, your checks and your claims system, and most teams have a first live run within days. To try the extraction yourself today, start for free and upload a 1500.
Bring a week of paper claims to the consultation
1500s, UB-04s, corrected claims and reimbursement forms, photocopies and faxes too. We process them live and walk through each field and check, and every claim that stops.