It reads what no one has time to read.

OwlVision is the document-AI layer under modern claims. It reads every file that lands, from medical records to police reports to handwritten notes, and turns it into one connected, cited evidence graph.

See OwlVision read a real file
- Session- With- Length
Live walkthroughClaims Engineer30 minutes
Vision, CLM-2026-04412reading
Source: p. 47

Patient presents with persistent lumbar pain (8/10) radiating to right lower extremity. Imaging 06/22: L4–L5 disc protrusion, 4mm. Plan: PT 2× weekly, follow-up 6 weeks. Work status: modified duty, 20 lb limit.

Extracted
Diagnosis
L4–L5 disc protrusion
p.47, 0.98
Pain score
8 / 10, right LE
p.47, 0.97
Plan
PT 2×/wk, 6-wk f/u
p.47, 0.95
Work status
Modified, 20 lb limit
p.48, 0.94
0+
document types
0.0%
extraction accuracy
0.0M
pages a day
0%
outputs cited
01The category leader, by volume

0

Claim documents read by OwlVision. And counting.

- As of- And
December 2025still climbing
  • Real documents.

    Every page counted is a real claim document, not a synthetic benchmark.

  • Real decisions.

    Each one helped an adjuster, underwriter, examiner or investigator decide faster, fairer, with more data.

  • A compounding flywheel.

    The next billion pages get processed better than the last. The lead widens.

02The ingest, live

Every document on the claim, converging into one cited file.

Claim
NWR-2026-04412
Medical records
IME, ER triage, MRI reads. 184 pp.
Demand letter
Plaintiff counsel, 19 pp.
Scene photos
47 JPGs from FNOL, EXIF intact.
Wage-loss tables
Employer payroll, 36 months.
FNOL recording
14:32 call, transcribed.
PDF
Police report
CHP 555, diagram + narrative.
Pharmacy fax
Rx history, image-only TIFF.
Adjuster notes
Field notebook scans.
Broker emails
38 messages, .EML thread.
PIP application
Insured-completed claim form.
Telematics clip
12s MP4, impact event.
W-2
Tax filings
W-2s and 1099s.
IME report
Defense IME, Dr. Patel, 24 pp.
Deposition
Plaintiff dep, 312 pp.
Vehicle damage
83 photos, multiple angles.
Medical billing
CPT-coded, 8,412 lines.
Recorded statement
Insured RS, 22:14 verbatim.
PDF
ER discharge
St. Mary's, 12 pp.
Field diagram
Adjuster sketch.
Counsel thread
Defense + plaintiff, 14 msgs.
Subro questionnaire
Third-party detail.
Dashcam footage
4:21 clip, forward-facing.
W-2
1099-MISC
Self-employment income.
PT records
Provider fax, 78 visits.
Medical records
IME, ER triage, MRI reads. 184 pp.
Demand letter
Plaintiff counsel, 19 pp.
Scene photos
47 JPGs from FNOL, EXIF intact.
Wage-loss tables
Employer payroll, 36 months.
FNOL recording
14:32 call, transcribed.
PDF
Police report
CHP 555, diagram + narrative.
Pharmacy fax
Rx history, image-only TIFF.
Adjuster notes
Field notebook scans.
Broker emails
38 messages, .EML thread.
PIP application
Insured-completed claim form.
Telematics clip
12s MP4, impact event.
W-2
Tax filings
W-2s and 1099s.
IME report
Defense IME, Dr. Patel, 24 pp.
Deposition
Plaintiff dep, 312 pp.
Vehicle damage
83 photos, multiple angles.
Medical billing
CPT-coded, 8,412 lines.
Recorded statement
Insured RS, 22:14 verbatim.
PDF
ER discharge
St. Mary's, 12 pp.
Field diagram
Adjuster sketch.
Counsel thread
Defense + plaintiff, 14 msgs.
Subro questionnaire
Third-party detail.
Dashcam footage
4:21 clip, forward-facing.
W-2
1099-MISC
Self-employment income.
PT records
Provider fax, 78 visits.
Medical records
IME, ER triage, MRI reads. 184 pp.
Demand letter
Plaintiff counsel, 19 pp.
Scene photos
47 JPGs from FNOL, EXIF intact.
Wage-loss tables
Employer payroll, 36 months.
FNOL recording
14:32 call, transcribed.
PDF
Police report
CHP 555, diagram + narrative.
Pharmacy fax
Rx history, image-only TIFF.
Adjuster notes
Field notebook scans.
OWLVISION, CLAIM NWR-2026-04412○ IDLE
Documents ingested
00 / 56
Ingest
Documents
0 / 56
Phase
Scattered
Cited fields
Contradictions

From a stack of PDFs to a cited claim file, in seconds.

OwlVision runs four stages on every claim that arrives. Ingest, classify, extract, reason. Every stage is deterministic, every output traces back to the page it came from, and the whole pipeline is finished before the file reaches an adjuster.

01Ingest

Every page, every file type, on arrival.

OwlVision picks up documents the moment they land, from claim portals, broker emails, EDI feeds, scanned faxes, attachments, handwritten field notes. PDF, TIFF, DOCX, MP3, MP4, JPG, EML, raw text. No manual sorting, no triage queue.

In this loop
  • PDF, TIFF, DOCX, MP3, MP4, JPG, EML

  • IDP/OCR + handwriting + recorded audio

  • Reads attachments inside attachments

01 / 04
03The extraction engine

Every value cites its page.

OwlVision returns structured fields, not predictions. Every diagnosis, date, dollar amount, and statement of fact carries the page number, paragraph, and confidence score behind it. Click any value and the source highlights.

  1. 01220+ document types out of the box, across every line of business
  2. 02Page-level citations on every extracted value
  3. 03Confidence scores per field, calibrated against ground truth
  4. 04Cross-document reasoning: contradictions, gaps, timeline conflicts
  5. 05Native write-back to the modern claims-management system you run
Source: p. 47● reading

Patient presents with persistent lumbar pain (8/10) radiating to right lower extremity. Imaging on 06/22: L4-L5 disc protrusion, 4mm. Plan: continued PT 2x weekly, follow-up in 6 weeks. Patient reports compliance with home-exercise program. No new injuries since prior visit. Work status: modified duty, 20 lb lifting restriction.

Extracted, structured
Diagnosis
Lumbar disc protrusion, L4–L5
p. 47, 0.98
Pain score
8 / 10, right lower extremity
p. 47, 0.97
Finding
Protrusion 4mm
p. 47, 0.96
Plan
PT 2×/wk, follow-up 6 wks
p. 47, 0.95
Work status
Modified duty, 20 lb limit
p. 48, 0.94
04The full-context principle

Why competitors miss what we catch.

Single-category tools read one kind of document well. OwlVision reads the whole claim (220+ document types across seven categories, every line of business) so the contradictions that only appear across documents actually surface.

Medical

  • ER triage
  • IME reports
  • MRI / imaging
  • Operative notes
  • PT progress
  • Pharmacy

Legal

  • Demand letters
  • Depositions
  • Pleadings
  • Mediation briefs
  • Arbitration awards

Financial

  • Wage-loss tables
  • Tax filings
  • Medical billing
  • EOB statements
  • Liens & invoices

Incident

  • Police reports
  • FNOL transcripts
  • Telematics
  • Dashcam / body-cam
  • Scene photos

Activity

  • Surveillance
  • Witness calls
  • Recorded statements
  • Field diagrams

Correspondence

  • Broker emails
  • Counsel threads
  • Insured messages
  • Reservation of rights

Identity & ops

  • IDs & licenses
  • Authorizations
  • Loss runs
  • Property schedules

220+

document types, one connected graph

05Cross-document reasoning

The contradictions that close a file.

Reading one document is table stakes. The value is in the seams: when the IME contradicts the deposition, when the FNOL audio contradicts the medical record. OwlVision links the whole file and surfaces the conflict, cited on both sides.

1 ContradictionCLM NWR-2026-04412
Medical record, p. 62

Patient reports no new injury since prior visit; symptoms attributed to original date of loss.

FNOL recording, 09:14

Claimant states they re-injured the back on 06/14 lifting at a second job.

Contradiction surfaced. Material to causation and apportionment; flagged before first human review.
06The compounding advantage

A claim that's read on day one is a claim that closes faster, fairer, and with less leakage.

OwlVision is the foundation under everything else Owl does. OwlSignal flags what matters because OwlVision read the file. OwlAssist answers because OwlVision structured it. Read the whole file, and the rest of the operating model unlocks.

Audited extraction accuracy across the full claim file.

Measured against a held-out test set of 18,400 documents, hand-labelled by senior examiners. OwlVision is benchmarked against the leading frontier models (GPT- and Claude-class) and against the single-category insurance tools carriers run today.

OwlVision
GPT-class frontier
Claude-class frontier
Mono-category tools
40%55%70%85%100%Medical recordsIMEs & expert reportsLegal pleadingsDemand lettersPolice reportsDamages & wage-lossRecorded statementsHandwritten field notes
METHODOLOGY: 18,400-DOC HOLD-OUT, BLINDED, Q1 2026Read the full report →
96%
OwlVision avg
70.8%
GPT-class avg
73.2%
Claude-class avg
82.2%
Mono-category avg
07Customer case study

How OwlVision was employed by a Tier-1 Disability Insurance to solve slowed productivity and inconsistent workflows

Following a successful pilot, a leading disability insurance carrier expanded OwlVision across the organization. Today, approximately 1,000 employees use OwlVision and OwlAssist to review claim documents, generate templates, and instantly find information across complex claim files. As adoption grew, manual work remained a major bottleneck. Completing claim templates typically took 10–15 minutes, while answering even simple questions required searching through hundreds of pages of medical records and supporting documents—slowing productivity and creating inconsistent workflows. OwlVision automated document-intensive tasks while OwlAssist enabled users to ask natural-language questions and receive evidence-backed, source-linked answers in seconds.

Result
+0×
employees Adopted OwlVision
0M
pages processed
0K
Templates Generated
0%
Template Conversion
08Frequently asked

The questions claims leaders ask first.

OwlVision is a domain-specific system, not a prompt over a frontier model. Every page is classified into an insurance-native taxonomy (220+ types), every value is extracted with deterministic, page-cited logic, and the whole pipeline is benchmarked against hand-labelled examiner ground truth, not generic OCR or web text.

09See it on your file

Bring us a closed claim. We'll read it back to you in fifteen minutes.

The fastest way to see what OwlVision does is to watch it read a claim you already know cold. Briefings are by request, held under NDA, and run with a claims engineer on the call, not a sales team.