Facial verification for health claims: is this the insured patient?
YIRA MedSense compares the photo on the member's ID with the photo taken at the hospital and returns a confidence level, a similarity score and an image-quality check for each photo, so claims teams and investigators can spot identity substitution before a claim is paid.
Medically reviewed by Dr. N. S. Santosh Varma, MBBS, FIIM, FIDM, FICCM · Medical Officer, YIRA MedSense · Updated 13 October 2026
{
"photo_1": "member ID card",
"photo_2": "hospital admission photo",
"confidence_level": "shown to reviewer",
"similarity_score": "shown to reviewer",
"image_quality": { "photo_1": "checked", "photo_2": "checked" },
"threshold": "set by your team",
"decision": "human reviewer"
}Illustrative structure. A score is a review signal, not validated accuracy.
Identity substitution is hard to see on paper.
In identity fraud, someone other than the insured person is admitted and treated under the member's policy. The claim file looks normal; only the face gives it away.
Borrowed policies
A family member or friend is treated under another person's cover.
Visual checks don't scale
Comparing ID copies by eye is slow and inconsistent across reviewers.
Poor photos
Low-quality scans and phone photos make manual comparison unreliable.
No audit trail
Manual checks rarely record what was compared and why it passed.
Two photos in. A clear review signal out.
Collect the photos
The member's ID photo and the photo taken at the hospital arrive with the claim documents, by API, mobile app or scanner portal.
Check image quality
Each photo gets a quality score, so a weak comparison caused by a poor image is visible, not hidden.
Compare the faces
MedSense returns a confidence level and a similarity score for the pair.
Flag for review
Pairs below your threshold go to an investigator, alongside prior-record matches and the structured medical file.
Every signal in one view.
| Output | What it tells the reviewer |
|---|---|
| Confidence level | How confident the model is in the comparison |
| Similarity score | How closely the two faces match |
| Image quality (each photo) | Whether a weak result could be caused by a poor photo |
| Threshold used | The cut-off your team has set for flagging |
| Linked claim file | Structured medical history, lab values and prior-record matches for the same claim |
A confidence level or similarity score is a review signal, not validated accuracy and not a fraud decision. Your team decides.
Part of the MedSense claims review.
Facial verification works best alongside MedSense prior-record matching, which flags documents reused from earlier claims, and medical document extraction, which turns the claim file into structured data.
Pre-authorisation
Check identity before treatment is approved.
Claim review
Confirm identity before payment.
Fraud investigations
Give SIU teams a recorded, repeatable comparison.
Audits
Re-run checks across past claims.
In production, at scale.
MedSense runs in production at Health Assist Insurance TPA Pvt Ltd. It has parsed more than 1 lakh (100,000+) medical documents, flagged over 5,000 fraud cases for review and supported 100+ insurance operations.
Sandbox
Test MedSense on your own sample documents.
Provision
Set up document types, fields and rules.
Integrate
Connect by API to your claims or payer system.
Go live
Run in parallel, then switch over.
Timing depends on scope, data access and integration readiness. Client-managed deployment is available, so MedSense can run inside your own infrastructure.
Questions from claims teams
How does facial verification work for health insurance claims?
YIRA MedSense compares the member's ID photo with the photo taken at the hospital and returns a confidence level, a similarity score and an image-quality check for each photo. Pairs below your threshold are flagged for a human reviewer.
Does MedSense decide that a claim is fraudulent?
No. Facial verification produces review signals. Investigators and medical officers make every decision.
What if one of the photos is poor quality?
MedSense scores the quality of each photo, so a weak comparison caused by a blurry scan or a dark phone photo is visible to the reviewer.
Can we set our own threshold?
Yes. The threshold for flagging a pair is configured with your team during provisioning.
Is facial verification used together with fraud detection?
Yes. It runs alongside prior-record matching, which flags documents reused from earlier claims, so investigators see both signals in one place.
Can MedSense detect edited or forged photos?
Tampering and forgery checks can be built for your requirements as a scoped custom module. They are not part of the standard product today.
Where is the data processed?
MedSense supports client-managed deployment, so it can run inside your own infrastructure. Hosting and data residency are agreed per contract.
Book your demo today.
See MedSense on the documents and fields that matter to your claims team.