What challenges did Zocdoc face before AI implementation?

Patients struggle to understand health insurance coverage and whether a doctor is in-network. The manual process required deciphering photos of insurance cards with varied layouts and image quality.

What was the impact of the AI solution?

Achieved accuracy of almost 90%. The system was more accurate than patients’ own input. Rolled the solution into production. Enabled real-time coverage verification, in-network checks, and estimated co-pay for faster appointments. Zocdoc said patients can get appointments within 24 hours, compared with a national average wait time of 24 days for new patients.

Source

Zocdoc Insurance Checker: OCR + deep learning on AWS to extract health insurance policy IDs

Use case typeIntelligent document processingUpdated Jun 13, 2026

Zocdoc built Insurance Checker to help patients understand and verify health insurance coverage and in-network status by scanning photos of insurance cards. The system uses deep-learning-based computer vision to extract member ID information, classify carriers and plans, and perform OCR on varied card images. Zocdoc trained multiple neural network models on millions of labeled insurance cards and rolled the solution into production on AWS.

Organization: Zocdoc
Industry: Healthcare
Location: United States
Published: January 2018

Reported outcomes

90%

accuracyQuality & accuracy

24 daysappointment wait time compared to national average

Strategic outcomes

New product / capabilityBuilt insurance card OCR verification toolNew product / capabilityEnabled real-time coverage verificationCustomer experience & trustImproved patient appointment accessNew product / capabilityRolled the solution into production

Catalog median for quality & accuracy deployments: +90% across 282 reported metrics. Compare benchmarks →

Primary read

Use case focus

Showing 3 of 4

1Computer Vision
2OCR
3Machine Learning

Patients struggle to understand health insurance coverage and whether a doctor is in-network.
The manual process required deciphering photos of insurance cards with varied layouts and image quality.

Zocdoc used TensorFlow deep learning models on AWS Deep Learning AMIs and Amazon EC2 p2.8xlarge GPU instances.
The Insurance Checker applies computer vision for image classification, alignment, cropping, and OCR.
The models extract policy ID fields, classify carrier and plan, and support real-time coverage verification, in-network checks, and estimated co-pay.
The team continuously improved the models after production rollout using additional labelled customer data.

Technologies

TensorFlow AWS Deep Learning AMIs Amazon EC2 Keras Computer Vision OCR

Achieved accuracy of almost 90%.
The system was more accurate than patients’ own input.
Rolled the solution into production.
Enabled real-time coverage verification, in-network checks, and estimated co-pay for faster appointments.
Zocdoc said patients can get appointments within 24 hours, compared with a national average wait time of 24 days for new patients.

Architecture

Zocdoc trained TensorFlow-based computer vision models on AWS Deep Learning AMIs using Amazon EC2 p2.8xlarge GPU instances. The architecture included separate models for classification, alignment, and OCR to process insurance card photos with blur, background, size, and orientation variability.

Sources & evidence1

Groundedness: 5/5Type: Blog PostPublished: Jan 30, 2018Publisher: AWSEvidence: VendorConfidence: Medium

Primary source

AI-generated summary. Verify important details with the linked sources before relying on this case.

Explore related AI use cases

Customer evidence