Case Study: Zocdoc achieves rapid, accurate insurance verification and faster patient appointments with Amazon Web Services

Zocdoc builds patient confidence using TensorFlow on AWS

Zocdoc helps patients find the right, in‑network care amid the complexity of U.S. health insurance, where many people struggle to understand coverage and want an easier way to check benefits. The company faced two linked challenges: extracting reliable policy details from a wide variety of insurance ID cards submitted as photos, and matching patients to appropriate, available doctors to reduce long wait times (Zocdoc reports patients can get appointments within 24 hours versus a national average of 24 days).

Zocdoc built Insurance Checker using TensorFlow on AWS: customers snap a photo of their insurance card, and a deep‑learning pipeline (a base CNN classifier, an alignment model, and an OCR model) classifies carriers/plans, locates and reads member IDs, and verifies coverage in real time. Trained on millions of labeled cards on AWS GPU instances (using the AWS Deep Learning AMI, Keras, and EC2 p2.8xlarge servers), the system achieved nearly 90% accuracy—better than patient input—was rolled into production, runs entirely on AWS, and continues to improve with ongoing training and more data.

ZocDoc

Serkan Kutan

Chief Technology Officer