Goal
Reconcile all transactions and achieve “touchless-claims” milestone
Challenge
Traditional claims processing involves receiving and manually processing several forms of documentation from multiple parties, evaluating the claim, and determining the benefit amount.
While other vendors provide touch-less claims solutions, they struggle to process poor-quality images such as those with noisy backgrounds, reflection, engraved letters / numbers on cards, pictures of scanned documents, etc.
Outcome
Trained model to process images of documents, VIN, and accident damage submitted through various input channels and of varying quality (noisy backgrounds, reflection, etc.)
Classify 56 different document types with > 98% accuracy Intelligently extract key data from each document, with 99.5%
word level accuracy
Prevent fraud by cross referencing across multiple documents