Insurance Business Process Automation Using Deep Learning Techniques

The insurance industry is one of the least digitized in our country, it is also one of the most inefficient segments of the financial services industry. Business processes are still based on paper documents which can be lost and where information can be incorrectly recorded, resulting in a huge delay in the processing of policyholders’ files. To face the accelerated evolution of an increasingly competitive market, insurance companies are starting to look for ways and tools to improve their services by acting on their responsiveness and efficiency. As the ubiquity of machine learning and artificial intelligence systems increases, they have the potential to automate operations in insurance companies, reducing costs, increasing productivity, and improving service quality. This paper aims to propose a system based on Deep Learning techniques for the classification of policyholders’ digital documents in three types and the extraction of text from them. The performance values achieved with the VGG-16 are the highest compared to other classifiers with an accuracy of 98.1%, a precision of 97.8%, 99.2% for a recall, and an AUC of 98.4%. For extraction, the AlexNet and ResNet-50 models exhibit fairly close performance with an outperformance of AlexNet with an accuracy of 97.2%, a precision of 96.7%, a recall of 97.4%, and an AUC of 98%.