Luisa F. Polanía, American Family
In the last few years, deep learning has revolutionized computer vision, speech recognition and natural language processing and it has already started to disrupt insurance. This is motivated by the increasing availability of data in the insurance industry and the remarkable performance of deep learning algorithms.
In this talk, we will present two of the deep-learning related projects we are working on at American Family Mutual Insurance Company. The first project is related to body mass index (BMI) prediction from selfie images. The motivation for this project is to offer life insurance to our customers without requiring an uncomfortable physical exam to measure weight and height and other physical condition indicators. The two core components of this project are an image guidance app to acquire good-quality facial images and a convolutional neural network-based model to predict BMI. The applications of deep learning at AMFAM are not only limited to life insurance, but also extend to insurance underwriting. For example, we are using roof images to predict roof age, which is one essential factor to calculate an age-based depreciated value of a roof for insurance purposes. This project will be presented in the second part of the talk. Predicting roof age from images alleviates the problem of relying on self-reported roof age, which is typically underestimated and leave carriers with unreliable data built into the claims process. For home insurers, having wrong roof ages is detrimental given the increased exposure to risk associated with older roofs.