Maximizing Customer Lifetime Value using Stacked Neural Networks: An Insurance Industry Application

Abstract

This paper proposes a recommender system based on two-stage neural network architecture that maximizes Customer Lifetime Value (CLV). The Stage-I neural network uses a self-attention mechanism and a Collaborative Metric Learning (CML) to generate product recommendations. The Stage-II neural network uses a neural network-based survival analysis to infer insurance product recommendations that maximize customer lifetime. The proposed stacked neural network model can be used as a generative model to explore different cross-sell scenarios. The applicability of the proposed recommendation system is evaluated using transactional data from an Australian insurance company. We validated our results against a state of the art self-attention recommendation system, successfully extending its functionality to include lifetime value.