ON MULTI-ARMED BANDITS AND DEBT COLLECTION

In this paper we consider minimisation of payment arrears in non-governmental organizations on the example of sport clubs. In the presented approach we focus on the optimisation of contact with customers, to motivate them for timely payments. We use multi-armed bandits to model the impact of different ways of contact on payment arrears. The method allows for efficient balancing between exploration and exploitation during runtime even for limited opportunity of customer contact. We present the architecture of the enterprise system, describe the simulations used for optimization and evaluation of the algorithms and provide design considerations. We discuss differences between considered problems and classical MABs. We propose batch learning and product arms as a way for improvement of the model performance.