Efficient Dynamic Methods for Predictive Action Strategy Optimization for Risk Driven Multi-channel Communication

In this work we examine how to optimally plan for multi-channel communication campaigns. We primarily consider proactive campaigns that can take advantage of predicted risk or other behavioral propensity measure relating to the intended communication recipient(s). These predicted or estimated measures identify portions of the population that the servicer/campaign originator is interested in targeting. For example, in the application of loan servicing various risk segments low, medium and high predicted risk can be delineated and targeted differently. Each group not only has a different propensity to pay back a loan or liability but also different levels of sensitivity/conversion rates (to current status) with regards to communication campaigns. First we derive risk models that can predict the payback risk associated with loans at different times. The prediction techniques are not covered in this work. We then formulate the risk-driven campaign optimization problem and sub-problems, and solve them using various efficient dynamic techniques. Two main sub-problems are proposed relating to either single-or multi-channel communication. Using empirical values of cost structures and response rates from pilots we conducted in financial services, we were also able to run simulations to delineate regimes of optimal operation that might be appealing to loan servicing entities. Our simulation results demonstrate the optimal operational conditions with respect to different risk profiles, and at the same time, show how these optimal operational conditions change with respect to user defined parameters such as preference of one communication channel over the other.