Dose-response signal estimation and optimization for salesforce management

Abstract

Estimating generalizable relationships between actions and results from historical samples, especially when there is a level of noise or randomness in that signal, is an important problem to address before making decisions on actions to take. Many business analytics problems require the optimal assignment of limited resources to actions and activities to maximize some result or objective such as profit. We present a novel approach to solving this class of analytics problems by modeling the relationship between resource effort and expected return as a dose-response signal and formulating its causal estimation as one of kernel regression. The estimated expected value and variance of the result are then used to optimize resource allocation so as to maximize expected response while minimizing the risk around response subject to business constraints. We apply this approach to the task of optimally assigning salespeople to enterprise clients using real-world data, and show that profit can be substantially increased with fewer salespeople and reduced risk.