Person:

Chung, Doug

This study provides a comprehensive model of an agent’s behavior in response to multiple sales management instruments, including compensation, recruiting/termination, and training. The model takes into account many of the key elements that constitute a realistic sales force setting: allocation of effort, forward-looking behavior, present bias, training effectiveness, and employee selection and attrition. By understanding how these elements jointly affect agents’ behavior, the study provides guidance on the optimal design of sales management policies. A field validation, by comparing counterfactual and actual outcomes under a new policy, attests to the accuracy of the model. The results demonstrate a tradeoff between adjusting fixed and variable pay; how sales training serves as an alternative to compensation; a potential drawback of hiring high-performing, experienced salespeople; and how utilizing a leave package leads to sales force restructuring. In addition, the study offers a key methodological contribution by providing formal identification conditions for hyperbolic time preference. The key to identification is that under a multiperiod nonlinear incentive system, an agent’s proximity to a goal affects only future payoffs in nonpecuniary benefit periods, providing exclusion restrictions on the current payoff.

This paper evaluates the short- and long-term value of sales representatives’ detailing visits to different types of physicians. By understanding the dynamic effect of sales calls across heterogeneous physicians, we provide guidance on the design of optimal call patterns for route sales. The findings reveal that the long-term persistence effect of detailing is more pronounced for specialist physicians, whereas the contemporaneous marginal effect is higher for generalists. The paper also provides a key methodological insight to the marketing and economics literature. In the Nerlove-Arrow framework, moment conditions that are typically used in conventional dynamic panel data methods become vulnerable to serial correlation in the error structure. We discuss the associated biases and present a robust set of moment conditions for both lagged dependent and predetermined explanatory variables. Furthermore, we show that conventional tests to detect serial correlation have weak power, resulting in the misuse of moment conditions that leads to incorrect inference. Theoretical illustrations and Monte Carlo simulations are provided for validation.