Harpreet Singh presented  a research paper on "Automation of Dynamic Marketing Attribution Models" at 40th Annual Informs Marketing Science Conference At Temple University.

Abstract

Marketing performance tracking and optimization can be modeled as a non-linear, dynamic problem solved through ensemble Kalman filters. The key challenge in a practical marketing mix setting is to be able to initialize the Kalman filter. The use of BFGS or EM type of algorithms does not provide robust solutions when estimating the model’s variance parameters. Applying MCMC techniques combined with Kalman filters, we have developed a robust reliable method that avoids manual tinkering of model estimation, thereby enabling significant automation. This method can be further extended and made computationally robust through multiple banks of filter ensembles. Our team has successfully deployed these robust innovations in real world applications for multiple CPG & Retail companies globally. We will present our methods and applications that significantly reduce computational times, thereby empowering brand managers towards agile decision-making. 

Keywords: Automation; Marketing Attribution; Kalman Filter; MCMC; Ensemble Banks; Filter Initialization

In Real-Time Attribution, Marketing Attribution, Artificial Intelligence