The intuition behind machine learning in marketing: Linear TV attribution

Abstract

Linear or broadcast TV continues to be an important channel for lead generation and user acquisition but precise attribution to an ad is not possible; attribution methodologies include time-based windows, keyword search within those windows, pixels that fire within the same Wi-Fi, panels, etc. This paper describes a forecasting methodology that uses machine learning to analyse recent and historical time series of new user registrations as well as additional factors and variables that affect new user acquisition. We then use this forecast to construct a baseline of expected new user volumes and how we attribute new users above that baseline to TV.