In this note, we describe the differences between deep learning and traditional methods in more technical terms and outline how gains in accuracy can be monetized. Here, we assume that the reader has a rudimentary understanding of statistics.
Traditionally, consumer response has been modeled by single-equation models relating sales (outcome variable) linearly to the price/promotion, advertising, and distribution measures (explanatory variables). The state of the art today uses choice data (buy, not buy, how much) in a single equation (logistic) model, where outcome variables are choices made by individual consumers. Explanatory variables include a range of marketing measures such as exposure to ads on the internet, search efforts, traditional demographics, loyalty, price, advertising and promotion spending, distribution, and product attributes. Data is used to find the model parameters that best…