PhD student Bruno Jacobs introduces groundbreaking methods to analyze purchase behavior in high-dimensional assortments
Many traditional studies in marketing consider household purchase decisions between 3 brands of ketchup, or the decision between a few brands of coffee. However, the used approaches break down for the many thousands of products most (online) retailers have. But then, how do I predict the next product someone is interested in for such large assortments? The driving factor in Bruno Jacobs’ research has been alleviating this limitation and thus enabling marketing analytics in high-dimensional assortments
In his dissertation, entitled ‘Marketing Analytics for High-Dimensional Assortments’, Bruno Jacobs introduces new research methods that enable marketing analytics using data from (very) large product assortments that consist of (tens of) thousands of products. Before, the size of the assortment was a limiting factor in the application of analytical methods. In particular, Jacobs’ research pays attention to the computational scalability of the method, which is important if it is implemented in practice, especially if results are required in real-time, for example for online product recommendations. This scalability is obtained by building on recently developed Machine Learning techniques.
Jacobs: ‘The methods I have developed in my dissertation research provide high-level insights in very large product assortments such as those encountered at large online retailers. The unique contribution of my dissertation is that it enables this analysis for the whole assortment of the retailer. In previous literature, it was often not possible to consider the complete assortment of a retailer, or the method resorted back to very basic analysis techniques, like counting the number of times two products are bought by the same person.’
Instead of directly considering the preferences over all products in the assortment, Bruno Jacobs proposes to reduce the dimensions of the problem by considering a relatively small set of latent factors that can be used to describe purchase behavior for all products jointly. ‘With the methods developed in my dissertation research, we are able to infer purchase patterns that span across the entire assortment. The discovery of these patterns is driven by actual observed purchase data, and not by product attributes. The resulting patterns make intuitive sense, such as a preference for diet related products, eco-friendly products, or products related to having a baby. In addition, I allow seasonality, customer-characteristics, and dynamics, to affect the relevance of these latent factors.’
About Bruno Jacobs
Bruno Jacobs (1988) holds a Master's degree in Econometrics and Management Science from Erasmus School of Economics. In 2012 he joined the Erasmus Research Institute of Management (ERIM) as a PhD student. He carried out his research within the Marketing department and the Econometric Institute at Erasmus School of Economics. In 2015, Bruno Jacobs was a visiting scholar at Columbia University in the City of New York, hosted by Prof. Asim Ansari. He currently works as an Assistant Professor in Marketing at the Robert H. Smith School of Business, University of Maryland, College Park.