Earl Robles

Earl Robles is writing his Master Thesis on the application of Machine Learning Algorithms to Conjoint Analysis, in the context of feature prioritization of a B2B Product with a large attribute set. He is an MBA Candidate at the University of Augsburg (Germany) and has recently completed his Executive Fellowship Program at Katz Business School, University of Pittsburgh. His Master Thesis is being supervised in Katz as an extension of the Program.  The abstract of his topic can be read below. 

Abstract

The application of conjoint analysis in consumer markets is well known. While the consumer decision process maps psychologically to a small set of features or to a multi-level hierarchy of small sets, B2B decision process is influenced by a large list of criteria, sometimes with weakly vertical hierarchical topologies. The manual selection therefore of representative attribute sets and profile generation for conjoint analysis survey design is a non-trivial activity.

In addition, large B2B customers most often have elaborate logical and systematic selection processes in place to justify their decisions.  Thus, the hidden non-articulated psychological criteria is less common in a B2B product and therefore can be good candidate applications for complementary machine learning algorithms.

A product platform, an enabler of other products for the B2B customer (e.g. an embedded operating system), is characterized by a large set of features, majority of which are invisible in a one page data sheet. In addition, each non-highlighted feature requires NRE (Non-Recurring Engineering) cost, which would need to be prioritized by R&D. The absence of customer value score for each feature makes prioritization a subjective activity.

The main topic of the master thesis is the exploration of usage of machine learning algorithms such as SVM to supplement conjoint analysis. Concretely, they are used to aid in feature selection in survey design and generation of prioritized topological graphs of features (i.e. tree, list, etc.) based on customer value weights derived from survey data.