Access Type

Open Access Thesis

Date of Award

January 2015

Degree Type

Thesis

Degree Name

M.S.

Department

Industrial and Manufacturing Engineering

First Advisor

Alper Murat

Abstract

Due to the increased competition in the auto industry, proliferation of the vehicle models and increased customer need for choice and customization, it has become more critical than ever to offer a variety of features and customization flexibility while at the same time restraining and, even better, cutting down the costs. Product complexity, in the automotive industry, can be measured by the size of the assortment offered, i.e., set of vehicle configurations a customer can choose from (e.g., for a given model of a brand). While complexity fosters growth with increased alignment of product characteristics and customer needs, it results in decreased revenue (e.g., cannibalization) and profitability (e.g., increased total supply chain costs). Companies that manage complexity by improving their products’ true profitability have seen savings of 10 percent to 15 percent on their cost of goods sold.

In order to determine the optimal complexity that should be offered, the company must first understand its customers buying behavior, and their response at the instances where their primary vehicle configuration choice is not offered or is stocked out. In this thesis, we develop a customer choice modeling framework that predicts the likelihood of an average customer to buy a specific vehicle configuration in a given assortment offering. Our modeling approach utilizes neural networks to predict, based on the historical dealership level sales and inventory data, how likely a given configuration will sell when it’s offered along with a set of configurations. These configuration level sale probability estimates are then used to estimate the attraction factor for each feature included in the vehicle configuration. The attraction factor of each feature represents feature’s individual contribution to the probability of sale of the configuration as a whole. With this feature level estimation, the probability of sales for any feature combination or vehicle configuration can be estimated (including those configurations not yet built or offered). We report on the performances of several modeling and neural network based estimation approaches using historical dataset from a major US automotive OEM. Our models are parametric and thus can be used within an assortment planning model to determine the optimal product assortment that optimizes complexity by considering true profitability of the configurations in the assortment.

Share

COinS