Customer preference modelling using projection to latent structures in an automobile industry

Abstract

The nature of engineering design today is based on decision-making. With the advancement in communication and information technology, markets are not only global and highly competitive; they are also notably customer-driven. It is imperative for companies to capture this customer-driven market information and translate them into key decisions which are often carried out in the early stages of modern product development cycle. This increased the importance of decisions in engineering design which led to studies revolving decision analysis and multi-attribute decision making methods (MADM) to support decision making. However, MADM are often based on rigorous mathematics, time-consuming and ambiguous in nature. The complexity in their application processes made them deficient in a fast moving market, hence not able to effectively capture changing customer trends and needs. The report aims to bridge this gulf by employing Projection to Latent Structures (PLS) and Customer-revealed value (CRV).PLS is a regression technique that enables the construction of predictive models using various highly collinear factors. CRV is able to capture customer preference by relating demand and prices of the product in a market segment. In this research, a study on the ability of PLS in modeling customer preference in the automobile market will be proposed. The study aims to select the best sedan car model out of 3 models in the U.S market. Predictive PLS models are calibrated based on existing car attributes (input) and CRV (output).New car attributes will then be used as input to these predictive PLS models to model CRV (output).Estimated standard errors calculated based on the difference in the predicted CRV and their real-life values will be used to judge the errors of predictions. In this study, the author proved the usefulness of PLS which may be used as a concept selection method with CRV as the selection parameter. Effective PLS modeling relies on accurate inputs which are within the resource capabilities of most companies. By relating these inputs to a meaningful value such as CRV, companies are able to meet the demands of a fast-moving market today by making quick selection-based decisions which shorten the product development cycle and capture customer needs. The drawbacks of PLS are inherent insensitivity to attributes changes and its inability to pinpoint a critical product attribute. In addition, PLS models are most relevant only in the year which its calibration set is based on. With the considerations of these drawbacks, future work can suggest PLS to be used with techniques such as forecasting to make PLS modeling dynamic in nature. This creates a forecasted PLS model which is able to select present day conceptual designs which could be launched within that forecasted timeframe. It can also be use with MADM on a complementary basis to overcome its insensitivity to attributes changes and pinpointing.