abstract: Understanding customer preference is crucial for new product planning and marketing decisions. This thesis explores how historical data can be leveraged to understand and predict customer preference. This thesis presents a decision support framework that provides a holistic view on customer preference by following a two-phase procedure. Phase-1 uses cluster analysis to create product profiles based on which customer profiles are derived. Phase-2 then delves deep into each of the customer profiles and investigates causality behind their preference using Bayesian networks. This thesis illustrates the working of the framework using the case of Intel Corporation, world’s largest semiconductor manufacturing company. / Dissertation/Thesis / Masters Thesis Industrial Engineering 2017

To learn the uncertainty of customer preference on the attribute of new product, usually a firm needs to do market research. Developing a product on an attribute which is less preferred by customer may lead to a failure. In addition, a firm used to take efforts to design the product. In recent years, we observed a new business model in which the firm does not take effort to design new product, nor does she do market research by herself. She provides rewards to attract outside designers to design new product. Some designers may take effort and design products based on their private information of customer preference. The firm receives designs with different quality and attribute, she chooses one to produce. By solving this game model, we get the equilibrium quality of the design offered by each designer based on their private cost parameter. And we obtain the following insights: When the market size is too small, the firm gives nothing to designers; when the market size is sufficiently big, the firm only gives reward to the designer whose design is produced; otherwise the firm gives both rewards to participated designers and the designer whose design is produced. We find that when the market size is big enough or the disutility is high enough, the new business model dominates the benchmark business model. When both the disutility and market size are small enough, the firm prefers the benchmark business model. And the relative attractiveness of new business model versus benchmark model keeps the same when the market size is small enough. The impact of extra reward on relative attractiveness of new business model versus benchmark model increases with extra reward. When the extra reward is high enough, the firm always prefers the new business model.

Background. Markets nowadays are increasingly competitive as many companies are focusing themselves on minimising their churn rate and maximising their growth rate of customers. Therefore, it is useful to understand the product preferences from the customers’ perspective which could help the companies to learn their products’ most attractive components. A similar approach can be used in terms of sustainability information, where sustainability in terms of products refers to the making of the products out of materials that are easily available and processed [42]. It might be the case that customers with limited knowledge are unaware of the product’s additional sustainability components and that this information would be interesting to them. Hence, the companies can expand their information or present it in a more easily perceiving form to increase customer’s awareness about the product’s sustainability attributes. Objectives. Firstly perform a literature review to select the appropriate data mining (DM) technique required to build a preference model for existing customers. Next, experiment to compare the machine learning (ML) algorithms and choose the appropriate algorithm best suited for the current data to build a predictive preference model. Later, analyse whether the preference models built in this research project show consistent patterns in customer data. Methods. To have a better understanding of the DM and ML techniques for modelling and predicting customer preferences, a literature review was conducted. With the use of an unsupervised learning approach, the different customers’ common behaviours and deviating behaviours were modelled. Additionally, analysing the models helped in understanding how customers behave concerning different products. Results. The Literature Review indicated k-medoid as a suitable algorithm for clustering customers and the experimental results indicated that the K-Nearest Neighbours (KNN) is a better performing classification algorithm based on accuracy. The outcome from modelling the customer preferences towards sustainability are analysed. The results indicated that the data mining and machine learning techniques are capable of modelling the sustainability preferences of customers. Additionally, the comparison between the cluster solutions provided useful information about customer behaviour patterns. Conclusions. The main goal of this study was achieved by proposing an approach that can model the common and deviating preference behaviours of the customers in terms of sustainability. This type of analysis is helpful for the companies in understanding their customers’ preferences towards sustainability which in turn helps them in building more sustainable products.