Research paper on Coors Beer Company
This paper looks at the case study of Coors Brewers Limited and their effort for increased market share through the adoption of neural network generated formula update. How effective is their adoption/ what are its failures? And how should the failures be addressed?
In order to achieve its affirmed goal of increased market share, Coors has to perfect favorable product that goes beyond social stigmas in spite of the venue or event that it is consumed in. This value proposition was further complicated by the fact that Coors was expected to design a product that compliments a ranging potential mood set during which it was to be consumed. Based on the market research conducted by the brewer, analytical points and impacts were identifiable. This move was geared towards increasing market share through increased consumer selection over current market shareholders across a wide range of consumption categories. The research to ensure the beer gained a great market share was well back up with facts, and it was successful. Neural networks also helped in predicting rating of the beer flavor and profitability in areas where neural networks have been successfully applied. The neural networks created a more general structure for connecting monetary information of a firm to the corresponding bond rating. However, neural networks are not readily interpretable-the end user must employ the insight in interpretation.
The current process for analyzing various flavors combination is cost and time expensive. Effects within the current process include data collection, human taste test sampling, time, and costs. All these are associated with the manufacture of the actual test product (Turban et al., 2011). The approach was both time and money inexpensive. Artificial Neural Networks (ANN) can simulate human choices like the human brain thus was able to parallel process multiple inputs and output combinations that have been designed to attain the weighted values to the particular processing question. The more the layers of a given network structure, the higher the profitability. This approach also added to time, cost and complexity to the problem which was factored into the design of the neural network construct.
The lesson learnt from Neural Network analysis was that Coors used a Multilayer Perceptron (MLP) architecture with two hidden layers constructed using the NeuroDimension tool suite (NeuroDimension Inc, 2012). The initial attempt of Coors at using an ANN to solve this issue did not succeed. This malfunction was attributed to constraints based on flavor (Turban et al., 2011). In turn, this produced invalid results because of noise issues and the quality factor resulted in reduced variation of the data sets. By focusing on only one, feature, other elements collected and availed to the ANN are reported to have caused underestimated factors to add complication while not impacting the problem being tested.
The method used by Coors lacked reliability in data sets. The data set favored flavor facts and these only considered internal testing (Turban et al., 2011). Based on these assumptions a similar conclusion will be drawn. If one considers the current trends within the spirits market, nontraditional infusion of flavors is gaining popularity in the market. A data set of complimentary flavors combinations already exist that describe binary ratings for bitterness, texture, taste and sweetness. Another issue was that there was no consideration given to the beverage color as part of the presentation impact. Color is usually achieved through the addition of components into formula and brewing process. Either of these aspects can change the flavor. According to Harrington, effervesce of carbonation has an effect on beverage odor, perception, and flavor (2008). These factors were not well-thought-out in the formulation process.
The choice of a suitable Artificial Neural Network (ANN) is footed on various criteria (Turban et al., 2011). What kind of decision structure/model is appropriate in the situation? What is the level of supervision that should be applied to the pre-learning models design? What is the fidelity and depth of the modeling data used for design and testing? Which constraints are applied to the modeling software used to construct the ANN? What are the computational impacts that limit the network layering and node depth? All these information is analyzed, and recommendations for the ANN model can then be derived. In Coors’ scenario, a great deal of the data is indefinite, making a perfect response unlikely (Turban et al., 2011). As a result, the best justification and guess recommendation, with no constraints is to be provided. ANNs are difficult to determine since they address computationally difficult issues. However, based on my research of sensory evaluation models that are likely to solve a given problem, I found one that works well. This model is known as the Multilayer Perceptron (MLP) currently selected by Coors. However, I would also recommend a sub-model called the Multiple Input Multiple Output (MIMO). This sub-model is a specific alternate of the Back-propagation design.
Harrington, R. J. (2008). Food and Wire Pairing: A Sensory Experience. Hoboken, NJ: WSiley and Sons Inc.
NeuroDimension Inc. (2012). Neural Network consulting. Retrieved August 10, 2013, from nd.com: http://www.nd.com/resources/partners2.html
Turban, E., Sharda, R., & Delen, D. (2011). Decision support and business intelligence systems (9th ed.). Boston: Prentice Hall.