Excerpt
Table of Contents
1. Introduction
2. Characteristics of a good model and a good modelling process
3. Evaluation of the FAME model and the modelling process
4. Evaluation of the success of the modelling exercise
References
1. Introduction
This essay is concerned with a Fuzzy Attractiveness of Market Entry (FAME) model, developed for the Bulgarian winery Vinprom Svishtov (VS). VS had to decide whether to expand two of its wines, a cabernet sauvignon and a chardonnay, into either a regional or a national market (Shipley et al., 2013).
The model’s purpose was to assist VS’s management in deciding whether the firm should expand two wines into a regional or a national market. One market had to be chosen for both wines. Because VS had never before operated in either of the markets in question, there was no historical data available. Accounting for this lack of data availability, the model builders used expert judgment to feed the model. Experts included vineyard owners, academicians, experienced wine producers, and managers from VS (Shipley et al., 2013). The key formula used in the model is stated below: denotes the current attractiveness of a particular market for a wine, e.g. the attractiveness of the national market for the chardonnay. represents the current “Best Market Fit for the Firm’s Marketing Mix” (Shipley et al., 2013). This variable indicates how similar VS’s marketing mix (price, product, advertising, distribution) for the wine is to that of the currently dominant competitor in the market in terms of its fit with customer preferences. A value close to one indicates that VS’s marketing mix is approximately as suitable to serve customer preferences as the marketing mix of the currently dominant winery. Since increases in line with , the underlying assumption is that the marketing mix of the currently dominant winery is particularly suitable to serve customer preferences. The remaining variables represent the current market environment (), i.e. economic and social conditions, political climate and infrastructure, and the perceived future strategic importance of the market (). The latter incorporates the expectable profit margin and sales growth resulting from a market entry (Shipley et al., 2013). The values of all independent variables were based on expert judgment generated through questionnaires. Once had been computed for each wine/market combination, VS’s management could compare the attractiveness scores across markets.
Since the relationships between the variables are expressed using a mathematical formula, the FAME model can be classified as being mathematical (Hull, Mapes and Wheeler, 1976). More precisely, it is linear because it only contains exponents equal to one and multivariate since is sensitive to changes in more than one independent variable (McWilliams, 1987). The model generates several decision alternatives, i.e. different wine/market combinations, which can be rank-ordered according to their attractiveness scores. Hence, measures the effectiveness associated with each combination, thereby allowing VS’s management to choose the most attractive market. Therefore, the model is normative as it provides a decision aid for VS’s management (Krajewski and Thompson, 1981). With , the model contains an independent variable, which is affected by time because it measures the perceived future strategic importance of a market. Hence, the model is dynamic (Krajewski and Thompson, 1981).
While the above classifications were largely unambiguous, the classification of the model as either probabilistic or deterministic is less clear. According to Hull, Mapes and Wheeler (1976), “a probabilistic model (…) recognises that the values of some variables are uncertain and deals with this, using concepts from probability theory”. Krajewski and Thompson (1981) define a probabilistic model as a model “in which at least one parameter or exogenous variable is assumed to be a random variable”. The FAME model is purely based on human judgment. Therefore, assuming that human beings are incapable of predicting the future with absolute certainty, the model contains a substantial amount of uncertainty. The model builders addressed this issue by applying elements of fuzzy set theory, a concept from probability theory (Shipley et al., 2013). Because different experts were questioned, the model also features a certain level of randomness, as the value of each independent variable may vary across experts. However, some of its randomness is taken from the model because individual expert judgments are averaged in order to obtain an aggregate value for each independent variable. This makes the model slightly more deterministic (Hull, Mapes and Wheeler, 1976; Krajewski and Thompson, 1981). With respect to the further discussion, it is sufficient to conclude that the model contains probabilistic elements due to the inclusion of human judgments.
The model “revealed” that the national market was more attractive than the regional market for both wines. In accordance with this result, VS expanded both wines into the national market. However, the relative attractiveness of each market was not the only result the modelling process generated. Because each element of VS’s marketing mix was assessed separately, VS’s management was able to detect potential weaknesses in the company’s marketing mix that could be an obstacle to a successful market entry (Shipley et al., 2013).
2. Characteristics of a good model and a good modelling process
First of all, a good model addresses the problem it is supposed to address. As Hull, Mapes and Wheeler (1976) point out: “The construction of a realistic model (…) is of no practical use to the decision maker unless the model can be used to solve the original problem.” While it appears to be obvious that a model should tackle the problem it has been designed for, a mismatch between model and problem can indeed be a major reason for perceived failure in management science interventions (Tilanus, 1985).
A good model also features appropriate levels of detail and complexity. On the one hand, a model cannot include every single characteristic of the system to be represented (Urry, 1991). Indeed, it would even be counterproductive to develop a model that represents reality in every detail. Such a model would likely be over-complicated and probably not easier to analyse than the original system, which a model should actually simplify (Salt, 2008). On the other hand, a model must not be over-simplistic in that it excludes variables and/or aspects that are essential to understand the system to be modelled (Williams, 2008). In accordance with Hull, Mapes and Wheeler (1976), a good model is therefore not over-complicated, but includes all aspects that are necessary to represent the system under consideration appropriately.
Another characteristic of a good model is flexibility. Flexibility, as defined in this essay, means that the model can be easily altered if necessary. This enables the model builder to make changes to the model should the model in its current form turn out to be inadequate given the problem at hand. Flexibility is likely to be high in comparatively simple models because, for instance, the removal of a variable is less likely to have a great impact on the overall model because of complex interrelations with other variables (Ward, 1989). The possibility to alter the model easily may also increase the client’s acceptance of the model because he theoretically has a greater chance of influencing its design.
Furthermore, a good model is easy to understand. That is to say, the variables used in the model should be defined unambiguously (Krajewski and Thompson, 1981). The possibility of the model being misinterpreted is, ceteris paribus, likely to be low if the meaning of the variables is clear.
Lastly, a good model is valid. A model is usually validated by comparing outputs generated by the model with data from the real situation (Krajewski and Thompson, 1981). If model outputs and real data are similar, the model can be considered to be a good representation of reality (Williams, 2008). However, there are situations in which no past data are available, e.g. when a model addresses a problem never before faced by the decision maker. In these situations, the model has to be examined carefully in order to detect logical inconsistencies, and the results of the model have to be checked for abnormities (Krajewski and Thompson, 1981).
In a good modelling process, the model builder gains an understanding of the system to be modelled prior to building the model (Hull, Mapes and Wheeler, 1976). Particularly because a model should address the problem for which it has been developed, it is important that the model builder understands the characteristics of the underlying system, so that he can identify those factors that should be incorporated in the model. In order to gain an understanding of the system to be modelled, the model builder can for example talk to people involved in the system (Hull, Mapes and Wheeler, 1976). However, the model builder may face a situation in which there is neither experience within the client organisation available nor any data to collect, e.g. when the model is to be used in a completely new decision situation. In this case, the model builder should seek guidance from models that were developed for similar problems, e.g. by reviewing relevant literature (Krajewski and Thompson, 1981).
Furthermore, in a good modelling process the client is involved in the design of the model. First, by involving the client the model builder can make use of the client’s knowledge of the system to be modelled. Therefore, he may be able to develop a more adequate model than he would be without involving the client. Second, client involvement is likely to increase the client’s understanding of the model. The importance of the client understanding the model cannot be overemphasised. After all, it is the client who is supposed to make use of the model and the information it generates. By involving the client in the development of the model, the model builder can make sure that the client is able to interpret the information he derives from the model and to evaluate the plausibility of the model output (Ackoff, 1968). Third, involving the client is likely to increase not only the client’s understanding, but also his acceptance of the model. If the client is involved in the setup of the model, he is more likely to accept the model because he has the opportunity to influence it to a certain extent according to his own ideas and requirements. Krajewski and Thompson (1981) summarise the importance of model acceptance vividly by saying: “A model builder may arrive at a model that can be shown to save thousands of dollars per year, yet it is worth nothing if the person who is to use it does not accept it.”
[...]