Using Reference Models for business Process Improvement: A Fuzzy Paradigm Approach.
Oliver Thomas, Otmar Adam, Peter Loos · 2006
In business practice the enterprise-specific adaptation of reference models for process improvement is characterized by the fact that decision-making premises do not exist in the form of mathematic models or numeric values. Decisions are characterized by consideration and creativity and are usually derived from fuzzy conditions. Although these conditions are not precise, additional and important information for the understanding of concrete business situations is connected with them. Thus, verbal information, as well as vaguely formulated statements, premises, objectives and restrictions are very important for reference model adaptation. The systematic consideration of fuzzy data in reference model adaptation can only succeed, when the models to be adapted themselves allow the consideration of fuzzy data. The fuzzy set theory-based extension of information modeling therefore provides the foundation for the development of a methodology, as well as the prototypical realization of a tool for reference model adaptation with regard to fuzzy data in this article. 1. Vagueness in Reference Model Adaptation Information models are defined as purpose-relevant representations of an information system designed by way of a construction process [vom Brocke 2003, p.16; Thomas 2005a, p.25]. They are simply referred to as models. A reference model—to be precise: reference information model—is an information model used for the construction of other models. This article is therefore based upon a use-oriented reference model term which focuses on the use of reference models for the construction of enterprise-specific models [vom Brocke 2003, p. 34; Thomas 2005b, p. 24]. The reference model terms often found in information systems literature, based upon attributes which characterize these reference models—in particular the attributes “universality” and 48 BUSINESS INFORMATION SYSTEMS – BIS 2006 “recommendation character” [vom Brocke 2003, p.31ff.]—will not be followed here. Every model resp.partial model which can be used to support the construction of another model can be seen in this sense as a reference model. The reutilization of reference models connected with this can be seen as a fundamental idea resulting from paperless, tool-supported data-processing consulting at the start of the 1990ies. The user‘s primary task in reference model-based construction, which can be supported by IT-tools, is the adaptation of reference models. The derivation of specific models from a reference model characterized by this term corresponds with the creation of variants of reference models [Schutte 1998, pp.207–209]. Thus, for example, the enterprise-specific models information model productoriented manufacturing enterprise 1 E or information model process-oriented manufacturing enterprise 2 E could be derived as variants of the reference model manufacturing. The adaptation of a reference model consists of two phases in accordance with the understanding of the term in this article [Schutte 1998, p.316]. In a first step, the reference model is approximated to the requirements of the enterprise being considered. In a second step, modifications are made to the model. While the first phase focuses on a semi-automatic adjustment process, the second phase represents a process which must be carried out manually by the model-user. The result of the adaptation process is an enterprise-specific to-bemodel, which can be put into practice, i.e. implemented in the company. Although a wide variety of guidelines, procedure models, modeling languages and tools for the development of to-be-models for business processes exist, reusable “know-how”, which considers process flow and data processing support in an integrated sense, is not sufficiently available. In many areas the need for the repeated use of work-pieces (for example: modular systems in industrial product design) exists, not least due to economic reasons, whereas process design usually represents industrial single-part production. Research activities which attempt to remedy this problem exist among other things for model supported business process construction with process particles [Remme 1995], for context-specific individualization of process models [Rupprecht et al. 2001], for the design and distribution of construction processes in reference modeling [vom Brocke 2003] or for the management of reference process models [Thomas, Adam, Seel 2004]. Many of these approaches focus on the user-friendly and intuitive usability of methods by approximating these with human ways of thinking. However, necessary decisions require the exact quantification and formalization of decision rules. Often though, only uncertain, imprecise and vague information is available concerning the frequently technically indeterminable procedures for business processes [Rehfeldt 1998; Forte 2002; Husselmann 2003]. By the same token, the target system which is the basis for the adaptation of reference models is generally characterized by imprecise formulations and implicit interdependences. This is for example, illustrated by the statement “the processing time for commissions with a priority of ‘very high‘ should be reduced by proportionately reducing processing intensity, while retaining a ‘high‘ processing quality”. In USING REFERENCE MODELS FOR BUSINESS PROCESS IMPROVEMENT 49 this example, neither the concrete specification of both goals concerning processing time and quality, nor the measures derived can be quantified without the loss of information and in doing so, made directly processable. Today, information models, especially reference models, as well as methods for their enterprise-specific adaptation still do not consider these forms of fuzziness adequately. In the following, the perception of the term “fuzziness”, as well as the consideration of fuzzy data with the help of the fuzzy set theory, which has established itself in research and practice as an adequate approach, will be accounted for (section 2). Then, the existing approaches for the integration of fuzziness in information modeling will be set forth (section 3). On the basis of this, a recommendation for reference model adaptation under consideration of fuzziness, based on the modeling language Event-driven Process Chain (EPC) will be outlined (section 4). The article ends with a conclusion (section 5). 2. From Crisp to Fuzzy Sets In this article, fuzziness is understood as the uncertainty in regard to data and its interdependences. The trigger for fuzziness can be reality itself, language as a builder of models for reality or the use of language. The fuzzy set theory attempts to overcome the separation of a technologically necessary precision on the one side, as well as the empirically desirable consideration of qualitative information on the other and to tolerate a certain lack of precision, as well as vagueness and uncertainty in modeling processes. The fuzzy set theory was developed in the middle of the 1960ies [Zadeh 1965]. The crucial point in the fuzzy theory is not only to evaluate conditions (of objects) with “true” or “false”, but also rather to allow intermediate stages. Subsequent to ZADEH‘s original idea, the classic theory of crisp sets is extended by the description and combination of fuzzy sets: The degree of membership for each element of a predetermined (crisp) basic set to a subset A is expressed by a value μA( ) of a mapping : [0;1] A . One selects these degrees of membership from the interval [0;1] and gives the following interpretation: the higher the degree of membership of an element with regard to a (fuzzy) set, the more it belongs to this set. μA is called the membership function of the fuzzy set {( ; ( )) | } A . With fuzzy sets, linguistic variables [Zadeh 1973] can be formulated, which adopt expressions in natural language—so-called linguistic terms—as values. Figure 1 shows the linguistic variable “Order Value”. It features the terms “low”, “middle” and “high”. The memberships of an object value to these fuzzy sets are expressed by the membership functions μlow, μmiddle and μhigh. The object value 70,000 € belongs for example, to 0.5 to the fuzzy set “middle” as well as to the fuzzy set “high”. This representation of crisp values on fuzzy sets is called fuzzification. In a crisp context it would only be possible for example, to characterize an object value up 50 BUSINESS INFORMATION SYSTEMS – BIS 2006 from 70,000 € as a “high” order value, while 69,999 € would already be considered as “middle”. 0.2 0.4 0.6 0.8 1.0 μ