Jean-Paul Van Belle

Department of Information Systems

University of Cape Town 

Cape Town, South Africa

Jean-paul.VanBelle@uct.ac.za

Meta-modelling; enterprise models; evaluation of models; ontology.

When examining the conceptual or logical models that form the basis of organisational information systems, one is often struck by their wide diversity. This is usually explained by differences in information needs and the diversity of business processes in different organisations. But even in a "laboratory situation", such as a graduate systems development project where a single explicit set of business requirements is provided to a large number of "development teams", one is likely to end up with as many different logical models as there are development teams.

Because most "in-house" system development methodologies develop system specifications "bottom-up" from user requirements, most systems will display a functional bias and definitely turn out to be department or organisation-specific. A user-driven development approach implies that systems need to be redesigned whenever user requirements or business processes change. In addition, these applications often fail to take into account the "wider picture" and present problems when they need to interface with other intra- or inter-organisational systems.

The opposite approach is the "top-down" approach whereby a generic model, assumed to be applicable to all or most organisations, is adapted or refined to fit the needs of a specific organisation. This model-driven approach manifests itself in packaged IT solutions such as off-the-shelf accounting packages or ERP systems, which are customised to a greater or lesser extent to cater for the specific needs of the organisation. It is also the fundamental principle underlying the template or pattern-driven methodologies used by most large IS consulting companies.

The main thesis of this paper is that a "model-driven" top-down approach is preferable in principle to the "user-driven" bottom-up approach. By extension, it is argued that it is desirable to standardise the semantic content of a generic high-level enterprise reference model, suitable as the basis for real-life information systems. Naturally, it is expected that a number of competing proposals will be subjected to a free market-like "survival of the fittest" process, or perhaps a more oligopolistic initiative. The envisaged standard could be formulated by private sector forces or, alternatively, international standards bodies can formulate a standard (e.g. IFIP, OMG). It is likely that it will be a co-operative effort. Although a lot of work has already been done (some of which will be surveyed below), there are concurrent efforts on many different levels.

In what follows, existing generic enterprise models will be identified. It will become apparent that there is indeed a clear de-facto trend in the market towards this generic enterprise model. The next section will look at an illustrative example where the current models still fall short. Finally, a motivation will be given that future methodologies and tools should endeavour to incorporate more "knowledge" in enterprise models by means of knowledge-based approaches.

Although system engineers, IS practitioners and academics seem to enjoy designing new modelling methods and techniques (already more than 1000 in 1993, according to Jayaratna), IS as a discipline has paid relatively little attention to the theory or philosophy of modelling. This is probably due to the fact that IS is really an applied discipline. Practitioners build models of real organisations in order to design real information systems; the only criterium of the value of the model is the functionality of the resulting system. This approach is the instrumentalist approach (Ruttkamp 1999), similar and equally appropriate as what architects or engineers do when they design their artefacts.

However, if generic enterprise models are to be developed, more attention will need to be given to what a model is, the limitations of modelling and the need for a scientific theory underlying the model. A number of people are already working on meta-modelling approaches but perhaps more philosophical and/or logic-mathematical contributions will be needed (see e.g. Barwise 1974).

This paper stems from the authors current research which is concerned with developing a framework for comparing modelling approaches (e.g. object-orientation as opposed to classic IE) as well as investigating criteria to compare models within a given modelling paradigm (e.g. the SAP R/3 reference model with Baans DEM Enterprise Reference Models).

There are many different generic enterprise models that have been or could be used as the basis for information systems. The following headings are an attempt to classify them according to broad approaches paradigms being perhaps too strong a word.

Paciolis introduction of the double entry accounting system in 1494 provided an almost universally adopted model for the business information systems of the last 500 years. Although many shortcomings of the model have been identified, and many improvement/enhancements suggested, the model still survives to this day as the core of most contemporary enterprise systems.

Although the model is five centuries old, it already incorporates many critical IS components: a taxonomy of entities (with semantic interpretations), business logic rules, a number of attributes or data dimensions, distinction between the TPS and MIS components, control procedures. In spite of its shortcomings (historical view, essentially two-dimensional, ignores many relevant entities and events etc.) it is a cornerstone of virtually all transaction processing systems.

The implementation of the accounting model is hardly uniform. Each country has its own Accounting Board and legal requirements. An interesting case study of the accounting model is the Belgian implementation of the EU (then still European Economic Community, EEC)s 4th directive, aimed at the harmonisation of accounting procedures and financial statements. The "Belgian Royal Decree on Accounting Standard and Financial Reporting" (1983) prescribes the accounting rules and a chart of accounts to a very detailed level many hundreds of accounts are prescribed and need to be reported on by all large businesses. (Sharp et al 1999)

Financial managers of most large organisations who have to handle large financial budgets use financial models. The larger the organisations (and budget), the more formal and elaborate the models. Although the financial managers disciplinary paradigm was the accounting model, its focus is the recording of historical business events. The accounting model does not support forward-looking decision-making processes such as planning, forecasting, long-term budgeting, what-if analysis, probabilities and scenarios. Hence the development of financial business models to facilitate the latter type of analysis. Unlike the accounting model, there are no statutory or de-facto standard financial models. Most organisations have developed or customised their own models though, there tends to be a significant overlap between the models.

Financial models range from simplistic user-developed once-off financial spreadsheets to extremely sophisticated mainframe software produced by the likes of IBM. Many of these financial models are really an extension of the accounting model into the realm of decision support systems (Van Belle 1988).

The existence of many separate legacy systems that needed some form of integration prompted organisations to look at standardising their information models. Although enterprise architectures are still a research topic in the systems engineering and information systems community, its basic relevance, issues and methodology seem to have been settled (Periasamy, 1993). A considerable number of research articles has already been published in various journals. However, the enterprise architecture has usually been approached from a systems methodology approach.

The requirement to have a integrate data across the organisation to facilitate management decision making and enable effective management of an organisations data resource resulted in the need for a data warehousing approach, which in turn relies on standardisation of the semantics of enterprise data models. An early local account of the South African status was given by Stevenson (1991); a more international, popular accounting in Inmon (1997). Bernus (1996b) edited a more scientific collection of papers on this topic.

The adoption of object-orientation and the development of consistent object interfaces (OMG/CORBA, DCOM) has prompted a flurry of activity. The technology and standards have matured sufficiently for business objects to become a real world technology (Tailor 1995; Prins 1996; Gale 1996; Eeles 1998). Vendors are scrambling to develop libraries of business objects with universal appeal and considerable work is also being done within OMG to stimulate this process (see OMGs RFPs). Most vendors, however, aim to develop business objects with reference to a specific industry.

Some OO CASE tools now come pre-loaded with a library of business objects while in other cases, more fully developed libraries can be purchased separately as an add-on. A good example is the BOMA™ (Business Object Management Architecture) library developed by SES Software for Rational Rose developers.

As intimated above, IT solution providers and consultants employ (often proprietary) methodologies. Many times, these methodologies are supported by tools that contain a rich encyclopaedia/dictionary of high-level data structures of almost universal applicability (Bytheway 1995). These are then "specialised" or customised to fit the needs of individual clients or customers. The methodologies are typically supported by a specific structural view of the organisation and IT, as embodied in an underlying framework.

Not surprisingly, since IS researchers excel at developing new methodologies, there are an almost equally large number of frameworks (a quick survey revealed more than thirty in 1992!). Perhaps the most widely known such framework is the Zachman framework (Zachman 1987; Everden 1996). A nice aspect of the Zachman framework is that it contains a number of empty cells, i.e. areas in which current IS technology provides no assistance (yet). Cook (1996) demonstrates how to use Zachmans framework to develop a high-level enterprise model.

Enterprise-wide information systems or ERPs such as SAP, BAAN and PeopleSoft (there are a number of lesser-known ERPs) has shown that a major portion of information systems needs can be met through non-specific systems. The user requirements are not used for the conceptual design but rather the final customisation of (the user view/implementation) of the generic IS. Although some claim to be based on deep conceptual models of the enterprise (Sheer 1994), they are more often distillations of industry implementations from which higher level business object models were developed through a somewhat iterative process (Curran 1997).

The most publicised model is the SAP R/3 Reference Model (implemented using ARIS; see Currant 1997), though the other ERPs also have one or several standard reference models, e.g. BAANs Enterprise Reference Models (in DEM).

Practitioners have realised that there is substantial conceptual overlap between the models they develop for different enterprises. "[It] is my experience in actually creating many models [that] problems repeat themselves." (Fowler 1997). This experience has been documented in publications that contain generic data models (Silverton 1997), or best practices in systems engineering (Reingruber 1994). With the move towards object-orientation, some of the slightly higher-level principles have also been identified and a flourishing research area in object patterns has emerged (Gamma et al 1995; Hay 1996; Fowler 1997). A standard for the petro-chemical industry was in fact already been agreed upon quite a while ago (POSC 1994).

Ontologies are an emerging and exciting sub-discipline within enterprise modelling. Initially, the focus was on capturing the semantic contents of a high-level enterprise model as formally as possible by means of formal mathematical languages. This level of formalisation allows for deductive reasoning within artificial intelligence contexts (e.g. the Enterprise Integration Laboratory at the University of Toronto; Fox 1993). Other researchers, who do not require automated deduction, developed less formal ontologies (e.g. the Enterprise Ontology Project at AIAI in Edinburgh; Ushold 1995), typically using higher-level tools such as those provided by the Knowledge Sharing Effort at Stanford. There are also some smaller ontologies, such as those suggested by Wand (1992?) and Bytheway (1995).

At the other extreme is the 10-year mammoth project to capture most human "common sense" knowledge, known as the CYC project. Understandably, a substantial subset of the current knowledge base is concerned with organisations. However, most of the knowledge publicly released in this part of the library seems to be captured at a fairly superficial level.

Consider the following business processes:

• booking a lecture venue for a teacher.
• scheduling a machine for a production run of gizmo XYZ
• borrowing a book/video from a library/ video rental shop
• reserving a hotel room for a prospective holiday guest
• making an appointment with a dentist

Conceptually all are instances of (almost) the same process: the (time) scheduling of the productive capacity of an organisational resource. In practice, the implementation of a manual or computerised system to perform the above processes is likely to differ dramatically from both a logical (architectural) as well as a practical perspective. The underlying data and process models of each application will probably show little resemblance despite the fact that a " naïve" conceptual view will readily identify many common characteristics.

An sufficiently high-level conceptual enterprise model would provide (or suggest) common business objects for, inter alia, time points and intervals, organisational (sub-)units, events, monetary values, enterprise resources, capacity, customers/clients (internal & external), services/products, the allocating/scheduling of resource capacity to events, costing of the use of a resource capacity etc. Applications would inherit most attributes and procedures from these objects so that it would be unnecessary to code e.g. the fact that a video (venue, seat, hotel room) can only be lent (booked, hired) out once for each period. Similarly, code (or methods) relating to "high priority" allocations, slack or dead time in between reservations/schedules and retooling times, recurring events, overtime, productivity, scheduling sequential to or dependent on other schedules, rescheduling, maintenance, block/group booking, costing, charging to user departments, etc. should already be inherited from the object classes.

Currently, object libraries exist for interfaces (e.g. forms, menus, graphics, communications, input/output, etc.) and algorithms (e.g. mathematical & financial functions, dates, security, encoding). Application-specific "business" objects for many industries are also being developed (refer OMG), but again typically from a "bottom up" perspective. There should be both a place for and considerable interest in a library of high-level "generic business objects" (conforming to e.g. OMG/CORBA standards).

Similar examples can be given for the even simpler area of data structures. No information systems known to the author would alert the user that a shareholder, customer and employee happen to be one and the same person, or that a cheque for payment of purchases is being sent to an employees address. One could argue that, in principle, this is a simple matter to resolve in object-oriented systems. This can be countered by the following example. Customers are often divided into corporate and "individual" customers. Ignore the fact that different systems in large organisations often use partly different definitions of customer Stevenson found more than twenty different interpretations of "customer" in a enterprise data architecture exercise within a large insurance company. Also ignoring the fact that some organisations may want to also want to have "households" as customers, how would one model the customer entity (or object) conceptually?

• The CUSTOMER as a SUPERTYPE of ORGANISATION and PERSON?
• The CUSTOMER as a SUBTYPE of ORGANISATION and PERSON?
• The CUSTOMER as a RELATIONSHIP between ORGANISATION and (ORGANISATION or PERSON)?

Each of the options is conceptually different and has both drawbacks and advantages (see Reingruber 1993 for a discussion).

From the discussion of the trends it may appear that there is no need for a conscious research pursuit towards generic, knowledge-based enterprise models: if that is what the market demands, market forces dictate that supply will eventually follow. However, the following comments may be relevant.

From the illustrative examples above, it is clear that the models developed to date are still very much "bottom-up" generalisations of "what works". An a priori conceptual approach is still the exception. The problem may partly be due to the tools we use: if all you have is a hammer, everything is apt to look like a nail.

What is needed is a good look at the differences between the various approaches mentioned above, and how they can be integrated. This will entail research into meta-modelling to identify assumptions and constraints of the various methodologies. It is likely that additional research in more fundamental modelling theory will be required i.e. the philosophy of information systems and modelling theory. A lot of work has already been done in the area of IS-specific modelling approaches, see for instance the research headed by Lyttinen or Bernus/Nemes.

The author is currently working to develop a representative sample from the universe of enterprise models, including those outside the traditional IS scope, more or less along the lines of those suggested earlier. Two critical research issues at the moment are the following:

1. An investigation of the different model dimensions or aspects. This boils down to developing a comprehensive listing of model attributes that are common to all enterprise models. So far, the following have been identified:

• Model description including name, author(s), summary description, source, historical annotation
• Model scope including specific organisational view, intended model users, purpose, scientific context (which discipline). These factors focus on the pragmatic model analysis.
• Main model entities or concepts: the root concepts used in the "top two or three layers" of the models. (All models investigated so far allow for hierarchical analysis.) This looks (qualitatively) at the high-level semantic contents of each model.
• Meta-modelling constructs: a listing of the type of constructs employed i.e. focussing on the syntax of the model.
• Model views: although these are really (non-exclusive) subsets of model constructs, but link to the model scope and notation i.e. combine syntactical, semiotic and pragmatic aspects.
• Model notation: a syntax/semiotics perspective.
• Model complexity and detail. A quantitative analysis of the full semantic content of the model, both in absolute number of concepts and in the relative complexity of the relationships between them. A number of statistical tools can be used for this analysis, many suggested by statistical or computational linguistics e.g. cluster analysis; factor analysis; network analysis; categorization analysis.
• IS artefacts built using the model. An evaluation of the actual information system(s), if any, that have been built on the basis of the model.

1. The uniform representation of both the meta-modelling constructs and the key model content elements. This requires a knowledge representation language. However, any language powerful enough to represent all model constructs will be powerful enough to represent model attributes, too i.e. it can be used to code or represent the above model "catalogue". But, for communication purposes, more intuitive and less formal equivalent notations will be required (English language with hypertext i.e. html; as well as graphical).

Whereas meta-modelling is concerned with more descriptive i.e. objective analysis, decision-makers and practitioners have a need for comparative i.e. evaluative analysis. Which model is "better" (for a given purpose and context)? To the authors knowledge, no explicit frameworks exist to compare models although a large number of candidates can be identified from related research e.g. linguistic analysis, IT methodology frameworks, ontology research and the like (which look at similar intellectual artefacts). Note that it is not just necessary to develop criteria to compare models within a given modelling approach with each other, there is a need to compare models across different methodologies too. The challenge is akin to selecting a Nobel prize e.g. in literature: how to compare an obscure Greek political play with a popular Chinese historical novel?

The authors ultimate aim is to present one single contingency-based evaluation framework. Its first iteration, developed by Price et al (1998) is a heuristic combination of elements from the following.

• The Seligman framework for analysing IS Development Methodologies. This includes the following five elements: way of thinking, way of controlling, way of modelling, way of working and way of supporting [Hommes, 1998].
• Parts of the NIMSAD framework for evaluating methodologies [Jayaratna, 1994].
• Williams criteria of necessary requirements for an Enterprise Reference Architecture (in Bernus, 1996).
• Fox (1996) list of characteristics which should be used to evaluate a generic model, including functional completeness, perspicuity (ease of understanding/self-documentation), generality, precision/granularity, efficiency and minimality.
• Hommes (1998) framework to evaluate the quality of a modelling technique, including consistency, correctness, usability and comprehensibility.

In the proposed model, the two major dimensions being measured are the expressiveness (or "way of modelling"), the degree to which a model corresponds to the organisation, and the arbitrariness (or "way of working"), the degree of freedom one has when modelling the organisation (or building a system from the model).

The two dimensions are then broken down into the following seven factors: efficiency, generality, consistency, correctness, usability, comprehensibility and completeness. The first four factors relate to the expressiveness and the last four to the arbitrariness (correctness thus relates to both dimensions).

Although the model has been applied to a case study (i.e. comparing the high-level SAP R/3 and BAAN enterprise models), much work still remains to be done.

It is not hard to argue the case for standards. Cook (1996) mentions the reply attributed to a small railroad owner proposing to link its line to those of a much bigger competitor who pointed out the differences in size: "It is true that my railroad may not be as long as yours, but it, sir, is just as wide". The following are but a few of the classic arguments in favour of standardised models.

• Increased system developers productivity because there is no need to re-invent the wheel many times over - similar to the arguments in favour of software (component) re-use and off-the-shelf systems.
• More scope for inter-operability of different IS (in the same organisation) and inter-organisational systems communications.
• More flexible information systems: given the universal character of the underlying deep structure of an organisations systems, it should be more resistance to superficial changes in organisation structure or business processes.

To a limited extent, some subsets of enterprise model have already been standardised: EDIFACT, the Universal Data Element Framework, ISOs Basic Semantic Repository, the Standard for the Exchange of Product Model Data, Petrotechnical Open Software Corporation and refer also to the current work in OMG.

It should be noted that it is unlikely that all, or even most, of specific organisations models can be generalised or standardised. Silverston et al (1997) claims that "in general, one third of a data model (corporate or logical) consists of common constructs that are applicable to most organisations and the other two thirds of the model are either industry- or enterprise-specific." Fowler (1997) equally "suspect(s) a small number of highly generic processes." However, he proceeds with the following comment "Many diverse businesses use a set of very similar abstract process models. This raises some significant questions about the promised development of vertical class libraries for industry sectors. I believe that true business frameworks will not be organized along traditional business lines but instead along abstract conceptual processes."

Although the author does not intend proposing a standard a work of much greater scope and complexity than can be handled by any one individual, or even institution the identification of a "common denominator" set of constructs will be instructive both from theoretical and practical perspective. An unresolved aspect is that of the "equivalence" matter: is it possible to have a number of fundamentally different (i.e. non-compatible) but equally valid "core models" on which one can built real-life information systems (akin to the different "types" of mathematical geometries: Euclidean, Riemann etc.). Early investigations seem to indicate much less divergence in the enterprise models. This relates to the "arbitrariness" dimension identified higher.

A lot of the knowledge embedded in a generic enterprise model will not fit neatly in any of the current system development paradigms: semantic interpretations, default frames, deductive dependencies, optional or conflicting rules, exceptions, contexts, scenarios, existential quantifiers, etc. are hard to represent in most of todays modelling tools (although OO approaches have solutions for some of these).

Perhaps the only models that are "rich" enough to capture these elements are the ontologies referred to higher, especially the work done on the TOVE project. It is the authors opinion that future methodologies will have to be able to represent knowledge about the enterprise more explicitly. An advantage of knowledge-based methodologies is the increased intelligence of the associated tools perhaps with the end destination of self-executing models?

• It allows for advanced knowledge representation constructs.
• It can interpret and execute the knowledge in its own database.
• It has sufficient support, a historical track record and wide variety of applications.
• A GUI-interface for ontology-building has been developed ("Ontosaurus").
• It is available for a variety of platforms and architectures (incl. multi-user distributed).

Examples of generic conceptual enterprise models range from the 500 year-old accounting model to the most recent "fashions" in IS. However, most of todays information systems do not exhibit a sufficiently deep underlying universal structure. The case can therefore be made that some more research should be devoted to comparing both models and modelling approaches. This could be used to motivated and develop/propose one (or more) standard enterprise model(s). These models would probably incorporate much knowledge that can currently not be accommodated by conventional modelling techniques. Hence, one would require the adoption of knowledge-based approaches and tools, probably inspired in part by those employed in enterprise ontology research.

Some of the material in this document has been adapted from:

VAN BELLE J.P. (1998) "Semantic Standardisation of High-level Enterprise Modelling. Arguments for an Ontological Approach." 28th Conference of the SA Computing Lecturers Association, Stellenbosch: 28-30 June.

Jean-Paul Van Belle © 1999. The author assigns to ACIS and educational and non-profit institutions a non-exclusive licence to use this document for personal use and in courses of instruction provided that the article is used in full and this copyright statement is reproduced. The authors also grant a non-exclusive licence to ACIS to publish this document in full in the Conference Papers and Proceedings. Those documents may be published on the World Wide Web, CD-ROM, in printed form, and on mirror sites on the World Wide Web. Any other usage is prohibited without the express permission of the authors.