ORGANISATIONAL DESIGN: DYNAMICALLY CREATING AND
SUSTAINING INTEGRATED VIRTUAL ENTERPRISES

(Draft. To appear in Proc IFAC99 World Congress, Beijing)

Peter Bernus^# and Laszlo Nemes^*

^#Griffith University, School of Computing and Information Technology,
Nathan (Brisbane) QLD 4111, Australia <bernus@cit.gu.edu.au>
^*CSIRO Division of Manufacturing Science and Technology
Locked Bag 9, Preston (Melbourne), VIC 3072, Australia <lnm@dmt.csiro.au>

Abstract: We investigate the management system of the enterprise as an agent maintaining a system of objectives. We then analyse the organisation as a set of individual autonomous co-operating agents so that agenthood of the entire enterprise becomes an emergent property of the organisation. Important questions include: under what condition can agenthood emerge? how to create such an organisation?, and how to guarantee that change preserves agenthood?.

Keywords: Enterprise integration, enterprise modelling, organisational factors

 
Introduction

It is increasingly important to devise faster and more reliable ways of designing purposeful, agile organisations (Bernus et al, 1997). We use the definition of the organisation as the human component of the enterprise, forming the human-organisational architecture (Williams, 1994).

An enterprise can be thought of as consisting of an operational and a decisional system (Doumeingts et al, 1998), each implemented partly by humans and partly by machines (Bernus and Nemes, 1994). We borrow the definition of the 'planning agent' from artificial intelligence to describe the desired quality of the organisation.

We demonstrate necessary conditions for the enterprise to behave as an agent, and also show the relationship of this view to concepts such as the fractal factory, holonic manufacturing, and others.

The organisation as an agent

Two crucial questions in organisational design are: 1) how to design the task structure of the enterprise to form a co-ordinated whole? and 2) what tasks allocation to humans (or groups) ensures that the enterprise will act to satisfy its objectives? The second question is typically not asked in business process engineering: it is assumed that the organisation will to what it is told to.

An organisation should conduct a system of activities managed and controlled to satisfy a set of organisational objectives. This requires purposeful behaviour so that the organisation can be characterised as a planning agent.

A planning agent determines a course of action to achieve its set of objectives. This course of action, or plan, is constrained by the agent's resources as well as the agent's own functional capabilities. The organisation follows, or appears to be following this plan while monitoring the effectiveness of the actions to actually satisfy the objectives for which the plan was made. If the plan fails in some way, then the plan or the objectives are modified. Enterprises are striving to achieve this ideal agent behaviour.

The system of objectives may be related to external relationships (satisfying customer demands, producing returns to investors,…), or internal ones (improving processes, co-ordination, resource usage,…). Simplified reference models for any organisation are shown in Fig.1 and in Fig.2. Notice that the model conforms to the simple view of systems utilised by control theory.

The operation of the organisation has an interface to the outside world through which consumed- and

Fig. 1 Reference model of an organisation (a simple GRAI model)

produced material and information services can flow while the management and control system has an information interface to the outside world.

The management system maintains a set of constrained objectives:

1. The management information flow between the organisation and the outside world partially determines what objectives are worth following, or realistic, e.g. what are the present or predicted needs, requirements, or orders to which the organisation should respond now or in the future;

2. Capabilities of the management and control system limit the ability of the organisation to commit to certain objectives, even if operational resources are otherwise present;

3. The material and information interfaces limit what is accessible for the organisation's resources and determine the feasibility of actual operations asopposed to potential ones;

4. Capabilities of the operational resources (usually called competencies) form a natural limit to the organisation's feasible objectives.
   

Autonomy and authority

In Fig. 3 the mechanism (resource) assigned to management and control is the same agent, which performs the operations. This is typical of e.g. biological organisms and reflects the organisation's ability to retain autonomy, operating and controlling itself at the same time. Also the organisation has the authority to commit its resources to objectives.

An autonomous individual agent therefore must co-ordinate its objectives and actions with other agents, and it must co-ordinate its own actions to satisfy its objectives under the negotiated conditions. This requires the agent to plan and control.

Fig. 2 Reference model of an organisation (in IDEF0) - including external interfaces

The internal co-ordination tasks are interrelated with the external ones given the need to rely on other agents. Thus an agent may give up some autonomy to achieve objectives otherwise outside its reach.

Since objectives are structured entities, agent negotiation may be performed on various levels of detail. E.g. a contingent strategy to act together may be agreed on, with the condition to find agreeable tactics, in turn contingent upon agreeable plans, etc.

Thus, based on negotiation protocols, agents can determine future joint action, and act on it.

The organisation as an emerging agent

Artificial organisations such as enterprises do not necessarily display the property of agenthood. For example, it is not automatically true that joint action on behalf of agents is necessarily satisfying some overall objective, or that joint action does not lead to some undesirable state. An important aim of enterprise integration is to build enterprises that do behave as agents. Therefore we analyse the conditions under which agent behaviour does emerge from joint action of agents.

The organisation as an agent consists of a set of agents (Fig 3), thus organisational design is the art of creating agents out of agents. Each agent may

Fig. 3 Multiple agents in the organisation

take multiple roles in the organisation determining the individual agent's functions in that organisation.

Fig.3 is a simple model of control / operation: management provides control to a resource to perform the desired activity. This model is too simplistic if the resource itself is an agent, such as a human. If 'c' must be performed by agent A then a preliminary negotiation between a management agent (F) and agent A must establish A's commitment to do 'c' (see Fig.4). Since F does not directly dispose of A's resources, F passes a requests to 'do c' to A's autonomous control, which in turn disposes over A's resources granting (or rejecting) the request.

The top part of Fig.4 shows the apparent control / feedback (dotted line); the bottom part shows the autonomous control and operation of A, and the interaction between F and A. Since F and A need not always re-negotiate the 'terms and conditions' under which A commits to do 'c': negotiation overhead is potentially low. But, if breakdown occurs, re-negotiation may avoid system level breakdown. Since what was a direct control instructing a resource to perform 'c' now becomes a request, treated as one of the objectives of A, the expectation that 'c' indeed gets satisfied depends on successful F - A negotiation. E.g., did F request some action that fits into A's framework of objectives? Is there a discrepancy between expected and feasible courses of action?

In section 5 we discuss fundamental requirements to achieve harmony between objectives of individual agents and those of the organisation.

Characterisation of Enterprise Organisation - the Harmony of Objectives

In this discussion we concentrate on the decisional (management and control) roles of agents in an enterprise. As explained above control can not be passed unguarded from management agent F to operational agent A but through agent' A's autonomous control. Such controls continuously compete with any other requests to A.

A request 'do c' therefore becomes one of the objectives of agent A, and it is executed because of previous commitment by A to grant such request - optimally after consideration by A of the feasibility of such request(s). The objectives in A's decision framework must be such that a negotiated request 'do c' is either granted, or appropriate corrective action is taken. Each role of an agent is a negotiated commitment to perform in the future any and all actions that the role requires.

Consistency of decision frameworks and their allocation to agents

In our model there are four consistency criteria to be maintained when roles are allocated to agents:

• Functional adequacy: both qualitatively and quantitatively the agent must have the abilities to perform the functions necessary in the agent's roles (Fig.5).

• No role conflicts: avoid conflicts among the roles played by any given individual agent; and conflicts between the set of roles allocated to one agent and the set of roles allocated to another agent, especially if their plans may be contingent on one another (Bernus and Uppington, 1998).

• Shared priorities, policies and values: the agent's individual objectives are part of the agent's decision framework and as tasks compete for the agent's resources the agent must resolve the competition of objectives through applying priorities, policies or values. If autonomously applied these must be shared such that an agent's decisions can be accepted by the others as valid and justified. Therefore a shared value system and subscription to policies is a necessary condition of negotiation and correct interpretation of commitments (Fig.6).

• Adequate level of autonomy: It is possible to negotiate with, and request from an agent the performance of each small step of a task or negotiate the entire task as one. If the agent's capabilities allow the definition of the task on a higher level there is no need to extend control to the details of the task (See Fig.7).

Unnecessarily detailed control limits the agent's ability to utilise its faculties and optimise resource usage. Intrusive control (lack of autonomy) has the potential to stop the agent from co-operating because its own priorities, policies and values negate the needlessly prescribed practices.

Fig. 4 Enterprise Organisation consisting of agents A, B, E and F

Fig. 5 Capabilities (abc) match agent's roles (a,b)

Fig. 6 If organisational policies and individual policies and values are inconsistent, then expectations fail, communication may break down, commitment is weakened

These criteria must extend to all elements influenc-ing the agent's behaviour in an organisation (e.g. personal, social, psychological determinants). Since there is little hope that these elements could be made explicit or that it would be socially acceptable to do so, structures must be devised allowing these hidden elements to 'take care of themselves' (co-ordination based on limited mutual knowledge).

Fig. 7 Over-controlled vs. appropriately controlled agent. Agent A can be controlled through specifying task as 'd' rather then over-specifying the control by taking over autonomous control and dictating a,b and c to achieve d

Static organisation:

We define the 'static' organisation as one created by management to establish a capability to achieve certain class of objectives. If day-to-day objectives fall in this class then the organisation functions effectively.

Organisational design becomes a means of indirect control because it determines the channels of decision making

Based on the properties of the given organisational design management can determine any current objectives as long as they are in the class for which the organisation was designed. The organisation responds by actions (to achieve the current particular objectives) and provides feedback for management, closing the control loop. Under many conditions the static organisation is effective and efficient. However, the organisation will still have spontaneity, since the individual agents' decisional frameworks contain, but are not limited to, the objectives passed on by the organisation.

Management will regularly realise that the class of objectives for which the present design was suitable starts covering less and less of actual objectives and adapts the enterprise through regular re-organisa-tion. However, no organisation can depend solely on static principles: when unexpected changes happen response must be dynamic (see 5.3), at least while re-design adapts to a new class of objectives.

In circumstances where the design by management is always short-lived the environment is perceived as turbulent - although turbulence is not a property of the environment alone, but of the relationship between the environment and the organisation.

Dynamic organisation:

The dynamic organisation is a configuration of agents designed, built and operated by the involved agents for a given class of objectives using explicit negotiated commitments and not operating based on a status quo. Individual agents build mutual commitment, and abide by established co-ordination and co-operation rules instead of rigidly segregating the authority to design the organisation from operational tasks.

Both static and dynamic organisations are built for a purpose or mission' but their agility is different. A static organisation tends to preserve the present state, a dynamic organisation releases purpose-based commitments when the mission is completed.

Three requirements for dynamic organisations are:

• It is necessary to have commonly available proven reference models of dynamic organisations and commonly accepted co-ordination protocols that can be adhered to in order to form temporary mutual commitments;

• It is necessary for the individuals in the enterprise to develop a working knowledge of these models and protocols, to enable them to dynamically build the organisation themselves.

• The above requirements need to be complemented by responsibility structures to ensure that lost management functions are substituted to retain control of the organisation.

Several experiments showed the success of this principle, e.g. in industrialised countries where worker commitment, uniform shared values, and of unwritten policies facilitated it. See successful socio-technical experiments in Nordic countries (van Houten, 1990, Gibson, 1973).

Recent designs with agent based integration (whether stated explicitly or not) include holonic manufacturing, agent co-ordination protocols, and the fractal factory. Agent based integration is not limited to computer implemented agents but can be applied in purely human implemented organisational structures. See for a comparison of a some models in (Tharumarajah et al, 1996)

Some models for the dynamic organisation

Matrix The matrix organisation was proposed with the aim of separating the rigid functional structure of traditional organisations from task-specific project organisation to achieve dynamic behaviour. In a matrix each individual at any one time belongs to a functional entity and to one or more projects. However, it has been reported that matrix organisations where allocation of tasks and commitment was based on authoritative manage-ment (as in a static organisation), the advantages of the matrix did not get realised (Evans, 1982).

Similar results are claimed about failures of workflow systems (Klein, 1996).

It has been very hard to establish sound proofs of these claims, because the problem can not be studied on models constructed by external observers; it needs sociographic immersion methods to draw reliable conclusions. Under-standably so, because the conflict between organisational and individual objectives can not be studied unless the normally covert, invisible individual objectives are understood. Such research requires the researcher to become familiar with the individual subjects in the organisation, gain their trust, and access the 'invisible' component.

Project Project organisations are designed for a specific task (project). The project organisation does not typically remain within the boundaries of a parent organisation's management. (Projects in the matrix organisation are a special case of project organisation).

A typical scenario is where a main contractor establishes a project through negotiating with sub-contractors. Engineering companies usually operate in this mode: the major activity of the engineering company is to identify, create and support project enterprises, based on multiple levels of contracts.

Because failure is not a possible outcome the establishment of a high level trust in the feasibility of task allocation is paramount. Large project organisations therefore base the project component on proven abilities and potentialities of individuals (Christiansen et al., 1995), while risk-taking, longer-term capability development activities are left to a functional part of the organisation.

Fractal The name fractal factory comes from the recognition that fractals have a self-similar structure, repeating on all neighboring levels of aggregation (Warnecke, 1993). This form of organisation is based on the fact that the complexity of organisational boundaries makes organisational design complex, error prone, and inflexible. If it was possible to design models of organisational entities (cells, teams, workshops, departments, etc.) so that all of these would have the same control protocol (including negotiation and commitment) then organisational design would be simpler and more flexible. If in addition certain combination principles are observed, then agenthood is preserved during aggregation, therefore adhering to these principles the desired organisational properties would automatically emerge.

Holonic The holonic manufacturing principle (Valckenaers et al, 1996) is based on the desire that all levels of aggregation have certain defined levels of autonomy. On each level (individual, cell, team, department, etc.) enterprise entities have a "shell" around them enclosing an autonomous whole (Koestler, 1989).

We define autonomy as a relative property: an entity is not "autonomous" or "not autonomous," autonomy is the ability to perform a set of functions in a space-time confinement, leaning only on a defined or implied ubiquitous infrastructure.

If such infrastructure is available for each holon then holons may be considered autonomous.

The holonic property is a necessary but not sufficient condition of enterprise organisation. E.g. only such holonic systems would be viable which satisfy the consistency criteria discussed.

Agent based integration Transactions among agents need uniformly interpreted co-ordination protocols such as COOL (Barbuceanu and Teigen 1998). However, transactional protocols are necessary but not sufficient to ensure that the organisation will behave as an agent.

Conclusion

We discussed conditions of how to construct and organisation as an agent from lower level agents.

The resulting organisation will use a combination of the reviewed approaches, and can be characterised as a holonic, complex agent which recursively consists of (holonic) agents dynamically organising themselves, using uniform agent co-ordination protocols, into fractal-like organisational constructs both in management and operation. In this "ideal" organisation operational task commit-ments are based on matrix-like project entities, and resource development tasks are based on functional entities with an authority structure. Individual organisations would use a mixture of functional and project tasks, the proportion depending on the need for dynamism and the need for high skill versus low skill. Organisational design transactions would allow the satisfaction of the described consistency criteria of systems of objectives.

References

Barbuceanu,M. Teigen,R. (1998) System Integra-tion through Agent Coordination. In P.Bernus, K.Mertins and G.Schmidt (Eds) Handbook on Architectures of Information Systems, Berlin: Springer, pp797-826.

Bernus, P., Nemes, L. (1996) A Framework to Define a Generic Enterprise Reference Architec-ture and Methodology. Comp. Integr. Manuf. Systems 9(3) pp 179-191.

Bernus,P., Espinasse,B., Fox,M.S., Goranson,H.T. (1997). Business Evolution and Enterprise Integration. In Proc ICEIMT'97, K.Kosanke and J.G.Nell (Eds.), Berlin: Springer, pp140-151.

Christiansen,T., Christensen,L., Jin,Y., Lewitt,R.E., Kunz,J. (1995) Enterprise Integration by Modelling and Simulation - a framework and an application in the offshore oil industry. In Modelling and Methodologies for Enterprise Integration, P.Bernus and L.Nemes (Eds), London: Chapman&Hall, pp282-304

Doumeingts,G., Vallespir,B., Chen,D. (1998) Decisional Modelling using the GRAI Grid. In P.Bernus, K.Mertins and G.Schmidt (Eds) Handbook on Architectures of Information Systems, Berlin: Springer, pp 313-338

Evans,G.S. (1982) An exploratory test of the matrix assumption in a highly differentiated research organization: structural design versus behavioral imperatives. IEEE Trans. on Eng. Mgmt. vol.EM-29, no.3, pp 78-81

Gibson,Ch.H., (1973) Volvo Increases Productivity Through Job Enrichment. California Mgmt Review, 15(4) p64.

Klein,M., (1996) Challenges and Directions for EI: a Distributed AI perspective. In Modelling and Methodologies for EI P.Bernus and L.Nemes (Eds), London: Chapman&Hall, pp319-332

Koestler,A.. (1989) The ghost in the machine. Arkana Books

Tharumarajah, A., Wells,J., Nemes,L., (1996) Comparison of the bionic, fractal and holonic manufacturing systems concepts. Int. J. on Computer Integrated Manuf. (3) pp.217-226

Uppington,G., Bernus,P., Assessing the Necessity of Enterprise Change: Pre-feasibility and Feasibility Studies in Enterprise Integration. Int. J. of CIM, 1998 11(5) pp 430-447

Valckenaers,P., VanBrussel,H., Bongaerts,L., Wyns,J. (1997) IMS test case 5: holonic manufacturing systems. Journal of Integrated Computer-Aided Engineering, 4(3), pp191-201.

VanHouten,D.R. (1990) The political economy and technical control of work humanization in Sweden during the 1970s and 1980s. Work and Occupations, 14, pp483-513.

Warnecke, H.J. (1993) The Fractal Company. Berlin: Springer.

Williams,T.J., (1994) The Purdue Enterprise Reference Architecture. Computers in Industry, 24 (2-3) pp141-158

Williams,T.J., Bernus,P., Brosvic,J., Chen,D., Doumeingts,G., Nemes,L., Nevins,J.L., Vallespir,B., Vlietstra,J., Zoetekouw,D., (1994) Architectures for integrating manufacturing activities and enterprises. Computers in Industry, 24(2-3) pp111-140