A FRAMEWORK FOR SITUATED DISTRIBUTED DECISION SUPPORT1

ABSTRACT

The increasing complexity of business environment makes centralized decision making inadequate in many cases. With the distributed computing paradigm gaining more support and attention investigation of new models for distributed decision support systems is an important direction of research. This paper looks to expand the recently introduced model for situated DSS to address distributed and coordinated decision support. To demonstrate the applicability of this framework, a prototypical application for lead time management has been implemented. Simulation experiments have been conducted to investigate the impacts of real-time information and information coordination on decision performance. Overall, the results support our expectation that both information delay and information coordination have significant impacts on decision performance.

Keywords: Distributed Decision Making, Situated DSS, Distributed DSS, Lead Time Management, Simulations

1. INTRODUCTION

The growing complexity of die business and technological environment makes centralized decision making difficult or even impossible in many domains due to the limited cognitive capacities of human decision makers. Consequently, decision making is often distributed over empowered employees within an organization as well as over suppliers and/or customers across organizations as evidenced by emerging business practices such as decentralization, customerization [30] and supply chain management [6]. Decision Support Systems (DSS) have been traditional means of complementing human judgment with computational models, data, and knowledge repositories to enhance decision making effectiveness. In particular, various types of DSS accommodating multiple decision makers include Group Decision Support Systems (GDSS) [2, 13], Organizational Decision Support Systems (ODSS) [1, 18], and Negotiation Support Systems (NSS) [14].

With the advance of technology-enabled practices, such as e-commerce, enterprise-wide computing, and supply chain management, the amount of rapidly changing information generated, processed and stored has increased dramatically. Modern effective decision making has to focus on timely information processing, situation assessment and decision and action generation in a dynamic fashion [10, 26]. This demand for dynamic decision making support has prompted interest in research on real-time DSS [25] and "situated" DSS [28].

This paper aims at proposing new DSS framework that facilitates development of flexible systems to support dynamic distributed decision making. The framework is an extension of a recently introduced situated DSS model [28]. In particular, it should provide a solid foundation for the development of DSS, which would be better aligned witìi organizational or interorganizational distributed decision structures.

2. DISTRIBUTED DECISION MAKING AND SUPPORT

2.1 Distributed Problem Solving/Decision Making Process 2.2 Related work

Eom's review of DSS research covering the period from 1970 to 1999 [7] shows that distributed decision making has been somewhat underinvestigated by researchers, though there have been significant efforts focusing on related systems, including Group Decision Support Systems (GDSS), Organizational Decision Support Systems (ODSS) and Negotiation Support Systems (NSS). In die discussion of past, present, and future of decision support technology, Shim et al. [23] pay little attention to DDSS when they discuss collaborative support systems.

The term Distributed DSS (DDSS), originally introduced by Scher [21], referred to a conference-based system in an organization. Ratiiwell & Burns define a distributed decision making system as a cooperative network facilitating communication and conflict resolution among equal decision makers [20]. Swanson treats DDSS as one perspective of organizational DSS [24]. Based on Swanson's definition of ODSS, Chung et al. state that DDSS, conceived as a network of decision making nodes in an organization, is a subset of ODSS [4]. In addition, they divide DDSS into two categories: Rigid DDSS and Flexible DDSS. Other researchers have devised various technical approaches to build DDSS. Chi and Turban proposed to use agents for distributed resources, such as knowledge base and DBMS to support executive decision making [3]. Jeusfeld and Bui proposed a script language to allow construction of DSS from components stored on various Internet sites [15]. Ju et al. proposed an agent-based architecture of DDSS and discussed how the agents coordinate to largely automate decision making [16]. Gachet discussed a decentralized technical architecture for distributed DSS [8]. Gachet and Haettenschwiler conducted a case study to identify die impact of single-user DSS vs. distributed DSS in the collective decision making process [9].

Based on past work, we find that die traditional view of DDSS is either too narrow to cover some important aspects of distributed decision making process, for example, the peer-to-peer relationship between two decision units, or too technical to consider die nature of distributed decision making process, i.e. die interdependence of preference structures of decision units, which could be dynamic and uncertain. To bring together diverse research on DDM from such disciplines as computer science, economics, organizational theory and psychology, Schneeweiss proposes a unified model of distributed decision making characterizing it as the design and coordination of connected decisions [22]. In the present work, a decision model is viewed as a set of decision criteria C, action space A, and status of information /. In its simplest form, decision making is distributed over two decision units. If these two units have an identical decision model, i.e. sharing same decision criteria, action space, and information, we may say that they are in perfect cooperation. On the contrary, if their decision models are completely different, they are in fact isolated from each other. Usually the relationship between two units is somewhere between these two extreme situations.

2.3 Situated decision support systems

Recently, there has been rise of interest in facilitating active decision support [27]. For example, in [27] an agent-based architecture for DSS organized around key decision-making phases had been proposed. Vahidov and Kersten argued that traditional DSS research ignored implementation and monitoring phases of decision making process [28]. Inspired by software agent technologies and research on active DSS, they proposed a new framework of DSS called "Situated DSS", or "Decision Station" architecture, which promotes close links with die problem environment and has capabilities to implement decisions through the "effectors" as well as monitor the change of problem environment by means of "sensors".

Sensors and effectors are key components diat differentiate situated DSS from traditional "toolbox" DSS. Sensors can be equipped with passive capabilities, such as connecting, transforming, querying and alerting users, and/or active capabilities, such as adapting and planning. Their primary task is to collect relevant information, which could be either structured or unstructured about the dynamic developments in the problem domain. For example, in [5] information collection agents have been proposed for gathering information from unstructured sources. Effectors have the capability to alter the state of affairs in the target domain. Their primary task is planning and execution of actions based on the decisions made by a human or agent decision maker. Additionally, the framework contains active user interface and DSS kernel. The latter includes data, models, and knowledge, as well as the active "manager" component that performs the task of monitoring and has limited authority for autonomous action. The advantage of this architecture is that it can provide human decision makers with timely and proactive decision support in dynamic and complex environment. We believe that distributed decision support based on the ideas of situatedness and proactiveness, could be an adequate response for the prevaiUng trends in modern distributed dynamic decision making.

3. DISTRIBUTED DECISION STATIONS

To propose the design of distributed decision stations (DDS), we need to identify key components of information exchanged between the decision units. Kim et al. present four kinds of inputs and outputs in modeling coordination subsystem for ODSS [17]. Based on their ideas, we divide information into four categories (in case of two decision units): I = I^sup L^, I^sub S^, I^sup E^, I^sup R^}, where IL represents local information of a decision unit (DU), 7s represents me commonly shared information between DUs, F is the control information from other DU, and I^sup R^ is the feedback information from the other DU. When a DU receives control information, it can do nothing but follow this information in its decision making process as an instruction. On the I^sup R^ contrary, the feedback information may or may not be taken into account for generating control information by a given DU.

Considering that information asymmetry is more common than information symmetry, Schneeweiss defines one DU as top level and die otiier one as base model [22], We tend to treat two interacting DUs somewhat equally, i.e. by default there is no inherent hierarchy between the units. This structure is shown in figure 1 with the decomposition of information status I.

Here C^sup L^^sub I^ refers to private criteria of top level, while C^sup E^^sup I^ stands for representation of base-level preferences and vice versa. The constitution of criterion C determines how a DU treats the other unit. For example, if C^sup E^^sub I^ = ∅, it means that top-level DU is egocentric; if C^sup L^^sub I^ = ∅ then it is altruistic. Normally, these criteria are non-null.

Based on the decomposition of criteria structure, we can arrive at six styles of coordination between two decision units, listed in table 1 . In this table, 0 stands for an empty set and 1 stands for a nonempty set. For example, regarding coordination style 1 , the local criteria of both decision units are empty, but their external criteria are not empty. With regard to information sharing, we can obtain eight types of information coordination, presented in table 2.

4. DISTRIBUTED DECISION STATIONS FOR LEAD TIME MANAGEMENT

This section describes a business problem domain to illustrate the application of die framework for distributed decision stations and describes an implemented prototype.

4.1 Lead Time Management

During the past decade, the increasing popularity of the Internet has promoted many manufacturing companies to adopt e-commerce business model, which enables them to directly interact with end consumers by eliminating costly intermediaries in the traditional supply chain. Correspondingly, their business philosophy has shifted from production-centric strategy to customer-centric strategy. This, in turn prompts the shift from mass production, to mass customization, and even to "customerization" [30] , an effort to integrate customer into internal business processes, such as collaborative product design. In mass production model products are usually Made-To-Stock (MTS) according to sale forecasts, whereas in mass customization model products are made to actual customer orders (MTO). As a result, MTO companies can enjoy the benefits of product flexibility and lower inventory cost, but at the same time they might suffer from longer and unstable Order-To-Delivery (OTD) lead times.

Research in production and operations management shows mat the popular production planning and control methods, for example Just-in-Time and Theory of Constraints are inappropriate for MTO companies due to considerable uncertainty of production and market environments [19]. This challenge faced by MTO firms stimulates much research on lead time management in MTO companies, and the emergence of workload control (WLC) concept [11, 12, 19]. Simply stated, the principle of WLC, is to achieve stable OTD lead times by adjusting the input to and output from production system, i.e. the load and capacity of production system, respectively [29].

According to input-process-output model, production can be treated as a transformation process that employs various resources, such as machine and labor to convert raw materials into final products [19]. WLC, a concept first introduced by Wight [29], is aimed at maintaining transformation time at a normal level by controlling input/output. Conceptually, workload is modeled as a queue, where jobs are waiting to be processed at a certain resource. As a whole, a production system can be modeled as a queuing network, and hence a computational solution is hardly feasible [H].

The main purpose of workload control is to manage lead time. In the formal analysis of workload control model, Kingsman identified four phases of order fulfillment process in MTO companies and pointed out four corresponding levels of workload control, including customer inquiry, order acceptance, job release, and priority dispatching [19].

4.2 A business case

A fictitious, yet realistic web-based MTO manufacturing company is considered here to implement lead time management using WLC concept. The sales orders of die company primarily arrive from e-customers. The company allows its customers to personalize products and, thus production processing time of each order is varied with the specific product configuration. These characteristics seem to best fit the application of WLC [12].

Based on Kingsman's WLC model, the order flow and lead time structure is depicted in figure 3. We assume that supplier can deliver required materials without delays, and hence, the waiting time for materials can be ignored in our model. In addition, we assume that total manufacturing lead time Tm is relatively small due to the high efficiency of manufacturing and that the default scheduling rule is first-in-first-out (FUFO). Under these assumptions, the total order-to-delivery lead time is largely dependent on the time orders remain waiting in the order pool 7\ Thus the management of OTD lead time is highly determined by the management of the order pool size.

The stable size of the order pool can be maintained by adjusting demand and/or production capacity. In our case, a demand/capacity planner is assigned to coordinate marketing and production departments and can set demand and production objectives to the respective departments to ensure an adequate customer service level while keeping down marketing expenses and production costs.

To provide customers with more flexibility, the company conceives a new idea that they not only can track order but also can switch their order positions in the order pool by negotiating a corresponding compensation among themselves. As a consequence, a customer can get his/her order processed earlier if he/she is willing to pay a certain amount to the other customer. System composition and architecture

In analyzing the above case, four different kinds of decision stations can be proposed (figure 4). Decision station DSC is designed to support demand/capacity planners. Its sensor fetches real-time information about the current size of orders in pool, computes average demand for recent period, for example last day or last hour, die current production capacity, and alerts users when actual and/or production capacity is beyond the user-defined limits. The unit's actions include instructing other stations to achieve desired levels of average demand and production capacity per shift.

Decision station DS^sup M^ is designed to support marketing managers for short-term marketing control. It captares the real-time information of sales orders and provides marketing managers with trend analysis. Marketing managers can control demand by changing product price, product mix, advertising, as well as by following other promotional strategies. Decision station DS^sub P^ is designed to production managers for short-term capacity control. It captures die real-time information of current production capacity and provides the managers with cost analysis. Production managers can control capacity by hiring temporary employees, adding shifts, and subcontracting.

Decision station DS" is designed for buyers, i.e. customers on the Internet. When a customer places an order on the Internet, a default standard order-to-delivery lead time is placed in this order. However, if the customer is eager to get the product, he/she can negotiate with the other customers to advance the order.

4.3 Prototype Implementation

In order to illustrate the approach and conduct simulation studies, we have developed a prototype for the chosen case. A sample Screenshot of the decision station for demand/capacity planner is shown in figure 5. The left panel of the screen is demand and capacity monitor that displays information provided by the sensor Sc. The sensor can capture and calculate required data in real-time or in a period of delay (it is used to support our simulation study). The right panel is a queuing model to support lead time decisions, where two decision variables, die mean demand (shown as order arrival rate) and production capacity are input parameters.

A (human) planner can use queuing model to generate various alternatives, i.e. different combination of order arrival rate and production capacity. When the planner is satisfied with analysis result, he/she can pass control information to the decision stations for marketing managers and production managers, triggering other decision making processes.

5. SIMULATIONS

In the current study we assume that decision makers will choose an alternative based on defined preferences and, tiius, their choices can be reasonably modeled. The two factors included in our study are information delay and information coordination. Information delay is a key independent variable since it allows distinguishing between the isolated "toolbox" model of DSS and Decision Station that emphasizes continuous monitoring and action.

Three levels of information delay are set in this simulation: real-time (50 minutes); delay 10 (500) minutes; and delay 20 (1000) minutes. Information coordination refers to information control and feedback by the aid of control information Ie and reference information I". In our experiment, we setup three types of information control:

* Controlling marketing DS only by setting up an expected normal demand;

* Controlling production DS only by setting up an expected normal production level;

* Controlling botii marketing and production DSs.

Regarding information feedback, we designed an experiment in which either marketing or production department may be unable to fulfill the instruction from demand/capacity planner. If information feedback is enabled, they can send this information (e.g. die lack of capacity) to the planner, and then planner can adjust the instructions correspondingly. For example, wiüiout information feedback, the planner assumes that marketing department is able to adjust demand to any level between 5 orders per hour and 7 orders per hour. But in fact, the marketing department might only be able to achieve a demand level between 5 orders per hour and 6 order per hour. Due to the lack of feedback, die planner might lose a chance to do the other corrective actions, e.g. decreasing production capacity. In the experiment, we set up four types of information feedback: from both marketing and production; from marketing only; from production only; and no feedback.

Our expectations are formulated through the following hypotheses:

Hl: The level of information delay is negatively related to decision performance.

H2: The level of information control is positively related to decision performance. In our case, it means that controlling botii marketing and production department will achieve better performance than controlling eitíier marketing or production department.

H3: The level of information feedback is positively related to decision performance.

In our case, it impUes receiving information feedback from both marketing and production department will lead to better performance than receiving information feedback from either marketing or production department, and receiving information feedback either from marketing or production will lead to better performance than receiving no information feedback.

We define two sets of simulations, case 1 and case 2. In case 1, the arrivals of purchase order from website follows Poisson distribution with the mean of six orders per hour. The production time of finishing one batch of orders follows exponential distribution with the mean of eight hours. In case 2, the distribution of order arrivals is the same as in case 1 , but the production time follows uniform distribution with lower bound as seven hours and higher bound as nine hours. In both cases, the standard production time is eight hours. In addition, we set up simple decision rules. If the current order pool size is bigger than the upper limit, the decision aid will either decrease the mean of order arrivals down to three orders per hours or increase production capacity up to 100 orders per shift. If the current order pool size is smaller than the lower limit, the decision aid will either increase the mean of order arrival up to nine orders per hours or decrease production capacity down to 50 orders per shift.

6. RESULTS

6.1 Case 1: Production Time Follows Exponential Distribution

We first consider the impacts of information delay and information control. Results of ANOVA test for the case 1 revealed no significant interaction between information delay and information control (p = 0.656). The deviation of order pool size is significantly related to the level of information control ( ? < 0.001), while information delay does not appear to have significant impact (p = 0.192).

Further, we do contrast tests to compare the means of the three levels of information control. Contrast 1 checks the difference between controlling marketing and controlling both marketing and production. Contrast 2 looks at the difference between controlling marketing and controlling production. Contrast 3 examines the difference between controlling both marketing and production and controlling production. The results are shown in table 3.

Thus, there was no significant difference between controlUng marketing and controlling both marketing and production. However, controlling marketing only will achieve significantly (p < 0.001) less fluctuation in order pool than controlling production only. This is due to the fact that in simulations we assumed that marketing initiatives could be enacted quicker than making changes to production. Similarly, controlling both marketing and production will get significantly (p < 0.001) less fluctuation in order pool than controlling production only. This provides overall support for hypothesis 2.

It is somewhat surprising that the impact of information delay was insignificant in the overall test. However, the interaction plot reveals some interesting insights (figure 7). If we drop die case of controlUng production only from consideration, it turns out that information delay is significantly related to the deviation of order pool size at ? = 0.001. Thus in this case the Hypothesis 1 also finds partial support. This is most likely due to the fact that adjusting production capacity requires more time and cannot be done immediately when a need for such change is being sensed.

To examine the impact of information feedback on decision performance, we do three contrasts test. Contrast 1 compares full information feedback with no information feedback, contrast 2 compares information feedback from marketing with no information feedback, while contrast 3 compares information feedback from production with no information feedback. The results are shown in table 6. Both Contrasts 1 and 3 are significant at ? < 0.05, but contrast 2 is insignificant. It means that receiving information feedback from marketing department only does not significantly improve decision performance significantly compared to no information feedback at all. Feedback from marketing may indicate that die respective DS may not be able to achieve given objectives. This would imply change to the production capacity. But since, again the latter cannot be executed rapidly, the overall performance could be lower than a desirable level. On the contrary, receiving information feedback from production or from both marketing and production can significantly improve decision performance compared to no information feedback. This provides an overall support for hypothesis 3.

6.2 Case 2: Production Time Follows Uniform Distribution

Similar to the previous case die here the deviation of order pool size is significantly related to the types of information control ( p< 0.001). The results of contrast tests (denoted in the same manner as previously) are shown in Table 7. As shown in the table, there is no significant difference between controlling marketing and controlling both marketing and production. However, controlling marketing only will achieve significantly (p<0.00 1 ) less fluctuation in order pool tiian controlUng production only. Similarly, controlling both marketing and production will get significantly (p< 0.001) less fluctuation in order pool than controlling production only.

Again, since the information delay did not appear to have significant impacts overall, here we wiU try to identify finer details by ignoring the case of controlUng production only. The results, are similar to that of Case 1 , i.e. information delay is significantly related to decision performance at ? < 0.001 .

Regarding information feedback we found that in case 2 only receiving information feedback from production appears to have a significant effect (p = 0.045). Thus, one can conclude that the findings from the case 2 are similar to those from Case 1.

7. DISCUSSION AND CONCLUSIONS

The paper has introduced the framework for distributed decision stations. The framework extends the notion of situated DSS to support distributed decision making. To iUustrate the idea a prototype system for workload balancing has been implemented. Overall, the simulation experiments have provided reasonable support for our expectation diat information recency and coordination improve decision quaüty. Some apparently counter-intuitive results relate to coordination with the production department. The lesson learned from this finding implies diat in particular settings one should carefuUy study the characteristics of the problem domain. In particular, the feasibihty of carrying out actions, as well as their timing should be adequately reflected in the knowledge components of die respective decision stations.

The work has important practical impUcations. With the increasing interdependeny of intra- and inter- organizational decision processes as weU as complexity and dynamics of business environment die framework offers a possible technological response to promote timely and effective decision making and decision enactment. Future work could be focused on developing a technical architecture to support the framework. For this reason, it is worth investigating potential appUcability of advanced technologies, such as software agents, to build a solid and flexible structure supporting the conceptual framework.

One limitation of the study is in the use of simulations instead of human subjects. Since DSS are human-computer systems the ideal way to evaluate their effectiveness is through involving human expert decision makers. However, we assumed rational decision making in the experiments, and used simple rules to imitate human decision making. Conducting human-subject experiments could be one of die future projects and it would require the subjects to have some level of experience in the area.

[Sidebar]

Received: October 20, 2009 Revised: December 23, 2009 Accepted: January 13, 2010

I. This work has been supported by the grant from Natural Sciences and Engineering Research Council of Canada

[Reference]

REFERENCES

1. Aggarwal, A. L, and Mirani, R. "Macro Issues in die Development of Organizational Decision Support Systems," 28th Hawaii International Conference on System Sciences. Maui, Hawaii, January 3-6, 1995, pp. 917-926.

2. Barkhi, R., Jacob, V S., Pipino, L., and Pirkul, H. "A Study of the Effect of Communication Channel and Authority on Group Decision Processes and Outcomes," Decision Support Systems (23:3), 1998, pp. 205-226.

3. Chi, R. T, and Turban, E. "Distributed Intelligent Executive Information Systems," Decision Support Systems (14:2), 1995, pp. 117-130.

4. Chung, H. M., Mahapatra, R., and Marin, G. "Distributed Decision Support Systems: Characterization and Design Choices," 26th Annual Hawaii International Conference on System Sciences. Wailea, Hawaü, January 5-8, 1993, pp. 660-668.

5. Conlon, S. J., Hale, J. G., Lukose, S., and Strong, J. "Information Extraction Agents for Service-Oriented Architecture Using Web Service Systems: A Framework," Journal of Computer Information Systems (48:3), 2008, pp. 74-83.

6. Davenport, T. H., and Brooks, J. D. "Enterprise Systems and the Supply Chain," Journal of Enterprise Information Management (17:1), 2004, pp. 8-19.

7. Eom, S. B. Decision Support Systems Research (1970-1999): A Cumulative Tradition and Reference Disciplines. The Edwin Mellen Press, Lewiston, NY, 2002.

8. Gachet, A. "A New Vision for Distributed Decision Support Systems," Decision Making and Decision Support in the Internet Age: Proceedings of the Dsiage 2002 Conference. Adam, E, Brezillon, P., Humphreys, P., and Pomerol, J. C. edited. Oak Tree Press, 2002, pp. 341-352.

9. Gachet, A., and Haettenschwiler, P. "A DecentraUzed Approach to Distributed Decision Support Systems," Journal of Decision Systems (12:2), 2003, pp. 141-158.

10. Haeckel, S., and Nolan, R. "Managing by Wire," Harvard Business Review Sept.-Oct., 1993, pp. 122-132.

11. Haskose, A., Kingsman, B. G., and Worthington, D. "Performance Analysis of Make-to-Order Manufacturing Systems under Different Workload Control Regimes," International Journal of Production Economics (90:2), 2004, pp. 169-186.

12. Henrich, P., Land, M., Gaalman, G., and Zee, D.-J. V. D. "Reducing Feedback Requirements of Workload Control," International Journal of Production Research (42:24), 2004, pp. 5235-5252.

13. Jarvenpaa, S. L., Rao, V. S., and Huber, G. P. "Computer Support for Meetings of Groups Working on Unstructured Problems: A Field Experiment," MIS Quarterly (12:4), 1988, pp. 645-666.

14. Jelassi, M. T, and Foroughi, A. "Negotiation Support Systems: An Overview of Design Issues and Existing Software," Decision Support Systems (5:2), 1989, pp. 167181.

15. Jeusfeld, M. A., and Bui, T. X. "Distributed Decision Support and Organizational Connectivity: A Case Study," Decision Support Systems (19:3), 1997, pp. 215-225.

16. Ju, C, Ling, Y, and Norman, T. J. "Agent-Based and Software Composite Ddss," Technology Object-Orieted Language System Tools. Xi'An, China, October 30-November 4, pp. 50-57.

17. Kim, G.-M. J., Graves, E. G., Burns, J. R., and Myung, R. "An Architecture for Organizational Decision Support Systems (Odss) That UtiUzes a Model Coordination Subsystem," Journal of End User Computing (12:3), 2000, pp. 23-34.

18. King, J. L., and Star, S. L. "Conceptual Foundations for die Development of Organizational Decision Support Systems," 23rd Annual Hawaii International Conference on System Sciences. Kailua-Kona, Hawaii, January 2-5, 1990, pp. 143151.

19. Kingsman, B. G. "Modelling Input-Output Workload Control for Dynamic Capacity Planning in Production Planning Systems " International Journal of Production Economics (68:1), 2000, pp. 73-93.

20. Rathweel M. A., And Burns A. Infromation Systems Support for Group Planning and Decision-Making Activities," MIS Quarterly (9:3), 1985, pp. 255.

21. Scher, J. M. "Distributed Decision Support Systems for Management and Organizations," First International Conference on Decision Support Systems, DSS -81. Atlanta, GA, 1981, pp. 130-140.

22. Schneeweiss, C. "Distributed Decision Making - a Unified Appraoch," European Journal of Operational Research (150:2), 2003, pp. 237-252.

23. Shim, J. P., Warkentin, M., Courtney, J. R, Power, D. J., Sharda, R., and Carlsson, C. "Past, Present, and Future of Decision Support Technology," Decision Support Systems (33:2), 2002, pp. 111-126.

24. Swanson, E. B. "Distributed Decision Support Systems: A Perspective," 23rd Annual Hawaii International Conference on System Sciences. Kailua-Kona, Hawaii, January 2-5, 1990, pp. 129-136.

25. Tseng, C-C, and Gmytrasiewicz, P. J. "Real Time Decision Support System for Portfolio Management," 35th Hawaii International Conference on System Sciences. Big Island, Hawaii, January 7-10, pp. 1348-1356.

26. Vahidov, R. "Empowering Modern Managers: Towards an Agent-Based Decision Support System," Organizational Data Mining. Nemati, H., and Barko, C. edited. Idea Group PubUshing, 2004, pp. 170-187.

27. Vahidov, R., and Fazlollahi, B. "A Multi-Agent Dss for Supporting E-Commerce Decisions," Journal of Computer Information Systems (XLIV:2), 2003-2004, pp. 87-94.

28. Vahidov, R., and Kersten, G. E. "Decision Station: Situating Decision Support Systems," Decision Support Systems (38:2), 2004, pp. 283-303.

29. Wight, O. "Input/Output Control: A Real Handle on Lead Time," Production and Inventory Managment (1 1:3), 1970, pp. 9-31.

30. Wind, J., and Rangaswamy, A. "Customerization: The Next Revolution in Mass Customization," Journal of Interactive Marketing (15:1), 2001, pp. 13-32.

[Author Affiliation]

HUABING HU

Concordia University

Montreal, Quebec, Canada H3G 1M8

RUSTAMVAHIDOV

Concordia University

Montreal, Quebec, Canada H3G 1M8