2. Conceptual framework
2.1. The decision making process
McDonald (2006) presents a ten-step strategic marketing planning process covering mission, corporate objectives, marketing audit, SWOT analysis, assumptions, marketing objectives and strategies, estimate expected results, identify alternative plans and mixes, budget, and the first year detailed implementation programme.
While delineating the logical and systematic planning process, guidelines are offered to ensure that the input to the marketing plan is customer focused and considers pertinent strategic dimensions.
Step-by-step international marketing decision making involving go vs. no go decisions, entry mode selection, and marketing strategy development has been emphasized by many authors (Chee & Harris, 1998; Doole & Lowe, 2004; Jain, 1990; Root, 1987). This study will focus on these three main stages. It is argued that the planning process involves extensive expertise (McDonald & Wilson, 1990), uncertainty (Levy & Yoon, 1995; Li, Kinman, Duan, & Edwards, 2000), and decision-makers’ judgement and intuition (Li, 2005; Li et al., 2000; Mintzberg, 1994a, 1994b).
2.2. Human judgement
Mellers, Schwartz, and Cooke (1998) highlight the importance and roles of human judgement in decision making while examining the limitations of rational choice theory. Mintzberg (1994a, 1994b) argues that because managers may have experience and some specific knowledge about their products and markets, strategic planning must be coupled with managers’ judgement and personal vision. This argument is confirmed by Li et al. (2000)’s mail survey findings: The majority of the managing directors or marketing directors of the 104 responding large UK companies strongly agreed or somewhat agreed that their intuition and judgement are important in marketing strategy formulation.
2.3. Supporting techniques and technologies
The use of DSSs in marketing has attracted interests since 1970s (Little, 1979). A DSS is an interactive computer-based system that helps managers utilise data and models to deal with semi-structured or ill-structured decision making problems (Turban, Aronson, & Liang, 2005). Conventional DSSs are not powerful at handling the cases where managers make decisions based heavily upon their own experience and expertise. Traditional DSSs may also be restricted to the quantitative models used and cannot cope with missing data.


ESs can be developed to capture domain expertise and performing as skilful and cost-effective experts to generate advice or instructions (McDonald, 1989). ESs are being deployed worldwide in myriad applications because of their capabilities in symbolic reasoning and explanation facilities (Metaxiotis & Psarras, 2003). The basic principle underlying ESs is to embody the knowledge from human experts in a specific domain and make that knowledge available in a computer program and produce recommendations for solving problems in that domain (Wierenga & van Bruggen, 1997). ESs thus may have an important role to play in storing expertise and advising alternatives for international marketing planning.
Fuzzy logic (Zadeh, 1988) is another technique which is proposed to handle imprecise fuzzy terms or linguistic concepts using membership functions. It aims at modelling the imprecise modes of reasoning that play an essential role in the remarkable human ability to make rational decisions in an environment of uncertainty and imprecision (Zadeh, 1988). It therefore may be useful in dealing with the ambiguity and fuzziness relating to the factors and options for marketing in the global contexts.
Simulation technology, as a popular decision support tool, may also contribute to global marketing decisions. The Monte Carlo simulation technique can be applied to model and explore stochastic permutations of uncertainties (Rezaie, Amalnik, Gereie, Ostadi, & Shakhseniaee, 2007) in the international market. Using Monte Carlo simulation we try to obtain an unbiased and consistent point estimator (Usable, 1998) for relevant factors influencing the decision making problem.
The development and advances of artificial intelligent software agents may bring opportunities for creating more effective support for international marketing planning. Intelligent agents are software programs that can pro-actively perform certain tasks on the users’ behalf autonomously or with little guidance (Bui & Lee, 1999). They may deliver good value to the international marketing decision making process through coordinating communications amongst the user and various agents, and linking diverse supporting techniques and technologies together.
The conceptual framework that combines the decision making process, human judgement and supporting techniques and technologies is given in Table 1.
3. AgentsInternational: a multi-agent hybrid intelligent system for international marketing decision making
On the basis of previous definitions for hybrid intelligent systems given by Goonatilake and Khebbal (1995) and Li and Li (2009a), we describe a multi-agent hybrid intelligent system as a multiple agents-based software system that is composed of different interacting functional elements and unites the advantages of various techniques and/or technologies including artificial intelligence technologies, for the purpose of improving the effectiveness and efficiency of decision making or problem-solving (Li & Li, 2009b).
Agent coordination issues for decision support applications have also been addressed by Chi and Turban (1995), Bui and Lee (1999) and Sokolova and Fernandez-Caballero (2009). We propose
a multi-agent coordination and integration method, entitled as intercommunication job-sharing hybridisation. With this approach, the overall international marketing decision making problem is divided into distinct subtasks or jobs. These subtasks are then assigned to relatively independent software agents that share and exchange information and knowledge; carry out different subtasks to generate alternatives or options; and link and blend human decision-makers’ inputs, judgement and intuition for joint solutions.
The software architecture of the multi-agent-based hybrid system for international marketing decision making is shown in Fig. 1.
Within the architecture, the roles of the chief coordination agent include managing interaction between the user and connected intelligent agents. It asks the user to make choices, provide answers and judgemental inputs to various questions, factors or criteria that affect international marketing decision making. It then takes and aggregates inputs from users, executes reasoning and generates intelligent advice on go vs. no go decisions, entry modes, and marketing strategies.

Go vs. no go agent makes ‘‘go” or ‘‘no go” recommendations by helping review the factors or criteria in different categories and at different hierarchies. For strategic intention, entry pressure (growth of global market, competitive global entry, and long-term corporate commitment) and resource availability (financial resources, and capacity utilization) are considered. For market opportunity, expected sales potential (GDP (Gross Domestic Product) level, GDP growth rate, competition) and expected profit potential (production cost advantage, and marketing cost advantage) are included. For payback risk, non-economic risk (political risk, and social risk) and economic risk (foreign exchange rate, and trade balance) are covered. For synergy effects, product synergy (R&D/engineering synergy, manufacturing synergy, and logistics synergy) and global business synergy (management experience, and marketing experience) are taken into account (Jain, 1990; Levy & Yoon, 1995; Root, 1987).
While human judgemental inputs to the above factors are acceptable, a Monte Carlo simulation element has been developed using Java programming language and is managed by the simulation agent to represent and capture the market changes and uncertainties.
The triangular probability distribution is used to model pessimistic, most likely and optimistic scores for the factors influencing international marketing decisions. Users are required to provide data entries for these variables. The computer is commanded to perform ten thousand times of simulations using different random numbers to ensure the independence of each simulation. The averaged results of these simulations are then passed to WIN Prolog go vs. no go agent. A screen copy of the multi-agent system with the chief coordination agent, simulation element and the go vs. no go agent are shown in Fig. 2. Please note that several agents of the AgentsInternational system have been arranged together in one single screen for demonstration. Two copies of the prototype system are launched on the same computer, with agent communications made between them.
The entry mode selection agent is created with a set of rules and a knowledge base storing documented domain expertise from Doole and Lowe (2004) and Chee and Harris (1998). It advises modes of entry in line with the levels of risk and control: Indirect exporting, direct exporting, foreign manufacturing, assembly operations, contract manufacturing, licensing operations, joint ventures, strategic alliances, wholly owned subsidiary (Chee & Harris, 1998; Doole & Lowe, 2004). A snapshot for choosing entry modes is given in Fig. 3.


The knowledge base elements embodies ‘‘if . . . then . . .” rules and fuzzy rules with certain degrees of confidence, which command the use of expertise to go vs. no go decisions, entry mode selection, guidelines generation and marketing strategy formulation.

All the fuzzy rules are based up trapezoidal membership functions (Levy & Yoon, 1995; Li, 2000; Li & Li, 2009a) that help work out the level of certainty or confidence for related options or alter-natives. Fuzzy logic has been implemented in the go vs. no go agent, McDonald marketing strategy agent, and Watson and Zinkhan Internet strategy agent.
The intelligent agents for strategy formulation represent the following analytical models and guidelines: McDonald (1996)’s four-box directional policy matrix for setting marketing strategies, Harrell and Kiefer (1993)’s work on global strategy formulation, Watson and Zinkhan (1997)’s four-cell matrix for developing Internet strategies, Ho and Choi (1997)’s framework for formulating winning strategies through Sun Tze’s Art of War guidelines, Martin and Larsen (1999)’s key success factors for trade with China, Porter (1980)’s generic competitive strategy model and Pearlson and Saunders (2004)’s guidelines for IT/IS strategies. A sample output of the related strategy formulation agents is provided in Fig. 4.


The whole AgentsInternational system is composed of about 1000 lines of Java programs and around 5000 lines of WIN Prolog and Chimera agent code.
The AgentsInternational system is designed to assist the international marketing decision making process. It is not intended to replace human decision maker who should be in control and act as the core. The decision maker provides judgemental entries to decision making criteria or factors by scoring their values and assigning weights to them. The user may use an external analytic hierarchy process (AHP) software tool to make pair-wise comparisons for deciding the relative importance of relevant variables influencing the generation of various alternatives (Li & Li, 2009a). The multiagent system then executes calculations, aggregates human inputs for the analytical models, performs intelligent reasoning and produces options or recommendations. Finally, the user makes a final choice or decision on the basis of the system’s advice and human intuition, experience, creativity and personal judgement.
It is necessary to mention that the advantages of intelligent agent technology are not fully utilised in the AgentsInternational system. Although intelligent agents are created to post messages, represent various analytical models, store relevant expertise, perform intelligent reasoning and manage different stages of international marketing decision making, other agent features such as automatic search for information, autonomy, interaction among group decision making members are not yet implemented. This is the main limitation of the research prototype software system.
4. Empirical evaluation and findings
The aim of empirical evaluation is to assess the multi-agent hybrid system’s effectiveness on improving both the process and outcomes of international marketing decision making.
The effectiveness is appraised in terms of the performance of the decision activity (Keen & Scott Morton, 1978), helping understand relevant factors affecting decision making (Li, 2000), providing domain expertise and analytical models (Li, 2005; McDonald, 1989; McDonald & Wilson, 1990), dealing with uncertainty (Levy & Yoon, 1995; Li et al., 2000), helping supplement or complement human judgment (Li, 2005, 2007), helping strategic analysis (Li, 2000), helping couple strategic analysis with human judgement and intuition (Mintzberg, 1994a, 1994b), helping generate relevant options or alternatives (Gao, Wang, Xu, & Wang, 2007), helping build up decision confidence (Gao et al., 2007; van Bruggen, Smidts, & Wierenga, 1996), helping improve the quality of decision making (Li, 2005; Li et al., 2002) and user satisfaction (Turban et al., 2005).
The multi-agent system was tested by carrying out evaluation work in October and November of 2008 with eight managers in London including one company marketing director, one company managing director, one company analyst, and five university course leaders/managers. The participants were asked to apply the AgentsInternational system to make international marketing decisions such as go vs. no go decisions, entry mode selection, and marketing and competitive strategy formulation in the global contexts for their products, services or university courses, using their own inputs for their own cases. They then answered an evaluation questionnaire that contained both close-ended questions and open-ended questions.
Responses to close-ended questions are summarised in Table 2.


We can see from the Table that managers’ responses are positive and supportive. The AgentsInternational system is judged as moderately, very or extremely effective in helping understand relevant factors, providing domain expertise, dealing with uncertainty, generating relevant options or alternatives, coupling analytical models with human judgement, complementing managerial intuition, and improving the quality of decision making, user confidence and satisfaction.
Managers’ responses to the open-ended questions are summarised below.
Support for the process of international marketing decision making is described as:
‘‘Provides many theoretical models.”
‘‘Includes factors that I would not normally have considered.”
‘‘Simulation is helpful.”
‘‘Good level of support.”
‘‘Reinforce my confidence and consideration in decision making.”
‘‘Provides focus for thinking.”
The output, advice or recommendations produced by the system are perceived as:
‘‘The go/no go component is appropriate for decisions with many factors and different alternatives.”
‘‘Quite comprehensive. Help me cover areas I may not have thought of.”
‘‘Provides useful ideas/suggestions for consideration.”
Improvements on the outcomes of international marketing decision making are reported as:
‘‘Compared with the Web-based system, this one provides more confidence in the recommendations given as it uses more advanced technologies.”
‘‘Helps confirm outcomes rather than improve them at this stage.”
‘‘Highlights some of the marketing shortcomings in the institution.”
‘‘Expands thinking of areas and issues to be considered.”
Finally, the participants would like the following changes or modifications to be made on the AgentsInternational system:
‘‘More specific to my line of business, so it would be good to have it more tailor-made.”
‘‘More customising.”
‘‘Perhaps a screen with a brief explanation of the models.”
‘‘Look & feel, readability, colour scheme.”
5. Conclusions and future research directions
The aim of this study has been to explore the use of multi-agent hybrid system in support of international marketing decision making. We have in the paper presented the conceptual framework, software architecture and evaluation findings.
One main benefit of the multi-agent hybrid approach is to combine the strengths of intelligent software agents, knowledge bases, Monte Carlo simulation and fuzzy logic. Another advantage is to couple human judgement with analytical models. The third salient feature lies in the system’s functionality in assisting the three key stages of international marketing planning. This is the first research effort to develop such multi-agent integrated system for international marketing planning and test its value.
Of the eight managers involved in the evaluation studies, most of them assess the AgentsInternational system as effective with regard to the following measures:
 Help understand the factors that affect decision making.
 Help decision making by providing relevant knowledge, analytical models and guidelines.
 Help strategic analysis.
 Help couple analysis with judgment, intuition and creativity.
 Help supplement judgment and intuition.
 Help deal with uncertainty in the process of decision making.
 Help generate relevant alternatives or options in the process of decision making.
 Help improve the quality of decision making.
 Improve performance of the decision activity.
 Confidence about the advice or recommendations generated by the system.
 User satisfaction about the system and its advice or recommendations.
Support for the decision making process is described as ‘‘reinforce confidence and consideration in decision making”, and ‘‘provides focus for thinking”. The output or advice produced by the system is perceived as ‘‘clear advice with additional confidence levels”, ‘‘help me cover areas I may not have thought of”, and ‘‘provides useful ideas/suggestions”. Improvements on the outcomes of decision making are reported as ‘‘listing different alternatives in a ‘one stop shop’ fashion”, and ‘‘expands thinking of areas and issues to be considered”.

This paper lays a solid foundation for further development and evaluation research in this direction. Future work on the AgentsInternational system will be conducted to develop and implement more features such as more customising and helps for the users, autonomy, automatic search for relevant information, and agentassisted interaction and communications for group decision making.
详细解读将在2010年4月19日晚上19：00~21：30.