Bayesian belief network for assessing impact of factors on army’s lean–agile replenishment system

Lean is a concept that works by reducing the muda or waste. Ohno (1988) has shown that lean practices can lead to cost savings by elimination of waste. Lean thinking, as introduced by Womack and Jones (1996), has taken the concept of waste elimination from merely manufacturing processes to the whole of business practice. A Lean Supply Chain strategy is one that is aimed at creating a cost-efficient supply chain, with a focus on reducing inventory lead times and waste (Wang et al. 2004). The strategy works well with stable and predictable demand (Fisher 1997; Qi et al. 2009).

The origins of agility as a business concept lie in flexible manufacturing systems. The concept of manufacturing flexibility was extended into the wider business context (Nagel and Dove 1991) and the concept of agility as an organizational orientation was born (Aitken et al. 2002). Just as the lean-thinking concept emerged out of lean manufacturing, agility has been used as an organizational orientation and not as only a manufacturing strategy (Nagel and Dove 1991). Christopher (2000) defines agility as “a business-wide capability that embraces organisational structures, information systems, logistics processes and, in particular, mindsets”.

Supply chain strategies can be classified as those that emphasize cost reduction (lean), quick response (agile) or a mix of both. Shewchuk (1998) suggested that one single strategy, be it lean or agile, cannot be equally suitable to all types of supply chains. Fisher (1997) was the first to segregate products based on their characteristics and proposed different supply chain strategies for each type. It was only a matter of time when the industry realized the potential of combining the two strategies. Naylor et al. (1999) demonstrated that the two strategies, lean and agile, are not mutually exclusive. There are several examples that show the need to develop hybrid strategies (Christopher and Towill 2000). It was confirmed by Krishnamurthy and Yauch (2007) that lean and agile can coexist and they illustrated it by using data from a company in the USA. Agarwal et al. (2006) used an analytic network process to integrate the various criteria of decision-making and concluded that leagile is a better supply chain management (SCM) strategy than lean or agile. Aitken et al. (2005) identified seven discrete pipelines (or seven different supply chains) and illustrated them by using the case study of a lighting company. The separation of the supply chains, however, was based on clustering products into different kinds and thereby taking decisions to manage the supply chain based on the family of products that it catered to. A similar methodology that marries lean and agile has been illustrated by Towill and Christopher (2005), wherein they have segregated the activities in time and in geographical space. The methodology has been explained in the context of the healthcare industry in the UK.

Peltz et al. (2008), in their study, proposed a methodology for achieving leanness by designing wartime distribution networks that exploit the strengths of airlift and surface transportation modes to meet combatant command requirements at the lowest possible total cost. Girardini et al. (2004) introduced a method for determining stock levels at forward locations during wartime to ensure a more agile logistics system. Breunig et al. (2006) highlighted that the military must not only be agile, flexible, robust and effective, but also lean and efficient. In order to optimize military logistics, the requirements of the military must be balanced with the budget. Yoho et al. (2013) underlined the requirement for further research that explores the tension between massing too much and becoming too lean, as well as understanding how resilience may be achieved at the lowest possible economic cost.

The BBN has been used to model various logistics problems and derive meaningful decisions. A summary of the literature is presented in Table 1.

There has been a wide variety of applications of BBNs in the military field. The summary of the literature on this topic is given in Table 2.

Most of the research into applications of BNs in the military field is limited to tactical situations. Problems related to weapon allotment against aerial threats (Cutler and Nguyen 2003; Oxenham and Cutler 2006; Johansson and Falkman 2008; Hou et al. 2010; Wang et al. 2012), for counterterrorism (Sun et al. 2005; Goldstein 2006; Hudson et al. 2001; Jha 2009), military decision-making (Laskey et al. 2000; Wright et al. 2002; Xiang et al. 2008) and sensor data collection (Gillies et al. 2010) have been addressed using BBN. Logistics application of BBN involves supply chain disruptions, which can be mapped using BNs, and arriving at the probabilities of these disruptions by considering the contributing variables (Anderson et al. 2004; Li et al. 2006; Li and Gao 2010; Soberanis and Elizabeth 2010; Sutrisnowati et al. 2015). BBN as a concept has been widely used and validated in a large number of applications. However, its application in the military logistics field is not widely researched. This study will bridge this gap in literature, particularly in relation to the use of BBN in military logistics.

In the next section, we describe the theory of BNs, including the concept of mutual information in sensitivity analysis.

The research was conducted with the help of an exhaustive literature review to arrive at the factors that have an influence on the ability of the system to be lean or agile and its capability to make the switch. In Step 2, a panel of four experts deliberated, to determine the causal relations between these factors and to assess their relative impact on the system’s capability. These experts were selected based on their experience in the field of military logistics. Each of these experts had a minimum of 18 years’ experience in this field and was also a postgraduate. This ensured that the expert had knowledge both of the domain as well as of the techniques used in academics. In Step 3, these experts decided on the meaningful states of these factors, with emphasis on maintaining the number of states limited. In Step 4, the various factors or the variables were assigned conditional probabilities. These probabilities were compiled and then sent back to the experts for further iterations. The model was qualitatively validated with additional information from practising managers of military logistics. A total of eight practising managers were selected based on a minimum experience of 10 years in the field. This validation was done by asking relatively simple questions about the differing impacts of each of the factors on the final state of the system. It was ensured that the model is not changed completely but is adapted to the inputs of these practising managers. The methodology is explained in Figure 1.

Figure 1

Chase et al. (2009) in their white paper, have listed certain characteristics of industry leaders. According to the authors, leaders are more likely to use demand analytics and reporting (e.g. simulation, what-if analysis and scenario planning tools). Best-in-class organizations consistently shared many characteristics such as the ability to include causal factors (e.g. weather, natural disasters, competitor actions, and so on) into demand forecasts. Furthermore, in that article, they highlighted that forecasters cannot rely exclusively on historical patterns as a good predictor for the future; hence, there is increased focus on effective, real-time access to consumption data for more accurate demand forecasting and planning. Leading companies use integrated collaborative forecasts with customers. Resounding consensus emerged across all industries that “access to timely consumer data and new product forecasting were their biggest challenges to effective demand management”. Lockamy and McCormack (2004) emphasize the requirement of advanced SCM practices, such as collaborative forecasting and planning with customers and suppliers. Lee et al. (2000) pointed out that by letting the supplier have visibility of point-of-sales data; the harmful effect of demand distortion can be ameliorated. Chen et al. (2000) indicated that providing each stage of the supply chain with complete access to customer demand information can significantly reduce increase in variability of the orders placed by the retailer.

Use of modern technologies such as radio frequency identification (RFId), bar coding and warehouse management systems (WMSs) has eased the problems associated with warehousing, such as inventory inaccuracies, product misplacement and so on (Sahin 2004). Bar codes, sensors and/or RFId are used for the track-and-trace functionality throughout all supply chain processes (supply, manufacturing and distribution) (Heinrich 2005). RFId, in combination with other systems, is becoming the basis for new solutions, contributing to better management of supply chains in terms of cost reduction and improvement of customer service levels (Sahin 2004). Benefits of using RFId include the reduction of labour costs, the simplification of business processes and the reduction of inventory inaccuracies (Rekik et al. 2008). The cause for the out-of-stock issue is the factor related to store shelving and replenishment practices, in which the products ordered are in the store but not on the right shelf. These factors may be related to shelf space allocation, shelf-replenishment frequencies, store personnel capacity, in-store execution errors, and so on (Vuyk 2003). The potential benefits of RFId tagging of individual items is huge because the identity, location and authenticity of these items can be easily monitored, thus resulting in increased efficiency and reduced costs (Lee et al. 2005). Inventory record inaccuracy, namely, the discrepancy between the recorded inventory quantity and the actual inventory quantity physically present on the shelf, is a recurring occurrence of, often, considerable proportions (Thiel et al. 2009).

A good distribution-and-inventory control system leads to an efficient system and satisfied customers. Lateral trans-shipment is one such distribution strategy that has a positive impact on a supply chain. Chiou (2008) highlights this as “One strategy in SCM to have an impact on cost, service level, and quality, commonly practiced in multi-location supply chain systems facing stochastic demand, allows movement of stock between locations at the same echelon level or even across different levels”. Ross (2002) describes enabling visibility to inventory as a real process value that needs to be achieved. Real-time communication and supply chain visibility are indicators of higher maturity (www-scf.usc.edu). Selective inventory control not only streamlines the inventory but also is helpful in reducing it to a significant level (Bhatia 2008). Meredith (1987) points out that local firms offer better service, are innovative, respond quicker and provide customization and variety. Perry and Sohal (2000) also identify supply from local resources as a good quick response practice. Sheffi (2001) summarizes the solutions to the supply chain problems. The author highlights that the problem can be tackled by focusing on known solutions, i.e. (a) improvement in shipment visibility; (b) improved collaboration between trading partners and across enterprises; and (c) better forecasting through risk-pooling methods. Vendor-managed inventory (VMI) is a tool widely used in industry to cut costs and increase efficiency. Evidence has shown that VMIs can improve supply chain performance by decreasing inventory levels and increasing fill rates (Yao et al. 2007). Achabal et al. (2000) state that the VMI system reduces inventory costs for the supplier and the buyer and improves customer service in terms of, e.g. reduced order cycle times and higher fill rates.

Human behaviour and organizational culture greatly influence the direction of an organization. Employee involvement schemes have significantly improved operational performance in many businesses (Hanna et al. 2000). Various authors have highlighted the importance of motivation of workforce, technical competence and multi-skilling (Dench 1997; Hopp and Van Oyen 2004; Thakkar et al. 2009). With reference to the workforce, Herzenberg et al. (1998) have pointed that workforce agility may provide a wide range of benefits, such as quality improvement, better customer service, learning curve acceleration, economy of scope and depth. Training activities not only develop employees and improve their skills and abilities but also enhance their satisfaction with the job and their commitment to the organization (Harel and Tzafrir 1999). In addition, human resource management practices such as development-oriented appraisal and comprehensive training show a significant positive relationship with organizational commitment (Paul and Anantharaman 2004).

A summary of the factors is listed in Table 3, along with the references from where they have been drawn. The table also indicates the relevance of each of the factors to each of the modes of the system.

In Step 2 of the research, the selected factors were presented before the panel of experts. The experts, after deliberations, finalized these factors. A total of 23 factors were selected and finalized by the panel of experts. A network highlighting the causal relationship between these factors was also constructed in the step. In Step 3, each of the factors was given possible meaningful states. Various factors, their states and the definitions are given below in Table 4. The table also highlights the factors that have an effect on other factors.

The BBN is constructed using Netica 5.18. It is a commercial software package by Norsys Software Corporation, which can be used to work with BBNs, Decision Nets and Influence Diagrams. The variables finalized in the previous steps and the causal relations between them are used to draw the network using the software. Various nodes are then connected with the arcs. Conditional probabilities are filled into the tables. The network can then be compiled to give out reports. Figure 2 is a snapshot of the BBN drawn in Netica 5.18.

Figure 2

Jones et al. (2010) proposed a three-axiom-based partial validation method for BNs. First, a change in the prior subjective probabilities of each parent node should result in a relative change in the posterior probabilities of the child nodes. Second, a change in the value of the parent node should have a consistent magnitude effect on the child node; and third, if both x and y have an influence on the child node, the magnitude of influence of x + y should always be greater than the influence of x and y separately (Cai et al. 2013). In our model, e.g. a change in the parent nodes “incentive”, “salary” and “job security” has a corresponding effect on the child node “motivation”. The magnitude of change also is consistent. When the “high” state of parent node “salary” is changed from 33.3% to 43.3%, the high state of child node motivation changes from 49.2% to 54.5%. Similarly, a move of high state of parent node salary down to 23.3% moves the high state of motivation down to 43.8%. Furthermore, the combined action of changing the values of the parent nodes incentive, salary and job security has a larger effect on the child node motivation than the effect produced if they are changed separately. The three axioms were checked on each of the nodes and were proved correct, thereby providing partial validation to the model. Combined quantitative and qualitative validation of the model provides the requisite credibility to use the results of the exercise for future research.

Mutual information is an indicator used to identify the variable that reveals the most information on a target node and, hence, possible minimum and maximum beliefs can be identified (Kjaerulff and Madsen 2008). The variables that have the maximum impact on the target node can then be selected for improvement. This analysis has been conducted using the “sensitivity to findings” tab in the Netica software menu. Figure 3 is a graphical representation of the analysis. It can be seen that “forecasting” (value of mutual information=8.84%) and “inventory management” (value of mutual information=4.7%) have the most impact on a dynamic lean–agile spare part replenishment system. Forecasting is greatly influenced by the presence of a sound “scientific forecasting” system and “collaborative forecasting”. It is also important to point out that the presence of “information and communication technology (ICT)” (value of mutual information=1.02%) has substantial influence on the target node in spite of it being a root node in the fourth tier. This is because it not only affects collaborative forecasting but also “inventory visibility”. Inventory visibility has major influence on inventory management, which in itself has considerable influence on the target node. Other variables that have larger impact on the target node are “training” and “use of VMI”. Training (value of mutual information=2.09%) is influential because of its simultaneous impact on forecasting, inventory management and “human resource management”.

Figure 3

In order to bring about big changes in an organization, it is necessary that key result areas are identified. These key result areas are influenced by some factors and, therefore, it becomes essential that these factors are identified and special focus given to them to achieve positive results. However, organizations often suffer from inadequate or incorrect data. There is a need for a technique through which these essential factors can be segregated. A BBN of the variables that drive a dynamic lean–agile spare part replenishment system in the Army is such a useful learning tool to distinguish more influential factors from the less influential ones in case of incomplete or imperfect data. It provides a quantified method to highlight factors that have maximum impact on the system, thereby providing a road map to focus on the most rewarding factors. The results of this study indicate that Forecasting has a stronger bearing on the system than other factors. Subsidiary factors such as Inventory visibility and ICT also have considerable impact on the system behaviour. These two factors have overcome the disadvantage of hierarchy (they are considerably away from the target node) and have been able to prove their impact on the target node with the help of sensitivity analysis. This is proof that sensitivity analysis can divulge more information than the network structure. It is, however, necessary to first validate the model, both qualitatively and quantitatively. This validation will provide more credence to the model.

The proposed BBN model has many benefits that enable us to apply it to our case discussed in the study. The BBN model works well even in the absence of complete factual information and is particularly helpful in “drawing conclusions” (Dogan and Aydin 2011). Any future knowledge can be updated into the model at a later stage and make the findings even more accurate. To begin with, the prior nodes are given uniform probability of occurring, which gives a certain result that is not very accurate. However, fresh inputs, as and when they become available, drive the result more towards accuracy. During qualitative validation of the model, explicit information from the practising managers was updated to further refine the model, which was till then based on parameterization by the experts. This added focused information into the model and aligned it a little more with the ground reality. The practising managers were, in essence, working on the model provided by the experts, who had laid down logically correct boundary conditions, thereby ensuring that any incorrect information from the practising managers is not able to drastically alter the model.

Modelling the spare part supply chain on a BBN and the ensuing sensitivity analysis of the factors reveals the key factors that affect the replenishment system. The BBN structure gives a rough estimation of the factors that are critical; however, sensitivity analysis using the mutual information criterion reveals the hidden influence of certain factors that may seem irrelevant or too far in hierarchy from the target node. From the study, it emerges that accurate forecasting and real-time communication among all the stakeholders has a relatively higher impact than other factors in producing a scheme that is adaptable to a dynamically changing time-separated lean–agile replenishment system of spare parts in the Army. The concept of VMI is another factor that has a considerable impact on the leanness of the system. The model, however, needs to be further refined through the input of more data as they become available. In other circumstances, there may even be some inclusion/deletion/modification of certain factors/relationships. Future work is required to bring out a tailor-made forecasting methodology for the spare parts of the Army, which accommodates all three situations separately, i.e. peace, training exercise and war. Work is also required to lay out a framework incorporating vendor’s logistics into the military logistics in a seamless manner, while at the same time ensuring security of information.