J Intell Manuf (2012) 23:91–108 DOI 10.1007/s10845-009-0337-z A supply chain performance analysis of a pull inspired supply strategy faced to demand uncertainties G. Marquès · J. Lamothe · C. Thierry · D. Gourc Received: 15 December 2008 / Accepted: 7 October 2009 / Published online: 24 October 2009 © Springer Science+Business Media, LLC 2009 Abstract Vendor Managed Inventory (VMI) is currently seen as a short-term replenishment pull system. Moreover, VMI is usually synonymous with a distribution context and stable demand. However, industrial partners are faced with uncertainty in the context of a B to B relationship. Thus, an adaptation of the actors’ planning processes is needed and the question is posed of the interest of VMI in a context of uncertain demand. The purpose of this paper is firstly to analyze the link between VMI and pull logic. Secondly, we explore the extension of VMI notions to the relationship between industrial partners and we confront VMI with uncertain demand in terms of trend, vision of the trend and variability in order to verify the usual stable demand assumption. We also present an integration of VMI into a simulation tool called LogiRisk that we have developed for the evaluation of risks of in supply chain collaboration policies, and a small case study. Keywords Vendor Managed Inventory · Pull · Simulation · Risk · Uncertainty G. Marquès (B) · J. Lamothe · D. Gourc Université de Toulouse, Mines Albi, Centre Génie Industriel, Campus Jarlard, route de Teillet, 81013 Albi CT Cedex 09, France e-mail: marques@enstimac.fr G. Marquès · C. Thierry Université de Toulouse, IRIT, 5 allées Antonio Machado, 31058 Toulouse, France e-mail: thierry@univ-tlse2.fr J. Lamothe e-mail: lamothe@enstimac.fr D. Gourc e-mail: didier.gourc@enstimac.fr Introduction During the past decade the industrial context has changed. The production economic has trended from a market where consumers buy standard products offered by manufacturers to a market characterized by the personalization and the uncertainty concerning the demand and its forecast. Therefore, using supply chain collaboration more strategically has become crucial. It enables the creation of new revenue opportunities, efficiencies and customer loyalty (Ireland and Crum 2005). Among these supply chain collaborations, Vendor Managed Inventory (VMI) is today used in industry and has inspired a large number of academic works. However, in terms of implementation it clearly appears that VMI is limited to particular situations. For example, VMI is today almost exclusively synonymous with a distribution context. So the focus must be on ways to extend DistributionVMI notions to the relationship between industrial partners. Furthermore, many authors agree with the idea that the VMI has to be set against stable demand. We can project this stability of demand on different planning horizons. In a strategic horizon, a forecast could be represented by a demand trend. It would be qualified as stable if the trend stays the same all over. Conversely, a change in the trend, such as an increase or a decrease, leads to instability. In a shorter horizon, the term stability characterizes real demand. For example, the stability of this real demand could be measured thanks to a standard deviation around a mean: a small standard deviation for stable demand and a large one for an unstable demand. In this paper, we focus on the strategic aspect in order to analyze the impact of this instability on the supply chain performance. We use a discrete events simulation tool called LogiRisk in order to simulate the strategic choices, exchanges and decisions in a given supply chain (Lamothe et al. 2007). This tool 123 92 enables risk evaluation of different collaboration policies. A collaboration policy being the gathering of: • the collaboration protocol that defines decisional processes between the partners. Here, the protocol is defined by two main aspects: the type of forecast (internal based on historical forecasts or external transmitted by the partner) and the type of supply (push, pull or VMI); • and the union of the partners’ decisional behaviours during their decisional activities. Here, we focus on the strategy of inventory security level (expressed in weeks). The purpose of this paper is twofold: on one hand, identifying the pull aspects of the VMI throughout the definition of its objectives and decision levers. On the other hand, we aim to study the impact of a changing market demand trend on these objectives and the decision levers from a risk analysis process point of view. Consequently, a literature review allows VMI objectives and decision levers to be defined. This first part particularly underlines the shared aspects and differences with classic push and pull systems. In a second part, we present the elements of our model. In a third part, we discuss the simulation results of a case study. Finally, we give several conclusions and present future research works. Literature review Many articles deal with VMI. But what VMI actually is and how can it be concretely implemented in the supply chain is not obvious. In this part we aim to emphasize the link between VMI and pull logic, through the analysis of its objectives and decision levers. VMI systems The Supply Chain Council (2008) defines VMI as “a concept for planning and control of inventory, in which the supplier has access to the customer’s inventory data and is responsible for maintaining the inventory level required by the customer. Re-supply is performed by the vendor through regularly scheduled reviews of the on-site inventory”. The traditional VMI implementation success story is the partnership between Wal-Mart and Procter & Gamble. Other sectors have been explored ever since: house hold electrical appliances (De Toni and Zamolo 2005), automobile (Gröning and Holma 2007), grocery (Clark and Hammond 1997; Kaipia et al. 2002; Deakins et al. 2008), others (Tyan and Wee 2002; Henningsson and Lindén 2005; Kauremaa et al. 2007; Claassen et al. 2008; Gronalt and Rauch 2008). These case study papers underline the fact that VMI is more than an operational replenishment system. First, VMI is part of a larger collaboration partnership that includes tactical 123 J Intell Manuf (2012) 23:91–108 and strategic exchanges between partners. Secondly, these exchanges imply information technology changes (Holmström 1998; Achabal et al. 2000; Vigtil 2007; Vigtil and Dreyer 2008). The main objectives of VMI have been widely studied. According to Tang (2006), the customer’s target is to ensure higher consumer service levels with lower inventory costs. The supplier’s target is to reduce production, inventory and transportation costs. Some authors identify common subobjectives which permit the building of a better collaboration between partners, thereby reaching the main objectives. These authors claim that VMI also aims at speeding up the supply chain (Holweg et al. 2005) and so at reducing the bullwhip effect (Disney and Towill 2003; Holweg et al. 2005; Achabal et al. 2000; Cetinkaya and Lee 2000). VMI concepts have been defined (Marques et al. 2008) as follows: • a replenishment pull inspired system; • where the supplier is responsible for the customer’s inventory replenishment; • within a collaborative pre-established medium- or longterm scope. Moreover, VMI introduces information sharing and common decision-making processes. The integration of VMI into partners’ planning and scheduling processes results in a new collaboration protocol. Three levels in this protocol have been highlighted. The partnering agreement specifies the integration of the planning processes of the partners into a “VMI replenishment planning process”. The Logistical agreement fixes the parameters, which regulate the management of each article (minimum maximum inventory level, minimum delivery quantity, transport schedule, etc.) (Gröning and Holma 2007). The Production and dispatch process monitors short-term pull decisions such as production dispatch and transport. Why is VMI pull inspired? The comparison between push and pull systems has been often studied in the literature (Benton and Shin 1998; Spearman and Zazanis 1992; Ho and Chang 2001). Benton and Shin (1998) define three ways to distinguish the nature of push and pull systems: • order release: removing an end item in pull and anticipating future demand in push. In other terms, the question: “What is the triggering event of the process?” is asked; • structure of the information flow: local and decentralized in pull, global and centralized in push. In other terms, the question: “What is the information we have to make the decision?” is asked; J Intell Manuf (2012) 23:91–108 • Work In Progress (WIP) management: open queuing network with infinite queue space in push and closed queuing network in pull. In other terms, the question: “Which control of the WIP?” is asked. The first objective of this paper is then to study in which way VMI inherits pull philosophy. Thus, we first subject VMI to these three questions. An order release based on the demand Lack of demand visibility has been identified as an important challenge for supply chain management, resulting in inefficient capacity utilization, poor product availability and high stock levels for each partner (Smaros et al. 2003). According to this, increasing the demand visibility for production and inventory control was a first step to improving this collaboration between members of the supply chain. In this view, Quick Response (QR) was born in the early 80s in order to reduce the delay needed to serve customers in the textile industry. The supplier receives point of sale data from the customer and uses this information to synchronize production. In the early 90s, Continuous Replenishment Policy (CRP) was developed: based on consumer demand, the CRP pull system, based on real product consumption rate (Ip et al. 2007) replaces historical push systems based on demand forecast. Between traditional supply, QR and CRP, suppliers’ decisional sphere gradually grew until VMI, which transfers responsibility for the totality of the customer’s inventory replenishment decisions to the supplier (Tyan and Wee 2002). VMI inherits this pull logic and ODETTE (2004) clearly underline this link: • a replenishment signal is sent after the product consumption; • delivery quantities and times are predefined based on consumed quantity (supplier reacts); • forecast/planned consumption is not taken into account to make the dispatch decision (but it is taken into account in the min and max calculation). A transfer of information for a transfer of a decision Whatever the type of classic protocol, push or pull, the demand received by the supplier is composed of two main dimensions: • the real requirements (or net requirement) related to the market demand requirements (through the production requirements): gross requirement less inventory level; • the indirect requirements related to the risk management (demand, supply or internal risks): safety stock (in pieces 93 or in days). They are added to the real requirements when the supply decision is made. In push or pull systems, the supplier can not differentiate these two types of requirements but simply has to meet the order. Regarding a customer characterized by a limited risk aversion, the security inventory level could be very large and the supplier could have some difficulties to respect all the orders. This is one of the primary interests of VMI. With VMI, the customer delegates ordering and replenishment planning decisions to the supplier (Tang 2006). As Disney and Towill (2003) argue, moving to VMI alters the fundamental structure of supply chain ordering. If the order release remains the customer’s demand, the principle of VMI rests on a transfer of responsibility for the customer’s inventory replenishment decision. Most authors agree on the interest of transferring the customer’s inventory responsibility from customer to supplier (Dong et al. 2007; Holweg et al. 2005; Kaipia and Tanskanen 2003; Tang 2006; Kuk 2004). Holweg et al. (2005) explain that the supplier has to base replenishment decisions on the same information that the customer previously used to make its purchase decisions. When VMI is implemented, the supplier has a better vision of the customer’s demand (Kaipia and Tanskanen 2003). Thanks to this improved visibility, the supplier is able to smooth the peaks and the valleys in the flow of goods (Kaipia and Tanskanen 2003). In other terms, it could reduce the bullwhip effect. Disney and Towill (2003) have demonstrated that VMI can reduce this effect by 50%, mainly thanks to the visibility of the demand through the in-transit and customer’s inventory levels. Yao and Dresner (2007) show that information sharing reduces the supplier safety stock, thereby reducing the average inventory level. Even if it is one of the main causes of VMI failure (Tyan and Wee 2002), this information sharing is the key aspect of VMI. Being cognizant of a better structure of the demand could have great consequences on dispatch decisions, and therefore on supply performance. These consequences should be explored more deeply. A min/max to control the WIP In pull systems, strategies as kanban or conwip aim at limiting inventory level, respectively, at each stage of a production process or at the whole production line (Gaury et al. 2000). With VMI, the supplier has to maintain the customer’s inventory level within certain pre-specified limits (Tang 2006) based on a minimum and maximum range (ODETTE 2004). These bounds allow the quantity sent to the customer to be limited and controlled. That is why, even if there are no kanban-style stickers, the supplier monitors in-transit and inventory quantities. The supplier must keep sufficient inventory at the customer’s site so that the customer’s service level 123 94 J Intell Manuf (2012) 23:91–108 Table 1 Push/pull VMI inspiration comparison Order release Push Forecast (future demand) Pull Structure of the information flow Global middle long term customer information Real product consumption rate (removing an end item) Local short term customer information VMI Real product consumption rate (end item stock level) Work in progress Infinite queuing Fixed quantity Local short middle term customer and supplier information Limited interval is unchanged (Yao and Dresner 2007). ODETTE (2004) emphasize the fact that min/max inventory levels have to be agreed mutually by the partners. They give an example of this calculation: assumptions about VMI, we have implemented this VMI process inside a simulation tool. This section is dedicated to the global approach and a description of the simulation models. A simulation and risk oriented approach • Average Planned Daily Usage (ADU) = (Forecast total/ (actual number of weeks with > zero planned usage))/5 • Min calculation = Days of safety stock * ADU • Max calculation = Min + (5/Weekly ship freq.*ADU) + (Transit days * ADU) In addition to the decrease of inventory levels, this minimum and maximum quantity of components constraint in the customer’s inventory implies more small quantities and higher delivery frequencies. Implementing VMI leads to higher replenishment frequencies with smaller replenishment quantities (Yao et al. 2007; Dong et al. 2007) and so to greater inventory cost savings (Cetinkaya and Lee 2000). With VMI, the supplier obtains a new degree of liberty. It has the liberty of making decisions on quantity and timing of replenishment (Rusdiansyah and Tsao 2005). In this study, we seek to help managers with strategic decision-making in order to define a collaboration strategy. This collaboration strategy is built up from both a specified collaboration protocol (or process) and the different partners’ local planning behaviors. According to this idea, we propose a simulation approach that helps the decision-maker to fix his/her choice on a collaboration strategy enabling the evaluation of the risks of different protocols and behaviors. After defining the structure of the supply chain, we identify possible decisions and events that can impact the performance of the chain. We distinguish three types of element: demand market scenario, collaboration protocols and actors’ local planning behaviors. This defines an experimental plan that is processed using a simulation. Each experiment in the plan is processed by a simulation tool and defined by several parameters. Two types of parameters are distinguished: Synthesis Table 1, below summarizes points underlined in the three last parts. The three columns represent the three previously identified questions: What is the triggering event of the process? What is the information we have to make the decision? Which control of the WIP? VMI clearly appears as a pull inspired strategy. The main difference of this supply strategy when compared to a pull strategy is the transfer of replenishment decision responsibility that modifies the structure of the information flow. We can add the fact that with VMI removing an end item does not imply obligatory an order as with pull. A degree of freedom is given to the new decision maker: the supplier. Consequently, even if the WIP is not fixed by a real quantity as in pull, it is controlled through a limited interval defined by the partners. • Structural parameters: These parameters are shared by all the experiments. They define the structure of the supply chain under study and its different products. Furthermore, they could define elements of the demand market. • Simulation parameters: Each simulation parameter has a given set of values for each experiment. They are defined according to the questions the manager formulates. We differentiate parameters that are decisions of actors or group of actors and the events that occur during the simulation. The global performance of the supply chain and performance of each actor are evaluated. Managers base their decision-making on these evaluations. When all the experiments are performed, the target is to analyze the impact of each simulation parameter on the performance. Simulation model and approach A simulation tool: LogiRisk We have established a direct relation between pull and VMI and extracted a VMI process. In order to test some common Our approach is based on an extension of a simulation tool dedicated to risk evaluation of supply chain planning 123 J Intell Manuf (2012) 23:91–108 95 Fig. 1 The generic LogiRisk representation of the supply chain actor’s planning processes processes. In this part, we first give a general description of the macro processes of the tool that have been the subject of detailed presentations in previous articles (Lamothe et al. 2007; Mahmoudi 2006). Then, as VMI processes have been implemented in the existing models, we study the impact of VMI implementation on these in greater detail. Actor’s planning processes Lamothe et al. (2007) propose a simulation tool called LogiRisk developed in Perl language. This simulator is based on a discrete event simulation modeling approach. They have established a generic representation of the different planning processes (SOP, MTP, STP and L&IM) for each supply chain actor. These four planning levels could be seen according to two points of view: internal (production SOP, production MTP, …) that expresses one’s own production decisions, and external that expresses the material requirement sent to the supplier (supply MTP, supply STP, …) or the delivery decisions (dispatch L&IM). The Fig. 1, below summarizes these planning processes. Dotted lines separating the different horizons illustrate the aggregation/desegregation transformations which are currently made in LogiRisk. The actor’s model is centred on the strategic (SOP) and, to a lesser extent, the tactical processes (MTP). LogiRisk does not simulate the short-term but only makes a weekly flow assessment in order to know, for each week, what the actors wanted to produce (STP) and what they actually produced (L&IM). In the Table 2 below, we have cited the main models that define each process (columns 1 and 2). Then, for each model, we particularly underline parameters associated to the actors’ behavior. Finally, we give main equations that take into account these parameters. For further explanations, we refer to Lamothe et al. (2007). In column 1, we see that SOP and MTP use the same models. In fact, if the models are the same, the input data taken into account are different (granularity, originated process). The Sales and Operations Planning (SOP) processes detail the various decisions taken throughout long-term planning. The most important outputs of these processes are the production capacities (production SOP) and long-term forecast of supply requirement (supply SOP) (see Fig. 1). This model includes the products sale forecasting model. If no demand forecast is transmitted, the production SOP process internally computes its forecasts using simple, double, triple or Holt and Winters Smoothing algorithm (Eq. (1) Table 2). In other cases, it sums up the forecasts transmitted by customers (Eq. (1’)). According to the demand forecasts, the workload is computed and smoothed over several time periods (Eq. (2)) in the infinite capacity net requirement model. The resulting workload defines a capacity plan that must be validated by the SOP manager (Eq. (3)). This latter has a specific planning behavior: s/he compares the proposed capacity plan to the one validated in the previous SOP process, and accepts a given percentage of capacity variation. From this capacity plan, a planned production is calculated that allows a long term raw material procurement plan to be computed (Eqs. (4)–(8)). The Medium-Term Planning (MTP) processes compute the estimated production release of final products, as well as the required raw materials to order from the suppliers, or inventory levels (Eqs. (2), (4)–(8)) in function of the actor’s behavior in term of production type (push or pull). As in the SOP processes, the demand forecasts are either updated internally or aggregated from the demand forecast information received from the customers (Eqs. (1) or (1′ )). 123 96 J Intell Manuf (2012) 23:91–108 Table 2 Details of actor’s planning processes (inspired from Lamothe et al. 2007) Processes P. SOP,P. MTP Process models Models Actors’ behaviors parameters Main equations Products sale forecasting Internal forecasting (and type of forecasting: F) Fi,t = F (Historic demand f or i) p (1) with a function F: Holt and Winters algorithm, simple, double or triple smoothing  p Fi,t = For ecasts transmitted f or i (1′ ) External forecasting p p p P. SOP P. MTP Infinite capacity net requirement Products Safety Inventory level (SSi,t ) NRi,t = Fi,t+li − R Pi,t+li − Ii,t+li −1 + SSi,t+li P. SOP Production capacities plan defining Capacity variation acceptation (δ), Algorithm smooth_1 CAPAt = δ × smooth_1 P. SOP, P. MTP Production and products inventory levels planning Production smoothing algorithm Smooth_2 X i,t S. SOP, S. MTP Supply requirement and component inventory level planning Desired production computing Component Safety Stock (SS j,t ) P. STP Push production Pull production Admissible production computing p p Ii,t  SOP N Ri,t i  Procurement order computing P. L&IM Effective production launching Planned receipt planning D. L&IM Deliveries computing p p p = Ii,t−1 − Fi,t + R Pi,t + X i,t−li p p G R j,t = X i,t × B O Mi, j p p p Yi,t = G R j,t+l j − R P j,t+l j − I j,t+l j −1 p I j,t p p = I j,t−1 − G R j,t MT P X Di,t = X i,t 123 (3) (4) (5) (6) (7) + SS j,t (8) + R P j,t (9) MT P − I X Di,t = Ii,t i,t + Di,t  XD ST P X i,t = min X Di,t ;  X i,t D i i,t (9′ ) × C A P Ai,t × βi,t  (10) (11) MT P Y j,t if protocol is push (12) D j,t = P + I MT P − I G R ST j,t if protocol is pull j,t j,t   L&I M = min X ST P ; DL j,t−l j +I j,t (13) X i,t i,t B O Mi, j L&I M RPi,t+li = X i,t tot = DLi,t C = DLi,t The Short-Term Planning (STP) and the Launch & Inventory Management (L&IM) processes both detail the various short-term decisions. The Short-Term Planning process takes into account the calculation of the desired production release (desired production computing model), the actor’s own constraints (i.e. breakdowns in admissible production computing model) and the demand sent to the suppliers (procurement order computing model) (Eqs. (9)–(12)). The Launch & Inventory Management process is responsible for taking into account the other actors’ constraints (i.e. insufficient delivery, etc.) and the products inventories update. It deduces the real production release and finally the quantities to be dispatched to each customer (Eqs. (13)–(16)). + (1 − δ) × pr eviousC A P At )    p i = smooth_2 N Ri,t , C A P At P = X ST P × B O M G R ST i, j j,t i,t S. STP (2) min  − Di,t +Ii,t (14) ; R Pi,t +I i,t  − Di,t +Ii,t   −,C C tot DL i,t × Di,t + Ii,t (15) (16) Hypothesis applied to express the models in Table 2: – Each actor manages a single resource: the bottleneck. Production lot sizes equal to 1 – Products are considered as families as seen from the SOP process point of view. Each item of each actor is composed of one unique component. – For a given process, all the actors use the same horizon, granularity, and replanning period. When disaggregating plans, quantities are equitably distributed over the time buckets of each planning period. Notations used in the Table 2: – i: product i. – j: component j (component of product i). J Intell Manuf (2012) 23:91–108 – C: customer of the actor – li : production lead time of the product i (resp. component j). – Fi : Sales Forecasted of the product i (resp. component j). p – N Ri,t : infinite capacity net requirement of product i at t, by the Process p (∈[SOP;MTP]).   p i – N Ri,t : set for all products i of associated net requirement. – C A P At : Capacity decided for period t. p – X i,t : Planned Production of product i (resp. component j), for period t, by the Process p (∈[SOP;MTP;STP; L&IM]). – R Pi,t : Planned Receipt of product i (resp. component j), for period t. p – Ii,t : Inventory level of product i (resp. component j) planned for the end of period t, by the Process p (∈[SOP;MTP;STP;L&IM]). – SSi,t : Safety Stock expressed in days of stock of product i (resp. component j) for period t. p – G Ri,t : Gross Requirement of product i (resp. component j) for period t, by the Process p (∈[SOP;MTP;STP; L&IM]). – B O Mi, j : Bill of Material link between the product i and its component j p – Yi,t : Planned supply requirement of product i (resp. component j) for period t, by the Process p (∈[SOP;MTP;STP; L&IM]). p – I i,t : Actual inventory position of product i (resp. component j). – Di,t : Total orders of product i (resp. component j) received by the actor for the time t. C : Total orders of product i (resp. component j) received – Di,t by the actor from the customer C for the time t. – βi,t : Availability rate of the capacity affected to the product i (resp. component j) at time t. Capacity less breakdowns. tot : Total deliveries of product i (resp. component j) – DL i,t at time t decided by the actor. C : Total deliveries of product i (resp. component j) – DL i,t at time t decided by the actor for customer C. − : Total of inventory shortage of product i (resp. com– Ii,t ponent j) at time t. −,C : Inventory shortage of product i (resp. component – Ii,t j) at time t toward customer C. Specific notations for VMI processes (part 3.2.2.) 97 real , D min , D max : Total real/min/max requirement seen – Di,t i,t i,t by the supplier. real,C min,C max,C : Customer’s C real/ min/max , Di,t , Di,t – Di,t requirement received by the supplier. Collaboration processes In this part we describe the collaboration process models that are simulated by the tool. In this study three different collaboration protocols are implemented: • Push: modelled by a medium-term component orders. • Pull: inspired from kanban method: short-term orders with kanban quantity revision associated to STP processing. • VMI: modelled with a medium-/ long-term agreement (LA) and a supply STP decision transfer from the customer to the supplier. The Figs. 2–4, below, illustrate the different collaboration processes considered in this study. Simulation processing order is as defined by the numbers (1 to 16 or 17). In the next part we detail the models of processes that are impacted by VMI implementation (shown in red in the Fig. 4). VMI impact on the strategic horizon: the min/max calculation On the strategic horizon, partners have to collaborate in order to fix the customer’s minimum and maximum inventory level in the LA. However, in reality VMI implementation is most of time originated by a powerful customer. In this case, there is no effective negotiation. A true negotiated LA is not realized. The min/max inventory levels only include customers’ constraints. The supplier has to choose a strategy between the min and the max in order to fix targeted inventory for his own production SOP and MTP processes. The model integrates this vision. In the model, for a time t, the min/max calculation is only based on the customer’s long-term forecasted gross requirement for components (j), expressed as G R SOP j,t . This G R SOP results from the customer’s production SOP j,t planning process. We introduce two parameters: cover_ min j and cover_ max j . They are two coefficients applied to the G R SOP j,t in order to obtain two levels of customer’s targeted inventory min/max. t=cover_ min j – VMI_min j,t VMI_ max j,t : targeted inventory min/max fixed by a customer. – α: supplier’s behaviour towards the interval [ min;max]. – α_VMI j,t : Targeted inventory fixed by the supplier for its SOP and MTP processes. G R SOP couv_ min ∈ ℜ+ j,t VMI_ min = j,t t=0 t=cover_ max j (17) G R SOP couv_ max ∈ ℜ+ (18) j,t VMI_ max = j,t j t=0 j 123 98 J Intell Manuf (2012) 23:91–108 Fig. 2 Push collaboration process Fig. 3 Kanban collaboration process Then the supplier has to express his behavior toward this min/max level. In consequence, we introduce a parameter, expressed as α, that translates the supplier’s behavior towards the interval [ min;max] that it receives. This variable defines the planned level of replenishment that the supplier wants to achieve. SS j,t = α_VMI j,t (20) Impact of VMI on the operational horizon: supply and dispatch decisions LogiRisk distinguishes three protocols α_VMI j,t = (1 − α) × VMI_ min + α × VMI_ max (19) j,t j,t This planned level of replenishment is taken into account in the customer’s supply SOP and MTP. It replaces the Safety Stock level in the equations (2): 123 • Push: The production MTP process defines planned production under capacity constraints expressed by the production SOP process. This planned production of item i at MTP . Then, the customer’s supply time t is expressed as X i,t J Intell Manuf (2012) 23:91–108 99 Fig. 4 VMI collaboration process MTP calculates the firm orders of component j at time t, expressed as D j,t . The Bill Of Material link between i and j is expressed as B O Mi, j . In the push protocol, D j,t is a direct expression of supply requirement planned at time t for component j defined by the customer’s supply MTP . MTP and expressed as Y j,t MTP D j,t = Y j,t (12) MTP integrates both real requirements related to the Y j,t market demand requirements and indirect requirements related to the risk management, as referred to in part 2.2.2. • Kanban: In kanban supply, D j,t is built thanks to the planned inventory level of j at time t defined by the customer’s MTP (I MTP j,t ), the customer’s actual inventory level (I j,t ) and the production requirements transmitted STP ). by the customer’s production STP (X i,t STP D j,t = I MTP j,t − I j , t + B O Mi, j × X i,t (12) I MTP represents the indirect requirements related to risk j,t management. The rest of the expression is real requirements related to market demand. • VMI: In VMI supply, the customer’s supply STP is replaced by a supplier’s dispatch STP (transfer of responsibility). Consequently, with VMI, customers’ requirement is not a quantity but an interval in which the supplier can express its new degree of freedom: the delivery quantity. In this case, customers’ requirements comprise three values: STP D real j,t = B O Mi, j × X i,t − I j,t D min j,t = VMI_ min j,t D max j,t = VMI_ max j , t As in kanban, we find the expression of the customers’ real requirements related to market demand (D real j,t ). However, indirect requirements related to risk managebut by the results of ment are not expressed by I MTP j,t min/max calculation (VMI_ min j,t and VMI_ max j,t ). real , D min , In the interval characterized by the triplet (Di,t i,t max Di,t ) the supplier’s dispatch STP process fixes a targeted dispatch level in order to organize the production. Thus, the Eq. (12) is replaced by the following equation: min D j,t = D real j,t + (1 − α) × D j,t + α × D max with α ∈ [0; 1] j,t (12′ ) The output of the suppliers dispatch L&IM process is a delivery quantity of i (a supplier’s item i is a customer’s component j) sent to the customer C at time t, expressed C . The structure of the demand transmitted to the as DL i,t supplier has an impact on this process: • Push/kanban: in the initial dispatch model, the supplier compares its actual end products inventory level and the 123 100 J Intell Manuf (2012) 23:91–108 tot,α tot,min tot,real in function of inventory level (I i,t ) and DL i,t , DL i,t Table 3 Possible values of DL i,t tot,real DL i,t tot,min DL i,t tot,α DL i,t I i,t = 0 real I i,t ≤ Di,t real < I min real Di,t i,t ≤ Di,t + Di,t real + D min < I Di,t i,t ≤ Di,t i,t Di,t < I i,t 0 I i,t real Di,t 1 1 1 1 1 0 0 0 0 real I i,t −Di,t min Di,t real −D min I i,t −Di,t i,t  min max α Di,t −Di,t 0 sum of all the demand from its customers. Two cases are distinguished. If the inventory level is bigger than the total C to each customer C. of demand, the supplier delivers Di,t If the inventory level is smaller, the supplier delivers a proportion of Di,t . This proportion is calculated and distributed to customers as follow: tot = DL i,t   − min Di,t + Ii,t ; R Pi,t + I i,t − Di,t + Ii,t   −,C C tot C DL i,t = DL i,t × Di,t + Ii,t (15) (16) real , D min , • VMI: With VMI, a better vision of demand (Di,t i,t max Di,t ) allows the process to be broken down. The supplier does not try to directly achieve Di,t . First, it tries to real , then D min and finallyD . In consequence, satisfy Di,t i,t i,t we have distinguished 5 cases to adapt the Eq. (15). they are a function of the actual inventory level. For example, if the inventory of i is not sufficient to satisfy the real demand of all the customers, this process gives each customer a part of its real demand. The details of cases are given in the following Table 3: Finally, we adapt the Eq. (16) as follow : C = DL tot,real × D real,C + DL tot,min × D min,C DL i,t j,t j,t j,t j,t   tot,α max,C min,C (16′ ) + DL j,t × α D j,t − D j,t Consequently, even if the contractual min is not always kept, the dispatch VMI increases the performance in terms of customer’s component stock out. 1 • a global performance, for example: a demand market stock out. • a local performance, for example: ◦ component and finished product inventory levels; ◦ quantity of production that customers have not made due to component stock out; ◦ amplitude of production capacity variations; ◦ , etc. LogiRisk allows all these performance to be measured for each period (week) and saved throughout the simulation. VMI simulation and risk evaluation: case study illustration In the present study, we want to test the demand trend stability hypothesis currently associated to VMI. Based on this, we examine two problem statements through the utilization of LogiRisk for a given supply chain: • PS1: Can we justify VMI implementation despite a change in the demand trend? • PS2: Which are the influential parameters of the VMI model implemented? Structural parameters Actors and products The supply chain considered (Fig. 5) comprises one supplier and two customers, where: Performance measurement during the simulation As a complex system the supply chain has to confront potentially conflicting objectives. On the one hand, the supply chain is globally evaluated through the final consumer service level. On the other hand, each partner has to monitor its production using local objectives and constraints. Consequently, two levels of performance could be analyzed: 123 • customer A makes three products (P11, P12 and P12) with three different components (CP1, CP2 and CP3) in a same pulled production unit with the same production time and the same production lead time: 2 weeks; • customer B has the same characteristics as customer A, except the three finished products are named P21, P22 and P23, and use the same components); J Intell Manuf (2012) 23:91–108 101 Fig. 5 Supply Chain structure • the supplier makes three products (CP1,CP2 and CP3) with a same component (C) in a same pushed production unit and a production lead time of 12 weeks. The Fig. 5 synthesizes the general structure of the relationship, production and delivery time and the Bills Of Materials (BOM). Table 4, below, gives the values for the initial inventories. They have been defined so that each actor can produce 5,00,000 units/week. There is no work in process initially. Real demand variability A normal distribution is used to represent uncertainty due to the gap between real and forecasted demand. The mean is the forecasted demand for the period and we introduce a standard deviation of 20%. We use the Mersenne Twister algorithm (Matsumoto and Nishimura 1998) to generate the real demand. This is a pseudorandom number generator that generates series of values from a seed. Each experiment has been replicated with ten seeds. In the following results, the performance values associated to each experiment are a mean of the performance values obtained with each seed. Simulation parameters The two problem statements (justification of VMI despite a change in the demand trend and the VMI parameter choice) involve distinguishing two types of parameter: general simulation parameters and VMI-specific parameters. Each type is associated to a particular problem statement. VMI-specific simulation parameters In order to characterize the VMI, we add three VMI-specific parameters. These VMI parameters illustrate the decision levers emphasized in the literature review and the definition proposed: Customer’s Decision lever Supplier’s Decision lever LA frequency (called S1: 4; 8; 12; 24), expressed in weeks; levels of minimum and maximum customer levels (expressed in weeks) used in min/max calculation: cover_ min (called S2) and cover_max (called S3). VMI coefficient expressed as α in our model (called S4), expressed by a real number inside [0; 1]. General simulation parameters Simulation duration All the experiments of the plan are simulated over 600 weeks. In order to build all historical data for each actor and to obtain a stable state of the supply chain, the first 156 weeks are devoted to an initialization step. All analyses below are based on performance measures made between t = 156 and t = 450. The final 150 weeks are not taken into account in order to prevent time limit effects. According to the approach described in “Simulation model and approach”, we identify decisions and events among the general simulation parameters. Decisions Different replenishment systems between the actors are considered in the simulation. Thus the first general simulation 123 102 J Intell Manuf (2012) 23:91–108 Table 4 Initial inventory levels Actor Product Id Type Initial inventory Supplier CP1 P 2,000,000 Supplier CP2 P 2,000,000 Supplier CP3 P 2,000,000 Customer A CP1 RM 84,000 Customer B CP1 RM 84,000 Customer A CP2 RM 84,000 Customer B CP2 RM 84,000 Customer A CP3 RM 84,000 Customer B CP3 RM 84,000 Customer A P11 P 334,000 Customer B P21 P 334,000 Customer A P12 P 334,000 Customer B P22 P 334,000 Customer A P13 P 334,000 Customer B P23 P 334,000 parameter is the type of supply (supply_type called G1: push; kanban; VMI). In terms of actors’ local planning behaviors, we introduce a general parameter: SS_coef. It allows different safety stock levels to be simulated (expressed in weeks). We differentiate two SS_coef: for the supplier’s finished product inventory, called SS_coef_FP_S (G2: 0, 3; 0, 4) and for the customer’s component inventory, called SS_coef_Cpt_C (G3: 0, 2; 0, 3). Events Whatever the type of replenishment, the demand market trend is always stable during a first period (t = 312). However, at t = 312, we simulate three different trends (demand_trend called G4): increased, stable and decreased. Figure 6, below, shows the demand we have simulated and the different periods we have distinguished for the analysis (T1 = [156; 291], T2 = [292; 364], T3 = [365; 450]). We also take into account the vision of this market change: when do the actors know the market trend has changed and modify their forecasts? In order to translate this potential lag, we introduce a third simulation parameter expressed in weeks: the market_vision_variation (called G5: −20w; 0w; 10w). It is negative if the actors know the variation before its appearance, and positive otherwise. The Table 5, below, summarizes the notations used: Figure 7 summarizes the experimental plan carried out. It generates 2664 simulations. 123 Performance indicators In terms of performance measurement, in this study we adopt the supply chain point of view. The whole analysis is based on two indicators: • demand market stock out (called C1): for each week, we sum the quantity of orders customers have not respected (all customers and products taken into account); • total inventory level in the chain (called C2): for each week, the sum of suppliers’ and customers’ component and finished product inventories (all products and components taken into account). Depending on the dimension of the manipulated figures and the granularity level of our model, we round up all results to the nearest thousand. Results and discussion We have broken down the problem analysis into two main steps. First, we have analysed the influence of VMI parameters faced with the two types of events in addition to real demand variability: the trend (increased, decreased or stable) and the vision of the trend change (−20, 0, 10 weeks). From this analysis, we have identified which VMI parameters were influential, in order to make the comparison to pull and push supply processes. This is the second step of the study in which we answer PS2, i.e. can we J Intell Manuf (2012) 23:91–108 103 Fig. 6 Market demand Table 5 Notations used in the analysis Decisions VMI G1 supply_type S1 LA frequency G2 SS_coef_FP_S S2 cover_ min G3 SS_coef_Cpt_C S3 cover_max G3′ SS_coef_Cpt_C or cover_max S4 coef_VMI Performance measures Events C1 Demand market stock out G4 demand_trend C2 Total Supply chain inventory level G5 market_vision_variation Fig. 7 Simulation parameters justify VMI implementation despite a change in the demand trend? Step 1: VMI parameters influence analysis (PS1) In this step of the analysis, we have analysed the effect of the VMI parameters (S1, S2, S3 and S4) on the performance levels. It represents an experimental plan where the factors are: S1, S2, S3, S4, G4 and G5. In this step, we have chosen to prioritise market satisfaction. In consequence, we first minimize market stock out (C1). In order to analyze the results of this experimental plan, we firstly used Tagushi’s method that allows effects of factors and interactions to be measured. In addition to this measurement, we applied Fisher Snedecor variance test to the model (with a probability of 0.05) which allows the significant effects of an experimental plan to be identified. 123 104 J Intell Manuf (2012) 23:91–108 Table 6 Results of the VMI experimental plan during T2 for G4 = increased G5 S4 0 S3 0, 2 0, 3 0, 5 0, 2 0, 3 1 0, 2 0, 3 Table 7 Results at T1 S2 C1 0, 05 C2 −20 0 10 −20 0 10 12,000 18,000 117,000 401,000 351,000 261,000 0, 1 11,000 16,000 110,000 430,000 378,000 283,000 0, 15 10,000 15,000 103,000 460,000 406,000 306,000 0, 05 12,000 18,000 117,000 401,000 351,000 261,000 0, 1 11,000 16,000 110,000 430,000 378,000 283,000 0, 15 10,000 15,000 103,000 460,000 406,000 306,000 0, 05 11,000 15,000 107,000 445,000 392,000 294,000 0, 1 10,000 15,000 103,000 460,000 406,000 306,000 0, 15 10,000 14,000 100,000 474,000 419,000 318,000 318,000 0, 05 10,000 14,000 100,000 474,000 419,000 0, 1 10,000 13,000 98,000 489,000 433,000 329,000 0, 15 10,000 13,000 95,000 504,000 447,000 341,000 0, 05 10,000 13,000 98,000 489,000 433,000 329,000 0, 1 10,000 13,000 98,000 489,000 433,000 329,000 0, 15 10,000 13,000 98,000 489,000 433,000 329,000 0, 05 9,000 11,000 87,000 547,000 489,000 377,000 G1 G2 G3′ C1 C2 Kanban 0, 3 0, 2 16,000 470,000 0, 3 14,000 519,000 0, 2 14,000 519,000 0, 3 12,000 568,000 0, 2 16,000 476,000 0, 3 14,000 519,000 0, 2 14,000 525,000 0, 3 12,000 568,000 0, 2 19,000 472,000 0, 3 15,000 520,000 0, 2 17,000 520,000 0, 3 13,000 569,000 0, 4 VMI 0, 3 0, 4 Push 0, 3 0, 4 Conclusion 1 the Fisher Snedecor variance test shows that the LA_frequency has no significant effect on the two performance measurements. each trend. All the following conclusions are the same for each time period and each trend. Conclusion 2 the minimum of market stock out is obtained for α = 1(S4 = 1). This result is proved by the current LA model. In the model, the min/max calculation is imposed by the powerful customer. No negotiation is done. Tables 6–10 below, summarizes the results obtained for the different experiments at T1, T2 and T3. We have analysed the results for each time period: T1, T2 and T3 for 123 α (S4) translates the replenishment level chosen by the supplier. When α is equal to 1, the supplier targets are all over maximum. Larger is the customer’s component inventory level; lower is level of market stock out. In consequence, we fix α = 1 in step 2. J Intell Manuf (2012) 23:91–108 105 Table 8 Results at T2 for C1 (market stock out) G5 ×1000 −20 0 10 G4 G4 G4 G1 G2 G3′ Kanban 0, 3 0, 2 18 15 10 17 15 0, 3 15 13 9 15 13 0, 2 15 13 9 15 13 12 0, 3 13 11 8 13 11 10 0, 4 VMI 0, 3 0, 4 Push 0, 3 0, 4 Decreased Stable Increased Decreased Stable Increased Decreased Stable Increased 15 17 15 82 12 14 13 70 14 13 70 13 11 61 0, 2 18 15 10 17 15 15 17 15 82 0, 3 15 13 9 15 13 12 14 13 71 0, 2 15 13 9 15 13 12 14 13 71 0, 3 13 11 8 13 11 10 13 11 61 86 0, 2 22 18 13 21 18 17 21 18 0, 3 17 14 10 17 14 13 16 14 73 0, 2 19 16 11 19 16 14 18 16 74 0, 3 15 13 9 15 13 12 15 13 63 Increased Table 9 Results at T2 for C2 (total chain inventory) G5 ×1000 −20 0 10 G4 G4 G4 G1 G2 G3′ Kanban 0, 3 0, 2 417 437 519 436 0, 3 461 485 578 481 0, 2 461 485 577 481 485 0, 3 506 535 636 527 535 0, 2 417 437 519 437 437 466 490 437 335 0, 3 461 486 578 482 486 522 536 486 384 0, 4 VMI 0, 3 0, 4 Push 0, 3 0, 4 Decreased Stable Increased Decreased Stable Increased Decreased Stable 437 467 489 437 335 485 524 535 485 384 524 535 485 384 582 582 535 436 0, 2 461 486 578 482 486 523 536 486 384 0, 3 506 535 636 528 535 580 583 535 436 0, 2 421 439 520 439 439 469 492 439 338 0, 3 463 486 578 483 486 525 536 486 387 0, 2 465 487 578 484 487 526 538 487 388 0, 3 508 536 636 528 536 582 583 536 438 Conclusion 3 results C1 and C2 show that, when α = 1, the minimum target level has no effect (S2). According to the model and the role of α in the choice made between minimum and maximum, when α = 1 is chosen, any minimum target could be fixed. In consequence, we fix the minimum to 0.1 in step 2. Conclusion 4 the effect of the maximum target (S3) is too significant to be ignored in step 2. It will be a variable of step 2. Step 2: collaboration processes comparison (PS2) In this stage of the analysis we seek to test the demand stability hypothesis. We therefore analysed the experimental plan comprising: G1, G2, G3, G4, G5 and S3. G3 and S3 play the same role in the LogiRisk model: the first when G1 is push or kanban, the second when G1 is VMI. So, in the rest of the analysis we consider a parameter called G3’ that brings them together. 123 106 J Intell Manuf (2012) 23:91–108 Table 10 Results at T3 G4 ×1000 C1 C2 G1 G2 G3′ Decreased Stable Increased Decreased Stable Increased Kanban 0, 3 0, 2 49 25 4 319 411 720 0, 3 43 21 3 349 459 808 0, 2 43 21 3 348 459 808 0, 3 38 17 3 379 507 896 0, 2 49 25 4 319 411 720 0, 3 43 20 3 348 459 808 0, 4 VMI 0, 3 0, 4 Push 0, 3 0, 4 0, 2 43 20 3 348 459 808 0, 3 38 17 3 379 507 896 0, 2 57 29 4 327 413 720 0, 3 47 22 3 353 459 808 0, 2 51 24 3 357 460 808 0, 3 42 19 3 384 508 896 The Fisher–Snedecor variance test shows that all parameters can be taken into account in our analysis, except G4 and G5 at T1 and G5 at T2. The results are summarized is the tables below. Minimum values appear in grey. Conclusion 5 Kanban and VMI are very close from the supply chain point of view (C1 and C2). In this case study, the partnership is dominated by powerful customers. We can find similar contexts in industry where the customer imposes VMI implementation. In this case, the relationship is imbalanced and no negotiation occurs between partners to fix the min/max levels. Furthermore, inspired by the industrial case, we modeled a supplier which does not exploit its degree of freedom—the interval within which it could choose the delivery quantities. This type of supplier checks the customer inventory weekly and always replenishes the inventory to the same level. Here, the level is the maximum defined in the LA. The different results show that a customer confronted with this type of supplier, and which has implemented a kanban-based supply, has no particular interest in switching over to VMI. Conclusion 6 Kanban and VMI are justified despite all the variability we have simulated: real demand variability (20%), change in the demand trend and vision in the change of trend. The different tables show that VMI and kanban allow better performance even if some G2 or G3’ adjustments give similar results for push, kanban and VMI. We can also stress 123 that performance is not disturbed by the change in the demand trend and vision in the change of trend. Conclusion In this study, we aimed at testing the assumption mainly made when implementing a VMI collaboration strategy: demand stability. Faced with a very large amount of literature covering this recent type of supply, we first studied the concept. Through the literature review we have emphasized the closeness of the reasonings underlying VMI and pull. Then, we have implemented a VMI model inside a simulation tool called LogiRisk. The case study illustrates that the similarities between pull and VMI are significant enough to particular implementations to provide similar supply chain performance. Nevertheless, it must be noted that the granularity level of our model does not allow particular operational VMI characteristics to be simulated. For example, the delivery frequency increase reported in the literature has not been tested here. The main target was to confront VMI with different types of variability: real demand variability but also change in the demand trend and vision of the change of trend. Our case study shows that in this particular context, VMI or kanban performance is justified despite demand variability. This result calls into question the widespread assumption of demand stability and suggests that study could be made of VMI in combination with promotional operations and other forms of instability. J Intell Manuf (2012) 23:91–108 However, in order to obtain more general conclusions about VMI we have to explore other research axes. On the one hand, in term of modeling improvement: • model negotiation in the LA. The actors have to organize a shared and common planning which is used to parameterize the customer’s inventory min/max level. This common plan is built around exchanges between the partners. The customer expresses its component requirement plan. The supplier gives a delivery plan. Each actor includes its constraints in its plan. The modelling of this common plan could rest on the collaboration planning proposed by Dudek and Stadtler (2005) based on an exchange process that help to achieve convergence between each actors’ point of view. • model utilization of the supplier’s degree of freedom in terms of delivery quantities. In other terms, authorize a variation of α over time. These improvements could help us to analyze another VMI aspect: backing up of stocks from the customer to the supplier warehouse, as reported by Blatherwick (1998). On the other hand we need to confront VMI with other sources of variability. Thus we also plan to: • simulate different real demand variability; • study the cumulative effects of increase and decrease instead of simple increase or decrease; • analyse the effects of actors’ internal constraints: breakdown, quality level, etc. • analyse the effects of external events as strikes, disasters, etc. References Achabal, D., McIntyre, S., Smith, S., & Kalyanam, K. (2000). A decision support system for Vendor Managed Inventory. Journal of Retailing, 76(4), 430–454. Benton, W., & Shin, H. (1998). Manufacturing planning and control: The evolution of MRP and JIT integration. European Journal of Operational Research, 110(3), 411–440. Blatherwick, A. (1998). Vendor Managed Inventory fashion fad important supply chain strategy. Supply Chain an International Journal, 3(1), 10–11. Cetinkaya, S., & Lee, C. (2000). Stock replenishment and shipment scheduling for Vendor Managed Inventory systems. Management Science, 46(2), 217–232. Claassen, M. J. T., Van Weele, A. J., & Van Raaij, E. M. (2008). Performance outcomes and success factors of Vendor Managed Inventory (VMI). Supply Chain Management: An International Journal, 13(6), 406–414. Clark, T. H., & Hammond, J. H. (1997). Reengineering channel reordering processes to improve total supply-chain performance. Production and Operations Management, 6(3), 248–265. De Toni, A. F., & Zamolo, E. (2005). From a traditional replenishment system to vendor-managed inventory: A case study from the household electrical appliances sector. International Journal of Production Economics, 96(1), 63–79. 107 Deakins, E., Dorling, K., & Scott, J. (2008). Determinants of successful Vendor Managed Inventory practice in oligopoly industries. International Journal of Integrated Supply Management, 4(3/4), 355– 377. Disney, S. M., & Towill, D. R. (2003). The effect of Vendor Managed Inventory (VMI) dynamics on the Bullwhip Effect in supply chains. International Journal of Production Economics, 85(2), 199–215. Dong, Y., Xu, K., & Dresner, M. (2007). Environmental determinants of VMI adoption: An exploratory analysis. Transportation Research Part E, 43, 355–369. Dudek, G., & Stadtler, H. (2005). Negotiation based collaborative planning between supply chains partners. European Journal of Operational Research, 163, 668–687. Gaury, E. G. A., Pierreval, H., & Kleinjnen, J. P. C. (2000). An evolutionary approach to select a pull system among Kanban, Conwip and Hybrid. Journal of Intelligent Manufacturing, 11(2), 157–167. Gronalt, M., & Rauch, P. (2008). Vendor Managed Inventory in wood processing industries—a case study. Silva Fennica, 42(1), 101– 114. Gröning, A., & Holma, H. (2007). Vendor Managed Inventory, preparation for an implementation of a pilot project guidance for an upcoming evaluation at Volvo. Master’s thesis. Lulea University of technology. Henningsson, E., & Lindén, T. (2005). Vendor Managed Inventory: Enlightening benefits and negative effects of VMI for Ikea and its suppliers. Master’s Thesis. Lulea University of Technology. Ho, J. C., & Chang, Y. L. (2001). An integrated MRP and JIT framework. Computers and Industrial Engineering, 41(2), 173–185. Holmström, J. (1998). Business process innovation in the supply chain—a case study of implementing Vendor Managed Inventory. European Journal of Purchasing and Supply Management, 4(2–3), 127–131. Holweg, M., Disney, S., Holmström, J., & Smarös, J. (2005). Supply chain collaboration: Making sense of the strategy continuum. European Management Journal, 23(2), 170–181. Ireland, R.K., & Crum, C. (2005). Supply chain collaboration, how to implement CPFR and other best collaborative practices. J. Ross publishing and APICS (p. 206). Ip, W. H., Huang, M., Yung, K. L., Wang, D., & Wang, X. (2007). CONWIP based control of a lamp assembly production line. Journal of Intelligent Manufacturing, 18(2), 261–271. Kaipia, R., Holmström, J., & Tanskanen, K. (2002). VMI What are you losing if you let your customer place orders?. Production Planning and Control, 13(1), 17–25. Kaipia, R., & Tanskanen, K. (2003). Vendor managed category management—an outsourcing solution in retailing. Journal of Purchasing and Supply Management, 9, 165–175. Kauremaa, J., Småros, J., & Holmström, J. (2007). Empirical evaluation of VMI: Two ways to benefit. In Proceedings of NOFOMA 2007. Kuk, G. (2004). Effectiveness of Vendor Managed Inventory in the electronics industry: Determinants and outcomes. Information and Management, 41, 645–654. Lamothe, J., Mahmoudi, J., & Thierry, C. (2007). Cooperation to reduce risk in a telecom supply chain, special issue managing supply chain risk. Supply Chain Forum: An International Journal, 8(2), 36–53. Mahmoudi, J. (2006). Simulation et gestion des risques en planification distribuée de chaînes logistiques: Application au secteur de l’électronique et des télécommunications. PhD. Ecole Nationale Supérieure de l’Aéronautique et de l’Espace. Marques, G., Lamothe, J., Thierry, C., & Gourc, D. (2008). Vendor Managed Inventory, from concept to processes, for a supply chain collaborative approach. International Conference on Information 123 108 Systems, Logistics and Supply Chain, Madison, United-States, May 2008. Matsumoto, M., & Nishimura, T. (1998). Mersenne twister: A 623dimensionally equidistributed uniform pseudorandom number generator. ACM Transactions on Modeling and Computer Simulation, 8(1), 3–30. ODETTE. (2004). Vendor Managed Inventory (VMI) (version 1.0). Rusdiansyah, A., & Tsao, D. (2005). Coordinating deliveries and inventories for supply chain under Vendor Managed Inventory system. JSME International Journal, 48(2), 85–90. Smaros, J., Lethonen, J. M., Appelqvist, P., & Holmstrom, J. (2003). The impact increasing demand visibility production inventory control efficiency. International Journal of Physical Distribution and Logistics Management, 33(4), 336–354. Spearman, M. L., & Zazanis, M. A. (1992). Push and pull production systems: Issues and comparisons. Operations Research, 40(3), 521–532. Tang, C. (2006). Perspectives in supply chain risk management. International Journal of Production Economics, 103, 451–488. 123 J Intell Manuf (2012) 23:91–108 Tyan, J., & Wee, H. (2002). Vendor Managed Inventory: A survey of the Taiwanese grocery industry. Journal of Purchasing and Supply Management, 9(1), 11–18. Vigtil, A. (2007). Information exchange in Vendor Managed Inventory. International Journal of Physical Distribution and Logistics Management, 37(2), 131–147. Vigtil, A., & Dreyer, H. (2008). Critical aspects of information and communication technology in Vendor Managed Inventory. Lean business systems and beyond, 257 (pp. 443–451). Berlin: IFIP International Federation for Information Processing, Springer. Yao, Y., & Dresner, M. (2007). The inventory value of information sharing, continuous replenishment, and vendor-managed inventory. Transportation Research Part E: Logistics and Transportation Reviews. (in press, corrected proof, available online 9 April 2007). Yao, Y., Evers, P., & Dresner, M. (2007). Supply chain integration in vendor-managed inventory. Decision Support Systems, 43(2), 663–674.