Lrf

Ya

b

c

a

AA

KCOEI

1

bVatbiscbCAifimi

0h

Journal of Operations Management 31 (2013) 285–297

Contents lists available at ScienceDirect

Journal of Operations Management

j o ur na l ho mepage: www.elsev ier .com/ locate / jom

earning curves in collaborative planning, forecasting, andeplenishment (CPFR) information systems: An empirical analysisrom a mobile phone manufacturer

uliang Yaoa, Rajiv Kohlib,∗, Susan A. Sherera, Jerold Cederlundc

College of Business & Economics, Lehigh University, Bethlehem, PA, United StatesMason School of Business, College of William & Mary, Williamsburg, VA, United StatesARRIS, Suwanee, GA, United States

r t i c l e i n f o

rticle history:vailable online 7 August 2013

eywords:PFRrganizational learningmpirical analysisnformation systems value

a b s t r a c t

While Collaborative Planning, Forecasting, and Replenishment (CPFR) information systems have beenincreasingly deployed to improve supply chain operations in a cross section of industries, the extant lit-erature has largely overlooked the learning effects within organizations, thereby resulting in incompleteassessment of their business value. Using an operational-level panel data for nine product lines over 2.5years, we empirically examine the learning curves in CPFR between Motorola, a mobile phone manufac-turer, and one of its U.S.-based national retail partners. We found that the two key components of CPFR,collaborative forecasting (CF) and collaborative replenishment (CR), exhibit distinct learning curves. Fore-

cast accuracy improves immediately following CPFR implementation but the rate of improvement slowsover time, whereas inventory levels increase at first and begin decreasing after a period. Further, wefound different learning effects in terms of inventory levels when products are later replaced with newform factors. Product replacements have lower inventory levels than their antecedents, at least for low-end products. We discuss important implications for theory and practice at the interface of informationsystems and operations management.

. Introduction

Collaborative Planning, Forecasting, and Replenishment (CPFR),ased upon supply chain collaboration standards established by theoluntary Interindustry Commerce Solutions (VICS) Association,re information systems that enable partnering firms to integrateheir inventory planning, forecasting and replenishment processesy sharing information, developing joint forecasts and jointly craft-

ng replenishment plans. Since 1998, when VICS first adopted aet of standards for CPFR information systems, more than 300ompanies have engaged in CPFR practices leading to substantialenefits to suppliers, such as Procter and Gamble and Kimberly-lark and retail chains, such as Wal-Mart and Best Buy (VICS, 2007).lthough conceptually simple, CPFR implementations are complex

n practice as they require exchange of large amounts of data fororecasting a wide range of products. They must account for vary-

ng promotional activities, involve multiple functional areas from

ultiple firms, take an extended period of time to implement, andntegrate possibly incompatible business processes between CPFR

∗ Corresponding author.E-mail address: rajiv.kohli@mason.wm.edu (R. Kohli).

272-6963/$ – see front matter © 2013 Elsevier B.V. All rights reserved.ttp://dx.doi.org/10.1016/j.jom.2013.07.004

© 2013 Elsevier B.V. All rights reserved.

partners (Doiron, 2004). Despite the benefits reported for informa-tion sharing in supply chains (Im and Rai, 2008; Klein and Rai 2009;Patnayakuni et al., 2006), some firms have questioned the benefitsof CPFR and even firms that embrace CPFR often limit the scale ofCFPR implementation, primarily due to the inability to assess itsbenefits (Aviv, 2002).

Although the literature on CPFR has been growing, most previ-ous CPFR studies have been design focused (e.g., Wang et al., 2010;Chen et al., 2009) or analytical (e.g., Fu et al., 2010; Aviv, 2002) andfew studies provide empirical validation of the analytical results,thus limiting our understanding of the payoff from this emerginginformation system (IS). Empirical validation is important in that itprovides deeper understanding of the phenomenon by linking the-ory with real-world cases (Fisher, 2007). In addition, little researchhas examined products in a dynamic business environment char-acterized by constant new product launches coupled with rapidlychanging customer demand where accurate forecasting is criticalto business performance.

Sanders (2008) compared studies of seller–buyer relationships

and concluded that operations management has focused uponthe impact of IT (e.g., CPFR) in specific contexts while the infor-mation systems (IS) discipline has generally focused on actualuse of the IT. She concluded that little attention had been given

2 ons M

tnse2olittmHipai(l

amiW(etw

bamtcsllCvc(rftdpptao

ierieltl(motrfts

CPFR, sharing of inventory and demand data can expose them tounwanted risks, such as leaks of confidential information to com-petitors. Unless firms are convinced of sustained and long-term

86 Y. Yao et al. / Journal of Operati

o the benefits accrued to suppliers from the use of these tech-ologies for their operations. Although a number of papers havetudied IT enabled inter-firm collaboration (e.g., Mukhopadhyayt al., 1995; Mukhopadhyay and Kekre, 2002; Patnayakuni et al.,006), most of them examined the business value or antecedentsf such collaboration. Few papers have approached the issue from aearning theory perspective. For example, Im and Rai (2008) exam-ned learning in knowledge sharing in supply chains and foundhat both exploratory and exploitative knowledge sharing leado relationship performance gains, enabled by the ambidextrous

anagement of the relationship and ontological commitment.ult et al. (2003) had earlier established organizational learn-

ng as a strategic resource and found learning to have had aositive impact on cycle time and overall firm performance. Weddress these gaps by empirically examining the benefits of learn-ng from CPFR for products in a dynamic business environmenti.e., mobile phones) through a theoretical lens of organizationalearning.

The organizational learning framework has been widely useds a theoretical foundation in economics, strategy and operationsanagement. For example, it has been used to explain learn-

ng curves in manufacturing (Argote and Epple, 1990; Smunt andatts, 2003; Li and Rajagopalan, 1997) and learning spillovers

Baum and Ingram, 1998). However, it has seldom been applied tompirically analyze the learning curve in supply chain collabora-ion programs. Doing so will help us understand the course throughhich these programs improve performance.

We examine a CPFR implementation for a number of productsetween Motorola, Inc. and one of its key U.S. retail customers. As

leading manufacturer of mobile devices, Motorola offers variousodels of cell phones that have a life cycle averaging little more

han a year, sometimes as low as six months, making accurate fore-asting a critical yet complex requirement. Using operational-levelupply chain panel data for nine mobile phone products that uti-ized CPFR spanning over 30 months, we evaluate organizationalearning curves emerging from CPFR implementation, as well asPFR learning spillovers between products. Consistent with pre-ious works that modeled two key components of CPFR, namelyollaborative forecasting (CF) and collaborative replenishment (CR)Aviv, 2001, 2002), we examine the learning curves for CF and CRespectively in our analysis. We use two performance metrics –orecast error for CF and inventory level for CR – in examining howhe business value of CPFR is realized through learning that occursuring a product’s life cycle as well as through spillover betweenroducts. The key objective of this research is to examine the CPFRost-implementation learning pattern and to address (i) whetherhe learning curve is non-linear on forecast error and inventory,nd (ii) whether the learning curve varies based on the sequencef product launch.

The two key findings of this study are: (i) forecast error andnventory levels exhibit distinct learning curves over time—forecastrror declines immediately following CPFR implementation but theate of improvement slows over time, whereas inventory levelsncrease at first and begin decreasing; and (ii) products launchedarlier exhibit different learning curves than products launchedater; that is, replacement products have lower inventory levelshan their antecedents, at least for low-end products. Our findingsend support to some of the analytical results in prior literaturee.g., Aviv, 2001, 2002) that CPFR may lead to performance improve-

ent in terms of forecast and inventory management. Additionally,ur findings complement previous research by demonstrating thathe learning curves through which the performance benefits are

ealized are non-linear and, more interestingly, the learning curvesor CF and CR exhibit different patterns. Our estimates show thathe inventory level may increase for a period of time before ittarts decreasing. Failure to recognize such learning curve patterns

anagement 31 (2013) 285–297

may lead organizations to draw premature and flawed conclusionsabout the value of CPFR.

The paper is structured as follows. In Section 2, we review theliterature on previous studies in supply chain collaboration in boththe operations management (OM) and information systems (IS) lit-eratures, as well as research on CPFR and organizational learningtheory. In Section 3, we develop a model and four hypotheses aboutthe learning curve of CPFR. In Section 4, we describe our data andresearch setting, develop an econometric model, and in Section 5we report the results of our analysis. In Section 6, we discuss ourfindings and implications for theory and for practice. In Section 6,we present our contribution, limitations of the study and areas forfuture research. Finally, in Section 8 we present our conclusions.

2. Prior literature and theory

2.1. Collaborative planning, forecasting, and replenishment

Previous research in OM and IS literatures has accumulateda significant body of knowledge pertaining to the benefits ofinformation sharing in supply chains as well as the benefitsof information technology (IT) enabled coordination informationsystems1. Several empirical studies compare performance pre- andpost-information sharing and collaboration. For example, Clark andHammond (1997) utilized grocery products supply chain data tostudy information sharing and continuous replenishment program(CRP) implementations and operations. They reported a 50–100%increase in inventory turns (i.e., number of times inventory issold and replaced) but found that information sharing alone with-out CRP does not significantly improve performance. Cachon andFisher (1997) researched the benefits from information sharing andCRP to the Campbell Soup Company in terms of inventory reduc-tions, and concluded that these benefits were achieved primarilythrough information sharing. Several studies have considered fac-tors that impact benefits from IT enabled collaboration. Klein andRai (2009) found that both trust and buyer IT customization areimportant antecedents to benefits. Im and Rai (2008) found thatknowledge sharing leads to benefits but is enabled by manage-ment and IT design, in particular, contextual ambidexterity, definedas the behavioral capacity of a long-term relationship to allowfor the simultaneous pursuit of alignment and adaptability, andontological commitment, defined as the reliance of partneringfirms on digital boundary objects to span their knowledge bound-aries.

Most of the research on benefits from CPFR has used a model-ing approach. Aviv (2002) modeled how the ability of partners toobserve market signals can improve forecasting performance andconcluded that the success of CPFR implementation depends onthe uniqueness of forecasting capabilities of partners. In a laterwork, Aviv (2007) concluded that the benefits of collaborationvary depending upon the partners’ ability to anticipate demandand suggested that supply chain partners must agree upon a ref-erence demand model that both parties can collectively observe.Raghunathan (1999) modeled the impact of retailers’ choosing toparticipate in CPFR and found a greater decrease in manufacturer’scosts when a second independent noncompeting retailer partici-pates compared to when only one retailer participates, while thenonparticipating retailer’s costs increase. While firms benefit from

benefits from collaboration, the prospect of investments and risks

1 For a review, please see Sahin and Robinson (2002) and Patnayakuni et al. (2006).

ons M

wiiip(

noamtsmtl

2

hcefeHiw

dmbanat1lp1Toi(drp2

oacrraEaiEirmmMnt

Y. Yao et al. / Journal of Operati

ill prevent them from investing in CPFR. A few empirical stud-es have examined the organizational issues surrounding CPFR anddentified factors that help or hinder its success. Most CPFR empir-cal studies use either a case study approach (Småros, 2007) orost-implementation surveys to assess drivers and benefits of CPFRBarratt and Oliveira, 2001; Stank et al., 1999).

The IT enabled supply chain collaboration studies suggest thatumerous factors affect the ability to achieve benefits from collab-ration such as relational asset specificity, long-term orientation,nd relational interaction routines (Patnayakuni et al., 2006). Theodeling studies of CPFR benefits suggest that firms need to learn

o adapt and accept the new collaborative systems. None of thetudies examine the post implementation learning curves, whichay provide insights into the process of how benefits accrue over

ime and therefore shed light on the findings of the impact of col-aboration on inventory.

.2. Organizational learning

Learning is at the heart of organizational success. Recent workas viewed learning in supply chain collaboration as a dynamicapability that is critical to a firm’s ability to create value (Allredt al., 2011). Kudyba (2006) concluded that knowledge capturedrom interaction between partners leads to accurate forecasts andnables firms to adjust to future and sporadic changes in demand.e proposed a segmented approach to managing the supply chain

n which data are gathered and analyzed for each segment, such asarehousing to retailer, for learning.

Organizational learning has been studied in a number ofisciplines including management, economics, operations manage-ent, and information systems, and has served as a theoretical

asis to understand inter-organizational relationships, such aslliance performance (Im and Rai, 2008). The theory of orga-izational learning argues that organizations can create, retain,nd transfer knowledge that they learn from experience and thathis knowledge contributes to performance improvements (Argote,999; Argote et al., 2003). Organizations vary in their capacity for

earning, likely due to factors that are cognitive (Senge, 1990), inter-ersonal (Argyris and Schon, 1978), structural (Duncan and Weiss,979), or managerial (Dutton and Starbuck, 1978; Dutton andhomas, 1984; Tucker et al., 2007). Several mechanisms facilitaterganizational learning, for example, simplification and special-zation (Levinthal and March, 1993) and organizational memoryHuber, 1991). While organizational learning by doing has beenocumented in diverse industries, understanding of how collabo-ative organizational practices in a supply chain affect learning anderformance improvement is still limited (Macher and Mowery,003).

A significant development in organizational learning is thatrganizations face trade-offs in learning between exploitationnd exploration, commonly represented as a U-shaped learningurve (March, 1991). While exploitation requires investment inefinement and efficiency to improve execution of a given set ofoutines, exploration requires investment in discovery, innovation,nd experimentation to address new demands or opportunities.xploitation creates reliable experience and thrives on productivitynd refinement, while exploration creates varied experience, thriv-ng on experimentation and free association (Holmqvist, 2004).xploration requires an additional search for new knowledge ornsights (Sidhu et al., 2007). Hence, exploration of new alternativeseduces the speed with which existing skills are improved. In theeantime, improvement in competence at existing procedures

akes exploration with new alternatives less attractive (Levitt andarch, 1988). In other words, the efforts involved in developing

ew systems can lower the efficiency of the operations, which leado poor performance in the short term and lower survival chances

anagement 31 (2013) 285–297 287

in the long term. As Haveman (1992) pointed out “Change maybe ultimately adaptive, but only after enough time passes for theorganization to repair the problems associated with disruption.”The learning curve literature provides evidence that organizationsimprove efficiency through the U-shaped learning curve (Argote,1999). For example, using data on consumer complaints againstthe 10 largest U.S. airlines, Lapré and Tsikriktsis (2006) found thatcustomer dissatisfaction follows a U-shaped function of operat-ing experience. Although these findings inform our understandingof learning, most previous work in organizational learning hasbeen in intra-organizational learning. Increasingly growing inter-organizational IT provides expanded opportunities to study howlearning patterns emerge in collaborative settings.

As discussed by Gupta et al. (2006), the orthogonality of explo-ration and exploitation is based on a number of assumptions suchas scarcity of organizational resources and organizational inertia.Because of the scarcity of organizational resources an organizationcan focus on either exploitation or exploration, but not on bothsimultaneously. In the meantime, if one of the assumptions is nolonger applicable, exploitation and exploration can occur on a con-tinuum. For example, if the scarcity of organizational resources isknowledge, which can be infinite (Gupta et al., 2006), exploitationand exploration can be complementary as demonstrated by Im andRai (2008).

In addition to learning from their own experiences, organiza-tions learn from others (Baum and Ingram, 1998; Ingram and Baum,1997; Irwin and Klenow, 2004; Thornton and Thompson, 2001),for example, from external joint ventures or strategic alliances(Holmqvist, 2004; Inkpen and Currall, 2004; Larrson et al., 1998;Schildt et al., 2005), or from their previous experiences with sim-ilar programs (Pisano, 1996), generally referred to as learningspillovers. Organizations capture the experiences of other organi-zations through the transfer of encoded experience in the form oftechnologies, codes, procedures, or similar routines (Dutton andStarbuck, 1978). The knowledge transfer may be facilitated throughgovernment agencies and trade associations or by the movementof personnel. For example, Song et al. (2003) found that work-force mobility is likely to result in inter-firm knowledge transfer(i.e., “learning-by-hiring”). Thus, organizational learning can ben-efit trading partners through a number of ways and it is usefulfor organizations to understand how learning occurs and how itmanifests in performance. Table 1 provides representative orga-nizational learning studies that study the performance benefits oforganizational learning, the U-shaped learning curve, and/or learn-ing spillovers.

2.3. Summary

Although previous research has examined CPFR and organiza-tional learning in their respective fields, gaps still exist in theseliteratures. First, there is little empirical research to validate theanalytical findings of CPFR benefits. Second, few empirical studieson supply chain collaboration programs or IT enabled inter-firmcollaborations have taken a learning perspective to provide a morecomplete picture on how the benefits are realized, and more impor-tantly, how they are realized for sequential launches of productreplacements. Third, most CPFR research has considered functionalproducts with relatively steady demand and long life cycles; fewhave studied innovative products with short product life cycles.Our study aims to fill these gaps in the CPFR literature.

3. Research model and hypotheses

Building upon prior literature on CPFR and organizational learn-ing theory, we propose a research model and four hypotheses

288 Y. Yao et al. / Journal of Operations Management 31 (2013) 285–297

Table 1Representative studies: Impact of organizational learning on performance.

Study Performance measures Context Operationalization of learning Explains

Argote et al. (1990) Tonnage produced Wartime shipbuilding Cumulative yard output Learning curves, learningspillovers

Baum and Ingram (1998);Ingram and Baum (1997)

Organizational survival rate Hotel industry Cumulative number of roomyears

U-shaped learning curve;learning spillovers

Darr et al. (1995) Food and labor costs; servicetimeliness (% late pizzas)

Pizza stores Cumulative number pizzasproduced

Learning curves; learningspillovers

Dyer (1996) Defects, inventory costs Auto industry Time spent knowledge sharing Learning spilloversIngram and Simons (2002) Profitability Kibbutzim Accumulative labor units Learning spilloversIm and Rai (2008) Relationship performance Supply chain dyads Knowledge sharing Knowledge sharing in

inter-organizationalrelationships

Irwin and Klenow (2004) Market price Semiconductor production Cumulative shipments Learning spilloversLapré and Tsikriktsis (2006) Consumer complaint rates Airline Cumulative number of flights U-shaped learning curveLevin (2000) Automobile reliability Automobile production Time Learning curvesMacher and Mowery (2003) Die yields, cycle times Semiconductor manufacturing Cumulative volume; HR and IT

practicesOrganizational learning

Pisano (1994) Lead time Process development Percent of total projectperson-hours

Learning spillovers

Pisano (1996) R&D development hours R&D development

Pisano et al. (2001) Surgical procedure time Health care

H4

H3

H2

CPFR

Learning

Inven tory Level s

Forec ast Error

Prod uct Launc h

Seq uen ce

H1

tamFbitaf2

iciattmaCaamsmhbo

Fig. 1. Research model and hypotheses.

hat link a supply chain’s CPFR learning with its performance. Wessess CPFR performance based on two commonly used perfor-ance metrics: forecast error for CF and inventory level for CR.

orecast error is measured as the absolute percentage error (APE)etween actual shipment quantity and planned quantity forecasted

n the prior period. Inventory level is the total number of units athe customer’s warehouse at the end of each period2. These metricsre widely used in previous literature to evaluate supply chain per-ormance (e.g., Aviv, 2001, 2002; Gaur et al., 2005; Terwiesch et al.,005). Fig. 1 summarizes our research model and hypotheses.

Previous research has modeled CF through sharing of forecastnformation and CR through managing of inventory and order pro-esses as two separate components of CPFR, and examined theirmpacts respectively (e.g., Aviv, 2001, 2002). In practice, CF and CRre implemented in different ways, thus having different implica-ions for collaboration and performance. CF requires participantso share information on demands and forecasts, both private infor-

ation (Klein and Rai, 2009), to develop a joint forecast plan, byligning their calendars and creating formal communication points.R involves the execution of the forecasting plan and requiresltering replenishment responsibilities. The altered responsibilitiesre expected to be proactive and customer focused, including align-ent and integration of replenishment processes by both partners,

o that replenishment and inventory are managed in a collaborativeanner. The fundamental difference between CF and CR, which

as significant implications for the learning curves, is the level ofusiness process reengineering required. While CF requires collab-rative interaction between partners’ representatives to formulate

2 Detailed definitions of these variables are provided in Section 4.2.

Timing of technology transfer Organizational learningCumulative volume Learning curves

a forecast plan, CR requires more significant process reengineering,often deconstructing extant decision-making routines of many par-ticipants across a number of functional areas and then creatingnew joint routines and decision processes (Cederlund et al., 2007).In addition, CF is directly subject to environmental changes suchas demand shocks, technological advancement, etc. These changesindirectly affect CR because CR is derived through the forecastplans (Aviv, 2007). These differences are summarized in Table 2,which distinguishes between the mechanisms, objectives and pro-cess changes between CF and CR.

When participants begin working together to develop collab-orative forecasts by synchronizing planning calendars, one wouldexpect forecasting accuracy to improve immediately due to timelysharing of complementary information which was previously pri-vate and unavailable (Klein and Rai, 2009). For example, if amanufacturer previously did not have the information on demandand the retailer did not have the information on production capac-ity, CF enables sharing such information with each other. Over time,supply chain partners improve accuracy in forecasts by workingcollaboratively (Devaraj et al., 2007). However, the improvementmay fall into a “competency trap”, a tradeoff between exploratoryand exploitive learning (Ingram and Baum, 1997; Levitt and March,1988). Emphasizing exploitive learning may lead an organizationto employ established routines beyond their point of usefulness,even though there has been an environmental change. As partic-ipants gain experience from sharing of information through CPFRuse, the prolonged experience with CF may lead partners to rely onpast experiences rather than current and changing market condi-tions, thus causing a decrease in the improvement rate of forecastaccuracy. Although one would expect that the learned experienceof forecasting will continue to be helpful in future forecasts, inpractice the process tends to go on “auto-pilot” and firms focus theirattention to exploitation until an unforeseen circumstance in theenvironment forces an intervention. This is because CPFR are gen-erally implemented under conditions of scarcity of organizationalresources (e.g., management resource) and can be deployed eitherfor exploitation or exploration, thus leading to the “U-shaped”learning curve (Gupta et al., 2006). This phenomenon is supportedby a recent reconceptualization of the competency trap which artic-ulates that while organizations benefit from increased competency

“. . .the improvement in organizational competence that accom-panies the accumulation of experience paradoxically exacerbatesthe decline in organization-environment fit, which eventually leadsto competency trap” (Liu, 2006, p. 146).

Y. Yao et al. / Journal of Operations Management 31 (2013) 285–297 289

Table 2Comparing collaborative forecasting and replenishment processes.

Component Collaborative forecasting (CF) Collaborative replenishment (CR)

Major elements Time based sharing of plans and forecast Time based decision making on shipments to thecustomer warehouse

Primary mechanism Schedule alignment and communication Changing routines and processesResults Prediction of orders Execution of order forecastsProcess changes to support collaboration • Alignment of planning calendars • Joint resolution of problems and issues

• Synchronizing forecasting schedules • Changing organizational structures to createcustomer focused teams at the supplier andsupplier focused teams at the buyer

• Establishing formal communication points • Changing job responsibilities from reactive toproactive replenishment roles

and s

cclamluicoT

Hi

btpracshiotb2tlcawtipktttatkma“tl

Our data for this research were collected from Motorola andone of its largest retail customers in the U.S. The retail customeris a national provider of telecommunication services and generally

Level of process reengineering Low; formal communication

Impact of change in the environment Direct and high

In CF firms consistently engage in exploitive learning in fore-asting such as tweaking forecasting parameters without using aompletely new forecasting model or technology (i.e., explorativeearning in forecasting). As a result, the forecast algorithms may beccurate at the beginning but over time, as the external environ-ent changes, become obsolete if they are not shifted to explorative

earning, especially as demand patterns change due to new prod-cts launched by competitors. Therefore, we expect that the rate of

mprovement in forecast accuracy will reflect a U-shaped learningurve. That is, forecasting error will decrease with the accumulationf CPFR experience but will start increasing after a certain point.his leads to Hypothesis 1 (H1) as follows:

1. The relationship between forecast error and CPFR experiences U-shaped.

As discussed above, CR requires significantly greater effort inusiness process reengineering than does CF. While in CF the par-icipants can improve forecast accuracy immediately by sharingreviously private information, CR requires much more reenginee-ing of processes. It includes greater involvement by partners toddress increased organizational and workplace disruption due tohanges in processes and work routines over a longer period of timeuch that it is likely to follow a different learning curve than CF. Itas been argued that when a new program, technology, or process is

ntroduced, it may disrupt existing work routines, and the adoptingrganizations must go through a learning process, making cogni-ive, inter-organizational adjustments that allow new routines toecome ongoing practice (Edmondson et al., 2001; Pisano et al.,001). Once the new routines are in place, the adopting organiza-ions can quickly improve their performance through exploitiveearning. The disruption of routines at the beginning can causeonfusion and mistakes, thereby resulting in a worse performances the organizations experience explorative learning. Participantsork together to create new routines and amend existing ones so

hat the worsening rate decreases and the performance begins tomprove (Scott, 2000). In addition, from a knowledge managementerspective, when participants work together their experiences ornowledge of inventory and replenishment are likely to vary fromhe others. This is because they have to understand and absorbhe knowledge from their partner before it can be used effec-ively (Holmqvist, 2004). This additional effort in understandingnd absorbing new knowledge is likely to lead, at least temporarily,o performance deterioration. Once this process is complete, thenowledge and experience is likely to result in improved perfor-ance. For example, prior studies on technology adoption, across

range of operations and new technologies, have documented aperformance dip” at the beginning of the implementation of a newechnology (e.g., McAfee, 2002). Once the new routines are estab-ished and tested, their performance, as measured by inventory

chedule alignments High; changes to organizational structure andemployees’ job responsibilitiesIndirect and low

level, will improve. Hence, we propose Hypothesis 2 (H2) as fol-lows:

H2. The relationship between inventory level and CPFR experi-ence is inversely U-shaped.3

In addition to learning from their own experience, organiza-tions may learn from others or learn from their own experiencewith prior products or projects. Experience may spill over fromone operating unit to another within the same organization (Argoteet al., 1990), or from one program to another (Levinthal and March,1993). Similarly, learning can spill over from managing one productto a subsequent replacement product or new form factor. Productslaunched later may reuse certain routines and processes previouslylearned, thus resulting in improved problem reorganization andless cognitive bias (Keil et al., 2007). The knowledge and experienceaccumulated in managing earlier product launches may be directlyuseful in later products’ planning and replenishment, allowingparticipants to bypass or minimize the need for trial-and-errorexperimentation and to avoid having to search for the best rout-ines among a pool of alternatives (Levitt and March, 1988), thereby“jump-starting” the learning curve. In addition to the spillover ofprior knowledge and experience, the information shared may stillbe relevant for forecasting of the replacement product resultingin improved forecast accuracy. Prior work has found that morehistorical data lead to better forecasting accuracy (Raghunathan,2001). Previous work has also found empirical support that prod-ucts introduced later tend to be more productive, or have higherquality due to learning spillovers. For example, Argote et al. (1990)found that organizations beginning production at a later date aremore productive than those with early start dates; and Levin (2000)found that car models with the latest debuts have the best quality.Hence, we propose H3 and H4 as follows:

H3. Product replacements will have lower forecast error than theirantecedents.

H4. Product replacements will have lower inventory levels thantheir antecedents.

4. Research methodology

4.1. Data and empirical context

3 Firms may also run into competency trap for CR after the performance improvesfor some time, similar to that for CF in H1. We are limited by data availability tocapture such a complex learning curve, leaving it for future research.

290 Y. Yao et al. / Journal of Operations Management 31 (2013) 285–297

Table 3Product replacement sequence.

High end products

Stage* 1 2 3Product HE1 HE2, HE3, HE4, HE5 HE6Product introduction time Fall 2002 2003 January 2004

Low end productsStage* 1 2Product LE1 LE2, LE3Product introduction time Prior to CPFR August 2002

* Each stage represents a new form factor or replacement product line. Multiple produ

Shipm ents

Motorola

Retailer’s Central

Warehou se

Inventor y

Foreca sts

Orders

Aggregate

Demand

Retail store 1

Retail store..n

Retail store 2

….

semMtitfrtowtbisnfifi

cmlcelpifptrpsfbm

ip

ch

in the literature. For example, Argote et al. (1990) used cumulative

Fig. 2. Research setting.

ells bundled service and devices through service contracts. Thend-customers (i.e., mobile phone users) have a choice of variousobile phone models from a number of manufacturers, includingotorola. The CPFR program between Motorola and its retail cus-

omer officially started in August 2002, when they started sharingnformation including demand, production, and inventory informa-ion, and implemented business process re-engineering including,or example, integration of replenishment processes, protocols foresolution of discrepancies, and a number of significant organiza-ional and job changes. We collected data from the very beginningf the CPFR implementation for more than 2 years. Because thisas the first implementation of CPFR between Motorola and a cus-

omer, it allows us to capture the learning curve from the veryeginning. Fig. 2 depicts our research setting and shows the points

n a supply chain where our data are collected. As many firms aretill considering implementing CPFR and the process of CPFR hasot significantly changed since our data collection (VICS, 2007), ourndings based upon Motorola’s experience are of relevance to otherrms that seek to implement CPFR.

Our data include transactions between Motorola and the retailustomer for nine active product models. Given that mobile phoneodels vary in price and functionality, and thus may have different

earning curves, the nine products were classified in two productategories: high end (HE) and low end (LE) product groups.4 Withinach group, later launched products were replacements for earlieraunched products; that is, the products in stage 2 replaced theroduct in stage 1 and the product in stage 3 replaced the products

n stage 2. For example, product HE2 is one of the replacementsor product HE1 in the high end product group.5 Although multipleroducts were managed through CPFR, each was managed consis-ently by the same project team using the same forecasting andeplenishment processes. That is, none of the products receivedreferential treatment and the single team approach helps pre-erve the experience and knowledge of managing a product, thus

acilitating learning spillovers. Table 3 describes the relationshipetween products in terms of launch sequence and model replace-ents.

4 Low end product group includes candy bar phones targeted to customers desir-ng everyday basic communications, whereas high end product groups includehones that sold at higher prices to customers desiring more complex options.5 Replacement phones were sleeker smaller phones with improved sound and

olor display, and other newly available and desirable features, particularly for theigh end phones, such as voice activated dialing, real time chat, etc.

Fall 2002

cts within a stage are minor variations of the same form factor.

We gathered operational and forecast data for each product.Operational data include monthly sales and inventory levels at thecustomer and shipment quantity from Motorola to the customer.The partners collaborated on a regular basis to update joint 13-month rolling forecasts. Once the joint forecasts were determined,the customer placed orders with Motorola to replenish inventory,while Motorola planned its production based on the forecasts. Thus,the unit of analysis is product-month, resulting in a data panel with120 observations6. Examples of process changes made by Motorolato support both collaborative forecasting and replenishment areshown in Table 4.

4.1.1. Dependent and independent variablesForecast error (ERROR) is one of the two dependent vari-

ables and the performance measure for CF. It is measured usinga modified absolute percentage error (APE) for a particular timeperiod between actual sales quantity (SALES), mean sales quan-tity (MSALES) and plan quantity forecasted in the prior period7

(Cattani and Hausman, 2000; Silver et al., 1998). It is calculatedas: ERRORit =

∣∣(CPFRit − SALESit)/MSALESit

∣∣, where i and t denote

product and month, respectively.8 The modified APE is in the formof a percentage term and is an appropriate measure of forecastingaccuracy because it allows the forecast errors between differentproduct lines to be comparable. As expected, higher forecastingerrors for either partner are associated with worse overall fore-casting accuracy (Silver et al., 1998).

Inventory level (INV) is the second dependent variable and theperformance measure for CR. It is measured as the total unit inven-tory at the customer’s warehouse for mobile phone for a month.This is consistent with performance measures utilized by previousresearch (Gaur et al., 2005; Rajagopalan and Malhotra, 2001). Lowerinventory levels are associated with greater operational efficiencyand supply chain performance (Gaur et al., 2005).

The organizational learning literature has operationalized learn-ing as a function of cumulative volume or time (Table 1). Becausethe CPFR experience is primarily learned from working with ordersor shipments, for this research, cumulative volume is a more appro-priate measure than cumulative time, especially when the ordersand shipments are not placed on a regular basis. In particular, weoperationalize experience (EXP) as an independent variable in twoways. First, we operationalize learning in terms of cumulative ship-ping volume from Motorola to the customer. Cumulative volume asan indicator of experience is consistent with previous studies cited

ship production output as the experience variable for shipyards.Second, we operationalize learning using cumulative number of

6 The nine products included have various life spans from 8 months to 19 months.We collected data for each product each month, resulting in 120 observations.

7 The firms retain 13-month rolling forecasts. For example, forecasts for July 2002were done in June, May, April, March, etc. 2002. We used the forecast made in theprior period; that is, June 2002.

8 MSALES is the mean sales for a product over its product life cycle. We use MSALESinstead of SALES to avoid zero values in the denominator.

Y. Yao et al. / Journal of Operations Management 31 (2013) 285–297 291

Table 4Examples of process changes to support collaboration at Motorola.

Collaborative forecasting (CF) Collaborative replenishment (CR)

Alignment of planning calendars: Instead of waiting until the end of theweek, the retailer now loads its forecast on weekday 1 in order to align withMotorola’s schedule

Joint resolution of problems and issues: Motorola and retailer cooperate tojointly review exceptions and escalation events, measure performance,manage volume/order commitment and realization to balance inventory andsell through, and order to cash process. Motorola’s business operations teamconsiders retail pricing and competitive intelligence to make replenishmentdecisions

Synchronizing forecasting schedules: Motorola loads its forecast on weekday 2, so that on week day 3, the two teams jointly resolve discrepancies lineby line

Changing organizational structures to create customer focused teams atthe supplier and supplier focused teams at the buyer: Motorola createdcustomer focused teams, added a director of customer operations, and movedfrom product and regional focused teams to customer focused teams. Theretailer created a supplier focused demand planning team

Establishing formal communication points at multiple points: Weeklymeetings were established, each with a specific agenda: week 1 operations

Changing job responsibilities from reactive to proactive replenishmentroles: Motorola eliminated reactive order managers, replacing them with

obpby

aiwaipc3spfo

4

iireefilaci2ttatp(p

i

aea

review, week 2: forecasting; week 3: process improvement, week 4:financial implications

rders between Motorola and the customer. CPFR experience maye accumulated through each order placement and may be inde-endent of the order quantity or purchase volume. Hence, we useoth operationalization methods as a robustness check in our anal-sis.

Product launch sequence is an independent variable used toccount for the learning that accrues each time a new products introduced. As discussed above and shown in Table 3, there

ere two product launch sequences, one for high-end productsnd the other for low-end products. As a result, we operational-ze it using two variables. Product launch sequence for high-endroducts (HSEQ) is coded 1 for the high-end product in stage 1,oded as 2 for products in stage 2, coded as 3 for products in stage, and coded as 0 for low-end products. Similarly, product launchequence for low-end products (LSEQ) is coded 1 for the low-endroduct in stage 1, coded as 2 for products in stage 2, and coded as 0or high-end products. Table 5 summarizes the operationalizationf these variables.

.1.2. Econometric modelWe construct an econometric model to test (i) whether there

s a non-linear learning effect of CPFR on forecast accuracy andnventory and (ii) whether the learning curve is different foreplacement products launched later. The model consists of twoquations, one for each of our two dependent variables—forecastrror and inventory level. We use the exponential functional formor both equations. The exponential functional form9 (i.e., y = e˛+ˇx),ntroduced by Levy (1965), has been used widely in estimatingearning curves in organizational learning literature (e.g., Laprénd Tsikriktsis, 2006; Lapré et al., 2000). It is more appropriate,ompared with the traditional power form, when used to estimatenitial downward concavity and the plateau effect (Lapré et al.,000), both of which are exactly what we propose to estimate inhis research. Following a widely used approach in the organiza-ional learning literature (Thornton and Thompson, 2001), we use

quadratic form of experience (EXP) as independent variable to

est the non-linear learning effect (i.e., H1 and H2). We used tworoduct launch sequence variables for high end (HSEQ) and low endLSEQ) to test the linear effect of the product launch sequence onerformance10 (i.e., H3 and H4). Given the two product groups, we

9 The exponential form is equivalent to log of the dependent variable but not thendependent variables.10 An alternative way to model product sequence is to include two dummy vari-bles, one for the stage in product sequence and the other for the product group. Westimated this model specification and the results for the learning curve estimatesre consistent with those reported in the paper.

proactive replenishment analysts who actively seek data from the retailer onsell through. Supplier roles changed from simply purchasing to planningdemand replenishment

include a dummy variable to control for the heterogeneity betweenhigh and low-end product groups. In the inventory equation, weadded monthly sales (SALES) to control for environmental, sea-sonal, and demand effects on inventory levels (Lee et al., 1999).All other things being equal, higher monthly sales will be associ-ated with higher inventory levels. Finally, because we utilize paneldata, we controlled the fixed-time effect that accounts for the fac-tors that evolve with time, such as product life cycle, seasonality,technological advancement and productivity improvement duringour sample time. Table 6 presents the descriptive statistics and thecorrelation matrix for the variables.

Let i denote product, j denote product group, and t denote cal-endar month. The econometric model is specified as11

Ln(ERRORit) = ˇ0 + ˇ1EXPit + ˇ2(EXPit)2 + ˇ3HSEQj + ˇ4LSEQj

+ ˇ5HIGHj + εit (1)

Ln(INVit) = �0 + �1Ln(ERRORit) + �2EXPit + �3(EXPit)2 + �4SALESit

+ �5HSEQj + �6LSEQj + �7HIGHj + �it (2)

where ERROR, INV, EXP, HSEQ, and LSEQ are defined in Section 4.2(also see Table 4). Monthly sales (SALES) is the unit sales by the cus-tomer for a product during a month. Product group dummy variable(HIGH) is the dummy variable coded ‘1’ for high-end product groupand ‘0’ for low-end product group. ˇs and �s are parameters to beestimated, and ε and � are the random disturbance terms.

5. Estimation and results

We used panel data techniques to estimate our model. Giventhat inventory is often affected by previous inventory levels andsubject to autocorrelation in the model, failure to account for auto-correlation may lead to inefficient estimation. We performed thetest described in (Wooldridge, 2003, pp. 282–283) for autocor-relation in panel data and found that first-order autocorrelation(AR1) does exist in our panel dataset (F = 5.24, p < 0.05 for ERRORequation; F = 9.59, p < 0.05 for INV equation). Furthermore, theAR1 process is likely to be different across products, resulting in

panel-specific AR1 (i.e., each panel has its unique AR1 coefficient).We performed a likelihood ratio test to check whether the AR1coefficients are common across panels (Cheng and Nault, 2007).

11 When the value for INV or ERROR is zero, we add 1 to avoid taking the log of 0(Ba and Pavlou, 2002).

292 Y. Yao et al. / Journal of Operations Management 31 (2013) 285–297

Table 5Operationalization of variables.

Variable Definition References in the literature

INV Unit inventory Gaur et al. (2005), Rajagopalan and Malhotra (2001), Shah andShin (2007)

ERROR Absolute percentage error for a particular period between actualshipment quantity and plan quantity forecasted one period before

Cattani and Hausman (2000), Makridakis and Winkler (1983)

EXP Experience measured as (1) cumulative shipping volume fromMotorola to the customer, and (2) cumulative number of orders

Argote (1999); Baum and Ingram (1998), Ingram and Baum(1997), Irwin and Klenow (1994)

cts

Ws

vut(1epbssap

mncevcmwvfs

atpbttpa

TD

between Motorola and the customerHSEQ and LSEQ Product launch sequence for high end and low end produSALES Unit sales at the retailer

e find the test statistics were significant, suggesting that panel-pecific AR1 is more appropriate.

Another diagnostic issue related to panel data analysis is theariance of the error term structure between panels (i.e., prod-cts). Given that products differ in demand pattern, it is likelyhat the variances of the error term are different across panelsi.e., data are heteroskedastic). The modified Wald test (Greene,997) provided significance in the test statistics in the inventoryquation (�2 = 726, p < 0.001) and forecast error equation (�2 = 294,

< 0.001), respectively, suggesting that heteroskedasticity coulde an issue. Therefore, we deployed the feasible generalized leastquares (FGLS) procedure in STATA with the specification of panel-pecific AR1 and heteroskedasticity to estimate our models. FGLSllows estimation in the presence of AR1 autocorrelation withinanels and heteroskedasticity across panels (Wooldridge, 2003).

The equations for inventory levels and forecast error were esti-ated as a simultaneous equation system. Due to the recursive

ature of the two equations, the equation system is identified andan be estimated individually (Greene, 1997). Table 7 presents thestimation results for the models using both cumulative shippingolume and cumulative number of orders as learning variables. Theoefficients for all variables are generally consistent for both esti-ations, demonstrating the robustness of our results. Therefore,e discuss only the results of models using cumulative shipping

olume. The Wald statistics are 36 (p < 0.01) and 18 (p < 0.001) fororecast error and inventory equations, respectively, indicating theignificance of the estimates.

For forecast error equation, the coefficient for EXP is negativend significant ( ̌ = − 8.55 × 10−7, p < 0.001), and the coefficient forhe squared term of EXP is positive and significant ( ̌ = 4.51 × 10−13,

< 0.001), indicating that CPFR learning reduces forecast errorut at a decreasing rate until the forecast error starts to increase;

hat is, a U-shaped learning curve. For the inventory equation,he coefficient for EXP is positive and significant ( ̌ = 5.83 × 10−6,

< 0.01), and the coefficient for the squared term of EXP is negativend significant ( ̌ = −3.33 × 10−12, p < 0.05), indicating that CPFR

able 6escriptive statistics and correlation matrix (N = 120).

Mean SD Min Max

1. Forecast error (ERROR) 0.56 0.72 0

2. Inventory levels (INV) 26,127 32,728 0 24,123. Own experience (EXP)

(cumulative shipping volume)42,2923 44,2751 0 1836,3

4. Own experience (EXP)(cumulative number of orders)

18.68 15.77 0

5. Monthly sales (SALES) 60,651 60,992 0 286,56. Product launch sequence for

high end products (HSEQ)1.18 1.03 0

7. Product launch sequence forlow end products (LSEQ)

0.59 0.83 0

8. Product group dummy (HIGH) 0.63 0.49 0

* p < 0.05.** p < 0.01.

*** p < 0.001.

Bhattacharya et al. (2003), Ramachandron and Krishnan (2008)Clark and Hammond (1997), Lee et al. (1999)

learning increases inventory levels but at a decreasing rate untilinventory levels start to decrease; that is, an inverted U-shapedlearning curve. These findings lend support to H1 and H2. We cancalculate the turning point in the learning curves for both forecasterror and the inventory levels by taking the first derivative of theestimated equation with regard to EXP and setting it to zero. Wefind that the forecast error begins to increase after a cumulativeexperience of 947,894 units, and the inventory begin to decreaseafter 875,375 units.

The coefficients for HSEQ and LSEQ are insignificant in ERRORequation, indicating that H3 is not supported. The coefficient forLSEQ is negative and significant in INV equation ( ̌ = −3.96, p < 0.05),indicating that, for low-end products, replacement products havelower inventory. The coefficient for HSEQ, however, is insignificant.Thus, H4 is partially supported for low-end products. These resultssuggest learning spillovers exist from the earlier launched productsto its replacements for the CR component of CPFR for low-end prod-ucts. It is inconclusive for the CF component and for the high-endproducts.

It is interesting to note that the coefficient for ERROR ininventory equation is positive and significant ( ̌ = 1.17, p < 0.05),indicating that greater forecast error leads to higher inventory. InCPFR, since replenishment plans are derived from forecast plans(Aviv, 2002), the performance of CF affects that of CR. Lower forecasterror may lead to lower inventory levels. From the classic inventorymodel, firms stockpile ‘safety’ inventory to guard against demanduncertainty. When the forecast error is high, the firm has to build inextra inventory as a buffer to manage demand uncertainty. Hence,improved forecast accuracy (i.e., reduced forecast errors) resultsin lower inventory. Previous research has demonstrated that fore-casting error is a major factor in poor inventory management (Chenet al., 2000; Lee et al., 1997). This result is consistent with prior find-

ings and exhibits the validity and robustness of our analysis. It isalso interesting to note that the estimated coefficients for prod-uct level variables, HIGH, are significant in the estimation usingcumulative shipping volume as the learning variable, suggesting

1 2 3 4 5 6 7

5.91 119 0.10 115 0.09 −0.20* 1

66 0.03 −0.24** 0.88*** 1

96 0.12 −0.08 0.44*** 0.33*** 13 0.01 −0.04 −0.05 −0.11 0.25** 1

2 0.05 0.14 −0.10 −0.12 −0.26*** −0.82*** 1

1 −0.09 −0.09 −0.03 0.02 0.22* 0.88*** −0.93***

Y. Yao et al. / Journal of Operations Management 31 (2013) 285–297 293

Table 7Estimation results (Standard errors in parenthesis).

Cumulative number of orders as learning variable Cumulative shipping volume as learning variable

ERROR INVENTORY ERROR INVENTORY

Intercept 0.61*** (0.12) 16.50*** (4.46) 0.65*** (0.13) 14.93*** (3.50)forecast error (ERROR) 1.46** (0.56) 1.17* (0.56)CPFR experience (EXP) −0.03*** (0.004) 0.17** (0.06) −8.55 × 10−7*** (1.55 × 10−7) 5.83 × 10−6** (2.17 × 10−6)Squared term of CPFR

experience (EXP2)0.39 × 10−3*** (0.09 × 10−3) −2.81 × 10−3* (1.23 × 10−3) 4.51 × 10−13*** (1.16 × 10−13) −3.33 × 10−12* (1.61 × 10−12)

Monthly sales (SALES) 3.99 × 10−6 (5.02 × 10−6) 2.06 × 10−6 (5.24 × 10−6)Product sequence for

high end (HSEQ)0.07 (0.09) 0.59 (1.00) 0.08 (0.08) 0.76 (0.92)

Product sequence for lowend (LSEQ)

0.04 (0.07) −5.05* (2.27) −0.02 (0.08) −3.96* (1.79)

Product category—highend (HIGH)

−0.18 (0.20) −11.17* (4.94) −0.26* (0.19) −9.24* (3.92)

Dummies for calendarmonths (MON)

Included Included Included Included

Model statisticsN 120 120 120 120Wald Chi2 72*** 21*** 36*** 18***

tbm

(tseaft

6

6

fFtaai

* p < 0.05.** p < 0.01.

*** p < 0.001.

hat high end products tend to have lower forecast errors. This maye because high end products are likely to receive greater manage-ent attention and resources that ensure forecast accuracy.To better illustrate these results, we deploy regression estimates

cumulative sales volume as learning variable’s model in Table 7)o graph the learning curves for forecast error and inventory. Fig. 3hows the learning curves for forecast error and inventory for high-nd products in stage 1 (the learning curves for low-end productsnd for those in other stages are similar and not presented here). Fororecast error, the learning curve is U-shaped, whereas for inven-ory levels, the learning curve is inverted U-shaped.

. Discussion

.1. Implications for theory

Our findings demonstrate that the two performance metrics,orecast error and inventory level, follow distinct learning curves.orecast error follows a U-shaped learning curve whereas inven-ory level follows an inverted U-shaped learning curve. Forecast

ccuracy increases with a supply chain’s experience in CPFR butt a decreasing rate, and inventory levels decrease before increas-ng. The finding of an inverted U-shaped curve for inventory level,

0.2

.4.6

Forecast_Error

67

89

10

Log(Inventory)

0 50000 0 1000000 150000 0 200000 0

E...

Inventory Forecast_Error

Fig. 3. Estimated learning curves for forecast error and inventory.

although in a different context, is consistent with prior studieson technology adoption that have documented a “performancedip” at the beginning of the implementation of a new technol-ogy (e.g., McAfee, 2002). The findings of a U-shaped learning curvefor forecast error is consistent with those in the literature (e.g.,Baum and Ingram, 1998; Lapré and Tsikriktsis, 2006) and may beattributed to the “competency trap” (Levitt and March, 1988), atradeoff between exploratory and exploitive learning (Baum andIngram, 1998; Ingram and Baum, 1997). Organizations may focuson exploitative learning and reduce their exploratory activity pre-maturely in a changing environment, despite the fact that newopportunities and threats are present (Baum and Ingram, 1998).It is important to note that we have specifically controlled for thetime specific effect such as the changes in performance associatedwith product life cycle by adding the fixed-time effect to our model.As a result, the learning variable captures the learning specificallyemanating from CPFR implementation over and above the effectfrom product life cycle.

Our findings provide empirical validation for the analyticalresults in Aviv (2001, 2002, 2007). Aviv (2001, 2002) found that col-laborative forecasting can provide substantial benefits to the supplychain but the magnitude of these benefits depends on the specificsetting (Aviv, 2007). Through our empirical analysis, we show thatCPFR information systems can add value in reducing forecast errorsand inventory levels, although the extent of the value is dependenton the point in the learning curve at which it is assessed.

Our findings complement previous literature that hasempirically examined the business value of supply chain tech-nologies, such as Subramani (2004), Mukhopadhyay et al. (1995),Mukhopadhyay and Kekre (2002), and Hsieh et al. (2011), andinter-firm collaborations, such as Patnayakuni et al. (2006), Cachonand Fisher (1997), Clark and Hammond (1997), and Lee et al.(2000). From the perspective of the business value of IS, ourstudy provides a real-world context of collaborative systemsthat are increasingly becoming a focus of IS research12. Our find-ings extend previous research that compared performance pre-and post-implementation, by demonstrating that performance

improvement is realized through a non-linear learning curve.These findings suggest that the course through which the valueis realized is not straightforward. Firms may have to endure

12 MIS Quarterly, special issue on co-creation (36 (1) 2012).

2 ons M

pctl

imatsh1rlfuitggmfvlseq

6

fioOtestf

CtumgtrrcOtoaMialphactfmil

94 Y. Yao et al. / Journal of Operati

erformance degradation before they derive benefits from partnerollaborations. To the best of our knowledge, our research ishe first attempt in this regard, and contributes to the empiricaliterature on collaboration among supply chain partners.

Another insight from our findings is that the learning curves fornventory reduction vary with product releases. Product replace-

ents have lower inventory levels than their antecedents, at leastmong low-end products. The lower inventory levels may be due tohe learning spillovers from previous products that help to jump-tart the management of the next product. Such learning spilloversave been documented in previous research (e.g., Argote et al.,990; Levin, 2000). The replacement products may utilize theoutines and processes that have been established for the earlieraunched products, thereby bypassing the “trial and error” searchor the best routines and processes. In our context, since Motorolased the same project team to manage all of the products, the learn-

ng spillovers between products were easily facilitated. The facthat this finding was only significant for low end products sug-ests that spillovers occur when there is more similarity betweenenerations of products where replacements add simpler enhance-ents such as improved color display. The complexity resulting

rom more functional features in the high end products, such asoice-activated dialing, may have resulted in less opportunity forearning spillovers. In addition, that learning spillovers were notignificant for forecast accuracy may be the result of this dynamicnvironment in which knowledge learned about forecasting mayuickly become obsolete and irrelevant.

.2. Implications for practice

Although some CPFR adopters have reported substantial bene-ts from widespread implementation, others have reported modestr no significant benefits from rather cautious implementations.ur findings have a number of managerial implications that address

his dichotomy. First, firms should continue to allocate resources tongage in explorative learning in better routines, processes, andtandard operating procedures in collaborative forecasting withheir partners so that they can alleviate the degradation of theorecast accuracy as the environment changes.

Second, firms should be patient when assessing the effect ofPFR implementation as inventory levels may become worse athe beginning. The finding of distinct learning curves reflects theniqueness of the stages in the CPFR program; that is, the imple-entation processes for CF and CR are different in that CF facilitates

reater and more accurate information exchange whereas CR fea-ures joint replenishment processes that require greater processeengineering. Different from CF, the reengineering involved in CResults in a greater degree of destruction of current business pro-esses, thereby leading to a transitory deterioration in performance.nce appropriate processes are in place, the performance is likely

o increase. A premature evaluation may lead to underestimationf the overall lasting value of the CPFR investment. These resultsre consistent with the process changes made at Motorola. For CF,otorola typically did not change their templates (i.e., forecast-

ng algorithms) for a particular product once they were developednd implemented. However, during the time period of data col-ection, Motorola faced competition not only from existing mobilehone manufacturers who were adding new capabilities such asigh resolution cameras, but also new competitors such as Palmnd Research in Motion with their introduction of smart phones,ombining phone, PDA, camera, and internet access. This is consis-ent with the U-shaped learning curve for CF. Clearly, the changes

or CR are about business process re-engineering which is much

ore disruptive, whereas those for CF are about integration ofnformation flow, which provide explanation for the two differentearning curves. The new roles and new processes resulted in some

anagement 31 (2013) 285–297

degradation in inventory performance at the beginning, which isconsistent with the inverted U-shaped learning curve for CR.

Our findings provide a possible explanation as to why somepractitioners question the value of CPFR citing the complexity anduncertainty in implementation (Doiron, 2004). Failure to considerthe learning curve may lead to a biased or incomplete assess-ment of a CPFR benefits, given the nonlinear nature of the learningcurve found in our study. Depending upon the point in time whenassessing the benefits, one can draw different or even opposite con-clusions about the benefits of CPFR. For example, Motorola mayhave initially been skeptical about CPFR benefits as the inven-tory levels increased for the first 875,375 units, on average, soldusing CPFR. Similarly, firms should not completely withdraw fromtheir explorative learning activities following a decrease in forecastaccuracy. For example, in the case of Motorola, improvements inforecast accuracy would no longer be apparent after about 947,894units sold, on average.

Third, knowledge or experience gained from previous productsmay be helpful in reducing inventory of later launched replace-ment products. Thus, firms may find it useful to build an effectiveknowledge management system to capture and use learning fromprevious collaborative experiences. One approach is to deploy thesame project team to manage all products so that such knowledgecan be maximally reserved and transferred.

Although most of our results are statistically significant, thequestion of whether these results are economically significant isalso important. We examine the economic significance by com-puting the potential effect using regression estimates (cumulativeshipping volume as learning variable in Table 7). We begin with anexamination of the total effects of the learning variable on inven-tory level. Since our model is a structural equation in nature, inaddition to affecting inventory level directly, the learning variablealso affects the inventory level indirectly through forecast accuracy.Therefore, the learning variable can have both direct and indirecteffects on inventory levels. We can calculate the cumulative effect(both indirect and direct) for our learning variable EXP. The totaleffect is:

ln (INV) = (1.17 × ln (ERROR) + 5.83 × 10−6 EXP

−3.33 × 10−12 EXP2 + ˇA)

where (1.17 × ln (ERROR) is the indirect effect, 5.83 ×10−6 EXP − 3.33 × 10−12 EXP2 is the direct effect, and ˇA is the vec-tor of control variables, constant and their respective estimates. Theindirect effect can further be written according to the estimatesfrom the ERROR equation as

1.17 × (−8.55 × 10−7 EXP + 4.51 × 10−13EXP2 + ˛B)

where ˛B is the vector of control variables, constant and theirrespective estimates. Collecting terms, therefore, the total effectis:

(4.83 × 10−6 EXP − 2.80 × 10−12 EXP2 + 1.17˛B + ˇA)

indicating the total effect on inventory level is still an inverted U-shaped curve.

Next, accounting for the total effect calculated for inventory lev-els, the potential effects are computed as the change to a dependentvariable when moving EXP from sample minimum to sample max-imum for INV and from sample minimum to 947,894 (the pointwith the lowest forecast error) for ERROR, while maintaining other

variables at their means. Therefore, the maximum total poten-tial effect is reduced inventory levels from 567,388 to 336,132,a 40.76% decrease. Because the total effect for inventory whichaccounts for both the direct and indirect effects is still positive (i.e.,

ons M

4aota

7

rlfiosicIowsvpdccfleuFteceat(af

gftfidgdmstaaa

iociuFwe

Y. Yao et al. / Journal of Operati

0.76%), we can expect that CPFR implementation leads to over-ll performance improvement. The estimate of the potential effectf CPFR shows that the performance improvements resulting fromhe implementation of CPFR are not only statistically significant butlso economically significant.

. Contribution, limitations and areas for future research

Our research makes a number of contributions. First, thisesearch is one of the early attempts to empirically examine theearning curves of CPFR, thereby providing validation for analyticalndings previously reported in the literature. Second, we conductur study from a learning perspective in assessing the benefits ofupply chain collaboration programs, which has not been addressedn the past. Our estimation of the learning curves provides a moreomplete picture as to how the performance benefits are realized.n this way, we extend the findings in prior empirical literaturen supply chain collaboration programs with richer details. Third,e find that CF and CR exhibit distinct learning curves; that is, U-

haped vs. inverted U-shaped respectively, demonstrating that thealue realization from CPFR may not be linear and uniform acrossrocesses. Although this finding is unlikely to end the CPFR valueebate among practitioners, it provides tangible evidence of theomplexity and contingencies in implementing CPFR. Fourth, weomplement previous work on CPFR that has generally examinedunctional products (e.g., consumer packaged goods) with extendedife cycles and relatively low demand uncertainty (Aviv, 2001) byxamining a dynamic business environment characterized by prod-cts with short life cycle and rapidly changing customer demands.inally, as prior studies in IT-supply chain collaboration litera-ure have conceptualized supply chain integration capability asncompassing integration of flows related to information, physi-al products, and finances (e.g., Rai et al., 2006) and has providedmpirical evidence that this supply-chain integration capabilityffects firm performance positively, our paper makes an impor-ant contribution by showing that while both information sharingCF) and physical flow integration (CR) in supply chains positivelyffect performance, the temporal evolution of their respective per-ormance benefits can be vastly different.13

Although our use of data from a single manufacturer and a sin-le retailer has significant advantages in the control of exogenousactors that may influence performance, it also leads to a limita-ion of this study; that is, the generalizability of the results to otherrms and to other industries. In addition, our study dealt with oneynamic product – mobile phones – and therefore we cannot claimeneralizability across other products that are in higher or lowerynamic environments (e.g., durable products such as washingachines and refrigerators). Therefore, results from this research

hould be generalized with caution. Another research limitation ishat we measure performance using only the inventory variable,nd lower inventory may adversely result in lower service levelsnd lost revenues, although service levels are quite consistent prend post CPFR implementation in our case.

Our findings provide several opportunities for future researchnto the learning curves of IS-based supply chain collaboration andrganizational learning. The differential organizational learningurves for the two performance metrics – forecast error andnventory levels – provide researchers with a fertile area to further

nderstand learning mechanisms in supply chain collaboration.or example, how does the non-linearity of the course throughhich performance is realized impact the short-term overall

fficiency of the supply chains? What implications does this have

13 We thank a reviewer for this suggestion.

anagement 31 (2013) 285–297 295

for implementation of IS? What mechanisms may facilitate cross-boundary learning spillovers? What role do IS play in extractingand exploiting information for sustained knowledge value?

For organizational learning, our findings present future researchopportunities pertaining to firm’s exploration and exploitation.For instance, are there interactions between the exploration andexploitation postures of a firm? What are the ways to alleviatethe time lag between the two given that in reality firms exploreand exploit opportunities simultaneously? What organizationalresources – human, structural, or process – should be in place totake advantage of simultaneous learning? We hope that our find-ings provide a basis for interdisciplinary research in supply chainmanagement and organization theory. After all, as our researchindicates, a business initiative involving such IS as CPFR is notrestricted to one area and can span several academic disciplinessuch as IS, operations management, and organizational learning.

Future researchers may collect both inventory and service leveldata to study the dynamics between them due to CPFR implemen-tation as well as develop specific measures for CPFR experience forexploration to further validate our findings. Also, the differentialfindings for high end vs. low end products should be further exam-ined to determine whether introduction time or product type is themost critical factor. Finally, future researchers can compare learn-ing curves of CPFR with other forms of collaborative relationships,such as VMI or CRP. It will be interesting to see the differences andcommonalities between them as they may have different learningmechanisms.

8. Conclusions

Our objective in this paper was to examine organizational learn-ing from the implementation of a CPFR information system. We findthat two aspects of CPFR – forecasting and replenishment – exhibitdifferent forms of organizational learning. Forecast error of productdemand decreases quickly while inventory levels rise. This makesreplenishment appear worse. We find that these trends reverseover time. As new products are introduced and business conditionschange, the forecast accuracy decreases. However, inventory lev-els decrease due to improvement in collaborative replenishmentbecause the supply chain is cleared of old inventory and partnerslearn to understand and fulfill market demand.

References

Allred, C.R., Fawcett, S.E., Wallin, C., Magnan, G.M., 2011. A dynamic collabora-tion capability as a source of competitive advantage. Decision Sciences 42 (1),129–161.

Argote, L., 1999. Organizational Learning: Creating, Retaining, and TransferringKnowledge. Kluwer Academic Publishers, Boston.

Argote, L., Beckman, S.L., Epple, D., 1990. The persistence and transfer of learning inindustrial settings. Management Science 36 (2), 140–154.

Argote, L., Epple, D., 1990. Learning curves in manufacturing. Science 23, 920–924.Argote, L., McEvily, B., Reagans, R., 2003. Managing knowledge in organizations: an

integrative framework and review of emerging themes. Management Science49 (4), 571–582.

Argyris, C., Schon, D.A., 1978. Organizational Learning: A Theory of Action Perspec-tive. Addison-Wesley, Reading, MA.

Aviv, Y., 2001. The effect of collaborative forecasting on supply chain performance.Management Science 47 (10), 1326–1343.

Aviv, Y., 2002. Gaining benefits from joint forecast and replenishment processes: thecase of auto-correlated demand. Manufacturing and Service Operations Man-agement 4 (1), 55–74.

Aviv, Y., 2007. On the benefits of collaborative forecasting partnerships betweenretailers and manufacturers. Management Science 53 (5), 777–794.

Ba, S., Pavlou, P.A., 2002. Evidence of the effect of trust building technology inelectronic markets: price premiums and buyer behavior. MIS Quarterly 26 (3),243–268.

Barratt, M., Oliveira, A., 2001. Exploring the experiences of collaborative planninginitiatives. International Journal of Physical Distribution & Logistics Manage-ment 31 (4), 266–289.

Baum, J., Ingram, P., 1998. Survival-enhancing learning in the Manhattan hotel indus-try, 1898–1980. Management Science 44 (7), 996–1016.

2 ons M

B

C

C

C

C

C

C

C

D

D

DD

D

D

D

E

G

GF

F

G

H

H

H

H

H

I

I

I

I

I

K

K

K

96 Y. Yao et al. / Journal of Operati

hattacharya, S., Krishnan, V., Mahajan, V., 2003. Operational technologyimprovements in product development decision-making. European Journal ofOperational Research 149 (1), 102–130.

achon, G.P., Fisher, M., 1997. Campbell Soup’s continuous replenishment program:evaluation and enhanced inventory decision rules. Production and OperationsManagement 6 (3), 266–276.

attani, K., Hausman, W., 2000. Why are forecast updates often disappointing? Man-ufacturing & Service Operations Management 2 (2), 119–127.

ederlund, J., Kohli, R., Sherer, S., Yao, O., 2007. Successful implementation of col-laborative planning, forecasting and replenishment: lessons learned at MotorolaInc. Supply Chain Management Review 11 (7), 28–35.

hen, F., Drezner, Z., Ryan, J.K., Simchi-Levi, D., 2000. Quantifying the bullwhip effectin a simple supply chain: the impact of forecasting, lead times, and information.Management Science 46 (3), 436–443.

hen, M.C., Chen, K.Y., Hsu, M.F., Yeh, C.T., 2009. A web services-based collabo-rative planning, forecasting, and replenishment (CPFR) framework for managingspare parts of semiconductor equipment. IEEE Transactions on SemiconductorManufacturing 22 (4), 596–600.

heng, J., Nault, B.R., 2007. Industry level supplier-driven IT spillovers. ManagementScience 53 (8), 1199–1261.

lark, T.H., Hammond, J., 1997. Reengineering channel reordering processes toimprove total supply chain performance. Production and Operations Manage-ment 6 (3), 248–265.

arr, E., Argote, L., Epple, D., 1995. The acquisition, transfer, and depreciation ofknowledge in service organizations; productivity in franchises. ManagementScience 41 (11), 1750–1762.

evaraj, S., Krajewski, L., Wei, J.C., 2007. Impact of eBusiness technologies on opera-tional performance: the role of production information integration in the supplychain. Journal of Operations Management 23, 1199–1216.

oiron, M., 2004. CPFR: myths and realities. ECR Journal 4 (1), 51–54.uncan, R.B., Weiss, A., 1979. Organizational learning: implications for organiza-

tional design. Research in Organizational Behavior 1 (4), 75–125.utton, J.M., Starbuck, W.H., 1978. Diffusion of an intellectual technology. In: Krip-

pendorff, K. (Ed.), Communication and Control in Society. Gordon and Breach,New York, NY.

utton, J.M., Thomas, A., 1984. The history of progress functions as a managerialtechnology. Business History Review 58 (2), 204–233.

yer, J., 1996. Specialized supplier networks as a source of competitive advantage:evidence from the auto industry. Strategic Management Journal 17 (4), 271–291.

dmondson, A., Bohmer, R.M., Pisano, G.P., 2001. Disrupted routines: team learn-ing and new technology implementation in hospitals. Administrative ScienceQuarterly 26, 685–716.

aur, V., Fischer, M.L., Raman, A., 2005. An econometric analysis of inventoryturnover performance in retail services. Management Science 51 (2), 181–194.

reene, W.H., 1997. Econometric Analysis. Prentice Hall, Saddle River, NJ.isher, M., 2007. Strengthening the empirical base of operations management. Man-

ufacturing and Service Operations Management 9 (4), 368–382.u, H.P., Chu, K.K., Lin, S.W., Chen, C.R., 2010. A study on factors for retailers imple-

menting CPFR—a fuzzy AHP analysis. Journal of Systems Science and SystemsEngineering 19 (2), 192–209.

upta, A.K., Smith, K.G., Shalley, C.E., 2006. The interplay between exploration andexploitation. Academy Management Journal 4, 693–706.

aveman, H., 1992. Between a rock and a hard place: organizational change andperformance under conditions of fundamental environmental transformation.Administrative Science Quarterly 37 (1), 48–76.

sieh, J.J., Rai, A., Xu, S.X., 2011. Extracting business value from IT: a sense-making perspective of post-adoptive use. Management Science 57 (11),2018–2039.

olmqvist, M., 2004. Experiential learning processes of exploitation and explorationwithin and between organizations: an empirical study of product development.Organization Science 15 (1), 70–81.

uber, G.P., 1991. Organizational learning: the contributing processes and the liter-atures. Organization Science 2 (1), 88–115.

ult, G.T.M., Ketchenand, D.J., Nichols, E.L., 2003. Organizational learning as a strate-gic resource in supply management. Journal of Operations Management 21 (5),541–556.

m, G., Rai, A., 2008. Knowledge sharing ambidexterity in long-term interorganiza-tional relationships. Management Science 54 (7), 1281–1297.

ngram, P., Baum, J., 1997. Opportunity and constraint: organizations’ learning formthe operating and competitive experience of industries. Strategic ManagementJournal 18, 75–98.

ngram, P., Simons, T., 2002. The transfer of experience in groups of organizations:implications for performance and competition. Management Science 48 (12),1517–1533.

nkpen, A., Currall, S., 2004. The coevolution of trust, control, and learning in jointventures. Organization Science 15 (5), 586–599.

rwin, D., Klenow, P., 2004. Learning-by-doing spillovers in the semiconductor indus-try. The Journal of Political Economy 102 (6), 1200–1227.

eil, M., Depledge, G., Rai, A., 2007. Escalation: the role of problem recognition andcognitive bias. Decision Sciences 38 (3), 391–422.

lein, R., Rai, A., 2009. Interfirm strategic information flows in logistics supply chain

relationships. MIS Quarterly 33 (4), 735–762.

udyba, S., 2006. Enhancing organizational information flow and knowledge cre-ation in re-engineering supply chain systems: an analysis of the U.S. automotiveparts and supplies model. International Journal of Innovation Management 10(2), 163–173.

anagement 31 (2013) 285–297

Lapré, M., Tsikriktsis, N., 2006. Organizational learning curves for customer dissat-isfaction: heterogeneity across airlines. Management Science 32 (3), 352–366.

Lapré, M.A., Mukherjee, A.S., Van Wassenhove, L.N., 2000. Behind the learningcurve: linking learning activities to waste reduction. Management Science 46(5), 597–612.

Larrson, R., Bengtsson, L., Henriksson, K., Sparks, J., 1998. The interorganizationallearning dilemma: collective knowledge in strategic alliances. Organization Sci-ence 9 (3), 285–305.

Lee, H.G., Clark, T., Tam, K.Y., 1999. Research report: can EDI benefit adopters? Infor-mation Systems Research 10 (2), 186–195.

Lee, H.L., Padmanabhan, P., Wang, S., 1997. Distortion in a supply chain: the bullwhipeffect. Management Science 43, 436–443.

Lee, H.L., So, K., Tang, C.C., 2000. The value of information sharing in a two-levelsupply chain. Management Science 46 (5), 626–643.

Levin, D.Z., 2000. Organizational learning and the transfer of knowledge: an inves-tigation of quality improvement. Organization Science 11 (6), 630–647.

Levinthal, D.A., March, J.G., 1993. The myopia of learning. Strategic ManagementJournal 14, 95–112.

Levitt, B., March, J.G., 1988. Organizational learning. Annual Review of Sociology 14,319–340.

Levy, F.K., 1965. Adaptation in the production process. Management Science 11,B136–B154.

Li, G., Rajagopalan, S., 1997. The impact of quality on learning. Journal of OperationsManagement 15 (3), 181–191.

Liu, W., 2006. Knowledge exploitation, knowledge exploration, and competencytrap. Knowledge and Process Management 13 (3), 144–161.

Macher, J., Mowery, D., 2003. Managing learning by doing: an empirical studyin semiconductor manufacturing. Journal Product Innovation Management 20,391–410.

Makridakis, S., Winkler, R., 1983. Averages of forecasts: some empirical results.Management Science 29 (9), 987–996.

March, J.G., 1991. Exploration and exploitation in organizational learning. Organi-zation Science 2 (1), 71–87.

McAfee, A., 2002. The impact of enterprise information technology adoption onoperational performance: an empirical investigation. Production and OperationsManagement 11 (1), 33–53.

Mukhopadhyay, T., Kekre, S., Kalathur, S., 1995. Business value of information tech-nology: a study of electronic data interchange. MIS Quarterly 19 (2), 137–156.

Mukhopadhyay, T., Kekre, S., 2002. Strategic and operational benefits of electronicintegration in B2B procurement. Management Science 48 (10), 1301–1313.

Patnayakuni, R., Rai, A., Seth, N., 2006. Relational antecedents of information flowintegration for supply chain coordination. Journal Management InformationSystems 23 (1), 13–49.

Pisano, G., 1994. Knowledge, integration, and the locus of learning: an empiricalanalysis of process development. Strategic Management Journal 15, 85–100.

Pisano, G., Bohmer, R., Edmondson, A., 2001. Organizational differences in rates oflearning: evidence from the adoption of minimally invasive cardiac surgery.Management Science 47 (6), 752–768.

Pisano, G.P., 1996. Learning-before-doing in the development new process technol-ogy. Research Policy 25, 1097–1119.

Raghunathan, S., 1999. Interorganizational collaborative forecasting and replen-ishment systems and supply chain implications. Decision Sciences 30 (4),1053–1971.

Raghunathan, S., 2001. Information sharing in a supply chain: a note on its valuewhen demand is non-stationary. Management Science 47 (4), 605–610.

Rai, A., Patnayakuni, R., Patnayakuni, N., 2006. Firm performance impacts of digitally-enabled supply chain integration capabilities. MIS Quarterly 30 (2), 225–246.

Rajagopalan, S., Malhotra, A., 2001. Have U.S. manufacturing inventories reallydecreased? An empirical study. Manufacturing & Service Operations Manage-ment 3 (1), 14–24.

Ramachandron, K., Krishnan, V., 2008. Design architecture and introduction timingfor rapidly improving industrial products. Manufacturing & Service OperationsManagement 3 (1), 13–24.

Sahin, F., Robinson, E.P., 2002. Flow coordination and information sharing in sup-ply chains: review, implications, and directions for future research. DecisionSciences 33 (4), 505–536.

Sanders, N.R., 2008. Pattern of information technology use: the impact on buyer-suppler coordination and performance. Journal of Operations Management 26(3), 349–367.

Schildt, H., Maula, M.V.J., Keil, T., 2005. Explorative and exploitative learning fromexternal corporate ventures. Entrepreneurship Theory and Practice 29 (4),493–515.

Scott, J.E., 2000. Facilitating interorganizational learning with information technol-ogy. Journal Management Information Systems 17 (2), 81–113.

Senge, P.M., 1990. The Fifth Discipline: The Art and Practice of the Learning Organi-zation. Doubleday/Currency, New York.

Shah, R., Shin, H., 2007. Relationships among information technology, inventory, andprofitability: an investigation of level invariance using sector level data. Journalof Operations Management 25 (4), 768–784.

Sidhu, J., Commandeur, H., Volberda, H., 2007. The multi faceted nature of explo-ration and exploitation: value of supply, demand, and spatial search for

innovation. Organization Science 18 (1), 20–38.

Silver, E.A., Pyke, D.F., Peterson, R., 1998. Inventory Management and ProductionPlanning and Scheduling, third ed. John Wiley & Sons, New York.

Småros, J., 2007. Forecasting collaboration in the European grocery sector: observa-tions from a case study. Journal of Operations Management 25 (3), 702–716.

ons M

S

S

S

S

T

Y. Yao et al. / Journal of Operati

munt, T.L., Watts, C.A., 2003. Improving operations planning with learning curves:overcoming the pitfalls of ‘messy’ shop floor data. Journal of Operations Man-agement 21 (1), 93–107.

ong, J., Almeida, P., Wu, G., 2003. Learning-by-hiring: when is mobility morelikely to facilitate interfirm knowledge transfer. Management Science 49 (4),351–365.

tank, T., Daugherty, P.J., Autry, C.W., 1999. Collaborative planning: suppor-ting automatic replenishment programs. Supply Chain Management 4 (2),75–85.

ubramani, M., 2004. How do suppliers benefit from information technology use insupply chain relationships? MIS Quarterly 28 (1), 45–73.

erwiesch, C., Ren, Z.J., Ho, T.H., Cohen, M., 2005. An empirical analysis of forecastsharing in the semiconductor equipment supply chain. Management Science 51(2), 208–220.

anagement 31 (2013) 285–297 297

Thornton, R., Thompson, P., 2001. Learning from experience and learning from oth-ers; an exploration of learning and spillovers in wartime shipbuilding. AmericanEconomic Review 91 (5), 1350–1368.

Tucker, A.L., Nembhard, I., Edmondson, A.C., 2007. Implementing new practices:an empirical study of organizational learning in hospital intensive care units.Management Science 53 (6), 894–907.

VICS, 2007. Implementing Successful Large Scale CPFR Programs and on BoardingTrading Partners Business Process Guide. 〈www.vics.org〉, p. 3.

Wang, H.Y., Li, L.S., Zhu, X., Tan, Z., Xu, B., 2010. Construction of the risk quanti-

tative evaluation system of the system management pattern of supply chain.In: Proceedings of the Proceedings of 2010 International Conference on Risk andReliability Management, pp. 192–197.

Wooldridge, J.M., 2003. Econometric Analysis of Cross Section and Panel Data. MITPress, Cambridge, MA.