Embed Size (px)
Citation preview
This article was downloaded by: [University of North Texas]On: 29 November 2014, At: 20:07Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
International Journal of LogisticsResearch and Applications: A LeadingJournal of Supply Chain ManagementPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/cjol20
Collaborative forecasting: a selection ofpractical approachesJohanna Smårosa Helsinki University of Technology, Department of IndustrialEngineering and Management , FinlandPublished online: 12 May 2010.
To cite this article: Johanna Småros (2003) Collaborative forecasting: a selection of practicalapproaches, International Journal of Logistics Research and Applications: A Leading Journal of SupplyChain Management, 6:4, 245-258
To link to this article: http://dx.doi.org/10.1080/13675560310001626981
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the“Content”) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoever as tothe accuracy, completeness, or suitability for any purpose of the Content. Any opinionsand views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Contentshould not be relied upon and should be independently verified with primary sourcesof information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoever orhowsoever caused arising directly or indirectly in connection with, in relation to or arisingout of the use of the Content.
This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions
International Journal of Logistics: Research and ApplicationsVol. 6, No. 4, 2003
Collaborative Forecasting:A Selection of PracticalApproaches
JOHANNA SMA�ROS*
Department of Industrial Engineering and Management,Helsinki University of Technology, Finland
ABSTRACT The Collaborative Planning Forecasting and Replenishment (CPFR) processmodel developed by the Voluntary Inter-industry Commerce Standards (VICS)association has received significant attention from both practitioners and academics.However, despite promising pilots, the adoption rate of CPFR has been slower thanexpected, especially in Europe. The reason seems to be that the proposed collaborationprocess is currently too labour-intensive for many companies. There is a need forstreamlined approaches to get collaboration started in Europe. This paper presents a casestudy involving a supplier and a retailer collaborating to increase the retailer’s forecastingaccuracy for new product introductions. The approach is based on creating one forecastthat is then shared within the supply chain. Three other streamlined approaches toplanning and forecasting collaboration are also presented on a more general level. Finally,there is a discussion as to how the product life cycle model can be used to select andcombine the most suitable approaches to collaboration in different market situations.
Introduction
In recent years, the concept of inter-company collaboration, especially in the area ofplanning and forecasting, has received significant attention. By developing pro-cesses that make it possible to adjust plans and forecasts in a collaborative fashion,supply chain members aim to make it easier to take into account events such as
*Correspondence: Johanna Sma�ros, Department of Industrial Engineering and Management, P.O. Box
9500, FIN-02015 TKK, Finland; E-mail: [email protected]
International Journal of LogisticsISSN 1367-5567 Print=ISSN 1469-848X online # 2003 Taylor & Francis Ltd
http:==www.tandf.co.uk=journalsDOI: 10.1080=13675560310001626981
Dow
nloa
ded
by [
Uni
vers
ity o
f N
orth
Tex
as]
at 2
0:07
29
Nov
embe
r 20
14
promotions, new product introductions and assortment changes that affect demandthroughout the supply chain (Barratt & Oliveira, 2001).
Collaborative forecasting makes it possible to take advantage of the expertise ofall, or at least several, supply chain members. One benefit that follows from this is areduced reliance on historical records (Helms et al., 2000). Time series methods thatbuild on historical data can forecast changes that follow continuous or recurringpatterns, but cannot accurately forecast the impact of events such as price changesthat happen irregularly (Bowersox & Closs, 1996; Mentzer & Schroeter, 1994).Through collaborative forecasting, a company or department can get access tobetter information on important drivers of future demand, such as promotions.This makes it possible to complement time series forecasting either with regressionanalysis, which examines the relationship between sales and other variables such asadvertising, or with subjective forecasting, which relies on expert opinion (Jain,2000; Mentzer & Schroeter, 1994). Furthermore, working based on one collabora-tively developed, shared forecast reduces the problems related to what Mentzeret al. (1997) call the ‘‘islands of analysis’’ phenomenon, where different groups,departments or companies develop their own forecasts independently of each otheraccording to their own specific needs, and risk ending up with actions based onconflicting plans (Helms et al., 2000).
The most visible undertaking and the igniting spark of the current collaborationboom is the Collaborative Planning Forecasting and Replenishment (CPFR) processmodel developed by the Voluntary Inter-industry Commerce Standards (VICS)association. However, despite successful pilots and significant support from con-sultancies, information technology (IT) companies and electronic market-places, theadoption rate of CPFR has been slower than expected, especially in Europe. Thisgives rise to questions concerning the applicability of CPFR in the European retailenvironment, as well as whether there are more suitable approaches to collabora-tive forecasting than the CPFR process model.
This paper starts with a brief presentation of the VICS association’s CPFRprocess model. Some problems related to the application of the model, especiallyin Europe, are highlighted. Next, the research problem of the paper—finding anddeveloping practical ways for companies to collaborate—is presented. A colla-borative forecasting process currently being piloted by a large European fast-moving consumer goods (FMCG) supplier and one of its retail customers ispresented as an example of an alternative approach to the CPFR process model.Three other approaches to collaboration are also presented on a more generallevel. Finally, there is a discussion as to how the product life cycle model canbe used to select and combine the most suitable approaches to collaboration indifferent market situations.
The VICS CPFR Process Model
The first CPFR project was initiated in the mid 1990s by Wal-Mart and Warner-Lambert and supported by IT companies SAP and Manugistics, as well as con-sulting firm Benchmarking Partners. The project was called CFAR. During theCFAR pilot, Wal-Mart and Warner-Lambert independently estimated demand 6months in advance and compared forecasts and resolved discrepancies on a weeklybasis. In addition, Wal-Mart began placing its orders for the pilot product group,
246 J. Sma�ros
Dow
nloa
ded
by [
Uni
vers
ity o
f N
orth
Tex
as]
at 2
0:07
29
Nov
embe
r 20
14
Listerine mouthwash products, 6 weeks in advance in order to match Warner-Lambert’s 6-week manufacturing lead time. This made it possible for the supplierto manufacture Listerine according to consumer demand and to follow a smootherproduction plan. Wal-Mart, on the other hand, saw an improvement in in-stockposition from 85 to 98% as well as significantly increased sales in combination witha substantial reduction in inventory. During the pilot, the VICS working groupoverseeing the project worked on developing a widely applicable model for col-laborative forecasting, which later evolved into the current CPFR process model(Seifert, 2002).
The CPFR process model contains nine steps (VICS, 1998):
(1) Develop front-end agreement: the parties involved establish the guidelines andrules for the collaborative relationship.
(2) Create joint business plan: the parties involved create a business plan that takesinto account their individual corporate strategies and defined category roles,objectives and tactics.
(3) Create sales forecast: retailer point-of-sales data, causal information and in-formation on planned events are used by one party to create an initial salesforecast, this forecast is then communicated to the other party and used as abaseline for the creation of an order forecast.
(4) Identify exceptions for sales forecast: items that fall outside the sales forecastconstraints set in the front-end agreement are identified.
(5) Resolve=collaborate on exception items: the parties negotiate and produce anadjusted forecast.
(6) Create order forecast: point-of-sales data, causal information and inventorystrategies are combined to generate a specific order forecast that supports theshared sales forecasts and joint business plan.
(7) Identify exceptions for order forecast: items that fall outside the order forecastconstraints set jointly by the parties involved are identified.
(8) Resolve=collaborate on exception items: the parties negotiate (if necessary) toproduce an adjusted order forecast.
(9) Order generation: the order forecast is translated into a firm order by one of theparties involved.
Although CPFR pilots between companies, such as Nabisco and Wegman’s,Kimberly-Clark and Kmart, and Wal-Mart and Sara Lee, have resulted insignificant improvements in product availability, increased sales and reducedinventory (VICS, 1999), the adoption rate of the CPFR process model has beenslower than expected (Barratt & Oliveira, 2001; Frantz, 1999; McCarthy & Golicic,2002). Most implementations involve only two or a few parties, and only a limitednumber of products. Completed large-scale implementations are rare, especially inEurope. Some European companies have even internally prohibited participationin CPFR pilots.
Several reasons have been presented to explain the difficulties in implementingCPFR. Frantz (1999) identified scalability issues, problems related to trust andinformation sharing by adversaries, change management and the difficulty inreaching critical mass as the main problems in implementing CPFR. Barratt &Oliveira (2001) added ineffective use of point-of-sales information, difficulties inmanaging the exception and review processes and lack of common planning ofpromotions and new product introductions to the list. However, according to the
Collaborative Forecasting 247
Dow
nloa
ded
by [
Uni
vers
ity o
f N
orth
Tex
as]
at 2
0:07
29
Nov
embe
r 20
14
three European retailers the author has been working with, the single mostimportant obstacle and the root cause of many of the problems mentioned is, infact, the retailers’ lack of forecasting processes and resources. Since Europeanretailers generally have been able to rely on short order-to-delivery lead times fromsuppliers, developing sophisticated forecasting approaches has not been necessary.Instead they have used efficient but simple time series models for the bulk of theirforecasting, and dealt with larger events, such as major promotions, in an ad hocfashion. The forecasting resources and processes required by the CPFR processmodel are often missing.
One can conclude that for companies such as Wal-Mart with high-quality fore-casting processes and resources already in place, CPFR is a logical and even rela-tively straightforward step to take. In contrast, for the many companies, especiallyretailers, lacking these necessary components, implementing CPFR by the bookoften requires too much work. The companies need to invest not only in colla-boration, but also in developing forecasting processes and capabilities that theycurrently do not have. There is, therefore, a clear need for more practical solutionsto get planning and forecasting collaboration started in Europe.
Research Focus and Methodology
Although there is a remarkable amount of academic literature discussing thebenefits of increased supply chain integration and inter-company collaboration(Andraski, 1998; Bowersox, 1990; Ellinger et al., 1997; Gustin et al., 1995; Kenderdine& Larson, 1988; LaLonde & Powers, 1993; Larson, 1994; Stank et al., 2001), much lesshas been written on how companies can achieve these benefits in practice.
In the area of collaborative forecasting, research has mainly focused on theimportance of developing forecasts in a collaborative fashion, the extent to whichcollaborative forecasting is used, as well as the important organisational issuesrelated to collaborative forecasting (Fosnaught, 1999; Helms et al., 2000; Mentzer &Kahn, 1997; Wilson 2001). Little academic research can be found on how inter-company planning and forecasting collaboration should be set up in practice, whatapproaches there are and how suitable different alternatives are in differentsituations. Although research such as Barratt’s & Oliveira’s (2001) survey on theenablers and inhibitors of implementing CPFR indicates that the suggested processmodel is somewhat problematic, little has been written on how successful com-panies have managed to solve these problems or what alternative ways of colla-borating they employ. Practice-oriented papers, such as McCarthy’s & Golicic’s(2002) case study on the collaborative forecasting processes used by three compa-nies and their supply chain partners, are rare.
In this paper, we try to narrow this gap in supply chain integration theory bypresenting some initial results of our research on finding and developing practicalapproaches to collaborative forecasting that build on information and capabilitiesalready present in consumer goods supply chains. By looking for approachesthat are based on existing information and capabilities, we wanted to avoid theproblems caused by many retailers’, especially European retailers’, limited fore-casting resources and capabilities. We also wanted to develop effective practices inwhich the amount of duplicate work is minimised.
248 J. Sma�ros
Dow
nloa
ded
by [
Uni
vers
ity o
f N
orth
Tex
as]
at 2
0:07
29
Nov
embe
r 20
14
The research reported in this paper consists of several case studies. The casestudy approach was selected to ensure that the research findings be both theore-tically valid as well as of proven practical value (Eisenhardt, 1989). The three casestudies involved a total of two European FMCG suppliers as well as three Europeanretailers. The research consisted of both quantitative and qualitative components.Quantitative analyses based on actual company data were conducted to measureand compare the companies’ forecasting accuracy. The qualitative research mainlyconsisted of repeated brainstorming sessions with the companies’ logistics man-agers, and purchasing and sales personnel. The research was constructive in nature,i.e. the current problems of the case companies were first explored, then potentialsolutions were tested using actual company data, and finally the approaches werepiloted in practice.
The research setting was two consecutive projects at Helsinki University ofTechnology. The first project, ECOMLOG (1999–2002), focused on e-business andIT opportunities in the supply chain. This is where the researchers first faced thesignificant difference between the CPFR and IT-driven collaboration hype andthe practical reality of the companies trying to implement collaborative practices.The ongoing project, NETLOG (2002–2004), focuses on supply and demandnetworks. One of the main research areas in this project is finding practical ways forcompanies to work together to ensure efficient planning, forecasting, material flowsand response to customer requirements in a business network environment.
Next, we shall present the results of a case study examining the forecastingcollaboration between one large European supplier and one of its retail customers.The research was conducted following a constructive case study approach. Theresearcher attended meetings in which the companies discussed their problems,and then helped the companies find one possible way of collaborating to improvethe situation. The approach was first validated using historical data and laterpiloted in practice.
After this, three other approaches to collaboration are presented on a moregeneral level (the details of the studies have or will be published in separate arti-cles). The first case—sharing of early sales data—was inspired by the researchconducted by Fisher et al. (2000) and has recently been piloted with two FMCGsuppliers and one of their retail customers. The second approach—sharing ofassortment information—is currently being tested and developed further with onelarge European FMCG supplier and has been implemented by a European con-fectionery company. The third approach—sharing of exception alerts—is based onstatistical process control theory. Initial tests using historical data have been con-ducted, but the idea still needs to be tested in practice.
Developing One Forecast and Sharing it
It is, indeed, possible to co-operate using a more streamlined approach than theVICS association’s CPFR model. Here one large European consumer packagedgoods supplier and one of its retail customers have started to co-operate to improveforecasting accuracy for new products.
When the companies first started discussing collaborative forecasting, theyquickly realised that their planning and forecasting processes were significantlydifferent. The retailer dealing with tens of thousands of products mainly relies on
Collaborative Forecasting 249
Dow
nloa
ded
by [
Uni
vers
ity o
f N
orth
Tex
as]
at 2
0:07
29
Nov
embe
r 20
14
statistical, time series forecasts, whereas the supplier’s much smaller product rangemakes it possible for it to use expert judgement to adjust statistical forecastsmanually. In addition, the companies work with different planning horizons. Asthe retailer’s order-to-delivery lead time is short—ranging from less than a day to afew days—the emphasis is on short-term forecasts needed for efficient purchasingand inventory management, i.e. for forming the basis of daily purchasing decisions.The supplier, on the other hand, needs to estimate demand several weeks ormonths ahead to cope with long production lead times.
The difference in processes is clearly visible in the way the companies deal withproduct introductions (Figure 1). To enable efficient production, the supplier’s salescompany has to deliver a sales forecast to the factory months before a productlaunch. This initial forecast is based on information on previous launches andexpert judgement. The forecast is then updated when the supplier’s retail custo-mers announce their assortment decisions and distribution plans for the product.After the product launch, the supplier updates its forecast on a weekly basis using acombination of time series methods and qualitative estimates. The retailer, on theother hand, needs the forecast at a much later stage, i.e. when it starts purchasingthe new product. If the product is a new variant of an existing product, the previousproduct’s sales history is used as a basis for time series forecasting. On the otherhand, if the product is completely novel, the retailer usually does not useany forecast at all, but instead purchases the product manually during the firstmonths following the introduction. After some months, the retailer starts usingstatistical forecasts as the basis for purchasing decisions.
The different processes also lead to different results. When comparing forecastsproduced by the supplier with those produced by the retailer, we found significantdifferences. In February 2002, forecast information extracted from the supplier’sand the retailer’s information systems was examined. The comparison seemed toreveal a pattern. The retailer’s forecast was much lower than the supplier’s mainlyfor recently introduced products (‘‘Jan 02’’ products in Figure 2). The retailer’sforecast was also significantly lower for products introduced approximately 5months earlier (‘‘Aug 01’’ products in Figure 2). The forecast was, to some extent,lower for products introduced approximately 9 months earlier (‘‘May 01’’ inFigure 2), although the retailer’s and the supplier’s forecasts for these productswere, in general, similar. The products for which the retailer’s forecast was higher
FIGURE 1. Different Processes for Product Introductions.
250 J. Sma�ros
Dow
nloa
ded
by [
Uni
vers
ity o
f N
orth
Tex
as]
at 2
0:07
29
Nov
embe
r 20
14
than the supplier’s were, on the other hand, mainly products that had been on themarket for 9 months or more (‘‘May 01’’ and ‘‘Old’’ products in Figure 2).
It was anticipated that there would be some differences in the companies’forecasts. After all, the retailer’s statistical forecast looked at history and wasexpected to react more slowly to changes than the supplier’s forward-looking,statistical and qualitative forecast. However, the magnitude of the differences wassurprising, especially for the recently introduced products.
Owing to the detected differences in the companies’ forecasts, it was decided totake a closer look at the products that had been introduced less than 6 months ago(‘‘Jan 02’’ and ‘‘Aug 01’’ in Figure 2) and examine how the forecasts correspondedto the actual outcome.
When examining the forecasting accuracies of the retailer and the supplier forthese products (Figure 3), it became clear that there was room for improvement byboth parties. However, the supplier’s forecast generally performed better than theretailer’s statistical forecast. Although the supplier’s forecast tended to be slightlyinflated, the often significantly understated forecast of the retailer was consideredmuch more problematic. In fact, for some of the most recently introduced products,the retailer did not use the forecast at all, but instead purchased the productsmanually (for these products the statistical forecast generated by the purchasingsystem is included in Figure 3 to enable comparison). It was also seen that thesupplier had more resources to invest in improving its forecasting accuracy.
Based on the results of the analysis, it was suggested that the companies try out anew forecasting approach in which the retailer would use the supplier’s forecast forproduct introductions. The companies agreed to set up a pilot project in which thesupplier would share its forecasts for its product introductions, and the retailer
FIGURE 2. The Retailer’s Forecast as a Percentage of the Supplier’s Forecast (A Low Percentage Meansthat the Retailer’s Forecast was Significantly Lower than the Supplier’s, a High Percentage That the
Retailer’s Forecast was Significantly Higher than the Supplier’s).
Collaborative Forecasting 251
Dow
nloa
ded
by [
Uni
vers
ity o
f N
orth
Tex
as]
at 2
0:07
29
Nov
embe
r 20
14
would use these forecasts as a basis for purchasing decisions until it had enoughhistorical data to enable a switch to time series forecasting.
There is not as yet quantitative data available from the pilot, but the companiesconsider it a step in the right direction and have already decided to continuetheir co-operation. Based on the pre-pilot quantitative analysis and the way theco-operation process is set up, both companies are expecting to benefit fromthe collaboration. Two main benefits for the retailer have been identified. Firstly, asthe supplier’s forecast can be used to set inventory management parameters fornew products, manual buying is no longer needed at the beginning of a product’slife cycle. Secondly, as the forecasting accuracy improves, the retailer’s service levelto its stores increases. The improved store service is also the greatest benefitfrom the supplier’s point of view. The supplier expects the reduced stock-out risk totranslate into increased sales and aims to reinforce this effect by further developingits forecasting capabilities.
Since the process is based on sharing existing information, minimal additionalwork is required from either party. The supplier’s forecasts can easily be extractedfrom the supplier’s information systems and sent to the retailer by e-mail or, asplanned in this case, using an existing information exchange portal. The idea is toattain a higher return on the forecasting effort performed by the supplier by usingthe available information more extensively.
Other Opportunities for Streamlined Collaboration
The example presented above of a supplier sharing forecast information on newproducts with one of its retail customers is not the only alternative to the CPFRprocess model. Other streamlined approaches include co-operation through thesharing of sales data, co-operation through the exchange of planning informationand co-operation through the sharing of exception alerts.
FIGURE 3. Supplier and Retailer Forecasts Plotted Against Actual Sales (Points Below the SalesLine Indicate That the Forecast was too Small, Points Above the Sales Line Indicate That the Forecast was
too High).
252 J. Sma�ros
Dow
nloa
ded
by [
Uni
vers
ity o
f N
orth
Tex
as]
at 2
0:07
29
Nov
embe
r 20
14
Sharing of Sales Data
Fisher et al. (2000) have done extensive research on the use of early sales data as ameans of rapidly adjusting forecasts for improved accuracy. Their research revealsthat early sales figures can show very accurately where sales are headed, making itpossible to attain significant improvements in forecasting accuracy by updatinginitial forecasts based on early sales information. For example, in a case from theapparel industry, it was found that forecasting accuracy could be increased from a45% accuracy level to a 92% accuracy level by updating the initial forecast based onthe first 2 weeks’ sales. (Fisher & Raman, 1996; Fisher et al., 1994; 2000)
The author has recently piloted a similar approach in co-operation with twoFMCG suppliers and one retailer. The retailer shared point-of-sales data, i.e. salesdata collected at the checkout counters in its stores, with the suppliers on their newproduct introductions 2, 4 and 6 weeks after the product launches. The suppliersused this information to monitor how fast demand for their products was growing,when the growth started to slow down and at what level demand finally settled.Based on this information, they were able to update their forecasts rapidly. Sinceboth suppliers tend to over-forecast rather than under-forecast the sales of theirnew products in order to assure supply, the point-of-sales information was, formany products, used to lower rapidly and in a controlled manner forecast andinventory levels. However, for a few products, forecasts were also raised, as theproducts’ sales surpassed expectations. The rapid detection of these surprisinglyhigh sales made it possible for the supplier as well as the retailer to avoid potentialout-of-stock situations and maintain their service levels. Both the suppliers and theretailer found the results of the information-sharing pilot very encouraging andthey are currently adjusting their forecasting processes to make this type ofco-operation part of their routines. The retailer is also planning to start co-operatingin a similar way with other suppliers.
The results of this pilot study indicate that an efficient way to co-operate is forthe retailer to share point-of-sales information on early sales of new products withits suppliers. This type of co-operation does not significantly increase the retailer’sforecasting workload as the necessary information already exists and the retailermerely needs to convey the information rather than analyse it. Yet, the co-operationenables notable improvements in the suppliers’ forecasting accuracy, leading tomore reliable and efficient supply of products. By taking advantage of the sup-pliers’ more accurate forecasts, as in the case example presented above, the retailercan also benefit from the improved forecasts in its own operations, for example incontrolling stock levels and store replenishments.
Exchange of Planning Information
Another way of co-operating is to exchange planning information on, for example,upcoming assortment changes, promotions and price changes. Although thistype of co-operation is present in most retailer–supplier relationships, importantdevelopment opportunities remain. By making relevant information available earlyenough, and by developing ways of dealing with the information effectively, i.e.translating it into accurate forecasts, the value of this kind of information sharingcan be significantly increased.
Collaborative Forecasting 253
Dow
nloa
ded
by [
Uni
vers
ity o
f N
orth
Tex
as]
at 2
0:07
29
Nov
embe
r 20
14
Researchers at Helsinki University of Technology have developed a methodfor translating plans into forecasts using sales profiles, i.e. mathematical func-tions describe how a category’s sales are divided between top–selling products,average sellers and niche products within the category. The profiles, thus, describehow a product’s position or rank within the category is related to its share oftotal category sales. Since the sales profiles of many categories are stable, i.e.they do not change although changes, such as promotions, occur within thecategory (see Figure 4), forecasts can be developed simply by ranking the pro-ducts within a category and using the category’s sales profile to establish thecorresponding shares of total sales. Instead of developing forecasts product byproduct, the forecaster is able to work with a whole category at a time, adjustingproduct ranks when necessary and using the sales profile to derive auto-matically product-level sales forecasts. Initial results indicate that the approachenables more accurate forecasts to be developed with only a moderate amountof work. This ranking approach to forecasting is described in more detail byHolmstro€m et al. (2002) and Fra€mling & Holmstro€m (2000). The method is currentlybeing tested and developed further in co-operation with a large EuropeanFMCG supplier and it has recently been implemented by a European confec-tionery company.
When suppliers improve their ability to translate planning information intoforecasts accurately, the focus of collaboration shifts from forecasts to plans. Whenplans, rather than forecasts, are discussed and compared, several steps in the CPFRprocess can be removed. The retailer does not necessarily need to use resources totranslate all plans into forecasts, but can instead transfer already available planninginformation to the supplier, and in this way improve supply chain visibility andability to respond to demand changes.
FIGURE 4. The Monthly Sales Profiles of This Example Product Category are Stable, Despite Promotionsand Product Introductions.
254 J. Sma�ros
Dow
nloa
ded
by [
Uni
vers
ity o
f N
orth
Tex
as]
at 2
0:07
29
Nov
embe
r 20
14
Sharing of Exception Alerts
Access to accurate sales information, ideally point-of-sales information, makes itpossible to construct statistical tools to monitor automatically the relationshipbetween actual and forecast sales in order to identify potential discrepancies.
Statistical process control has been used in quality control for decades andseveral different monitoring tools, e.g. different types of control charts, have beendeveloped to identify when processes are out of control (see, e.g. Mitra, 1993). Thesame tools have also been applied to forecasting, although to a lesser extent. Inforecasting, control charts can be used to detect automatically statistically sig-nificant differences between actual and forecast sales, to detect when the cumula-tive forecast error has grown alarmingly large or, for example, when the differencebetween forecast and actual sales starts to follow a distinct pattern (Gardner, 1983;Stevenson, 2002). Statistical process control is especially suited for mature productswith stable demand for which forecasts can be developed statistically and sub-stantial changes in demand are unlikely but possible.
So far, statistical process control and control charts have mainly been used incompanies’ internal operations. However, the approach also presents an opportu-nity for co-operation. Retailers typically have access to the most accurate and, often,least fluctuating sales information (Lee et al., 1997). By monitoring these sales dataand conveying alerts generated by the monitoring tools to its suppliers, a retailercan, without having to commit its own resources to forecasting, offer the suppliersan opportunity to react to changes in demand or incorrect forecasts. In exchange forits investment in the monitoring system, the retailer can, as described in the pre-vious cases, demand that suppliers analyse the monitoring information and pro-vide more accurate forecasts to the retailer. Initial tests of the approach usinghistorical data have produced promising results, but practical testing still needs tobe done before the value of this collaboration approach can be established. Theinterest among both suppliers and retailers is, though, considerable.
Conclusion: One Size does not Fit all, Several CollaborationModels are Needed
By developing a common, standardised collaboration model, the VICS associationaimed to make large-scale, computer-aided collaboration easier. However, as themodel was based on work done by Wal-Mart, perhaps the most industriallyoperating retailer in the world, and further developed in the North American retailclimate where retailers have also previously engaged in forecasting, the applicationof VICS style CPFR in Europe is far from simple. Applying the CPFR process modelin Europe would, in many cases, require that retailers replace their basic statisticalforecasts with more sophisticated forecasting models and qualitative forecasting,something that requires significant resources. The CPFR model is also based oncomparisons and checks of independently developed forecasts rather than onidentifying who has access to the best information and knowledge and leveragingthis in the rest of the supply chain. This means that the model includes a certainamount of duplicate work. It is, therefore, questionable whether it will pay forEuropean retailers to commit resources to CPFR when they can achieve goodresults through a more streamlined co-operation with suppliers. Regardless of
Collaborative Forecasting 255
Dow
nloa
ded
by [
Uni
vers
ity o
f N
orth
Tex
as]
at 2
0:07
29
Nov
embe
r 20
14
where one stands on this issue, it is clear that developing the retailers’ forecastingprocesses and resources takes time. So, be it a temporary transition phase or theultimate goal, there is a need for a streamlined process to get retailer—suppliercollaboration started in Europe.
The aim of the research was to answer a question posed by several companies:‘‘The CPFR process model is too heavy for us—do there exist more practicalalternatives?’’ The examples presented in this paper show that improved fore-casting accuracy can be achieved through very streamlined co-operation processes.These processes focus on sharing the most useful information available, translatingthe information into accurate forecasts and sharing the forecasts in the supplychain. What is the most useful information depends, among other things, on theproduct’s life cycle phase. Access to point-of-sales information can, for example, beextremely useful for suppliers when introducing new products, but the value ofthis information is much smaller for mature products with stable demand.Similarly, discussions and meetings between suppliers and retailers are usefulwhen changes, such as assortment changes, price changes or promotions areplanned, but not all products need the same amount of attention all of the time.
The product life cycle model offers a natural framework for focusing on theessentials (Figure 5). When new products are introduced, the supplier needs toproduce accurate forecasts to enable efficient use of limited manufacturing capa-city. This means that the supplier is willing to invest a lot of time and effort inforecasting even before the product has been launched. This forecast can in manycases be used by the retailer as well. After the launch, the supplier still has a strongincentive to monitor sales and update forecasts, and is usually more than willing toanalyse point-of-sales data and turn them into accurate forecasts that can be usedby the retailer as well. In the maturity phase of the life cycle, co-operation can focuson managing special events, for example by sharing assortment and promotioninformation. For products with stable demand, statistical monitoring and exchangeof exception alerts are good approaches.
A product’s life cycle phase is, of course, not the only determinant of what kindof collaboration retailers and suppliers should engage in. The supplier’s cap-abilities, the relationship between the collaborating companies, the criticality andvalue of the products, among other factors, have an important impact on what
FIGURE 5. There are Different Opportunities for Collaboration in a Product’s Different Life Cycle Phases.
256 J. Sma�ros
Dow
nloa
ded
by [
Uni
vers
ity o
f N
orth
Tex
as]
at 2
0:07
29
Nov
embe
r 20
14
collaboration approach, if any, to choose. However, the life cycle framework pre-sents an important illustration of why several collaboration approaches are needed,and why companies should apply them flexibly.
Limitations and Further Research
The constructive case study approach proved fruitful in this research. Since the aimof the research was to identify and construct practical approaches to collaboration,the opportunity to test ideas on company representatives, to use actual companydata and pilot the ideas was invaluable. The case study approach, however, alsointroduced limitations. Owing to the labour-intensive and time-consuming natureof the research, only a small number of case companies could be included.Additional research is needed to establish the generalisability of the results todifferent markets and industries as well as to different types of companies.
The findings of our research support earlier work by, for example, McCarthy &Golicic (2002). Both studies show that collaboration can be beneficial and thatcompanies need to find their own ways of collaborating. Rather than developingone single process model, a framework for selecting the right approach is needed.Still, considerable research efforts are required before we can reach the goal ofbeing able to understand what kind of collaboration is successful in differentsituations. There is currently very little descriptive research available on howcompanies collaborate in logistics planning and forecasting. Practice-oriented casestudies as well as surveys from both the USA and Europe would be very valuable.In addition, further testing and development of the collaboration approaches pre-sented here as well as completely new ones are needed.
Finally, we would like to encourage critical discussion regarding both the VICSassociation’s CPFR process model and the idea of combining more streamlinedcollaboration approaches presented here. In particular, the practical applicability ofthe different approaches should be examined thoroughly. In this paper, forexample, the investments required by the presented collaboration models have notbeen discussed in detail. The labour and technology requirements have only beentouched on since it was assumed that the resource requirements would anyhow besignificantly smaller than if the full CPFR process had been employed. However, toenable comparison and adoption of the different collaboration models, thoroughinvestment and benefit calculations will be of great importance.
REFERENCES
ANDRASKI, J.C. (1998) Leadership and the realization of supply chain collaboration, Journal ofBusiness Logistics, 19, (2), pp. 9–11.
BARRATT, M. & OLIVEIRA, A. (2001) Exploring the experiences of collaborative planning initiatives,International Journal of Physical Distribution & Logistics Management, 31, (4), pp. 266–289.
BOWERSOX, D.J. (1990) The strategic benefits of logistics alliances, Harvard Business Review, 68, (4),pp. 36–45.
BOWERSOX, D.J. & CLOSS, D.J (1996) Logistical Management: The Integrated Supply Chain Process(Singapore, McGraw-Hill).
EISENHARDT, K.M. (1989) Building theories from case study research, Academy of ManagementReview, 14, (4), pp. 532–550.
Collaborative Forecasting 257
Dow
nloa
ded
by [
Uni
vers
ity o
f N
orth
Tex
as]
at 2
0:07
29
Nov
embe
r 20
14
ELLINGER, A.A., DAUGHERTY, P.J. & GUSTIN, C.M. (1997) The relationship between integratedlogistics and customer service, Transportation Research Part E: Logistics and TransportationReview, 33, (2), pp. 129–138.
FISHER, M.L., HAMMOND, J.L., OBERMEYER, W.R. & RAMAN, A. (1994) Making supply meet demandin an uncertain world, Harvard Business Review, 72, (3), pp. 83–93.
FISHER, M.L. & RAMAN, A. (1996) Reducing the cost of demand uncertainty through accurateresponse to early sales, Operations Research, 44, (1), pp. 87–99.
FISHER, M.L., RAMAN, A. & MCCLELLAND, A.S. (2000) Rocket science retailing is almost here: Areyou ready?, Harvard Business Review, 78, (4), pp. 115–124.
FOSNAUGHT, K. (1999) The strategic power of consensus forecasting: Setting your organization upto win, Journal of Business Forecasting Methods & Systems, 18, (3), pp. 3–7.
FRANTZ, M. (1999) CPFR pace picks up, Consumer Goods, www.consumergoods.com=archive=JanFeb99.
FRA€MLING, K. & HOLMSTRO€M, J. (2000) A distributed software for collaborative sales forecasting,Proceedings of the Management and Control of Production and Logistics MCPL 2000 Conference,Grenoble.
GARDNER, E.S., JR (1983) Automatic monitoring of forecast errors, Journal of Forecasting, 2, pp. 1–21.GUSTIN, C.M., DAUGHERTY, P.J. & STANK, T.P. (1995) The effects of information availability on
logistics integration, Journal of Business Logistics, 16, (1), pp. 1–21.HELMS, M.M., ETTKIN, L.P. & CHAPMAN, S. (2000) Supply chain forecasting—collaborative
forecasting supports supply chain management, Business Process Management Journal, 6, (5), pp.392–407.
HOLMSTRO€M, J., FRA€MLING, K., KAIPIA, R. & SARANEN, J. (2002) Collaborative planning forecastingand replenishment: new solutions needed for mass collaboration, Supply Chain Management:An International Journal, 7, (4), pp. 136–145.
JAIN, C.L. (2000) Editorial: Which forecasting nodel should we use?, Journal of Business ForecastingMethods & Systems, 19, (3), pp. 2–4.
KENDERDINE, J.M. & LARSON, P.D. (1988) Quality and logistics: A framework for strategicintegration, International Journal of Physical Distribution and Materials Management, 18, (7),pp. 5–10.
LALONDE, B.J. & POWERS, R.F. (1993) Disintegration and re-integration: Logistics of the 21stcentury, International Journal of Logistics Management, 4, (2), pp. 1–12.
LARSON, P.D. (1994) An empirical study of inter-organizational functional integration and totalcost, Journal of Business Logistics, 15, (1), pp. 153–169.
LEE, H.L, PADMANABHAN, V. & SEUNGJIN, W. (1997) The bullwhip effect in supply chains, SloanManagement Review, 38, (3), pp. 93–102.
MCCARTHY, T.M. & GOLICIC, S.L. (2002) Implementing collaborative forecasting to improve supplychain performance, International Journal of Physical Distribution & Logistics Management, 32, (6),pp. 431–454.
MENTZER, J.T. & KAHN, K.B. (1997) State of sales forecasting systems in corporate America, Journalof Business Forecasting Methods & Systems, 16, (1), pp. 6–13.
MENTZER, J.T., MOON, M.A., KENT, J.L. & SMITH, C.D. (1997) The need for a forecasting champion,Journal of Business Forecasting Methods & Systems, 16, (3), pp. 3–8.
MENTZER, J.T. & SCHROETER, J. (1994) Integrating logistics forecasting techniques, systems,and administration: The multiple forecasting system, Journal of Business Logistics, 15, (2),pp. 205–225.
MITRA, A. (1993) Fundamentals of Quality Control and Improvement (New York, MacMillan).SEIFERT, D. (2002) Collaborative Planning Forecasting and Replenishment—How to Create a Supply Chain
Advantage (Kevelaer, Galileo Press GmbH).STANK, T.P., KELLER, S.B. & DAUGHERTY, P.J. (2001) Supply chain collaboration and logistical
service performance, Journal of Business Logistics, 22, (1), pp. 29–48.STEVENSON, W.J. (2002) Operations Management (New York, McGraw-Hill).VICS (1998) Collaborative Planning Forecasting and Replenishment Voluntary Guidelines,
www.cpfr.org=Guidelines.htmlVICS (1999) Roadmap to CPFR: The Case Studies, http:==208.143.22.52=cpfr_pdfWILSON, N., JR (2001) Game plan for a successful collaborative forecasting process, Journal of
Business Forecasting Methods & Systems, 20, (1), pp. 3–6.
258 J. Sma�ros
Dow
nloa
ded
by [
Uni
vers
ity o
f N
orth
Tex
as]
at 2
0:07
29
Nov
embe
r 20
14
