See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/225215972

Applying Six Sigma methodology to collaborative forecasting

Article in The International Journal of Advanced Manufacturing Technology · November 2007

DOI: 10.1007/s00170-007-1276-5

CITATIONS READS

37 1,914

2 authors, including:

Fu-Kwun Wang
National Taiwan University of Science and Technology
187 PUBLICATIONS 4,805 CITATIONS

SEE PROFILE

All content following this page was uploaded by Fu-Kwun Wang on 07 January 2015.

The user has requested enhancement of the downloaded file.

Int J Adv Manuf Technol (2008) 39:1033–1044
DOI 10.1007/s00170-007-1276-5

ORIGINAL ARTICLE

Applying Six Sigma methodology to collaborative forecasting

K. K. Chang & F. K. Wang

Received: 26 February 2007 / Accepted: 9 October 2007 / Published online: 9 November 2007
# Springer-Verlag London Limited 2007

Abstract Collaborative planning, together with forecasting Collaboration between trading partners has been recognized
and replenishment (CPFR), is a processing tool that as one of the most efficient methods for improving forecast
improves supply chain management (SCM). The majority accuracy and increasing service level, while reducing costs.
of studies done on CPFR have focused on identifying the The collaborative planning, forecasting and replenishment
advantages and future developments of CPFR, along with (CPFR) model from the voluntary inter-industry commerce
providing explanations for the implementation stages of standards (VICS) is a process tool to improve supply chain
CPFR. Six Sigma methodology is a method that can lead to management (SCM). The concept of such a model is to
a continuous decrease in process variances. In this paper, share information and risk among the supply chain partners
we applied Six Sigma methodology and proposed a and improve forecasting accuracy in doing so. The CPFR
continuous improvement model on different phases of model includes collaborative planning, execution and
CPFR. We used a case study to demonstrate how to monitoring by all members. The benefits of implementing
improve the performance of collaborative forecasts. The CPFR for trading partners are mentioned in VICS [1]. For
results show that the proposed improvement model can examples, CPFR implementation can improve forecasting
effectively improve the accuracy of collaborative forecasts. accuracy by 10%–40%, reduce inventory cost by 10%–
40%, save between 0.3% and 1% in transportation cost and
Keywords Six Sigma methodology . Collaborative increase customer service level by 0.5%–4%. CPFR has
forecasting also been known to reduce out of stock occurrences,
shorten the lead time from the time of order to replenish-
ment, reduce purchase returns and strengthen relationships
1 Introduction between supply chain members. As one might expect,
CPFR is coveted by decision-makers and engineers
The majority of supply chain models employed these days primarily because it improves forecast.
lack a coordination framework; they are unable to match Crum and Palmatier [2] explored why businesses are
the supply and demand that occurs between the enterprises. cautious in embracing collaboration, and subsequently
Downstream firms fail to provide the right forecasts, while investigated the conditions that will need to be met if
upstream firms are forced to raise inventory levels to demand collaborations and CPFR become the standard
compensate for such changes; the result is an increase in business practices among partners today. Based on field-
inventory and a reduction in operational performance. work conducted on hundreds of supply chain partners,
Finley and Srikanth [3] identified the requirements of
successful supply chain collaborations, collaborations that
K. K. Chang : F. K. Wang (*) required strategic alignment among contiguous supply
Department of Industrial Management, chain participants within a unified channel. They found
National Taiwan University of Science and Technology,
No. 43 Keelung Road, Sec. 4,
that the partners should apply quasi-real-time connectivity
Taipei, Taiwan 106, Republic of China and channel-wide metrics focused on downstream demand
e-mail: fukwun@mail.ntust.edu.tw in order to enable the delivery of tangible business benefits.
1034 Int J Adv Manuf Technol (2008) 39:1033–1044

Hutchins [4] pointed out that quality tools can be applied [1]. After a revision was done of the six-year-old model, the
to supply chain areas, such as mapping of supply chain execution process contained four stages and eight steps.
processes, process standardization, process variation con- The new CPFR business model is a continuous improve-
trol, supplier certification, total customer satisfaction, ment model and focuses on effects. Table 1 shows that the
auditing, preventive and corrective action and supply differences between CPFR process models of 1998 and
performance measurement. Moreover, Snee [5] pointed 2004. Under CPFR, both trading partners develop a joint
out that Six Sigma is an action oriented and focuses on business plan, which includes a promotion calendar. The
processes used in customer service and defect reduction buyer and the seller agree on a joint sales forecast and a
through variation reduction and improvement goals. Based joint order forecast. The joint sales forecasts can drive
on the above discussion, we found that Six Sigma can be production scheduling, distribution planning and store
employed to improve forecasting accuracy in CPFR. activity planning. Any deviations from any of the forecasts
Therefore, we proposed a continuous improvement model that are beyond an agreed-upon threshold are defined as
in CPFR. This paper is organized as follows. Section 2 exceptions; they induce collaborative actions by both
reviews the related researches on CPFR and Six Sigma parties to realign the planning for the channel. Order
methodology. Section 3 describes the proposed model forecasts are also checked for exceptions and then realigned
which is a continuous improvement model on different to generate the actual replenishment orders. Many CPFR
phases of CPFR implementation. Section 4 reports a case pilot initiatives involved a limited number of items and
study to reveal ways in which one can improve the suppliers, with both the buyer and the seller providing
performance of collaborative forecasts. The final section forecasts. Table 2 lists four scenarios of collaboration role
presents the conclusions and directions for future research. alternative.
Fliedner [6] explained the CPFR processes, cited the
benefits achieved when employing the CPFR processes,
2 Literature review and identified possible obstacles that may occur when
utilizing the CPFR process. A collaborative forecasting
A set of CPFR guidelines have been prepared and model can be implemented with an enterprise resources
published by VICS. In the guidelines defined by VICS in planning (ERP) system. Lin et al. [7] proposed a suitable
1999, the execution process had three stages and nine steps CPFR implementation process for the mechanical wood

Table 1 The CPFR process models of 1998 and 2004 [1]

Int J Adv Manuf Technol (2008) 39:1033–1044 1035

Table 2 Collaboration role alternatives [1] 4. Improve: This module focuses on how the Six
Scenario Sales Order Order Sigma improvement technology could be developed.
forecast forecast generation The module also identifies critical factors that may
arise from the control process. The method of
Scenario A (conventional order Buyer Buyer Buyer solving the problem is tested and a plan is devised
management)
to mitigate it.
Scenario B (supplier managed Buyer Seller Seller
5. Control: This module identifies the controls that must be
inventory)
Scenario C (co-managed inventory) Buyer Buyer Seller in place to sustain the benefit of the new process. It
Scenario D (vendor-managed Seller Seller Seller standardizes the improved steps to keep the process
inventory) stable.
Wyper and Harrison [12] proposed applying the Six
Sigma methodology to the human resource (HR) function
carving industry. In addition, they suggested that CPFR
in order to urge improvement in HR processes. Ehie and
should focus on collaboration between the buyer and the
Sheu [13] integrated Six Sigma and the theory of
seller to establish collaborative forecasting procedures.
constraints (TOC) to improve the production system. It
Holmstrom et al. [8] presented a forecasting solution for
was estimated that the project resulted in a total saving of
CPFR that is based on category forecasting using the
$200,000 per year, exceeding the average savings from a
retailers’ rank and share information as initial data to
typical Six Sigma project at the plant. Motwani [14]
forecast the stock keeping unit (SKU) level. Holweg et al.
reported that the Dow Chemicals Company applied Six
[9] concluded that the effectiveness of supply chain
Sigma to improve its process, and the project resulted in a
collaboration depends upon two major factors: the level to
total saving of $1.5 billion from 1999 to 2003. Blakeslee Jr.
which it integrates internal and external operations, and the
[15] indicated that aligning organizational components
level to which the efforts are aligned to the supply chain
within the company (leadership, strategy, people and
setting in terms of geographical dispersion, demand pattern
technology) is vital in implementing Six Sigma. Moreover,
and product characteristics.
he provided seven key principles critical to ensuring that
Six Sigma methodology is a logical and systematic
companies reap the benefit of improved business perfor-
approach to achieving continuous process improvements.
mance from their investment in Six Sigma. Wang et al. [16]
This process improvement methodology was developed
applied Six Sigma methodology to supplier development in
in the 1980s in Motorola’s high-volume manufacturing
SCM. In their study, they developed an application
environment. Note that the sigma quality level offers an
guideline for the assessment, improvement and control of
indicator of how often defects are likely to occur when
quality in SCM using Six Sigma methodology. The
using a particular process; a higher sigma quality levels
methodology can increase the supplier’s performance and
indicate a process that is less likely to create defects. The
improve the quality of all supply chain processes, thereby
framework consists of five integrated modules: define,
reducing costs and improving customer service. Tong et al.
measure, analyze, improve and control (DMAIC).
[17] applied the DMAIC approach to improve the sigma
Chowdhury [10] pointed out that Six Sigma is a manage-
level of the screening process in the manufacture of surface
ment philosophy that emphasizes how to avoid mistakes
mounted printed circuit boards. Mahesh et al. [18]
and overlapped work. Hoerl [11] considered the difference
presented a Six Sigma methodology for benchmarking of
between Six Sigma and previous management methods.
rapid prototyping & manufacturing processes. Knowles et
The definitions of the DMAIC modules are given as
al. [19] integrated Six Sigma, the balance scorecard and
follows.
SCOR into an SCCIM model. The model has two phases:
1. Define: This module defines the required activities and strategic cycle and operational improvement cycle. The
the key process. DMAIC cycle was introduced completely or partially
2. Measure: This module measures the capability of the depending on the operational feature. The purpose of the
process. It measures the defects created by the process strategic cycle is to ensure maximum benefits from the
and quantifies the difference between the output of the process. The purpose of the operational cycle is to
process and the goal of the improvement goal. continuously decrease the variance in the supply chain.
3. Analyze: This module analyses when and where defects Spector [20] reported that auto parts production could be
occur. It applies statistical methods to verify the cause improved by the application of lean_Six Sigma. In
of the process problem. The purpose of this module is particular, work in process (WIP) inventory was reduced
to evaluate the current performance as the foundation of by more than 70%, cycle time was halved and customer
solving the process problem. service level was increased up to 95%.
1036 Int J Adv Manuf Technol (2008) 39:1033–1044

3 Improvement model to produce the historical and holdout forecasts. To evaluate

the forecasting accuracy, the mean absolute percentage
There are four stages in CPFR implementation. The first error (MAPE) is identified as a critical-to-quality (CTQ)
stage is to create a joint business plan. It is extremely characteristic that needs to be controlled by the collabora-
important to determine the items necessary for the tive forecasting process.
collaboration, along with a willingness to conform to
the collaboration rules. The second stage is demand and 3.2 Measure module
supply management, which is at the core of CPFR. The
major tasks involved in the second stage include The purpose of this module is to collect the sales data of all
forecasting point-of-sales (POS) demand and making the products involved in the collaborative forecasts from a
order shipment requirements that will need to be met POS system. It also includes cause and effect information
during planning. Consumption data are used here to and event data. Only when such is collected can we proceed
generate sales forecasts. Using information from the sales to the next module.
forecasts and inventory policies, we can determine when
each store needs to release an order to the distribution 3.3 Analyze module
center. This information is then used to generate an order
forecast for the manufacturer. The third and fourth stages The purpose of this module is to evaluate the current
are execution and analysis, respectively. In this section, performance and to re-evaluate the standards to be met for
we developed an improvement model that applies Six forecasting accuracy. A highly regarded standard such as
Sigma methodology to CPFR in the second stage of MAPE is used to examine the sales data collected in the
CPFR (see Fig. 1). The details of the DMAIC cycle are measure module. In the analyzing model, we analyze the
given as follows: data pattern by using the autocorrelation function (ACF)
and the partial autocorrelation function (PACF); in doing
3.1 Define module so, we can determine whether or not the data collected is
stationary and of value. After analyzing the data pattern, we
The major purpose of this module is to improve the find the time series data by way of a simple exponential
forecasting accuracy of collaborative forecasts. In the smoothing model, which allows us to forecast future sales.
collaborative forecasting, we need to define what type of The MAPE values of all products on the control chart are
data from the product sales data are needed for forecasting, used to monitor the forecasting accuracy. Once the process
and the time period in which such a sale was done in order is under control, the current data can be used to evaluate the

Fig. 1 An improvement model

for CPFR
Stage 1 Stage 2 Stage 3 Stage 4
Strategy & Planning Demand & Supply Management Execution Analysis

Define : the key

process
(collaborative
forecast)

Control : sustain
the
Measure : collect
collaborative
forecast with the sales data
high accuracy

Improve : adopt
advanced Analyze : examine
forecasting the forecasting
model to improve accuracy by a
forecasting control chart
accuracy
Int J Adv Manuf Technol (2008) 39:1033–1044 1037

process performance. Further, this module identifies and 4 Case study

prioritizes improvement areas that are out of control. Once
these areas have been prioritized, we should consider the 4.1 Company and processes
areas that must receive immediate attention and proceed to
the next module. This case study was conducted in a company called
company-A which is a part of the paper industry in Taiwan.
3.4 Improve module This company has three different types of products —
cultural paper, family paper and industrial paper. These
This module focuses on identifying and implementing different types of products have different supply chain
changes that may occur, which would improve the models (see Fig. 2). Since industrial paper is customized
overall forecasting accuracy. The analysis of the fore- according to a customer’s needs, it is difficult to forecast
casting accuracy showed that some products’ MAPE each customer’s order. Cultural paper and family paper
values were not within the limits. That is, the collabo- have similar production and marketing processes. Further,
rative forecasts for some products were not satisfactory. their combined revenues were merged over 60% of the total
In order to improve the forecasting accuracy, ARIMA or revenue. Therefore, this case study focuses on cultural
some advanced forecasting models was used as a core paper and family paper.
tool. For example, we can change the original forecasting To improve partnership with its retailers and competition
method from a simple exponential smoothing model to in this paper market, this company used the concept of
an ARIMA model. After the first improve cycle, we CPFR to establish its own collaborative forecast process.
return to the measure module to repeat this cycle until Following business reengineering, the CPFR processes
the collaborative forecasts for all products are under from the downstream retailer to the upstream supplier (see
control. Then, we can formulate the final producing plan Figs. 3 and 4) are given as follows:
and collaborative replenishment plan according to the 1. The retailers transfer the sales forecasts of all collabo-
final collaborative forecast. rative products from the POS system to the sales
database.
3.5 Control module 2. The regional salespersons aggregate the sales of all
collaborative products and report them to the regional
In order to improve forecasting accuracy in demand and manager.
supply management, the variation between what is fore- 3. The regional manager gathers the sales forecasts and
casted and the actual sales should be monitored using produces the regional order forecast using the system
control charts. If the difference value exceeds the control and personal experience.
limit, a revision of the forecast model should be performed 4. The head office gathers the order forecast of all
immediately to formulate a suitable collaborative demand collaborative products in every region and then
and supply course of action. produces the whole order forecast.

Fig. 2 The supply chain model

of company-A A company Demand plan
(Sales Forecast)

Cultural paper Product schedule and

replenishment plan Retailer
department
MTS

Demand plan
(Sales Forecast)

Family paper Product schedule and

Wood pulp replenishment plan Retailer
department
MTS

Certainly Order
(size, strength,numbers)

Industrial paper Enterprise

Waste paper pulp Product schedule
department customer
MTO
1038 Int J Adv Manuf Technol (2008) 39:1033–1044

Fig. 3 The CPFR processes of

company-A Sub-regional sales Retailer
forecast sales forecast

Summary sub-
regional sales
forecast

System
forecasting

Produce regional Regional

sales forecast Target plan

Meeting
result

Summary regional
sales forecast

System
forecasting

Produce final Webplan

forecast

Meeting
result
Forecast posting to
APS

Order generation

Replenishment
maintain

End

5. The regional managers transfer the region order to the From Table 2, we can identify that company-A belongs
retailers according to each retailer’s past sales data. The to Scenario B. In this situation, company-A is in charge of
retailers need to confirm the order. the order forecasting and order generation. Even after
6. After the order is confirmed, the data are transferred to executing CPFR, the accuracy of the collaborative forecasts
an advanced planning and scheduling (APS) system failed to meet the target. Thus, company-A decided to apply
that calculates the inventory and route to determine the DMAIC approach to improve its forecasting accuracy.
when the order should be produced. Finally, the Currently, the forecasting models are fitted by a simple
replenishment plan is established and sent to the model such as moving average, weighted moving average
individual retailers. or simple exponential smoothing.

Fig. 4 Product’s forecast model 70.00%

control chart before 1st improve
60.00%
cycle
50.00%
History
40.00% MAPE

UCL
30.00%

20.00%

10.00%

0.00%
Product1 Total

Product2 Total

Product3Total

Product4 Total

Product5 Total

Product1 Combine

Product2 Combine

Product3 Combine

Product4 Combine

Product5 Combine
Table 3 MAPE values before 1st improve cycle

Item Product1 total Product2 Product3 Product4 Product5 Product1 Product2 Product3 Product4 Product5 Mean Standard UCL
total total total total combine combine combine combine combine deviation
Forecast Weighted Exponential Exponential Exponential Exponential Exponential Exponential Weighted Weighted Exponential
model moving smoothing smoothing smoothing smoothing smoothing smoothing moving moving smoothing
average average average
Weighted Weighted Exponential Weighted
moving moving smoothing moving
average average average
Exponential
smoothing

History 14.54% 20.91% 1.99% 57.27% 2.53% 13.02% 19.99% 1.92% 58.78% 2.74% 19.37% 20.55% 39.92%
MAPE
Int J Adv Manuf Technol (2008) 39:1033–1044

Holdout 16.61% 13.37% 3.49% 49.92% 2.25% 8.70% 23.38% 3.51% 51.75% 4.02% 17.70%
MAPE

Table 4 MAPE values after 1st improve cycle

Item Product1 total Product2 Product3 Product4 Product5 Product1 Product2 Product3 Product4 Product5 Mean Standard UCL
total total total total combine combine combine combine combine deviation
Forecast Weighted Exponential Exponential Holt- Exponential Exponential Exponential Weighted Holt- Exponential
model moving smoothing smoothing winters smoothing smoothing smoothing moving winters smoothing
average average
Weighted Weighted ARIMA Weighted
moving moving moving
average average average
Exponential
smoothing

History 14.54% 20.91% 1.99% 6.73% 2.53% 13.02% 19.99% 1.92% 6.40% 2.74% 9.08% 7.08% 16.16%
MAPE
Holdout 16.61% 13.37% 3.49% 6.68% 2.25% 8.70% 23.38% 3.51% 7.16% 4.02% 8.91%
MAPE
1039
1040 Int J Adv Manuf Technol (2008) 39:1033–1044

The details of the DMAIC approach adopted in this case combine forecasts for product_#4 are above the UCL. Thus,
study are as follows: these two forecasts needed improvement. The sales fore-
casts of product_#4 from areas #2 and #3 were fitted by an
Define module MAPE was used as a performance measure exponential smoothing model. When we evaluated the time
to evaluate the forecast accuracy. The goal was for the series data by using ACF and PACF graphs, we found that
MAPE values of the forecasts of all products to be less than the data were not stationary. That is, this data set could not
10%. Here, we selected five products from a particular be fitted by an exponential smoothing model and, thus,
product family to improve their collaborative forecast. The needs to be fitted by another type of model.
selling areas of all five products could be divided into three
areas: area_#1 (products #1, #2 and #3), area_#2 (products
#2, #3, #4 and #5) and area_#3 (products #3, #4 and #5). 4.2 Improvement using advance forecasting models
We collected 50 weeks worth of sales data for the forecast.
Data from the first 45 weeks were used to build a When a product’s forecast was above the UCL, an ARIMA
forecasting model, while the data from the last 5 weeks model or another advanced forecasting model was selected
was used to verify the forecasting model. We used a control for the forecasting. Table 4 shows that the MAPE values of
chart to monitor the forecast accuracy of all products. The the combine and total forecasts of product_#4 in historical
upper control limit (UCL) on the control chart was set as periods are 6.40% and 6.73%, respectively. Further, we
the average MAPE value of all products plus one standard found that the MAPE values of the combine and total
deviation. If a product forecast’s MAPE value was greater forecasts of product_#4 in holdout periods are 7.16% and
than the UCL, then the forecast was deemed as needing 6.68%, respectively. Thus, the UCL was reduced from
improvement. 39.92% to 16.16%.
The first improvement cycle shows the total and
Measure module The area manager produced the forecasts combined forecasts in Fig. 5 for product_#2, which is
of all products based on the area retailer’s POS system. We above the UCL. Thus, these two forecasts needed improve-
aggregated the forecasts from the three areas to obtain the ment. The sales forecasts of product_#2 from areas #1 and
final forecast result of all the products; this forecast is called #2 were fitted by an exponential smoothing model. When
the ‘combine forecast’. In addition, we generated the final we evaluated the time series data by using ACF and PACF
forecast of all products from the company’s sales database; graphs, we found that the data were not stationary. That is,
this forecast is called the ‘total forecast’. As one may this data set could not be fitted by an exponential
expect, ten time series data were collected for further smoothing model and needed to be fitted by a different
analysis. model. Thus, an ARIMA model was selected for the
forecasting. Table 5 shows that the MAPE values of the
Analyze module All time series data were fitted by using combine and total forecasts of product_#2 in historical
one of three models — moving average, weighted moving periods are 11.15% and 10.81%, respectively. Further, we
average or simple exponential smoothing. MAPE values found that the MAPE values of the combine and total
obtained by using the best fitted forecasting model for all forecasts of product_#2 in holdout periods are 6.78% and
products are shown in Table 3. The UCL was obtained as 9.17%, respectively. Thus, the UCL was reduced from
39.92%. Figures 3 and 4 shows that both the total and 16.16% to 11.81%.

Fig. 5 Product’s forecast model 25.00%

control chart after 1st improve
cycle 20.00%

History
15.00% MAPE

UCL
10.00%

5.00%

0.00%
Product1 Total

Product2 Total

Product3Total

Product4 Total

Product5 Total

Product1 Combine

Product2 Combine

Product3 Combine

Product4 Combine

Product5 Combine
Table 5 MAPE values after 2nd improve cycle

Item Product1 total Product2 Product3 total Product4 Product5 total Product1 Product2 Product3 Product4 Product5 Mean Standard UCL
total total combine combine combine combine combine deviation
Forecast Weighted ARIMA Exponential Holt- Exponential Exponential Holt- Weighted Holt- Exponential
model moving average smoothing winters smoothing smoothing winters moving average winters smoothing
ARIMA Weighted ARIMA Weighted
moving average moving average
Exponential
smoothing

History 14.54% 10.81% 1.99% 6.73% 2.53% 13.02% 11.15% 1.92% 6.40% 2.74% 7.18% 4.62% 11.81%
MAPE
Holdout 16.61% 9.17% 3.49% 6.68% 2.25% 8.70% 6.78% 3.51% 7.16% 4.02% 6.84%
MAPE
Int J Adv Manuf Technol (2008) 39:1033–1044

Table 6 MAPE values after 3rd improve cycle

Item Product1 Product2 Product3 total Product4 Product5 total Product1 Product2 Product3 Product4 Product5 Mean Standard UCL
total total total combine combine combine combine combine deviation
Forecast Holt- ARIMA Exponential Holt- Exponential Holt-winters Holt-winters Weighted moving Holt-winters Exponential
model winters smoothing winters smoothing average smoothing
ARIMA Weighted moving ARIMA Weighted moving
average average
Exponential
smoothing

History 7.72% 10.81% 1.99% 6.73% 2.53% 7.72% 11.15% 1.92% 6.40% 2.74% 5.97% 3.34% 9.31%
MAPE
Holdout 6.51% 9.17% 3.49% 6.68% 2.25% 6.51% 6.78% 3.51% 7.16% 4.02% 5.61%
MAPE
1041
Table 7 MAPE values after 4th improve cycle
1042

Item Product1 Product2 total Product3 Product4 Product5 Product1 Product2 combine Product3 Product4 Product5 Mean Standard UCL
total total total total combine combine combine combine deviation
Forecast Holt- Decomposition Exponential Holt- Exponential Holt- Decomposition Weighted Holt- Exponential
model winters (multiplicative simple smoothing winters smoothing winters (multiplicative simple moving winters smoothing
exponential smoothing) exponential smoothing) average
Decomposition Weighted ARIMA Weighted
(multiplicative simple moving moving
exponential smoothing) average average
Exponential
Smoothing

History 7.72% 8.26% 1.99% 6.73% 2.53% 7.72% 8.01% 1.92% 6.40% 2.74% 5.40% 2.60% 8.00%
MAPE
Holdout 6.51% 7.69% 3.49% 6.68% 2.25% 6.51% 8.49% 3.51% 7.16% 4.02% 5.63%
MAPE

Table 8 MAPE values after 5th improve cycle

Item Product1 Product2 total Product3 Product4 Product5 Product1 Product2 combine Product3 Product4 Product5 Mean Standard UCL
total total total total combine combine combine combine deviation
Forecast Holt- Decomposition Exponential Holt- Exponential Holt- Decomposition Weighted Holt- Exponential
model winters (additive ARIMA) smoothing winters smoothing winters (additive ARIMA) moving winters smoothing
average
Decomposition Weighted ARIMA Weighted
(additive ARIMA) moving moving
average average
Exponential
smoothing

History 7.72% 8.13% 1.99% 6.73% 2.53% 7.72% 8.01% 1.92% 6.40% 2.74% 5.39% 2.59% 7.98%
MAPE
Holdout 6.51% 8.59% 3.49% 6.68% 2.25% 6.51% 8.49% 3.51% 7.16% 4.02% 5.72%
MAPE
Int J Adv Manuf Technol (2008) 39:1033–1044
Int J Adv Manuf Technol (2008) 39:1033–1044 1043

Table 9 Final collaborative forecasts’ MAPE values

Item (X, Y) Product1 (1, 0) Product2 (0.67, 0.33) Product3 (0.64, 0.36) Product4 (0.71, 0.29) Product5 (0, 1) Mean

History MAPE 7.72% 7.78% 1.85% 6.39% 2.53% 5.26%

Holdout MAPE 6.51% 7.13% 3.50% 6.54% 2.25% 5.18%

X is the optimal weight of the Product’s Combine forecast model, and Y is the optimal weight of the Product’s Total forecast model.

We repeated these improvement cycles until it was found that the average MAPE value of all products in
impossible to further improve the collaborative forecasts of historical periods is 5.26%. The average MAPE value of all
all products. After the fifth DMAIC cycle, the MAPE products in holdout periods was calculated as 5.18%. The
values of all products could not be decreased further (see result shows that final collaborative forecasts can offer the
Tables 6, 7 and 8). We found that the average MAPE value highest forecasting accuracy for the product as a whole.
of all products in historical periods decreased from 19.37%
to 5.39%. Further, the average MAPE value of all products Control Based on the above results in Table 10, we can
in holdout periods decreased from 19.15% to 5.72%.The conclude that the optimal collaborative forecast can be
result shows that the proposed model is effective in generated using the forecast models that yield the smallest
improving the forecasting accuracy. The model can also MAPE values of all products. A control chart can then be
effectively reduce the forecast variability of each product. utilized to contrast the forecasted sales from the real sales.
Although the MAPE values of the combine and total If the difference value exceeds the control limit, the
forecasts of product_#2 in historical periods are 8.13% and forecasted model is revised immediately so that it produces
8.01%, respectively, the forecast of all products cannot be a suitable collaborative forecast. The purpose of this
improved and the goal cannot be met. Thus, we can break module is to sustain the forecasting accuracy in demand
the improve cycle. We can produce the collaborative and supply management.
forecast using the forecast models.
For each product, we have two models that offer two In addition, the financial benefit of the proposed method-
forecasts. We can find the best fitted weight between the ology in this case company is significant. For instance, the
two forecasts to produce the final collaborative forecast. monthly revenue of the case company is about US$ 53
For example, we found that the MAPE values of the million. Using the DAMIC approach to improve the forecast-
combine and total forecasts of product_#3 in historical ing accuracy for five weeks, the shortage cost and material
periods are 1.99% and 1.92%, respectively. We adopted the shortage cost for the supplier can be reduced from US$
regression method to find the initial weights and use the 138,669.39 to US$ 9,603.34. That is, it reduced by 93%. On
genetic algorithm (GA) to find the best fitted weight. We the other hand, the shortage cost and holding cost for retailers
found that the best fitted weights of the combine and total can be reduced from US$ 166,324.14 to US$ 87,437.70. That
of product_#3 are 0.64 and 0.36, respectively. Further, we is, it reduced by 47%. This approach can be applied into other
found that that the MAPE value of product_#3 in the stages of CPFR to improve their performance.
historical period is 1.85%. Moreover, the MAPE value of a
product_#3 in the holdout period is 3.5%. We repeated the
above method for all the products. The results are shown in 5 Conclusions
Table 9. They show that the two models can be integrated
to form one final collaborative forecast model. Past studies on CPFR focused mainly on identifying the
This model enables us to find the smallest average advantages of CPFR, recognizing the uses of CPFR in the
MAPE in both the historical and the holdout periods. We near future, and explaining CPFR implementation stages.

Table 10 The comparison of collaborative forecast models

Collaborative forecast models Item Product_#1 Product_#2 Product_#3 Product_#4 Product_#5 Mean

Original model (choice the best one between two History MAPE 13.02% 19.99% 1.92% 57.27% 2.53% 18.95%
forecast models) Holdout MAPE 8.70% 23.38% 3.51% 49.92% 2.25% 17.55%
DMAIC model (choice the best one between two History MAPE 7.72% 8.01% 1.92% 6.40% 2.53% 5.32%
forecast models) Holdout MAPE 6.51% 8.49% 3.51% 7.16% 2.25% 5.58%
DMAIC model (combining two forecast models History MAPE 7.72% 7.78% 1.85% 6.39% 2.53% 5.26%
using GA to allocate the weights) Holdout MAPE 6.51% 7.13% 3.50% 6.54% 2.25% 5.18%
1044 Int J Adv Manuf Technol (2008) 39:1033–1044

Six Sigma is a method that can lead to a continuous 2. Crum C, Palmatier GE (2004) Demand collaboration: what’s
holding us back? Supply Chain Manag Rev 8(1):54–61
decrease in process variances. In this paper, we applied Six
3. Finley F, Srikanth S (2005) 7 imperatives for successful
Sigma methodology and proposed a continuous improve- collaboration. Supply Chain Manag Rev 9(1):30–37
ment model on different stages of CPFR implementation. 4. Hutchins G (2002) Supply chain management: a new opportunity.
We discussed a case study to demonstrate how to improve Qual Prog 35(4):111–112
5. Snee RD (1999) Why should statisticians pay attention to Six
the performance of collaborative forecasts using the Six
Sigma? Qual Prog 32(9):100–103
Sigma methodology. We found that the average MAPE 6. Fliedner G (2003) CPFR an emerging supply chain tool. Ind
value of all products for historical periods can be reduced Manage Data Syst 103:14–21
from 19.37% to 5.26%. We additionally concluded that the 7. Lin JT, Yang CH, Lin TM (2004) A CPFR implementation
methodology study: a carpenter mechanical industry case study.
average MAPE value of all products for holdout periods Int J Electron Business Management 2:172–178
can be reduced from 17.7% to 5.18%. 8. Holmstrom J, Framling K, Kaipia R, Saranen J (2002) Collabo-
Regardless of the type of SCM style adopted, the most rative planning forecasting and replenishment: new solutions
important obligation would be the sharing of information needed for mass collaboration. Supply Chain Manag: International
J 7(3):136–145
and the building of trust among trading partners involved in
9. Holweg M, Disney S, Holmstrom J, Smaros J (2005) Supply
collaborative forecasting. It is highly probable that the Six chain collaboration: making sense of the strategy continuum. Eur
Sigma methodology will be incorporated into CPFR in the Manag J 23(2):170–181
near future, a methodology that will become the standard 10. Chowdhury S (2001) The power of Six Sigma. Dearborn Trade,
Chicago
practice for any e-business application that seeks to excel in
11. Hoerl RW (1998) Six Sigma and the future of the quality
this highly competitive globalization era. The analyzed case progression. Qual Prog 31(6):35–42
study is in fact, limited in that it was drawn from a 12. Wyper B, Harrison A (2000) Deployment of Six Sigma
relatively short span of time. Scenario B in Table 2 also methodology in human resource function: a case study. Total
Qual Manage 11(4–5):720–727
only consisted of the paper industry. Research done in the 13. Ehie I, Sheu C (2005) Integrating Six Sigma and theory of
future should pull data from a longer period of time, and constraints for continuous improvement: a case study. J Manuf
should include other industries belonging to the other Technology Management 16(5):542–553
scenarios (either A, C, or D) in Table 2. In addition, one 14. Motwani J, Kumar A, Antony J (2004) A business process change
framework for examining the implementation of Six Sigma: a case
referee suggested that we should investigate how to manage
study of Dow Chemicals. TQM Mag 16(4):273–283
demand smoothing and supply arrangement in CPFR. 15. Blakeslee JA Jr (1999) Implementing the Six Sigma solution.
Qual Prog 32(7):77–85
16. Wang FK, Du TC, Li EY (2004) Applying Six Sigma to supplier
Acknowledgement The authors wish to gratefully acknowledge the development. Total Quality Management & Business Excellence
referees of this paper who helped to clarify and improve the 15(9–10):1217–1229
presentation. The authors are grateful for partial research grant from 17. Tong JPC, Tsung F, Yen BPC (2004) A DMAIC approach to
the National Taiwan University of Science and Technology, Taiwan. printed circuit board quality improvement. Int J Adv Manuf
Technol 23(7–8):523–531
18. Mahesh M, Wong YS, Fuh JYH, Loh HT (2006) A Six Sigma
approach for benchmarking of RP&M processes. Int J Adv Manuf
Technol 31(3–4):374–387
References
19. Knowles G, Whicker L, Femat JH, Canales FD (2005) A conceptual
model for the application of Six Sigma methodologies to supply
1. VICS, CPFR Guidelines, Voluntary Inter-industry Commerce chain improvement. Int J Logist: Research Applications 8(1):51–65
Standards. Available online at: http://www.cpfr.org/ (Accessed 1 20. Spector RE (2006) How constraints management enhances lean
Oct., 2006) and Six Sigma. Supply Chain Manag Rev 10(1):42–47

View publication stats