-
International Journal of Production Research,Vol. 44, No. 23, 1
December 2006, 50655087
Efficiency analysis of supply chain processes
G REINER*y and P. HOFMANNz
yDepartment of Information Systems and Operations, Vienna
University of Economics andBusiness Administration, Nordbergstrae
15, 1090 Vienna, Austria
zSAP Research, SAP Labs Palo Alto, 3410 Hillnew Avenue, Palo
Alto, CA, USA
(Revision received September 2005)
We present an integrated benchmarking approach. To analyse the
performance ofinter-organizational (supply chain) processes at
company level we combinedependency analysis and data envelopment
analysis (DEA). DEA has beenproven to be a powerful tool for
benchmarking processes and for identifying themost efficient ones.
Before using DEA analysis the inputs and outputs as well asthe
relevant dependencies have to be identified. To support the
determination ofinput and output variables we propose to use
dependency analysis. We illustratethe application of this
integrated approach by analysing the results of anempirical
benchmarking study of 65 European and North American companies.The
study shows that make-to-stock is still the predominating
manufacturingstrategy of the analysed industries. Therefore, we
utilize different DEA modelswith weight restrictions only for
companies with a make-to-stock strategy. Theresults support our
basic hypotheses that efficient supply chains lead to highfinancial
performance. Furthermore they indicate improvement potential for
thebenchmarked supply chain processes.
Keywords: Supply chain performance measurement;
Benchmarking;TETRAD; DEA
1. Introduction
The wealth of nations and companies depends on a continuous
growth in
productivity. To sustain long-term growth and profitability in a
competitive
environment, industrial enterprises must continuously improve
their efficiency
(Sudit 1995). Therefore questions about the measurement of
efficiency are
increasingly important. The search for potential improvement of
efficiency has
also been spurred on by the realization that not only do single
enterprises compete
against each other but also entire supply chains (Christopher
1992).The present study examines the entire supply chain processes
of companies. We
intend to identify dependencies between operational performance
and financial
success. We are also looking for dependencies between several
key performance
measures of supply chain processes and the financial performance
of a company or
division respectively.
*Corresponding author. Email: [email protected]
International Journal of Production Research
ISSN 00207543 print/ISSN 1366588X online 2006 Taylor &
Francishttp://www.tandf.co.uk/journals
DOI: 10.1080/00207540500515123
-
In general, the partial efficiency of a supply chain is measured
by different
performance indicators pertinent to specific aspects of the
performance of an
enterprise. For the purpose of benchmarking supply chains
against each other we
propose a composite measure of the total efficiency, the
productivity measured by
data envelopment analysis (DEA), to identify the best practice
supply chains.
In this context one needs to answer questions such as: Does
enterprise X, having
twice as many inventory turns as enterprise Y, but having 2%
lower delivery
performance to request date, perform better; i.e. does it have a
higher productivity?
One has to establish which values of input and output indicators
characterize an
efficient supply chain. Multiple input and output indicators
have to be considered.We propose an integrated framework applying
TETRAD (a dependency analysis
approach and software tool) and data envelopment analysis (DEA)
to analyse the
results of an empirical benchmarking study of 65 European and US
companies.
Benchmarking helps businesses to identify potential targets for
performance
improvement. This is the initial step that a company must
undertake in order to
plan and implement improvement methods for achieving better
performance.
Although the literature proposes a lot of benchmarking methods,
research in this
area lacks rigorous analytical tools for effective analysis of
benchmarking results.
Several performance measures integrated with best practice
processes must be
analysed (Talluri 2000).A review of the literature shows that
there are not many applications of this kind
in the area of supply chain management. An interesting work in
this context was
presented by Talluri and Baker (2002) who introduced a
multi-phase mathematical
programming approach for supply chain design. In general,
different elements of
supply chain processes have been analysed with DEA models for
specific industries.
A lot of work has been done to analyse the suppliercustomer
relationship. In this
context supplier evaluation is a core topic (Narasimhan et al.
2001). Other literature
deals with the efficiency/measurement of manufacturing processes
(Talluri and
Sarkis 2002). The analysing of distribution systems is another
core application area
of DEA (Ross and Droge 2004).In this paper we will show an
application of DEA for the entire supply chain
processes of companies, i.e. the SCOR processes source, make and
deliver. These
supply chain processes within one company we call a decision
making unit (DMU).
We use a widely accepted industry standard, the Supply Chain
Operations Reference
(SCOR) model (SCOR 2005). Only such a commonly used model offers
the chance
that there are enough non-disclosure (intra-company) data
available and the data is
comparable for different companies and industries. The SCOR
model is a cross-
industry process reference model designed for supply chain
management. The SCOR
model was designed with the intention to support better software
systems,
benchmarking (use of common measurements and terms), as well as
recognizing
and adopting best practices rapidly, regardless of their origin
(Meyr et al. 2002). The
SCOR model is a tool to build standardized, comparable and
measurable processes
for supply chains. Not all areas of supply chain management are
included,
e.g. product development and marketing.In section 3 our
integrated approach presented in section 2 will be illustrated
by
analysing a benchmarking study of selected industries. We will
analyse dependencies
between performance measures and we will also show how DEA can
be used
5066 G Reiner and P. Hofmann
-
to identify reference sets, i.e. companies with high
productivity. We will illustratehow to gain further managerial
insight for the analysed supply chain processes.
2. Integrated approach of analysing benchmarking results
Benchmarking is a technique of finding potentials for
improvement of processes.It can be used to determine best practice
supply chain processes that result in bettersupply chain
performance. The technique is also able to identify efficient
supplychain processes as a reference for the improvement of
inefficient processes.
The increased application of business excellence models (e.g.
EuropeanFoundation for Quality Management Excellence Model, Malcolm
BaldrigeNational Quality Award) during the last years has drawn
significant attention tobenchmarking (EFQM 2004, MBNQA 2005). Also
the SCOR model calls explicitlyfor the application of benchmarking
(SCOR 2005).
Benchmarking methods for process improvements are mostly
developed andintroduced by practitioners. Many practitioners use
simple techniques rather thananalytical methods. Therefore there is
strong demand for effective methods ofanalysing benchmarking
results, which can be used for design, analysis andimprovement of
processes (Talluri 2000).
DEA, a multifactor nonparametric productivity analysis
technique, has beenproven to be a powerful tool for methodical
benchmarking of processes andidentifying the most efficient ones.
It has been widely used in a variety of settings(Tavares 2002).
The idea behind benchmarking is to transfer results (e.g. best
practices) fromone company to others or to other industries
irrespective of the characteristics(complexity, mechanics,
demographics, etc.) of the company (Dahlgaard et al. 1998).Using a
quantitative method such as DEA for recognition of best practices
requirescomparable DMUs, i.e. supply chain processes in a company.
The followingrequirements have to be fulfilled to apply DEA, i.e.
to enable comparison of DMUsregardless of the industry in which the
DMU operates:
. Processes have to be comparable with regard to a standardized
processmodel. In our case we use the definition of processes of the
SCOR model.
. All DMUs have to use measures with the same definition and
calculation forcomparable processes. This is fulfilled by using the
performance measuresof the SCOR model for benchmarking.
Productivity of the DMUs within the DEA model is measured by the
weightedsum of outputs over the weighted sum of inputs. In order to
employ DEA one mustdefine for each performance indicator (e.g.
delivery performance) if it is input oroutput. Uncertainty as to
which measure is used as input or output for a DMU oftenoccurs
(Post 2001). The performance measures of the SCOR model are not
definedwith respect to input or output vectors of efficiency
analysis. The SCOR variables aredefined from a practitioners point
of view. The SCOR best practices rely on theexperience of supply
chain experts.
We propose a framework integrating dependency analysis and DEA
to overcomethis difficulty. Figure 1 shows the proposed approach
for analysing the results ofan empirical benchmarking study at
enterprise level. A dependency analysis of the
Efficiency analysis of supply chain processes 5067
-
benchmarking data to identify the input and output variables for
DEA is conductedby means of TETRAD. The result of TETRAD is a
decision support for theclassification of performance indicators
into input and output variables. The finaldecision about the
selection has to be made by the user of DEA taking into accountDEA
requirements. Then DEA is applied to the input and output variables
toevaluate the performance of the supply chain processes the data
of which has beencollected by the empirical study.
The results of DEA have to be further analysed before best
practice DMU can beidentified and managerial implications derived.
For example, one has to see ifthe DEA results deliver efficient and
inefficient DMUs in a sensible mix. The DEAresults should be
verified, e.g. by analysing the direction and strength of
thecorrelation coefficients of the measures of the efficient and
inefficient DMUs.
Figure 1. Integrated approach.
5068 G Reiner and P. Hofmann
-
DEA suggests best practice DMUs (companies or divisions within a
company)for benchmarking and shows quantitatively how much a
performance indicatorcould improve in order to bring the
underperforming process to the best practicelevel. The main
advantage of using a productivity analysis like DEA forbenchmarking
is its ability to deal with aggregate process information. This
allowsevaluating and controlling the overall process performance
without using detailedprocess information. However, comparing only
aggregate inputoutput vectors ofDMUs does not provide sufficient
advice to an inefficient DMU on how to improveits processes.
Further detailed process analysis is necessary for concrete steps
towhere and how to change the supply chain process. Our integrated
approach (seefigure 1) only shows that there are best practice
processes possible that are moreefficient. We consider the
underlying physics only on an aggregate level; forexample,
make-to-stock strategy, number of warehouses, number of suppliers,
etc.,and look primarily at the processes from a performance point
of view. Our integratedapproach does not point out a general
limitation of DEA. In general, with DEA boththe aggregate and the
detailed comparison are feasible. The latter would requiremodelling
the relevant sub-processes explicitly (Homburg 2001).
Benchmarking yields little insight into best practices if a
simplistic approach isused not considering that multiple inputs of
supply chain processes can producemultiple outputs; e.g. inventory
and resources contribute to delivery performance aswell as cash to
cash cycle time. The effects of changes of each individual measure
onthe overall efficiency would be difficult to evaluate without a
quantitative methodlike DEA to generate a composite efficiency
measure (Sudit 1995). DEA extends thetraditional concept of
efficiency to make it suitable for the multi-input
multi-outputcase. This is done without reducing several inputs to a
simple one-dimensional value(e.g. cost). Instead, on the basis of
the DMUs inputoutput vectors, a productionfrontier is established
that can be viewed as best practice. In the context of
activity-based management Homburg (2001) shows that a DMU needs
only determine thetotal resource usage of all its activities
because DEA does not require determiningthe physics of producing
activities and using resources on a detailed level.
2.1 Dependency analysisTETRAD
This section describes a dependency analysis of the data to
identify the inputoutputvariables for DEA. The method we propose is
represented by the TETRAD-Project(2004). The standard way to define
cause and effect for a statistical model isto apply theoretical
knowledge for the assumptions of the model. In caseof insufficient
theoretical knowledge for specification of a unique model
depen-dency analysis can be a tool for searching the unknown model
space (Castilloet al. 1997).
TETRAD is a computer program used for analysis of dependency.
TETRADhelps to find statistical models that explain the data and
are compatiblewith user-entered background knowledge. The program
output is a (partially)directed acyclic graph (DAG) with nodes
(e.g. performance measures) and directededges denoting direct
influences between the nodes (Spirtes et al. 1993). The programis
based on a set of tests of conditional independence between the
measuredvariables. The basis for the TETRAD algorithm is a match of
graphical and
Efficiency analysis of supply chain processes 5069
-
probabilistic methods. There are several search algorithms
differing in the assump-tions they make.
For discovering causal relationships we applied an algorithm
that assumes thatthere is no latent (i.e. not measured) variable
that contributes to the association oftwo or more measured
variables (TETRAD-Project 2004). Figure 2 shows the basictypes of
edges that could appear as result of the algorithm. In the case of
a directededge the used algorithm deducted that A is a direct cause
of B. An undirected edgemeans the algorithm cannot reveal if A
causes B or if B causes A. The absence of anedge between any pair
of nodes means that they are independent, or that the causaleffect
of one node on the other is intermediate through other observed
variables.Finally, a triplet of nodes with undirected edges (A and
B are connected by anundirected edge, A and C are connected by an
undirected edge whereas B and C arenot connected by an edge) means
that B and C cannot be both cause of A. This leadsto three possible
interpretations:
1. A is a cause of B and C.2. B is a direct cause of A, and A is
a direct cause of C.3. A is a direct cause of B and C is a direct
cause of A.
An illustrative example of TETRAD results could be found in
Jammernegg andKischka (2005).
The TETRAD-Project has had many successful applications so far,
e.g. inmedicine and econometrics. We propose to use it for support
of the identification ofinput and output data for a DEA analysis.
Furthermore it could be used to gaindeeper insight into the causal
dependencies of the data set.
Figure 2. Dependency graph (cf. TETRAD-Project 2004).
5070 G Reiner and P. Hofmann
-
2.2 Data envelope analysisDEA
Data envelopment analysis (DEA) is the most widely used
programming method formeasuring the productivity of decision making
units (DMU). DEA is a mathematicalprogramming technique that
provides upper bound estimates of the actual totalfactor
productivity of DMUs as well as estimates of production frontiers
(Sudit1995). DMUs using multiple inputs to produce multiple outputs
can be evaluatedand compared. The inputs and outputs may have
different measurement units andthey can be under certain
assumptions even qualitative (Cooper et al. 1999). DEA isused to
calculate the efficiency of the units relative to each other as
well as theoptimal weights of input and output. In our research the
term decision-making unit(DMU) describes supply chain processes
within one company, i.e. the SCORprocesses source, make and
deliver.
The basic model for DEA is called the CCR (Charnes, Cooper and
Rhodes)model and was developed by Charnes et al. (1978).
Illustrated below is the dualmodel of the CCR, which is often
preferred over the primal model for usage incalculations. is the
aggregate efficiency score for DMUO, the supply chain
underobservation. yr0 is the output r generated by DMUO . . . xiO
is the input i used byDMUO. Xj (x1j, x2j, . . . , xmj) is the
vector of actual inputs used by DMUO.Yj (y1j, y2j, . . . , ysj) is
the vector of actual outputs generated by DMUO. si is theamount of
slack in input i for DMUO. s
r is the amount of slack in output r for
DMUO. j is the dual multiplier (the weights assigned to the
inputs and outputs atDMUj). s is the number of outputs. n is the
number of DMUs and m is the number ofinputs.
minj,,s
i,sr
Xmi1
si Xsr1
sr
!1
s:t:Xnj1
Xjj si xi0, 8i 1, . . . ,m2
Xnj1
Yjj sr yr0, 8r 1, . . . , s 3
, j, si , s
r 0, 8i, r, j 4
The CCR model assumes constant returns to scale. In the context
of supply chainprocesses the assumption is not appropriate that all
DMUs are operating at optimalscale. For example, imperfect
information flows, imperfect competition andconstraints on
resources may cause a DMU to be not operating at optimal
scale.Therefore we use the BCC (Banker, Charnes and Cooper) model
assuming variablereturns to scale. Models with variable returns to
scale differ from the CCR model bythe following additional
constraint (Banker et al. 1984).
Xnj1
j 1 5
Efficiency analysis of supply chain processes 5071
-
This constraint ensures that an inefficient DMU is only
benchmarked againstDMUs of similar set up. Whereas using the CCR
model could lead to benchmarkingDMUs against other DMUs with
substantially different set up.
A DMU is called efficient if the following is true for the
optimal solution:
. 1 and
. all slack variables si , sr equal zero.
The optimal solution shows for all input variables the fraction
of their value thatis sufficient to generate the desired output. 1
is the necessary reduction of allinputs in order for a DMU to be
efficient. Therefore the value of is a radialefficiency
measure.
DEA is a valuable tool not only for classifying DMUs as
efficient and inefficientbut for finding improvement potentials for
inefficient units as well. Each inefficientDMU (
-
make-to-stock (MTS) and configure-to-order (CTO). About 45% of
the companiesare in consumer goods, 38% operate globally, 67% have
an MTS manufacturingenvironment and 60% have revenue generated by
the supply chain of no more than$500M.
The team collected data from the various supply chains by
sending questionnairesfocusing on supply chain planning and
processes. We applied the widely acceptedindustry standard SCOR
model to assess the current state of a supply chainsbusiness
processes.
The SCOR model measures internal facing and customer facing
businessperformance metrics. Internal facing metrics include:
. inventory days of supply,
. inventory carrying cost, and
. cash-to-cash cycle time.
Customer-facing metrics include:
. on time delivery to request date, and
. on time delivery to commit (promised) date.
Table 1 shows the relevant SCOR performance indicators selected
for ouranalysis. For a detailed description of these indicators see
SCOR (2005). To answerthe question which of these indicators should
be used for the analysis of the empiricaldata we used the TETRAD
dependency analysis as decision support. This way wealso determined
input and output variables.
Figure 3. General information.
Efficiency analysis of supply chain processes 5073
-
Table
1.
Perform
ance
Measures(cf.SCOR
(2005)).
No.
Perform
ance
measure
Definition
a1
Deliveryperform
ance
torequestdate
Ordersthatare
delivered
onthecustomersrequesteddate.
a2
Fillrate
byorder
Ordersshipped
from
stock
within
24hours
oforder
receipt.
a3
Fillrate
byline
a4
Accounts
receivable
Representsalesthathavenotyet
beencollectedascash.
a5
Totalinventory
Raw
materialsandwork
inprogress,finished
goods,fieldsamples,other.
a6
Cash-to-cash
cycletime(days)
Inventory
daysofsupplydayssalesoutstandingaveragepaymentperiodfor
materialsinventories/C
OGSaccounts
receivable/revenuevendorliabilities/material
costs(raw
materialsandpurchasedparts).
a7
Cost
ofgoodssold
(COGS)
Labour,materialandoverheadexpenses.
a8
Profitability(EBIT
)Revenueexpenses(excludingtaxandinterest).
a9
Expenses(sellinggoodsandadministration)
Includes
marketing,communication,customer
service,
salessalaries
andcommissions,
occupancy
expenses,unallocatedoverheads,etc.
Excludes
interest
ondebt,domesticor
foreignincometaxes,depreciationandamortization,extraordinary
item
s,equitygainsor
losses,gain
orloss
from
discontinued
operationsandextraordinary
item
s.
a10
Net
asset
Are
calculatedastotalassetstotalliabilities;wherethetotalassetsare
madeupoffixed
assets(plant,machineryandequipment)
andcurrentassetswhichisthetotalofstock,
debtors
andcash
(alsoincludes
accounts
receivable,inventory,prepaid
assets,deferred
assets,intangiblesandgoodwill).Thetotalliabilitiesare
madeupin
much
thesamewayof
long-term
liabilitiesandcurrentliabilities(includes
A/P,accrued
expenses,anddeferred
liabilities).
5074 G Reiner and P. Hofmann
-
a11
FTEsin
productionandmanufacturing
Number
ofem
ployeesconvertedinto
fulltimeequivalents
(FTE).Figuresforthenumber
of
personsworkingless
thanthestandard
workingtimeofafull-yearfull-tim
eworker
should
beconvertedinto
fulltimeequivalents,withregard
totheworkingtimeofafull-tim
efull-yearem
ployee
intheunit.
a12
FTEsin
warehousingandtransportation
a13
FTEsin
purchasingandreceiving
a14
TotalFTEs
a15
Number
ofship-from
locations
Number
oftier
1suppliers.
a16
Number
ofproductionlocations
a17
Number
ofwarehousingfacilities
a18
Number
ofship-tolocations
Number
oftier
1customers.
a19
Inventory
carryingcost
Sum
ofopportunitycost,shrinkage,
insurance
andtaxes,totalobsolescence
forraw
material,
work
inprogress,andfinished
goodsinventory,channel
obsolescence
andfieldsample
obsolescence.
a20
Total,order
fulfilmentleadtime
(maketo
stock)
Customer
signature/authorizationto
order
receiptorder
receiptto
completionoforder
entrycompletionoforder
entryto
start
manufacturestart
manufacture
tocomplete
manufacturecomplete
manufacture
tocustomer
receiptofordercustomer
receiptof
order
toinstallationcomplete.
Formake-to-stock
products,thelead-tim
efrom
start
manufacture
tocomplete
manufacture
equals0.
a21
Revenue
Thetotalvalueofsalesmadeto
externalcustomersplusthetransfer
price
valuationof
intra-companyshipments,net
ofalldiscounts,coupons,allowances,andrebates.
Efficiency analysis of supply chain processes 5075
-
3.1 Strategy selection
We decided to use DMUs with make-to-stock (MTS) manufacturing
strategy sinceit is the prevalent strategy for most industries as
one can see from figure 3. A mixtureof DMUs with different
strategies would lead to wrong DEA results since DMUswith different
strategies have different dependencies between their
performancemeasures. Therefore, the TETRAD analysis was conducted
only for DMUs withMTS scenario.
In supply chain processes the positioning of the customer order
decoupling point(CODP) determines the trade-off between inventory
costs and costs of resources.In the case of MTS production the
decoupling point is at the finished goodsinventory, i.e. one puts
up with higher inventory costs. In make-to-order (MTO)production
the CODP is located at the raw material, i.e. higher cost of
resources areaccepted (van der Vorst et al. 1998). Olhager (2003)
identified two major factors thataffect the strategic positioning
of the order decoupling point, the productionto delivery lead time
ratio and the relative demand volatility (standard deviationof
demand relative to the average demand). If the production lead time
is longer thanthe delivery time for a customer order MTO production
is not advisable becauseof poor customer service.
3.2 Process description
The definition and calculation of measures for comparable
processes require astandardized process model. Therefore we use the
standardized process definition ofthe SCOR model for MTS with the
relevant process categories and process elements.In figure 4 we
describe the relevant process categories.
P1. Plan supply chain: the development and establishment of
courses of action overspecified time periods that represent a
projected appropriation of supply chainresources to meet supply
chain requirements.
P2. Plan source: the development and establishment of courses of
action overspecified time periods that represent a projected
appropriation of material resourcesto meet supply chain
requirements.
P3. Plan make: the development and establishment of courses of
action over specifiedtime periods that represent a projected
appropriation of production resources tomeet production
requirements.
P4. Plan deliver: the development and establishment of courses
of action overspecified time periods that represent a projected
appropriation of delivery resourcesto meet delivery
requirements.
S1. Source stock product: the procurement, delivery, receipt and
transfer of rawmaterial items, subassemblies, product and or
services.
M1. Make-to-stock: the process of manufacturing in a
make-to-stock environmentadds value to products through mixing,
separating, forming, machining, andchemical processes. Make to
stock products are intended to be shipped from finishedgoods or off
the shelf, are completed prior to receipt of a customer order,
andare generally produced in accordance with a sales forecast.
5076 G Reiner and P. Hofmann
-
Figure
4.
SCOR-m
odelsselected
process
categories
andprocess
elem
ents
(SCOR
2005).
Efficiency analysis of supply chain processes 5077
-
D1. Deliver stocked product: the process of delivering a product
that is maintained
in a finished goods state prior to the receipt of a firm
customer order.
3.3 Results of TETRAD analysis
We obtained 28 DMUs with MTS strategy and a complete data set.
It is not
surprising that we found no DMU from the automotive industry and
high-tech
industry in this data set. JIT concepts play an important role
in the automotive
industry and construct to order is very important for the
high-tech industry. The
selected sample consists of 18 DMUs from the consumer good
industry (CG), four
from the chemical industry (CH), three from the electronic
equipment manufacturer
(EE), two from the medical devices industry (MD) and one from
the mill industry
(MI). Figure 5 shows the result of the search by dependency
analysis for the causal
structure between the following quantitative variables (see
table 1).The qualitative representation of dependencies in figure 5
supports the selection
of input and output variables. Using TETRAD avoids
classification problems as in
the relevant DEA literature. The variable inventory for example
is classified in some
supply chain management DEA research works as input variable
(Gropietsch 2003)
and in other articles as output variable (Ross and Droge
2004).
Figure 5. Tetrad graphical results.
5078 G Reiner and P. Hofmann
-
A level of significance () of 0.2 has been chosen for this
TETRAD analysisbecause it gives the most reliable output for a
sample size 100 or smaller (Scheines
et al. 1994). Since we want to analyse productivity the input
variables have to act on
the output variables. Dependency among the input variables or
dependency between
the output variables is not required.From figure 5 we can see
that a8, a9 and a19 fulfil the conditions for output
variables whereas a3, a12, a15, a16, a17 and a18 fulfil the
conditions for
input variables. The variables a1, a2, a6, a13 and a20 show
dependencies to
other variables but these dependencies are not directed (no
arrows). The other
variables show directed dependencies but can not be classified
explicitly as input or
output variables for DEA without resorting to supply chain
management knowledge.We use supply chain and DEA knowledge to
finally pick the input and output
variables from the TETRAD results.The profitability (EBIT) a8
cannot be used for DEA as output as suggested by
TETRAD since it is calculated from input and output variables.
Instead of a8 we
select a21 (revenue) as output variable. Instead of the
suggested output variables
a9 (selling goods and administration) and a19 (inventory
carrying cost) we use the
corresponding cost drivers as input variables since costs cannot
be used as output
variable for a productivity measure such as DEA. As replacement
for a19 we will
use a5 (total inventory) as input variable because it is also a
driver of a19 and
shows a strong correlation with a21 (revenue). As the drivers of
a9 (logistics cost)
we use the TETRAD suggestions a17 (number of warehousing
facilities) and a18
(number of ship-to locations) as input variables. A further
driver of logistic costs
a15 (number of ship-from locations) is also confirmed as input
variable. We replace
the suggested input variable a16 (number of production
locations) with the classical
input variable for DEA a11 (FTEs in production and
manufacturing) because these
variables show a high correlation with each other.Besides
revenue as output variable for DEA we look for a
customer-oriented
output variable. We have to identify the relevant variable out
of the group of SCOR
level one service measures, a1 (delivery performance to request
date), a2 (fill rate
by order) or a3 (fill rate by line). The TETRAD results suggest
a3 as input variable
therefore we cannot use it as output variable. a2 shows a
dependency with a3 and
therefore cannot be used either. a1 was not classified
explicitly as input or output
variable by TETRAD. Nevertheless we choose a1 as customer
oriented output
variable in view of the fact that it is the most frequently used
measure for customer
service in the SCOR model.For final determination of the input
and output variables we checked if the
following precondition is fulfilled. The value of input
variables should decrease
whereas the value of output variables should increase for the
desired result
(Scheel 2001).Using TETRAD analyses as decision support we have
determined the following
variables as input variables: the driver of production costs
(a11 FTEs in production
and manufacturing), the driver of inventory costs (a5 total
inventory) and the
drivers of logistics costs (a15 number of ship-from locations
(Tier 1 suppliers), a17
number of warehousing facilities, a18 number of ship-to
Locations (Tier 1
Customers)). As output variables we used Delivery Performance to
Request Date
(a1) and revenue (a21). Table 2 lists the means and standard
deviations.
Efficiency analysis of supply chain processes 5079
-
3.4 Results of DEA analysis
The number of DMUs for DEA analysis confines the number of
variables. As a rule
of thumb: numbers of DMUs 4 [ input variables output variables].
In somepublications a factor of 2 instead of 4 is considered
sufficient (Dyson et al. 2001,Homburg 2002). Looking at make to
stock as the primary manufacturing strategy
we have chosen 28 DMUs excluding the ones with incomplete data
sets.We analysed several DEA models by varying different input
variables with the
same set of output variables. Table 3 specifies these process
models by listing which
inputs were included in the corresponding model. All models were
analysed byassuming variable return to scale (BCC) and additional
weight restrictions. The
choice of the appropriate model for benchmarking should be left
to the ultimate
decision maker, the supply chain manager (Ross and Droge 2004).
Model 1 includes
only the drivers of production and inventory costs. Model 2
includes in addition thedrivers of supplier and customer facing
logistics. Whereas Model 3 includes,
in addition to Model 1, a driver of intra-company logistics
costs.Empirical research results on the importance of performance
measures show that
the weights used in a DEA model could not be unrestricted
(Talluri and Sarkis 2002).
To choose appropriate bounds we applied a method suggested by
Roll et al. (1991).We ran the BCC model in its unbounded version
and looked at the weight matrix.
Then we eliminated the virtual zero weights so that the outcomes
would not be
affected directly by bounding. We defined lower (0.1) and upper
(10) bounds for the
marginal rate of substitution of variables. Each combination of
ratios within a group
of input and output variables can be represented as a matrix. We
allowed a relativelyhigh variation of weights since we compared
DMUs with make to stock production
from different industries. The following matrix W of weight
restrictions was applied
for the DEA calculation using the BCC model for process model
1:
W
1 0:1 0 01 10 0 00 0 1 0:10 0 1 10
0BBBB@
1CCCCA 7
Table 4 shows the results of the DEA analysis. The column Ind.
indicates the
industry for each analysed DMU. The numbers in the columns BCCm
i are the
efficiency indicators with the above weight restrictions. The
factor (1 ) forj 1, . . . , 28 indicates the reduction of input
needed for the DMU to reach theefficiency frontier. Assuming model
1 we find seven efficient DMUs. With model 2
we find four efficient DMUs. Model 3 shows eight efficient
DMUs.
Table 2. Descriptive statistics.
Inputs Outputs
a5 a11 a15 a17 a18 a1 a21
Means 714 968 502 58 610 86 654Standard deviation 843 1227 1273
256 2308 14 874
5080 G Reiner and P. Hofmann
-
The following example illustrates how to read table 4. We find
an efficiencyindicator of 0.287 for the DMU 17 in model 1. The
inputs are 71.3% (1 ) higherthan the inputs of the virtual DMU that
is a combination of the dominating efficientDMUs 4, 16 and 22. In
the column reference setmodel 1 of the table 4 we find
thecoefficients of this combination, the variables j of the BCC
model. For inefficientDMUs the reference DMUs with the
corresponding j are shown in brackets in thecolumn, reference set,
of table 4. For each efficient DMU the number of inefficientDMUs
containing this DMU in their reference set is noted; e.g. 7 in the
case ofthe efficient DMU 16.
At this point we are able to identify best practice companies as
targets forbenchmarking. As mentioned, the reference set of an
inefficient DMU contains theefficient DMUs that are efficient with
the optimal weights for the inefficient DMU.In order to move to the
efficient frontier inefficient DMUs should orient towards
theefficient DMUs that are part of their reference set. A unit that
is part of manyreference sets is a good performer among the
efficient DMUs. It is an excellentbenchmarking partner for many
inefficient DMUs. Such a DMU shows true efficientperformance
(Boussofiane et al. 1991). We assert that it belongs to a company
withbest practice supply chain processes. Efficient DMUs that are
seldom found inreference sets of inefficient DMUs are probably not
true efficient performers;at least in the sense of being targets
for improvement by inefficient DMUs(Ramanathan 2003).
The distributions of the selected DEA models presented in table
5 allowadditional insight. We expect that the models 2 and 3
capture more operationalcharacteristics of the SC processes and are
more comprehensive in measuringperformance due to more input
variables compared to model 1. Ross and Droge(2004) stated for CCR
and BCC models without weight restrictions that thedistributions of
efficiency indices for models with more input variables should
betighter. We used weight restrictions for our models. Therefore
the oppositecharacteristic would be expected (Roll et al. 1991).
Models with fewer inputvariables should show a tighter
distribution, a higher mean and median efficiencyvalue. The
standard deviation of model 1 and 2 is 0.258 and 0.324
respectively. Themean and median efficiency scores for model 2 are
the lowest in all cases.Consequently the distributions of model 1
and model 2 are in line with ourconjecture above. Other than
expected the distribution of model 3 has the highest
Table 3. Summary table of alternative process models.
Model 1 Model 2 Model 3
Inputsa5 x x xa11 x x xa15 xa17 xa18 x
Outputsa1 x x xa21 x x x
Efficiency analysis of supply chain processes 5081
-
Table
4.
TheDEA
resultsforeach
DMU.
DMU
Ind.
BCCm1
Reference
set m
odel1
BCCm2
Reference
set m
odel2
BCCm3
Reference
set m
odel3
1CG
0.858
4(0.147)8(0.534)24(0.261)
27(0.058)
1.000
17
0.855
4(0.150)8(0.532)24(0.261)
27(0.057)
2CG
0.716
4(0.533)22(0.029)27(0.438)
0.679
1(0.447)16(0.095)27(0.458)
0.730
4(0.236)22(0.037)24(0.252)
27(0.475)
3CH
0.802
16(0.762)22(0.054)25(0.184)
0.188
16(0.799)25(0.201)
0.756
16(0.790)22(0.013)25(0.197)
4CG
1.000
16
0.697
1(0.901)16(0.099)
1.000
13
5CG
0.383
4(0.479)16(0.030)24(0.491)
0.544
1(0.986)16(0.014)
0.456
4(0.626)24(0.374)
6CG
0.901
4(0.216)22(0.007)27(0.777)
0.151
1(0.614)16(0.386)
0.899
4(0.218)22(0.007)27(0.774)
7CG
0.894
25(0.256)27(0.744)
0.485
16(0.202)25(0.230)27(0.568)
0.897
25(0.256)27(0.744)
8CG
1.000
20.805
1(1.000)
1.000
29
EE
0.384
4(0.403)16(0.054)24(0.543)
0.651
1(0.943)16(0.057)
0.418
24(0.968)25(0.020)28(0.012)
10
CH
0.427
24(0.999)27(0.001)
0.439
1(1.000)
0.428
24(0.999)27(0.001)
11
CG
0.651
4(0.335)22(0.321)27(0.344)
0.991
16(0.650)25(0.019)27(0.331)
0.657
22(0.332)24(0.290)27(0.378)
12
CG
0.684
4(0.210)22(0.043)25(0.396)
27(0.351)
0.576
16(0.354)25(0.353)27(0.293)
0.693
4(0.300)22(0.046)25(0.402)
27(0.252)
13
CG
0.248
4(0.874)16(0.079)22(0.047)
0.179
1(0.748)16(0.248)27(0.004)
0.255
4(0.312)22(0.063)24(0.601)
25(0.023)
14
CG
0.457
4(0.959)25(0.024)27(0.016)
0.588
1(0.700)16(0.154)27(0.146)
0.455
4(0.961)25(0.024)27(0.014)
15
CH
0.976
4(0.104)24(0.209)27(0.687)
0.017
1(0.690)16(0.310)
0.938
4(0.387)22(0.001)27(0.613)
16
CG
1.000
71.000
19
1.000
317
CG
0.287
4(0.813)16(0.141)22(0.046)
0.173
1(0.695)16(0.305)
0.294
4(0.454)16(0.002)24(0.486)
25(0.058)
18
EE
0.497
4(0.570)16(0.207)22(0.223)
0.008
1(0.253)16(0.747)
0.495
4(0.569)16(0.208)22(0.222)
19
CG
0.684
8(0.353)24(0.556)27(0.092)
0.711
1(1.000)
0.698
8(0.353)24(0.556)27(0.092)
20
CG
0.650
4(0.221)25(0.056)27(0.723)
0.675
16(0.146)25(0.020)27(0.834)
0.666
4(0.258)25(0.059)27(0.683)
21
MD
0.543
22(0.912)25(0.088)
0.434
25(0.442)27(0.558)
0.544
22(0.912)25(0.088)
22
CG
1.000
11
0.298
16(0.695)25(0.305)
1.000
11
23
CH
0.404
4(0.679)22(0.038)27(0.283)
0.697
1(0.653)16(0.196)27(0.151)
0.404
4(0.679)22(0.038)27(0.283)
24
CG
1.000
60.876
1(1.000)
1.000
925
CG
1.000
81.000
71.000
10
26
MP
0.461
4(0.843)22(0.066)25(0.038)
27(0.053)
0.675
1(0.542)16(0.396)27(0.062)
0.476
4(0.891)22(0.067)25(0.041)
27(0.001)
27
EE
1.000
13
1.000
10
1.000
13
28
MD
0.916
4(0.650)16(0.346)25(0.004)
0.043
1(0.570)16(0.430)
1.000
1
5082 G Reiner and P. Hofmann
-
mean and median efficiency scores, as well as the lowest
standard deviation (i.e. atighter distribution). Process model 3
includes number of warehouses as extension to
model 1. Our results suggest that the pooling method of
warehouse centralization(i.e. a smaller input variable a17) has
improved the calculated efficiency for mostcompanies and is no more
a performance differentiator. In other words mostcompanies have
already realized the potential benefits of pooling to reduce risks
forthe supply chain processes. Our advice to the supply chain
managers of the analysedDMUs to further improve efficiency would
be, to put more emphasis on productcommunalities, postponement and
transport processes. There it seems that hepotential for
competitive differentiation is still big.
Table 6 shows further input for decision makers. We want to
verify the DEAresults with our theoretical knowledge about supply
chain processes. Good supplychain processes could be characterized
by short cash-to-cash cycle time (a6) andhigher profitability
(EBIT) as a percentage of revenue (a8/a21) (Hammel et al. 2002).The
efficient DMUs of this empirical study should be characterized by
highprofitability (EBIT) as a percentage of revenue (a8/a21) and
short cash-to-cash cycletime. The results of model 1 and 3 (see
table 6) verify this conjecture. The efficientDMUs exhibit shorter
average cash-to-cash cycle time and higher averageprofitability
(EBIT) as a percentage of revenue than the inefficient DMUs.
Model
2 supports only partially the conjecture since the average EBIT
as% of revenue forefficient DMUs is lower.
The correlation coefficient calculated for all relevant DMUs for
a6 and a8/a21 is0.196; i.e. relatively weak. In addition, the
coefficient is positive. This is the expectedresult for a
make-to-stock strategy. An increase in inventory leads to an
improvedservice level at the price of an increased cash-to-cash
cycle time. The higher servicelevel has a positive impact on
customer satisfaction and retention. This leadsgenerally to higher
revenue and profitability.
On the other hand, detailed analysis of correlation shows for
all efficient DMUsa strong negative correlation coefficient between
a6 and a8/a21 irrespective of theDEA process model. This confirms
clearly our main hypothesis, efficient supplychains (short
cash-to-cash cycle time) lead to high financial
performance.Considering the weak and positive correlation of
average DMUs strongly supportsour assertion that efficient
companies differentiate themselves from the rest byhaving best
practice supply chain processes established. The correlation
coefficientfor efficient DMUs is the most negative for model 2, the
most comprehensive processmodel (number of input variables) that
allows for major differentiation.
The results presented in table 6 have a limitation though. They
are not statisticalsignificant, due to the small sample size of
each group analysed.
Table 5. DEA efficiency results of alternative process
models.
BCCm1 BCCm2 BCCm3
Mean 0.708 0.566 0.715Median 0.700 0.651 0.714Standard deviation
0.258 0.324 0.252Coefficient of variation 0.364 0.573 0.353
Efficiency analysis of supply chain processes 5083
-
Table
6.
DEA
perform
ance
resultsofalternativeprocess
models.
Model
1Model
2Model
3
EfficientDMU
Non-efficientDMU
EfficientDMU
Non-efficientDMU
EfficientDMU
Non-efficientDMU
Key
perform
ance
measures
a6(M
ean)
69.4
105.0
78.4
99.0
88.9
98.0
a6(Standard
deviation)
49.8
69.3
70.3
66.4
72.0
65.2
a8/a21(M
ean)
0.16
0.13
0.11
0.14
0.15
0.13
a8/a21(Standard
deviation)
0.07
0.07
0.05
0.08
0.06
0.08
Coefficientofcorrelation
a6
a8/a21
0.47
0.37
0.78
0.30
0.47
0.42
5084 G Reiner and P. Hofmann
-
4. Conclusions
In this article we illustrated an integrated approach by
combining dependency
analysis and data envelopment analysis (DEA) to explore
benchmarking results of an
empirical study of supply chain processes employed by companies
from different
industries. The study used the process definition and
performance measures of the
SCOR model, a widely accepted industry standard. A decision
making unit (DMU)
was defined by the supply chain processes (source, make and
deliver) within one
company.The analysed companies are part of different industries
and geographies. They
generate a broad variety of annual revenues and follow a
make-to-stock manufac-
turing strategy. Using a balanced supply chain performance
scorecard the study
evaluated the supply chain performance.A dependency analysis of
the data from the empirical study was conducted by
means of TETRAD to identify the input and output variables for
DEA. The result of
TETRAD was used as decision support for the classification of
performance
indicators into input and output variables.We used the DEA model
with variable returns to scale (BCC) for analysing
DMUs with a make-to-stock production strategy. We illustrated
how to suggest best
practice processes as targets for improvement by DEA analyses.
Examples of best
practice performers among the efficient DMUs were shown. Supply
chain processes
represented by outputinput relations of such DMUs belong to the
best practice
companies.Besides suggesting best practice DMUs for
benchmarking, DEA showed
quantitatively how much a performance indicator could improve in
order to bring
the underperforming process to the best practice level.We showed
the DEA results for three process models with different number
of input variables. In particular, the distributions of the
efficiency scores were
presented. The results suggest that the pooling method warehouse
centralization
improves the calculated efficiency similarly for most companies.
In other words most
companies have already realized the potential benefits of
pooling to reduce risks for
the supply chain processes. Our practical advice to the supply
chain managers of the
analysed DMUs would be to put more emphasis on product
communalities,
postponement and transport processes in order to further improve
efficiency and
differentiate themselves from competitors.Finally we identified
dependencies between operational performance and
financial success at company level. The analysis of correlation
coefficients supports
our basic hypotheses that efficient supply chains lead to high
financial performance.
For the efficient DMUs we found for all three process models a
strong negative
coefficient of correlation between the DEA input variable
cash-to-cash cycle time
and the EBIT as a percentage of revenue. Taking further into
account that on the
other hand average DMUs show a weak and positive correlation
strongly supports
our assertion that efficient companies differentiate themselves
by having best practice
supply chain processes established. For the most comprehensive
process model (most
number of input variables) the correlation coefficient for
efficient DMUs is the most
negative. The latter suggests that the drivers of transport
logistics costs are essential
for productivity analysis.
Efficiency analysis of supply chain processes 5085
-
The diversity of results and possible analysis demonstrates the
usefulness of ourintegrated approach.
A limitation of our approach is the restriction to make to stock
productionstrategy. Further research activities could apply this
approach also to differentmanufacturing strategies. Another
constraint was the relatively small sample size dueto the limited
access to non-disclosure intra-company data.
It would be interesting to integrate customer satisfaction
measures (Reiner 2005)to interpret the relationship between the
different cost drivers, customer satisfactionand profitability with
DEA.
References
Banker, R., Charnes, A. and Cooper, W., Some models for
estimating technical scaleinefficiencies in data envelopment
analysis. Manage. Sci., 1984, 30(9), 10781092.
Boussofiane, A., Dyson, R.G. and Thanassoulis, E., Applied data
envelopment analysis.Euro. J. Op. Res., 1991, 51(1), 115.
Castillo, E., Gutierrez, J.M. and Hadi, A.S., Expert Systems and
Probabilistic NetworkModels, 1997 (Springer: New York).
Charnes, A., Cooper, W.W. and Rhodes, E., Measuring the
efficiency of decision makingunits. Euro. J. Op. Res., 1978, 2(6),
429444.
Christopher, M., Logistics and Supply Chain ManagementStrategies
for Reducing Costs andImproving Services, 1992 (Pitman Publishing:
London).
Cooper, W.W., Park, K.S. and Yu, G., IDEA and AR-IDEA: Models
for dealing withimprecise data in DEA. Manage. Sci., 1999, 45(4),
597607.
Cooper, W.W., Seiford, L.M. and Tone, K., Data Envelopment
Analysis: A ComprehensiveText with Models, Applications, References
and DEA-Solver Software, 2000 (KluwerAcademic: Boston).
Dahlgaard, J.J., Kristensen, K. and Kanji, G.K., Fundamentals of
Total QualityManagementProcess Analysis and Improvement, 1998
(Chapman & Hall: London).
Dyson, R.G., Allen, R., Camanho, A.S., Podinovski, V.V.,
Sarrico, C.S. and Shale, E.A.,Pitfalls and protocols in DEA. Euro.
J. Op. Res., 2001, 132(2), 245259.
EFQM, EFQM excellence model. Available online at:
http://www.efqm.org (accessed17 December 2004).
Gropietsch, J., Erfolgsfaktoren im Supply Chain Management der
Konsumguterindustrie.Doctoral thesis, University of Munster
(Westphalia), 2003.
Hammel, T., Kuettner, D. and Phelps, T., The re-engineering of
Hewlett-Packards CD-RWsupply chain. Supply Chain Manage.: An
International Journal, 2002, 7(3), 113118.
Homburg, C., Using data envelopment analysis to benchmark
activities. Int. J. Prod. Econ.,2001, 73(1), 5158.
Jammernegg, W. and Kischka, P., Dynamic, customer-oriented
improvement of supplynetworks. Euro. J. Op. Res., 2005, 167(2),
413426.
MBNQA, Baldrige National Quality Program. Available online at:
http://www.quality.nist.gov (accessed 5 January 2005).
Meyr, H., Rohde, J. and Stadtler, H., Basics for modeling. in
Supply Chain Management andAdvanced PlanningConcepts, Models,
Software and Case Studies, 2nd ed., edited byH. Stadtler and C.
Kilger, pp. 4570, 2002 (Springer: Berlin).
Narasimhan, R., Talluri, S. and Mendez, D., Supplier evaluation
and rationalisation viadata envelopment analysis: An empiric
examination. J. Supply Chain Manage., 2001,37(3), 2837.
Olhager, J., Strategic positioning of the order penetration
point. Int. J. Prod. Econ., 2003,85(3), 319329.
Post, T., Performance evaluation in stochastic environments
using mean-variance dataenvelopment analysis. Op. Res., 2001,
49(2), 281292.
5086 G Reiner and P. Hofmann
-
Ramanathan, R., An Introduction to Data Envelopment AnalysisA
Tool for PerformanceMeasurement, 2003 (Sage: London).
Reiner, G., Customer-oriented improvement and evaluation of
supply chain processessupported by simulation models. Int. J. Prod.
Econ., 2005, 96(3), 381395.
Roll, Y., Cook, W.D. and Golany, B., Controlling factor weights
in data envelopmentanalysis. IIE Transactions, 2004, 23(1), 29.
Ross, A. and Droge, C., An analysis of operations efficiency in
large-scale distributionsystems. J. Op. Manage., 2004, 21(6),
673688.
SAP and PRTM, Pittiglio, Rabin, Todd and McGrath (PRTM)/SAP
Benchmarking Study20022003: Supply Chain Planning. Available online
at: http://www.sap.com/company/press/press.asp?pressID2425 and
http://www.sap.com/company/press/pdf/PRTM_SCM_Benchmarking.pdf
(accessed 6 April 2005).
Scheel, H., EMS: Efficiency Measurement System, Users Manual,
Version 1.3, 2000(Dortmund).
Scheel, H., Undesirable outputs in efficiency valuations. Euro.
J. Op. Res., 2001, 132(2),400410.
Scheines, R., Spirtes, P., Glymour, C. and Meek, C., TETRAD II:
Tools for Discovery, 1994(Lawrence Erlbaum Associates: New
York).
SCOR, Supply Chain Council. Available online at:
http://www.supply-chain.org (accessed22 August 2005).
Spirtes, P., Scheines, R. and Glymour, C.N., Causation,
Prediction and Search, 1993(Springer: New York).
Sudit, E.F., Productivity measurement in industrial operations.
Euro. J. Op. Res., 1995, 85(3),435453.
TETRAD-Project, The TETRAD Project: Causal Models and
Statistical Data.Available online at:
http://www.phil.cmu.edu/projects/tetrad/index.html (accessed17
December 2004).
Talluri, S., A benchmarking method for business-process
reengineering and improvement.Int. J. Flex. Manuf. Syst., 2000,
12(4), 291304.
Talluri, S. and Baker, R.C., A multi-phase mathematical
programming approach for effectivesupply chain design. Euro. J. Op.
Res., 2002, 141(3), 544558.
Talluri, S. and Sarkis, J., A methodology for monitoring system
performance. Int. J. Prod.Res., 2002, 40(7), 15671582.
Tavares, G., A Bibliography of Data Envelopment Analysis
(19782001). RUTCOR ResearchReport, RRR 0102, January, Piscataway,
New Jersey, 2002.
Van der Vorst, J.G.A.J., Beulens, A.J.M., De Wit, W. and Van
Beek, P., Supply chainmanagement in food chains: Improving
performance by reducing uncertainty.Int. Trans. in Op. Res., 1998,
5(6), 487499.
Efficiency analysis of supply chain processes 5087