Sie sind auf Seite 1von 10

Computers in Industry 60 (2009) 104–113

Contents lists available at ScienceDirect

Computers in Industry
journal homepage: www.elsevier.com/locate/compind

Quantitative relationships between key performance indicators for supporting


decision-making processes
Raul Rodriguez Rodriguez *, Juan José Alfaro Saiz, Angel Ortiz Bas
Research Centre on Production Management and Engineering, Polytechnic University of Valencia, Camino de Vera S/N, Edificio 8G, Acceso D Planta 1 Valencia, 46022 Spain

A R T I C L E I N F O A B S T R A C T

Article history: Performance measurement systems (PMS) are tools widely used by enterprises for managing and making
Received 20 December 2007 strategy-based decisions. A PMS defines a group of strategic objectives and associated performance
Received in revised form 26 June 2008 indicators (KPIs) that provide information as to whether the upstream objectives are being reached or
Accepted 7 September 2008
not, but with no further information about the causes. Up to now, if an objective is not being reached
Available online 28 November 2008
managers do not have further information regarding the causes; in terms of accurate information they are
limited to the associated KPI. However, regarding the decisions to be made: What would they be based
Keywords:
on? How and where to dig to find cause-effect relationships? And, even more difficult: How to make it
Decision making
Performance measurement
objective? This study presents a unique proposal able to objectively – not based neither on experience nor
Relationships subjective judgments – identify and quantify relationships between performance elements defined
Information analysis within a PMS, offering additional information to managers to make cross-enterprise decisions. Finally,
the paper presents the main results obtained from applying the proposal to a real world enterprise and
future research lines.
ß 2008 Elsevier B.V. All rights reserved.

1. Introduction an approach under four pre-established perspectives. However, lots


of inter- and intra-perspective relationships, initially hidden, stay
For some time ago, enterprises have been using performance latent and have the potential to provide very relevant information
measurement systems (PMS) to try to measure their performance for any given enterprise’s management.
and as a management tool. Using the methodology inherent to their In general terms, a PMS defines some objectives derived from
own PMS, analysts define performance indicators that the enterprise strategy and one associated set of key performance indicators (KPIs),
think are the most appropriate for reflecting performance. It is following a top-down process. Then, once the PMS is implanted and
intended that the process of obtaining the results from these working, the different measures gathered by the KPIs inform
performance indicators will be adequate in terms of costs (resources whether the upstream objectives they are associated to are being
and time). However, at this point in time, all the effort invested is reached or not, following a bottom-up process. Thus, when a KPI has
neither totally repaid nor well-used by the management. This is due a value out of its allocated range, it is said that the upstream
to the fact that the performance indicators are used, almost objective to which it is associated is not being reached, without any
exclusively, for measuring (with higher or lower success) the further information or analysis. In this sense, both researchers and
degree of accomplishment of the defined objectives based on the managers [2–6] have recently pointed out the KPI level as a source of
expected results. Performance indicators may additionally provide additional meaningful information for organisations. They argue
very important information regarding the existing relationships that analysis at this level will lead to a better understanding of
between them that allow the re-planning of the objectives they are business relatedness by identifying relationships that take place
associated to, as well as improvement of the decision-making between the different KPIs. The immediate effect of discovering
process. Some of the most evident relationships are handled almost these relationships would be to clearly establish possible KPI
unconsciously by enterprises. Furthermore, some PMS, such as the redundancy. However, the fact of discovering relationships between
Balanced Scorecard [1], establish some relationships when defining KPIs is potentially much more profitable for an organisation if, once
identified, analysts projected them upstream towards the strategic
objectives and discovered the latent relationships that occur
* Corresponding author. Tel.: +34 645832160.
between objectives of the PMS. Then, cause-effect relationships
E-mail addresses: raurodro@upvnet.upv.es (R.R. Rodriguez), between objectives could be explicated and managers would have
jalfaro@omp.upv.es (J.J.A. Saiz), aortiz@omp.upv.es (A.O. Bas). additional decision-making information.

0166-3615/$ – see front matter ß 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.compind.2008.09.002
R.R. Rodriguez et al. / Computers in Industry 60 (2009) 104–113 105

Thus, the purpose of this paper is to present a proposal able to Looking at Fig. 1, it seems clear is that a previous condition for
quantify relationships in the performance measurement system applying QRPMS is that the PMS framework defined must support
context (QRPMS), which objectively identifies and quantifies and show a clear traceability between the performance measure-
relationships between KPIs and then projects them upstream in ment elements. Researchers have developed different PMS for
the PMS, establishing meaningful cause and effect relationships at managing organisations such as [8–11], etc. All of them coincide in
the objectives levels. defining objectives and some sort of indicators to monitor the
accomplishment of such objectives.
2. Research methodology However, none of the above proposals incorporate any
mechanism for identifying relationships between KPIs and poster-
The research followed a constructivist approach, based on the ior upstream projection. Thus, the search should focus on
following activities: recompilation, analysis and study of scientific techniques/frameworks that have been or could be applied at
knowledge, acquisition of main postulates and construction of the PMS context to objectively find relationships between KPIs.
initial framework [7]. At the performance management context, [12] affirms that the
The initial elements considered in the present constructivist multi-criteria decision aid methods (MCDA) are the most
approach were the following: commonly used, classifying them as follows:

 The PMS existing within this ambit.  Objective programming.


 The existing techniques used within the performance measure-  Scoring models.
ment context that have been used or could be used for the  Hierarchical techniques.
present research.  Deployment techniques.
 The requirements that should be covered by frameworks/
techniques to address this research. In general, MCDA establish a ranking of the different
competitive priorities for maximizing performance according to
The aim to accomplish all these requirements has been the base some given criteria [12]. In general, all these techniques carry out
for creating the QRPMS framework. subjective decisions at any stage or are simply not adequate to be
This paper is structured with an initial literature review of applied in the context of the present work. However, there is one
performance measurement, reviewing what has been already done framework inside the hierarchical techniques called quantitative
in this ambit for the identification and projection of relationships model for performance measurement systems (QMPMS) devel-
between KPIs. As a result of this analysis, the research space for the oped by Ref. [13], that needs to be further commented upon. This
present paper was established. Next, the quantitative relationships framework provides a method for quantifying relationships
performance measurement system (QRPMS) proposal is described. between factors that affect performance, not between KPIs.
Then, and to best clarify all the concepts associated with QRPMS, a Furthermore, it applies the Analytic Hierarchical Process (AHP)
practical application based on real data from a manufacturer [14] technique to quantify the established relationships, which is
organisation is presented. another hierarchical technique and subjective. For these reasons it
Finally, the most important conclusions reached, coming from is discarded for the present research.
both the theoretical proposal and from the practical application, At the performance measurement context, there are some
are established. frameworks that aim to identify any sort of relationship between
performance elements. Thus, [15] apply a correlation analysis to
3. Literature review quantify relationships between pairs of KPIs defined within the
design of a PMS. This work uses the quantified relationships for
Fig. 1 illustrates the idea of introducing an activity of identifying assessing one model designed to react to external changes that
relationships between KPIs of a PMS and their posterior upstream might affect one organization instead of projecting them upstream
projection within a PMS. in the PMS. Additionally, the usage of correlation techniques do

Fig. 1. Identification and projection of KPIs relationships.


106 R.R. Rodriguez et al. / Computers in Industry 60 (2009) 104–113

limit the analysis, as this technique carries out pair-wise From the above literature review it is possible to conclude that:
comparisons between pairs of KPIs instead of analyzing all the
KPIs at the same, losing the possibility of identifying crossed  There are a multitude of methods and frameworks applied to
relationships. both performance measurement and management, but almost
Ref. [16] focuses on establishing the relationships between KPIs all make subjective considerations at any step.
from two perspectives (internal and learning and growth) defined In practice, there are few frameworks that consider the idea
inside a balanced scorecard designed according to Ref. [1]. It underlying this research. The few ones whose research aim is
establishes the relationships between these indicators by applying close to the present work stop when they find relationship
the MIC-MAC method [17], which is a subjective process and between KPIs, not projecting them upstream in the pertinent
therefore will not be used in this research. PMS. Besides, the method used for finding these relationships is
Some other works [18–20] have in the past carried out different either inappropriate (subjective) or has got inherent limitations
correlation analysis in order to reduce the spectrum of KPIs to be (pair-wise comparisons) that makes impossible the finding of all
controlled and monitored by organisations. Again, the usage of the relationships.
correlation techniques makes the approach inadequate for this  Though not applied to this ambit extensively, the multivariate
work. techniques are the most appropriate. Then, down to data
Looking at the statistical literature, the multivariate analysis characteristics, either the principal component analysis or the
has got the potential to identify relationships between variables structural equation model for identifying relationships between
over time [21,22]. From these, the main a priori techniques to be KPIs, and the partial least square model for quantifying them are
applied are: factorial analysis, principal component analysis, the most appropriate methods.
structural equation models and the analysis of variance.  To resume, there is not any PMS framework that explicatively
From all of these, the factorial analysis should be discarded incorporates any mechanism for identifying, quantifying and
since it is necessary to specify the model, in other words, to projecting upstream the existing relationships between KPIs.
establish what are the subjacent variables, the observed Thus, there is a gap within these frameworks regarding efficient
variables, the error terms as well as possible relationships and effective guidelines and tools for analytically and graphically
[21]. Additionally, the analysis of variance needs special managing and monitoring the identification, quantification and
requirements such as independence, approximation to normal upstream projection of the relationships between KPIs in the
distribution and homoscedasticity, which are, a priori, very context of the PMS. Hence, a proposal for identifying quantitative
difficult to accomplish regarding the type of variables in this relationships at the performance management context (QRPMS)
work and associated data. has been developed. This proposal aims to overcome the
However, when assessing the other techniques, the decisional detected weaknesses in the above-mentioned frameworks. The
factor has to do with the proportion of the number of variables main particularities of this proposal are described more deeply in
for the study (number of columns in the initial data matrix: the the next point.
KPIs) and the number of observations available for each of these
variables (number of rows in the initial data matrix). Thus, the 4. QRPMS proposal
structural equation model should be only applied when
the proportion of variables/observations is, at least, 3:1 [22]. In the light of the previous points, this point presents the
On the other hand, the principal component analysis gives good methodology called quantitative relationship at the performance
results when this proportion does not take place and even when measurement system that is the main contribution of this paper.
there are missing values in the initial data matrix [23,24], The main goals of QRPMS are the following:
especially when applying a further model that uses partial least
squares (PLS) in the calculation of principal component analysis  To be a standard methodology to be applied to any PMS that
[25,26]. Additionally, within this context [27] offers some light, clearly establishes traceability between their objectives and
as they carried out an analysis between possible techniques to be associated KPIs.
applied for identifying relationships between variables, con-  To identify and quantify relationships between KPIs objectively.
cluding that the principal component analysis was the most This is very important, as the intention is to manipulate real data
appropriate. gathered by the KPIs and applying objective mathematical
But once relationships between KPIs have been stated, it is techniques to thus obtain objective relationships, where no
necessary to quantify, in magnitude and sense, such relationships subjective opinions can bias the achieved results.
in order to assess their importance. This is one of the core  To construct KPI cause-effect maps based on the identified and
objectives of the present methodology, as the quantification of quantified relationships. This will directly lead to the discovery
relationships between the KPIs will allow determining their of possible KPI redundancy when the relationship between the
importance depending on whether the relationship is weak (low KPIs is very high and then to find out about non-evident, a priori,
value) or strong (high value). This will lead, as it will be shown relationships between the KPIs. To do this, visual support
later, to discriminate for the study those relationships found to be graphics will be provided.
weak. To do this, different models can be applied, though the  To provide the necessary mechanisms for projecting these KPI
regression ones have probed to be appropriated, and more cause-effect relationships upstream in the PMS towards the
concretely the partial least squares models as they overcome objectives ambit.
the main problem of the classical regression models, the co-  To construct objective cause-effects maps based on such a
linearity [21–25]. This combination of principal component projection. Visual supporting graphics will be also developed in
analysis/structural equation model plus PLS model has been this phase. This will provide with additional information to
already applied in several fields, even in the performance decision makers in the most strategic context, as relationships
management one, when [28] used them to identify and quantify between objectives will become established and demonstrated
relationships between KPIs defined within a PMS and then using based on the analysis of real data.
these relationships to draw scenarios regarding an organization’s  To identify these main KPIs whose variation could potentially
future performance. provoke the non-achievement of objectives different to the ones
R.R. Rodriguez et al. / Computers in Industry 60 (2009) 104–113 107

with which they are associated. Then, an organisation could therefore should be considered for the study instead of being
focus its efforts on closely controlling and monitoring these rejected.
causal KPIs improving therefore its understanding about how its Since the data will have been collected over different time
PMS evolves and having a better management tool for supporting intervals (yearly, monthly, daily, etc.), it is necessary to homo-
decision-making processes. genise it to a common frequency in order to have it all in the same
temporal frequency. At this stage, the decisional time horizon of
QRPMS has got the following four phases (Fig. 2): the analysis should be taken into account. For instance, for an
annual analysis, all the data coming from the KPIs, from all the
1. Design and analysis of the PMS. different time-periods, will be homogenised to a yearly frequency.
2. Initial data treatment. Different homogenisation operations might be applied to each KPI
3. Identification and projection of KPI relationships. depending on the sort of distribution its data presents. Then, a
4. Presentation and analysis of results. simple mean will be applied for normally distributed data and
either the median or the geometric media could be applied to very
Each phase has different main activities to be carried out and asymmetric or with high kurtosis coefficient distributions.
they are all described and justified next. The last task of this phase implies the application of centring
operations, which will allocate the same degree of importance to
4.1. Phase 1. Design and analysis of the PMS each variable (KPI). Since the KPIs are very heterogeneous with
regard to their measurement units (s, time, %, etc.) the centring
In the first step the enterprise designs and implements a PMS. technique to be applied in this proposal is that of auto-scaled, as
As commented above, there are different frameworks for designing this technique combines centring and standardisation techniques
a PMS. In order to apply this proposal the only condition to be met [21].
by the PMS used is that it provides a clear traceability between its As a result of applying this phase, the data coming from the KPIs
performance elements; in other words, it must clearly establish the will form a matrix of data filtered, homogenised in frequency and
links between the objectives and the associated performance standardised, and ready for application of the proper mathematical
indicators. techniques for identifying relationships between the KPIs. This
Once the PMS has been designed, it has to be implemented and task is carried out in the next phase.
analysed, then decision makers should check out the consistency of
the different objectives for each perspective and associated KPIs in 4.3. Phase 3. Identification and projection of KPIs relationships
order to detect possible failures and correct them.
This phase identifies and quantifies relationships between KPIs
4.2. Phase 2. Initial data treatment by applying adequate techniques to the initial data matrix reached
in the previous phase. The question now is: What mathematical
In this phase the data collected by the KPIs is brought together technique/s is/are suitable to be applied to identify relationships
in order to begin its analysis. Enterprises keeping electronic between KPIs? This question was solved above, in the literature
registers will much more easily complete these operations, review, which concluded that the principal components analysis
whereas those enterprises with low IT levels will have to invest (PCA) could be, a priori, applied to any initial data matrix
more resources. Once the data is available three operations will be regardless of the proportion between the number of variables
applied: filtering, homogenisation and centring. and the number of observations. Besides, the structural equation
The application of filtering operations will lead to the model (SEM) could be applied every time that the proportion above
identification any anomalous behaviour of the data that could pointed out occurs. Consequently, and with regard to the initial
potentially bias the analysis. Then different statistics such as: data matrix, the analyst will apply one or another of these two
mean, standard deviation, mode, range, maximum and minimum techniques.
values, percentiles, histogram and normal graphic representation, Then, an initial exploratory analysis takes place when applying
etc., could be applied to each KPI [21]. Should this analysis be either the PCA or the SEM to all the KPIs of the study. This
conducted by external analysts, the participation of internal exploratory analysis will identify relationships between KPIs. In a
experienced personnel is necessary. This is due to the fact that second step, a confirmatory analysis is conducted applying again
what an external analyst could consider as abnormal behaviour it either the PCA with least squares or the SEM, but with the
might represent a normal situation of the organisation and difference that PCA or SEM is applied only to these KPIs that the

Fig. 2. QRPMS methodology: phases.


108 R.R. Rodriguez et al. / Computers in Industry 60 (2009) 104–113

exploratory analysis pointed out that kept any sort of cause-effect Once the Graphic of the BDKPI relationships has been
relationship. This is done to confirm the results of the exploratory constructed, it is time to project the identified cause-effect
analysis and it only takes into account related KPIs, with the aim to relationships upstream in the PMS. Then, a new graphic, called
avoid the possible noise that other non-related KPIs could Graphic of projections deployment, is built from the Graphic of
introduce into the system. Finally, those KPIs maintaining BDKPIs relationships. Since the traceability between the objectives
cause-effect relationships are identified and they are named and their associated BDKPIs is clear and unequivocal, the modus
‘‘Business Drivers Key Performance Indicators’’ (BDKPI) due to the operandi focuses on substituting, in the Graphic of BDKPI
importance they have in the management and evolution of the relationships, each BDKPI for its associated objective, maintaining
enterprise. the relationships. Consequently, the Graphic of projections
Since it will be possible to apply the PCA with least squares deployment will show the cause-effect relationships between
technique in most of the cases, regardless of the characteristics of objectives, and therefore analysts will either confirm suspected
the data, as was stated in the literature review, the process of relationships they had conducted based on experience, or to
identifying the BDKPIs by applying the PCA is highlighted next. become aware of new relationships that take place within the
The results achieved from applying the PCA are graphically organisation.
interpreted. First, the analyst needs to decide how many principal Thus, supported by the quantified cause-effect relationships
components to retain for the study. This is done based on the between objectives, analysts have available new additional
percentage of variability that the different principal components information about the behaviour of the organisation, and how
explain, applying the called Kaiser’s criterion [29]. Analysis of the important relationships take place between their main strategic
principal components retained for the study is made pair-wise, and objectives. Without doubt, this new source of information, created
there the BDKPIs will be the ones that best explain the total following objective analysis of real data of the organisation will
variance, all these contained within the two concentric ellipses in a open new alternatives to managers when making decisions. Then,
correlation loadings graphic. for instance, when one objective is not being reached, looking at
Once all the BDKPIs have been identified it is time to quantify, in the Graphic of projections deployment, managers will be able to
magnitude and sense, all the identified relationships. To this end, quickly find out whether this specific objective (effect objective)
partial least squares models are applied, as justified above. This has being influenced by another (cause objective) or not. If so, it
model comes represented by a typical regression equation, which will be possible to investigate whether the action of this cause
shows how good the model is in predicting either one effect objective is the one responsible for this non-accomplishment of the
variable from one or more cause variables (called PLS1 models) or objective effect, by analysing in the Graphic of BDKPIs the
more than one effect variable from more than one cause variable relationships which are the corresponding BDKPI effect and BDKPI
(called PLS2 models). At this stage, analysis has to decide which PLS cause. From these two, the BDKPI cause will be called ‘‘Causal
models they want to construct, as they need to specify what the business driver key performance indicator’’ (CBDKPI), and the
cause variables are and what the effect ones are. However, analysts organisation will have to put special emphasis on monitoring and
have an important source of information in the correlation controlling all these CBDKPIs, as changes of the values of them may
loadings matrixes, as these variables, symmetric by one axis or lead to variations of other BDKPIs with the consequent domino
in diagonal through the centre (intersection point of both principal effect upstream in the PMS.
components), are willing to maintain the highest correlation [30]. As a resume, QRPMS is a generic methodology, as it can be used
Additionally, though not least important, this design of PLS models by any organization whose PMS has traceability between the
should be conducted with the enterprise’s managers, as they, in the objectives and associated KPIs. Additionally, QRPMS is able firstly
light of the resulted BDKPIs, will be more interested in quantifying to objectively identify cause-effect relationships between the KPIs
specific relationships between BDKPIs than in other. Consequently, and secondly to project such relationships towards the objectives
the design of PLS models is an important task to be developed level. Moreover, and due to the operations carried out in phase 2 of
conjointly by external analysts (if any) and enterprise managers. the methodology, it is possible to affirm that the QRPMS can
Then, with the completion of this phase, all important cause- identify and quantify objectively relationships between variables,
effect relationships between KPIs have been identified and KPIs, independently of the nature of the data gathered by these
quantified. Then, the Graphic of the BDKPI relationships can be KPIs (percentage, money, numbers, etc.). On the other hand, the
drawn. Such a graphic is included in the next and last phase of the data gathered by the KPIs will come defined in the PMS, according
proposal. to their associated objectives, and it will be related to any process
that the enterprise is interesting in monitoring, controlling and
4.4. Phase 4. Presentation and analysis of results measuring, as derived from its strategy.
Then, QRPMS helps to identify objectively (based on the
This phase builds on the work carried out in the previous ones. evolution of real data gathered by the KPIs) important relation-
Then, it first shows a Graphic of BDKPI relationships, which ships between all the elements of a PMS that before to the
represents all the cause-effect relationships between BDKPIs, application of QRPMS were unknown (latent) or objectively
indicating the sense and the intensity of the relationship. This indemonstrable for the organization, as managers perceived that
intensity will be classified as either very strong or strong-average there were relationships between the different performance
and represented by a continuous or discontinuous arrow elements, but they could only based on experience and subjective
respectively. Additionally, all the BDKPIs are presented within judgments to establish them.
the perspective they belong to, which makes easier their posterior
projection upstream and also gives a more complete global picture 5. Application
of the different relationships. Looking at this graphic, analysts can
study the different cause-effect relationships and confirm QRPMS has been applied to one enterprise with 230 people that
expected relationships, which have been demonstrated with this manufactures baby clothing, in the top 5 of the Spanish market.
study, and also about non-expected relationships that may arise. At This enterprise had already implemented a PMS in previous years,
this stage, redundancy between BDKPIs might be identified as well and in this case the application of QRPMS took around 3 months,
as to clearly identify what the main BDKPIs causes are. obtaining very interesting results, which are presented below.
R.R. Rodriguez et al. / Computers in Industry 60 (2009) 104–113 109

5.1. Phase 1. Design and analysis of the PMS define, for instance, four strategies to reach one associated
objective but, at the time of defining the PMS, it might not have
At the moment of implanting QRPMS, the enterprise had the necessary resources to implement all four, having to
already one PMS implanted and fully working. This PMS was implement for example only two of them.
constructed based on the PMS IE-GIP proposal [11].  Key performance indicators: These indicators might be defined
PMS IE-GIP clearly provides traceability between all the for measuring either one objective or one strategy.
performance elements, drawing some supporting graphics for
representing traceability called partial deployment graphics. The 5.2. Phase 2. Initial data treatment
usage of this type of graphics, or similar, is highly encouraged as
they facilitate the posterior upstream projection of the identified In this phase the data collected by the KPIs is brought together
relationships between KPIs. Additionally, PMS IE-GIP also distin- in order to start its analysis. Table 1 shows the set of KPIs defined
guishes between the next performance elements: for this PMS, having a total of 42 KPIs and data corresponding to 11
years, from 1995 to 2006. Operations of filtering, homogenisation
 Objectives: What to measure and control within the time horizon and centring were applied to the data coming from these KPIs.
for which the PMS has been designed. This is equivalent to the Thus, all the 42 KPIs overcame the filtering operations, as no
classic ‘‘strategic objectives’’ of Ref. [1]. abnormal behaviour was detected when applying the above
 Strategies: How to reach the defined objectives. One objective pointed out statistics.
will have one or more strategies. The frequency of homogenisation chosen was annual, as the
 Critical success factors: This element will be constituted by the managers had defined the PMS for a 1 year term and they decided
objective it is associated to and by the strategy/ies from the ones that the results of this analysis would be more profitable shown on
above defined that the enterprise wants to use at the moment of an annual basis. Then, for instance, the data coming from the
defining the PMS. This is due to the fact that an enterprise might customer KPI CP9 (number of orders per week) was aggregated by

Table 1
KPIs of the PMS.
Financial perspective
PF1 Net benefit (sales)
PF2 Turnover increment
PF3 Costs of structure
PF4 Financial indebtedness
PF5 Final stock of cloths per campaign

Customer perspective
PC1 Customer average order
PC2 Number of customers that ask for repetition/number of total customers
PC3 Number of repeating requests per model and customer/number of customers that ask for repetition
PC4 Percentage of annulled models
PC5 Efficacy of the model
PC6 Global efficacy of each collection per intermediary
PC7 Average order per intermediary
PC8 Number of new customers captured
PC9 Number of orders per week (sales speed)
PC10 Increment of customers
PC11 Number of captured customers that are already customers of other enterprise of the organisation/number of new customers
PC12 Number of non-accomplished orders/total number of orders
PC13 Number of customers that make a claim/total number of customers
PC14 Number of claims per model and customer/total number of customers that make a claim
PC15 Increment of incidences occurred between the last two years
PC16 Average reaction time to one customer’s claim

Internal perspective
PP1 Stock variability
PP2 Percentage of return item per model
PP3 Percentage of customer acceptation
PP4 Total number of exited models/total number of produced models
PP5 Number of clothes with deviations respect to the standard scaled
PP6 Customer accomplishment degree
PP7 Deviation between the order date and the reception date
PP8 Total number of innovative clothes/total number of produced clothes
PP9 Number of innovative successful clothes/total number of innovative clothes
PP10 Number of re-processed clothes/total number of clothes
PP11 Number of customers buying a determined style/total number of customers
PP12 Number of annulled catalogues/total number of catalogues

Learning and growth perspective


PA1 Number of critical processes totally documented/total number of critical processes
PA2 Number of critical processes controlled by the follow-up tool/total number of critical processes
PA3 Number of yearly training hours/number of yearly work hours
PA4 Hours needed of real training (detected by the enterprise)/hours needed of theoretical training (that employees think they should have)
PA5 Number of established inter-areas work/communication protocols/number of essential processes
PA6 Establishment of inter-areas work/communication protocols (Yes/No)
PA7 Number of inter-areas personnel interchanges/number of areas susceptible of interchange within the enterprise
PA8 Development of an inceptive plan (Yes/No)
PA9 Number of employees that know exactly the objective to be reached/total number of employees
110 R.R. Rodriguez et al. / Computers in Industry 60 (2009) 104–113

Fig. 3. Correlation loadings for principal components 1 and 2.

making a simple mean and transformed to annual frequency, which This technique is first applied to the initial data matrix in an
was done for all the 11 years. This operation was extended to all the exploratory analysis, keeping then only the three first for this
KPIs that came under other frequencies different to the annual. analysis, as they together explained 97.4% of the total variability of
Finally, centring operations were applied to the data of the KPIs the initial data matrix.
already filtered and homogenised in frequency. Then, it was possible to determine the BDKPIs, which are those
As a result of this phase, an initial data matrix with 42 KPIs (in situated between the two concentric ellipses in Fig. 3 (for the pair
columns) and 11 observations/years (in rows) was formed. of principal components 1 and 2) and in Fig. 4 (for the pair of
principal components 1 and 3).
5.3. Phase 3. Identification and projection of KPI relationships The BDKPIs obtained with the analysis are shown in Table 2. It is
important to point out that there are BDKPIs that appear in both
This phase applies a suitable technique to the initial data matrix Figs. 3 and 4, this is because the analysis is made under a pair-wise
to identify relationships between the KPIs. In this particular case, base, as stated above in the paper, and then the principal
the proportion between the number of observations and the component 1 is present in both figures.
number of variables is 11/42 and thus, as justified above, the Then, principal component analysis is again applied but only to
feasible technique to be applied is the principal component the BDKPIs in the confirmatory analysis in order to avoid the
analysis with partial squares. possible noise that the other KPIs could introduce to the analysis.

Fig. 4. Correlation loadings for principal components 1 and 3.


R.R. Rodriguez et al. / Computers in Industry 60 (2009) 104–113 111

Table 2
BDKPIs of the study.

Perspectives BDKPIs

Principal component 1 Principal component 2 Principal component 3

Financial PF1, PF2 PF3 PF5


Customer PC1, PC8, PC9, PC10 PC15 PC12, PC13, PC16
Internal PP1 PP3, PP8, PP9 PP8, PP9, PP10
Learning & growth PA2, PA3, PA8 PL1, PL3, PL4, PL6, PL7

The results obtained concluded that the same BDKPIs were placed
again in the area of maximum variance explained, and therefore
the final BDKPIs of the study are those presented in Table 2.
An important fact is that, from a total of 42 KPIs at the beginning
of the study, the analysis has identified relationships between 24
KPIs (the BDKPIs showed in Table 2). This is a first important result
for managers, as the analysis allows them to establish new
relationships between indicators as well as to either demonstrate
or reject other suspected, based on experience, relationships.
At this point, PLS models are applied to quantify the relation-
ships between the identified BDKPIs. In this regression models, it is
necessary to set what is/are the cause-effect variable/s and what is/
are the effect variable/s. Enterprise managers and analysts agreed
on the next general models:

General model 1:
Explicative (cause) variables: PF2, PP8, PP9, PP1, PC8, PC9, PC10,
PA3.
Fig. 5. Graphic of BDKPIs relationship.
Effect variables: PF1, PF5, PC15, PP3.
General model 2:
Explicative (cause) variables: PA1, PA3, PA4, PA6, PA7. For instance, and focusing on the financial objective 1 (to
Effect variables: PF3, PF5, PC12, PP9. increment the net profit by 10%), FO1, this is the most important
General model 3: objective for the enterprise, and this study has pointed out several
Explicative (cause) variables: PC3, PC7, PC9. objectives and strategies that cause strong effect both directly and
Effect variables: PF1, PP1. indirectly, as well as positively and negatively to this FO1. All these
relationships are shown in Table 3.
PLS1 models and PLS2 models were conducted for all these Then, analysts have additional information about what
three general models, carrying out 13 totals PLS models, and strategic performance elements are affecting FO1; these are the
obtaining important results that were analysed both individually following:
and conjointly to have a global picture of the differing quantifica-
tion of these cause-effect variables. Thus, all the relevant  PO1 ( ): This process objective of ‘‘Reduction of the excess
relationships, those with a high regression coefficient, were taken between the produced and the sold to not exceed the 15% of the
into account for drawing the Graphic of BDKPIs relationships, real sales’’, is directly, strongly and negatively affecting FO1. This
which is shown in the next phase. means that, negative variations of PO1 (achievement of this
objective) will cause positive variations on the effect objective
5.4. Phase 4. Presentation and analysis of results (FO1). Looking at the BDKPIs level, this means that positive
variations of PP1, BDKPI associated to the cause objective PO1,
With all the analysis carried out in the previous phases, it is
possible to build the correspondent Graphic of BDKPIs relation-
ships (Fig. 5).
From this graphic of BDKPIs it is possible to set cause-effect
relationships between different BDKPIs, which are also quantified
in both sense and magnitude. From the results obtained in the last
phase, only those strong relationships between BDKPIs are
represented in the graph. However, inside this category of strong
relationship, two subcategories are represented in the graph with
very strong (continuous line) and strong (discontinuous line).
At this point, the projection of this Graphic of BDKPIs
relationships is carried out, by substituting each BDKPI by its
correspondent objective/strategy in the PMS (Fig. 6).
This Graphic of projections deployment shows the projection
upstream of the previous graphic of BDKPIs relationships. Analysts
can now establish cause-effect relationships between the main
strategic performance elements of the PMS: the objectives and the
strategies. Fig. 6. Graphic of projection deployment.
112 R.R. Rodriguez et al. / Computers in Industry 60 (2009) 104–113

Table 3
Relationships for the effect objective FO1.

Objective Objectives with direct Strategies with direct and Objectives/strategies with indirect CBDKPI
effect and strong relationship strong relationship and strong relationship

FO1 PO1 ( ) PP1


LO3 (+) PL3
LO1 ( ) PL1
CS1.3 (+) PC3
CS1.4 (+) PC7
PS3.2 (+) PP9
LS3.1 ( ) PL4

will directly and negatively impact PF1, BDKPI associated to the training needs (LS3.1). This negative impact on the enterprise’s profit
effect objective FO1. Therefore, PP1 becomes a CBDKPI. may be due to the investment needed to be made in order to put
 LO3 (+): This learning and growth objective of ‘‘Development of a them into practice within the enterprise. However, the impact is
customised training plan’’ is directly, strongly and positively important but relative as they maintained an indirect relationship
affecting FO1. Positive variations of LO3 will produce positive with the FO1. It would be acceptable to think that, in the future, this
variations of FO1. Therefore, at the BDKPIs level, positive negative incidence could either turn into a positive one or even
variations of PL3 will lead to positive variations of PF1. disappear once the actions they propose have been carried out.
Consequently, PL3 becomes a CBDKPI. At the indicators level, and for the FO1, mangers should mainly
 CE1.3 (+): This customer strategy of ‘‘To improve the relationship focus on five CBDKPIs: PP1, PL3, PP9, PC3 and PC7; taking also into
quality-price’’ is directly, strongly and positively affecting FO1. account the two CBDKPIs of PL1 and PL4. If the PMS reveals that
Then, positive variations of CSE1.3 will lead to positive variations FO1 is not being reached this might be because any of these
of FO1, and at the BDKPIs level, positive variations of PC3 will CBDKPIs is affecting PF1 and forcing it out of its range. Managers
produce positive variations of PF1. PC3 becomes then a CBDKPI. should then look at the evolution and the accomplishment of these
 CS1.4 (+): This customer strategy of ‘‘To set clear objectives to seven CBDKPIs and associated objectives/strategies.
intermediaries regarding the date/volume/sale’’ affects directly, Conducting similar analysis for all the other effect objectives/
positively and strongly FO1. Positive variations of CS1.4 exercise strategies (FO3, FS1.4 and CO3) shown in the Graphic of projections
positive variations of FO1. Thus, the associated BDKPI of CS1.4, deployment (Fig. 6) the study has detected other CBDKPIs, that
PC7, becomes a CBDKPI. added to the above presented, provide the next final list of
 PS3.2 (+): This internal process strategy of ‘‘To find internal and CBDKPIs: PF2, PC3, PC7, PP1, PP8, PP9, PL1, PL3, PL4, PL6 and PL7.
external innovation sources’’ affects directly, strongly and Hence, from 42 KPIs the study has revealed that there are eleven
positively to FO1. This means that positive variations of PS3.2 especially important KPIs (the CBDKPIs) to be controlled and moni-
will produce positive variations of FO1. Thus, positive variations tored over time, as their variation will lead to variations on important
of PP9 will lead to positive variations of PF1, with PP9 becoming a effect performance elements of the PMS, as well as to provide mana-
CBDKPI. gers with important additional information for making decisions.
 LO1 ( ): This learning & growth objective of ‘‘To document the Deriving from the application of QRPMS to this enterprise, and
critical process of the enterprise and develop specific procedures what it has offered to managers, it is possible to establish the
for determined processes/work places’’ affects indirectly, following practical comments:
strongly and negatively FO1. Then, positive variations of LO1
will produce negative variations of the FO1. At the BDKPI level,  Mangers found that the application of QRPMS was, in practice, a
positive variations of PL1 will lead to negative variations of PF1, non-difficult task. This was due, mainly, to the action of an
with PL1 becoming a CBDKPI. external analyst conducting the analysis and to the fact that their
 LS3.1 ( ): This learning & growth strategy of ‘‘To detect training presence and interaction on the analysis was nonnegotiable,
needs’’ affects strongly, indirectly and negatively FO1. Therefore, becoming an active part of such an application.
positive variations of LS3.1 will cause negative variations of FO1.  Managers of the enterprise showed their enthusiasm and
Positive variations of PL4 will lead to negative variations of PF1, support to the results achieved by the application of QRPMS,
with PL4 becoming a CBDKPI. as they at least could have a solid framework that served to
demonstrate real interactions between KPIs and also to project
To sum up, managers analysing this effect objective FO1 (to them upstream the PMS, with consequent additional informa-
increment the net profit by 10%) and its different cause objectives tion about both the PMS elements in particular and of the
and strategies could come to several conclusions. Thus, in order to enterprise in general.
reach the objective FO1, the enterprise should mainly focus its  In practice, the main problem this enterprise has claimed to have
resources on reducing its manufacturing variability (PO1); fostering had to overcome when applying QRPMS is the fact of having all
customised training plans (LO3); improving their quality–price the data coming from the indicators available.
relationship (CS1.3), cutting costs and improving quality level;
improving relationships with its intermediaries by setting better and 6. Conclusions
clearer objectives regarding volume, date and sale data (CS 1.4);
finding internal and external innovation sources (PS3.2), as this stra- This paper has described a new methodology called QRPMS,
tegy, in principle not very important, with this analysis has revealed which is able to objectively find and quantify relationships
to be of key importance to the economic results of the enterprise. between KPIs defined within a PMS and then project them
Additionally, managers should also take into account the upstream towards the strategic level of the PMS, then becoming a
negative impact on FO1 that the investment on documenting the management support tool providing additional information to
critical process of the enterprise and developing specific procedures decision makers. After analyzing the most applied techniques and
for determined processes/work places (LO1) as well as on detecting frameworks in the performance measurement and management
R.R. Rodriguez et al. / Computers in Industry 60 (2009) 104–113 113

area that have been or could be applied in this ambit, and having [20] I. Lange, O. Schneider, M. Schnetzler, L. Jones, in: J. Olhager, F. Persson (Eds.), IFIP
International Federation for Information Processing, Volume 246, Advances in
probed the research opportunity, QRPMS was defined to offer a Production Managment Systems, Springer, Boston, 2007, pp. 379–386.
solid, complementary and efficient framework to organizations to [21] J.F. Hair, R.E. Anderson, R.L. Tatham, W.C. Black, Multivariate Data Analysis,
improve their management. QRPMS can be applied to any Prentice Hall International Inc., New York, 1995.
[22] J.E. Jackson, A User’s Guide to Principal Components, Hoboken, New Jersey, 2003.
organization that has a PMS that assures traceability between [23] P.P.C. Nelson, P.A. Taylor, J.F. MacGregor, Missing data methods in PCA and PLS:
their performance elements (objectives, strategies and KPIs), score calculations with incomplete observations, Chemometrics and Intelligent
applying different techniques for identifying (principal component Laboratory Systems 35 (1993) 45–65.
[24] S. Wold, K. Esbensen, P. Geladi, Principal component analysis, Chemometrics and
analysis or structural equation model) and quantifying (partial
Intelligent Laboratory Systems 4 (1987) 37–52.
least squares) the exiting relationships between KPIs. [25] P. Geladi, B. Kowalski, Partial least-squares regression: a tutorial, Analytica
Among the main characteristics of QRPMS are the following: Chimica Acta 185 (1986) 1–32.
[26] H. Martens, M. Martens, Multivariate Analysis of Quality, John Wiley & Sons Ltd.,
New York, 2001.
 Confirmation of a priori, and based on experience, some intuited [27] R. Rodriguez, J.J. Alfaro, A. Ortiz, Relationships among key performance indicators
relationships, quantifying them in sense and magnitude. within the performance measurement system context: literature review, in: PMA
 Discard of a priori, and based on experience, some intuited Conference, London, (2006), pp. 689–701.
[28] B. Patel, T.J. Chaussalet, P.H. Millard, Balancing the NHS balanced scorecard,
relationships, quantifying them in sense and magnitude. European Journal of Operational Research 185 (2008) 905–914.
 Objective demonstration of the existence of other important [29] H.F. Kaiser, The application of electronic computers to factor analysis, Educational
relationships between performance elements, quantifying them and Psychological Measurement 20 (1960) 141–151.
[30] M. Aluja, D. Marineau, Practical Application of the Principal Component Analysis,
in sense and magnitude. Palais, Paris, 1999.
 Identification of the CBDKPIs, whose monitoring and control will
Dr. Raul Rodriguez Rodriguez (Industrial Management
help to manage the enterprise more efficiently, as they will offer Engineer, MBA, PhD) is a lecturer in Operations
new additional information when associated objectives/strategies Management and Operations Research at the Poly-
upstream are not being reached. Additionally, managers might cut technic University of Valencia (UPVL). He has worked in
costs, as they can focus on closely monitoring only these CBDKPIs, several projects about performance management,
operations management, supply chain management
reducing the number of indicators to be controlled.
and information systems in different activity sectors
 Better knowledge about their enterprise, which will directly such as automotive, retail, consumer goods, ceramic or
affect re-definition activities of the PMS as well as management textile. Additionally, he is a full member of the CIGIP
in general. (Research Centre on Production Management and
Engineering) at the UPVL, having participated on
Future research should focus on enriching the proposal by several research projects at European level (ECOSELL,
applying QRPMS to others organizations, as well as to adapt the GPM-SME) as well as on national research projects (INPREX, SP7). He lectures in
both graduate and undergraduate courses and his research preferences are on
framework to inter-enterprises contexts, where the application of performance measurement/ management, operations management and decision-
QRPMS can help managers to make better conjointly taken decisions. making sciences. He is a Certified Quality Engineer and a full member of the
American Society for Quality. He has published several papers in books,
References international journals and delivered some conferences in these fields.

[1] R.S. Kaplan, D.P. Norton, The balanced scorecard. Measures that drive perfor- Dr. Juan José Alfaro Saiz is an assistant professor in
mance, Harvard Business Review (January/February) (1992) 71–79. Operations Management and Operations Research at the
[2] D. Itner, D. Larker, Coming up short on nonfinancial performance measurement,
Polytechnic of Valencia. He is an Engineer in Business
Harvard Business Review (November) (2003) 91–105.
Organization and he received his doctoral degree in
[3] J.J. Alfaro, PMS IE-GIP, Doctoral Thesis, Universidad Politécnica de Valencia, 2003.
Business Organization at the Polytechnic University of
[4] K.A. Merchant, Measuring general managers’ performances market, accounting
and combination-of-measures systems, Accounting, Auditing and Accountability Valencia in 2003. He is a member of the CIGIP (Research
19 (6) (2006) 893–917. Centre on Production Management and Engineering). He
[5] C. Wolf, P. Harmon, The State of Business Process Management, Business Process has worked as a Researcher in several Spanish Govern-
Trends, 2006 June, Accessed 10th March 2007, Available at www.bptrends.com. ment Projects (CICYT), one ESPRIT Project, V-CHAIN
[6] H.S. Jagdev, A. Brennan, J. Browne, Strategic Decision Making in Modern (Virtual Enterprise for Supply Chain Management),
Manufacturing, Springer, New York, 2003. ECOSELL (Extended Collaborative Selling chain), and
[7] P. Coughlan, D. Coughlan, Action research: action research for operations manage- others. He is a teacher in the Polytechnic University of
ment, International Journal of Operation and Productions Management 22 (2) Valencia. His research interests include performance
(2002) 220–240. measurement systems, supply chain management, integration enterprise and
[8] K.F. Cross, R.L. Lynch, The SMART way to sustain and define success, National modelling process business. He has published several papers in books, journals
Productivity Review 8 (1) (1988) 23–33.
and conferences in these fields.
[9] A.A. Atkinson, J.H. Watherhouse, R.B. Wells, A stakeholder approach to strategic
performance measurement, Sloan Management Review 38 (3) (1997) 25–37.
Dr. Angel Ortiz Bas is an assistant professor in
[10] A.D. Neely, C. Adams, P. Crowe, The performance prism in practice, Measuring
Business Excellence 5 (2) (2001) 6–12. Operations Management and Operations Research at
[11] J.J. Alfaro, A. Ortiz, R.R. Rodriguez, Performance measurement system for enter- Polytechnic University of Valencia. He is an industrial
prise networks, International Journal of Productivity and Performance Manage- engineer and he received his doctoral degree in
ment 56 (2007) 305–334. Industrial Engineering from the Polytechnic University
[12] G.J.C. Da Silveira, Improving trade-offs in manufacturing: method and illustration, of Valencia in 1998. He works as a consultant in several
International Journal Production Economics 95 (1) (2005) 27–38. projects about production management, supply chain
[13] P. Suwignjo, U.S. Bititci, A.S. Carrie, Quantitative models for performance measure- management, information systems and enterprise
ment system, International Journal of Production Economics 64 (2000) 231–241. modelling and integration in metal mechanic, ceramic
[14] T.L. Saaty, The Analytic Hierarchy Process, McGraw-Hill, New York, 1980. and automotive enterprises. He is a member of the CIGIP
[15] A.D. Youngblood, T.R. Collins, Addressing balanced scorecard trade-off issues (Research Centre on Production Management and
between performance metrics using multi-attribute utility theory, Engineering Engineering). He works as a Researcher in five Spanish
Management Journal 15 (1) (2003) 11–18.
Government Projects (CICYT) and one ESPRIT Project
[16] D. Cardona Siado, A. Garcia, Modelo de Indicadores para el despliegue de la
and is co-leader of the GROWTH Project, V-CHAIN (Virtual Enterprise for Supply Chain
Estrategia de Calidad, 2005, Accessed 14th June 2007, Available at http://hdl.han-
dle.net/1992/300. Management). He is a teacher at the Polytechnic University of Valencia and in the
[17] M. Godet, From Anticipation to Action—A Handbook of Strategic Prospective, FORD Spain Industrial Engineer School, and also teaches several Masters courses
UNESCO, Paris, 1993. (MBA). He is member of the IFAC/IFIP Task Force on Enterprise Integration. His major
[18] K. Bauer, KPI: reduction the correlation way, DM Review (February issue, 2005) 1–3. research interests are production planning and control, supply chain management,
[19] J. Fraser, Beyond KPIs: impacting supply chain performance, Manufacturing enterprise integration, information management, business process modelling. He has
Business Technology (May 2005) 34–35. published several papers in books, journals and conferences in these fields.

Das könnte Ihnen auch gefallen