A Concept for the Automatic Customizing of an ERP System with Data Mining
Rene Schult and Gamal Kassem
Institute of Technical and Business Information Systems
Otto-von-Guericke-University Universityplace 2, Magdeburg, Germany
ERP systems, reference model, data mining, workflow.
The implementation of an ERP system is a long and cost intensive process. Functions of the ERP system,
which are delivered in an enterprise neutral but sector specific fashion need to be adjusted to the specific
business requirements of an enterprise. Exact knowledge of the ERP system is required because each ERP
system has its own technical concepts and terminologies. Therefore many enterprises employ ERP system
experts in order to customise the ERP system to be introduced as well as to further enhance the customisation
after its introduction. A concept for the implementation of a Self-Adaptive ERP System should allow for
the automatic customisation of an ERP system on the basis of the of enterprise process models provided and
analysis of the ERP system usage.
Customising is the main phase of an ERP implemen-
tation process. The behaviour of the ERP applications
is controlled by the customisation. Changes within
the conversion of the ERP application parameters are
executed at this level (Peßl, 2004). First workflow
models of the enterprise are modelled. The work-
flow, according to Gierhake, represents a work pro-
cess, which is extensively supported by technology
and is based on a triggered event, leading along a de-
fined chain of part steps towards a defined work re-
sult whereby the completion level of the work result
increases with each step. A workflow is the descrip-
tion of a specification that defines which elementary
functions or tasks, that can be put into a standard, are
to be executed by a person, who has been assigned
a specific role at a given process state. If necessary,
required data processing application for object pro-
cessing are to be provided (Gierhake, 2000), pp. 57
ff.). At the implementation level, Rautenstrauch de-
fines workflows as ”...a part of a business process ...,
that consists of sequentially or parallel arranged op-
eration sequences (activities). Therefore it describes
subprocesses of the sequence organisation within en-
terprises. The activities themselves are implemented
as functions within applications systems. The cross-
applications are controlled and therefore can be bro-
ken out of the application systems ... (Rautenstrauch
and Schulze, 2003), p. 269). The workflow model en-
compasses a description of all relevant aspects of the
workflow, which includes all incidental tasks, their se-
quences and dependencies. The model of workflows,
specially business processes of an enterprise, is a pre-
condition or basis for the customisation of an ERP
system. During the implementation of an ERP system
enterprise neutral sectoral functions are adjusted util-
ising the customisation on the basis of the modelled
enterprise workflows to the specific business require-
ments of an enterprise. ERP system reference models
are available as guideposts for experts in customising.
Reference models generally describe objects and re-
lations of the business reality. ERP reference mod-
els are pre-modelled on standard enterprise models
and shapes the business architecture of an ERP sys-
tem. Reference models often provide ”best practice”
business processes in the form of a software library
(Peßl, 2004). Therefore, the customisation is used for
the adjustment of the ERP system to the enterprise
requirements. The adjustment itself is done by the
specific ERP customising experts (consultants for the
ERP system to be introduced). With the help of ref-
erence processes all business operations that are sup-
ported by an ERP system are mapped. The separate
processes show tasks of a business situation that be-
long together logically. This can include a purchase
order processing with the focus set on the lowest
priced material, the vendor selection and order trac-
Schult R. and Kassem G. (2008).
SELF-ADAPTIVE CUSTOMIZING WITH DATA MINING METHODS - A Concept for the Automatic Customizing of an ERP System with Data Mining
In Proceedings of the Tenth International Conference on Enterprise Information Systems - ISAS, pages 70-75
DOI: 10.5220/0001676800700075
ing or a complete sales order processing from the cus-
tomer inquiry to the delivery and the billing. Through
the dependency of the processes on each other an un-
derstanding of cross-enterprise processes can be cre-
ated with the user and on the other hand the inter-
action of operational function units can be depicted.
The main task of customising experts is the imple-
mentation of the enterprise specific workflow models
in the ERP System. Reference models may be used
as a pattern for the system adjustment. The reference
models are to be considered here as supporting infor-
mation that helps customising experts to try and (re-
ality true) map the enterprise specific workflow mod-
els in the ERP system. With this experts should ac-
complish (resolve) two main problems: the terminol-
ogy problem that can arise through the different terms
used and their meaning in an enterprise on the one
hand and the terminology of the ERP system on the
other hand. A further problem may arise at the selec-
tion of the adequate processes from the ERP reference
models. The ERP processes should be selected and
adjusted so that the enterprise processes are depicted
in the system without developing new ERP functions
or changing the structure of the enterprise itself. In
this paper a self-adaptive customising concept is in-
troduced that leads to the solution of the above men-
tioned problems through the usage of different data
mining methods and which automates the customis-
ing procedure. Furthermore the concept should pro-
vide for the automatic execution of further customis-
ing procedures in the future. For that purpose the
structure of a self-adaptive customising middleware
system (SACMS) is presented that acts as a means of
mediation between the workflow models of an enter-
prise and the ERP system. The SACMS should pro-
cess different process modelling languages and ERP
systems of different vendors.
The idea of the concept consists of the creation of
an ”intelligent” layer between the workflow models
of an enterprise, which should be mapped into a par-
ticular ERP system, and the ERP system itself. This
layer provides for the automatic customisition of the
ERP system in which two customising types are to
be distinguished: the initial customising will be done
immediately at the implementation of an ERP sys-
tem, while the ”post-customising” is continually per-
formed within the scope of a Continuous Improve-
ment Process (CIP) and provides persistent improve-
ment of the ERP customising.
2.1 Initial Customising
At the time of the implementation of an ERP system
the enterprise neutral and sectoral specific function-
ality is adjusted to the specific business requirements
of an enterprise during the initial customising. The
initial customising covers three phases: conversion,
evaluation and implementation (see figure 1).
2.1.1 Conversion Phase
In this phase enterprise workflow models as well
as ERP reference models are converted in formal
workflow model formats such as e.g. workflow nets
(van der Aalst et al., 2003). Workflow nets stem from
Petri nets (Baumgarten, 1990). The enterprise spe-
cific terms used in enterprise workflow models are
translated in the terminology of the ERP system in
the conversion phase. With this special text mining
methods are used to recognise the meaning of a term
and translate it into a terminology of the ERP system.
Details to those text mining methods are shown at the
end of this section. As the sequence of activities or
functions of a workflow define its structure the main
task of the text mining is to recognise which workflow
activity is equivalent to an ERP transaction. Therefore
not only the activity term is to be considered, the busi-
ness objects involved in the transaction, the so called
”workflow reference objects” are to be considered as
well. The workflow reference object encompasses all
business objects processed by a function of a work-
flow. The workflow reference objects can be divided
into following:
Output Objects. The result of a function is a trans-
formation of data for the processing of one or
more workflow objects. This object or these ob-
jects are denoted as output objects.
Input Objects. For the processing of output objects
by a function, data from other objects may be re-
quired. These objects are denoted as input objects.
For example during creating a sales order (an out-
put object), data about articles and the customer is
Workflow Activity Object. A workflow activity ob-
ject is a workflow reference object that uniquely
denotes a workflow. For example at the execu-
tion of the workflow ”Production order process-
ing” the functions ”Create a production order”,
”Confirm a production order”, ”Posting goods re-
ceipt for production order” and ”Release a pro-
duction request” are performed. The workflow
activity object in this workflow is the object ”Pro-
duction order”, as it is part of all functions of
an ERP System with Data Mining Methods
Figure 1: Self-Adapting Customising plan for the first customising.
the workflow: as output object in ”Creating pro-
duction order”, ”Confirm a production order” and
”Release a production order” and as an input ob-
ject in ”Posting goods receipt for production or-
Workflow reference objects can be classified into the
following object types:
Organisation Objects. are business objects that rep-
resent the organisation units of an enterprise,
e.g. purchasing department, warehouse, company
code, plant, etc.
Resource Objects. are all business objects other
than the organisation objects. Such as customer,
vendor, order, invoice, purchase order, article, etc.
A function of an enterprise workflow model can be
converted via text mining methods if the function term
(notation) and the involved workflow reference ob-
jects are equivalent to the ERP transaction term and
its workflow reference objects.
Text mining use syntactical rules for the recogni-
tion of term structures and mining methods for the de-
tection of the semantic meaning of terms. During the
usage of text mining large data sets are used which
describe the terms from the real enterprise world as
well as from the ERP technology world. One exam-
ple for such large dataset could be an ontology with
the content of words of the real world and the map-
pings to the words of the ERP technology world. But
it should not be limited to the vocabulary of only one
ERP system. It is also possible that an ontology for
both workflow models exists, the enterprise and the
ERP specific workflow model, or one for each model.
If we have both ontolgies, it is possible to match both
ontologies to one ontology for the formal workflow
model (Euzenat and Shvaiko, 2007).
Another example for such large data set is the us-
age of synonym lists. In which each word of the
ERP technology world and the enterprise world are
mapped to the equivalent of the formal model termi-
nology. An algorithm search for each word the equiv-
alent synonym from such a list to create an uniform
vocabular for the enterprise and the ERP workflow
2.1.2 Evaluation Phase
In the evaluation phase an converted ERP reference
model is compared with the converted ERP workflow
model and a suitable workflow or a workflow from the
workflow combinations form. The reference model is
constructed for the implementation phase as an ”ex-
tended workflow model”. The workflow selection is
based on the comparison of model objects as ERP
workflow structures and the functions and objects in-
volved on the one side, and how they are mapped in
the ERP reference model and whether they are linked
to other model objects on the other side. The link-
ing of objects in the ERP reference model can lead
to the cascading of further objects that have to be
considered in the implementation phase. An exam-
ple is the selection of a workflow from the ERP ref-
erence model for ”Processing Sales Order”. Here it
is considered whether the sales order is linked to fur-
ther sub-workflows, for example for economic and/or
technical checks, and whether these sub-workflows
are linked to further sub-sub-workflows. Through the
consideration of the linked objects all ERP transac-
tions necessary for the execution of a workflow are
provided and adjusted. Attention should be paid to
the fact that not all transactions can actually be used
in the future. In the post-customising phases the ERP
system can be further improved and customised after
the actual usage of the system.
ICEIS 2008 - International Conference on Enterprise Information Systems
2.1.3 Implementation Phase
At the implementation phase ERP customising object
data are adjusted based on the selected workflows (the
extended workflow models) from the ERP reference
model. ERP customising objects are ERP repository
objects which are necessary for the adjustment of an
ERP system. For this purpose historical ”best prac-
tice” information, stored for the customising of the
ERP reference models, are used. This information
allow that user roles, authorisation profiles, organi-
sation units, currency unit and all other customising
objects are automatically implemented in this phase
based on the extended workflow models.
2.2 The Post-Customising
During the implementation of an ERP system the
functionality provided within the initial customising
is adjusted to the specific business requirements of an
enterprise. With the provision of the linked model
objects or customising objects, the system parameter
settings are dimensioned in a large-scale fashion, so
that all possible functionalities of the enterprise can
be supported by the ERP system. Parts of the ERP
functionality set are not used in the daily operational
work which may lead to greater system complexity.
The underlying Self-Customising System takes
place in the post-customising phases. The system is
modified, respectively improved and adjusted to the
needs of the enterprise on the basis of the system us-
age over time. In this process methods of Application
Usage Mining (AUM) are used. AUM is a system
technically realisable procedure for the reconstruction
of actual workflow models of an ERP system. AUM
should be an auxiliary means for the automatic mod-
elling of a detailed actual workflow model of an en-
ERP systems support enterprise-wide business
processes and implement them as workflows. During
the work with an ERP system, the ERP users leave
traces in the form of ERP trace data that reflect the
user interactions with the system. A supporting tool
for this purpose is the methodology of the Application
Usage Mining which automatically reconstruct actual
workflow models from ERP traces that vulnerabili-
ties in the system usage can be detected and removed
through customising adjustments.
The post-customising is constantly executed in the
frame of a Continuous Improvement Process (CIP) in
order to provide a constant improvement of the ERP
customising. The post-customising covers, similar
to the initial customising, three phases: provision of
the actually executed ERP workflow models (actual
workflow models), evaluation phase and implementa-
tion phase.
2.3 Provision of ERP Workflow Models
Instead of using the conversion phase of the initial
customising in order to provide ERP workflow mod-
els, in the post-customising phase the actual workflow
models are automatically extracted using the AUM
methodology from the ERP system and are provided
for the evaluation phase (see figure 2).
The existing workflow models represent the ac-
tual ERP workflow models used within the system.
The AUM itself is a method that uses a KDD process
based on ERP trace data. The AUM process covers
three phases (Kassem, 2004):
In the preparation phase interaction data from
different sources of the ERP system are integrated in
a database (data pool). This are mainly trace data that
protocols the user interactions with the ERP system.
Additionally the data pool contains meta interaction
data that describes the trace data.
In the pattern detection phase the workflow
cases of a workflow are detected using special min-
ing algorithms (Kassem and Rautenstrauch, 2006). A
workflow case displays the sequence of the perfor-
mance of ERP functions of a workflow instance. The
algorithms address conflicts that may arise at the as-
signment of functions to the workflow cases. Based
on theses algorithms task steps (function cases) can
also be detected as a screen sequence at the execution
of a function. It displays which work steps of a task
were performed.
In the pattern analysis phase the workflow cases
detected in the pattern detection phase are analysed
with the help of workflow algorithms (Agrawal, 1998;
van der Aalst and van Dongen, 2002; van der Aalst
et al., 2002) and are extracted formally as well as
graphically using actual workflow models.
The extracted actual workflow models can now be
evaluated in the evaluation phase.
2.4 The Evaluation and Implementation
At the evaluation phase only the actual used func-
tionalities are considered in order to assure that the
workflow structures are customised by the requiren-
ments of the enterprise and therefore the model is be-
ing improved consequently. Here analysis methods
can be used for the evaluation of the generated actual
workflows and the interaction data, collected in the
data pool. The analysis should address the following
an ERP System with Data Mining Methods
Figure 2: Self-Adapting Customising plan for the ”post” customising.
The connection between the workflow process (its
structure) and the workflow attributes (i.e. costs, run-
time or attributes of workflow used business objects
like customer name or material weight) can contribute
to the detection of vulnerabilities at a workflow or,
amongst others, illegal processes. Such kind of infor-
mation can support the process management of an en-
terprise. Therefore mining methods can be developed
to classify the workflow structures (like parallelism,
branches, sequenciellity or iterations/constellations of
workflow structures) and for comparison with work-
flow attribute values for detection of connections as a
type of patterns.
An the evaluation phase user behaviour can be
analysed. The access paths of the users during
the workflow processing are represented as graphs
(Kassem et al., 2003). Different mining methods can
be used to find answers to the following questions:
Path Analysis:
At which step of equal user access paths an error
message appears?
At which step of user access paths transactions are
often aborted?
At which step the access paths reach dead ends?
To answer such questions from the workflow model
different graph mining methods like classification of
graphs (Borgelt and Kruse, 2002) can be used.
Association Analysis: The connection between
events (e.g. input of datafields and the pressing of a
button) and the screen sequence should be recognised
through the employment of association rule methods.
Here the customising of the system is checked ac-
cording to its correctness. The deviation of the user
access paths from the ideal access paths can be recog-
nised. The business process should be analysed by
such methods. The well known apriori algorithm
from Agrawal and Srikant in 1994 (Ester and Sander,
2000), for instance, can be used in order to create as-
sociation rules from the screen sequences.
Temporal Analysis: Time referenced data is needed
for the temporal analysis of the data. One example for
the goal of such analysis is to examine the effectivity
of the system usage of a user. Following questions can
be asked (and hopefully answered) after the analysis
How quickly new users are enabled to use the system?
How much time is needed for a user to process a task?
Which text fields always contained the same values and
which values precisely were used?
Which screens of a transaction were not used?
From which transactions and from which screen other
transactions were called?
From which transaction screen help programs were nec-
essary and which topic did they address?
The temporal mining methods like sequence anal-
ysis, detection of pattern evolutions or cluster mon-
itoring gives us the possibility to find changes and
movements at the workflow model during runtime and
so to adjust the system over time to customise the sys-
tem according to the new requirenments of the enter-
prise (Aggarwal, 2007).
The User behaviour Analysis can be used for per-
sonalisation and for checking the correctnes of the
system customising and for checking the efficency of
the system usage at user tasks. As one result of such
analysis it can be seen, how familiar a user is with the
system and if a user or a user group needs some train-
ing or if the system should be customised or person-
alised for some users. to their needs through personal-
ization. The extended ERP workflow models are pro-
vided for the implementation phase. The analysis of
ICEIS 2008 - International Conference on Enterprise Information Systems
the workflow structures and the user behaviour forms
the basis for the models. These extended workflow
models are after the system usage customised work-
flow models.
The ERP customising objects are modified by us-
ing the extended ERP workflow models at the imple-
mentation phase.
The post-customising process can be seen as a re-
curring process which customises the business pro-
cess of an enterprise dynamically at any given time.
The degree of customisation of an ERP system de-
pends on the flexibility potential of the system. Here
the flexibility of an ERP system means how cus-
tomisable the ERP system is to the business process
changes of an enterprise and how effectively the busi-
ness process activities can be processed by the users
or user groups depending on their roles and knowl-
The concept of a self-adaptive ERP system is an idea
for an automatic implementation and a permanent
customisation of an ERP system to the changes of
business processes and user behaviours within the en-
terprise. To implement this idea, we have shown a
structure of a system infrastructure, where the en-
terprise and the ERP-system layer are separated by
an adaptive middleware. The adaptive middleware
should be the interface between both layers. The
function of the adaptive middleware SACMS is to en-
sure the adaption of the system to the requirements of
the enterprise and users through intelligent methods
like data mining methods.
A condition for the successful implementation of
this concept is to have ERP-repository-data from the
ERP system with customised objects like business-
objects, transactions, screens screen-elements. Stan-
dardisation problems of workflow models, refer-
ence models and ERP-tracedata-formats inhibit gen-
eral rules for handling ERP-systems of different
providers. One possible solutions of this standardi-
sation problem can be a layer specific interface be-
tween the SAC-layers which enables the data transfer
between the different layers depending on the layer
standards. Therefore it is necessary to solve the stan-
dardisation problems of workflow models and ERP-
tracedata formats in the future.
Aggarwal, C. C. (2007). Data Streams, Models and ALgo-
rithms. Springer Verlag, Berlin, Germany.
Agrawal, R.; Gunopulos, D. L. F. (1998). Mining process
models from workflow logs. In Advances in Database
Technology - EDBT’98, 6th International Conference
on Extending Database Technology, pages 469–483,
Valencia, Spain.
Baumgarten, B. (1990). Petri-Netze: Grundlagen und An-
wendungen. BI-Wiss-Berlag, Heidelberg, 2. edition.
Borgelt, C. and Kruse, R. (2002). Graphical Models, Meth-
ods for Data Analysis and Mining. John Wiley and
Sons, England.
Ester, M. and Sander, J. (2000). Knowledge Discovery in
Databases. Springer Verlag, Berlin, Germany.
Euzenat, J. and Shvaiko, P. (2007). Ontology Matching.
Springer Verlag Berlin, Berlin, Germany.
Gierhake, O. (2000). Integriertes Geschftsprozessmanage-
ment, effektive Organisationsgestaltung mit Workflow-
, Workgroup- und Dokumentenmanagement-
Systemen. Vieweg, Braunschweig, 3. edition.
Kassem, G. (2004). Application usage mining as an instru-
ment of business process management. In Proceed-
ings of the IV International Conference of Applied En-
terprise Science (CICE’ 2004), Santa Clara, Cuba.
Kassem, G., Marx Gomez, J., Rautenstrauch, C., and
Melato, M. (2003). Analysis of the user’s behaviour in
very large business application systems with methods
of the web usage mining - case study sap r/3. In Ad-
vances in Web Intelligence (AWIC 2003), pages 329–
338, Madrid. Springer Verlag Berlin.
Kassem, G. and Rautenstrauch, C. (2006). The ability
to generate as-is-model from an erp-system based on
system user traces. In Proceedings of the 7th Annual
Global Information Technology Management Confer-
ence (GITMA), pages 133–136, Orlando, Fl, USA.
Peßl, E. (2004). Customizing von erp-systemen - ereignis-
gesteuerte prozessketten und das erp-referenzmodell.
Technical report, University of Applied Science,
Rautenstrauch, C. and Schulze, T. (2003). Informatik
fr Wirtschaftswissenschaftler und Wirtschaftsinfor-
matiker. Springer-Verlag.
van der Aalst, W. M. P. and van Dongen, B. F. (2002). Dis-
covering workflow performance models from timed
logs. In Han, Y., Tai, S., and Wikarski, D., ed-
itors, International Conference on Engineering and
Deployment of Cooperative Information Systems (ED-
CIS 2002), pages 45–63. Springer-Verlag Berlin.
van der Aalst, W. M. P., van Dongen, B. F., Herbst, J.,
Maruster, L., Schimm, G., and Weijters, A. (2003).
Workflow mining: A survey of issues and approaches.
Data and Knowledge Engineering, 47(2):237–267.
van der Aalst, W. M. P., Weijters, A. J. M. M., and Maruster,
L. (2002). Workflow mining: Which processes can be
rediscovered? Technical Report BETA Working Paper
Series, WP 74, Eindhoven University of Technology.
an ERP System with Data Mining Methods