AN EVOLUTIONARY APPROACH
FOR BUSINESS PROCESS REDESIGN
Towards an Intelligent System
Mariska Netjes
1
, Selma Limam Mansar
2
, Hajo A. Reijers
1
and Wil M. P. van der Aalst
1
1
Department of Technology Management, Eindhoven University of Technology, Eindhoven, The Netherlands
2
Business Sciences College, Zayed University, Dubai, UAE
Keywords:
Business Process Redesign, Process Modelling, Business Process Management, Workflows, Best Practices.
Abstract:
Although extensive literature on BPR is available, there is still a lack of concrete guidance on actually chang-
ing processes for the better. It is our goal to provide a redesign approach which describes and supports the
steps to derive from an existing process a better performing redesign. In this paper we present an evolutionary
approach towards business process redesign and explain its first three steps: 1) modelling the existing process,
2) computing process measures, and 3) evaluating condition statements to find applicable redesign “best prac-
tices”. We show the applicability of these steps using an example process and illustrate the remaining steps.
Our approach has a formal basis to make it suitable for automation.
1 INTRODUCTION
Although numerous papers and books on Business
Process Redesign (BPR) were published during the
past 15 years (e.g. (Grover et al., 1995; Kettinger
et al., 1997; Al-Mashari and Zairi, 1999)), guidance
for concrete process redesign is scarce. Valiris and
Glykas (Valiris and Glykas, 1999) identify that “there
is a lack of a systematic approach that can lead a
process redesigner through a series of steps for the
achievement of process redesign”. This paper fits
within our aim to fill this void by providing a redesign
approach which describes and supports the steps to
derive from an existing process a better performing
one.
A key element of the approach that we propose in
this paper is the generation of diagnostic information
on an existing business process. Such a diagnosis is
performed using so-called process measures, which
provide a global view on the characteristics of the
process. Actual values for the process measures may
reveal weaknesses in the process. Identified weak-
nesses are removed by the application of one or more
redesign best practices. A redesign best practice de-
scribes a well-tried way to remove a particular prob-
lem from a process to improve its performance. In
our earlier work we have described 29 redesign best
practices and presented a qualitative description, the
potential effects and possible drawbacks for each best
practice (Reijers and Limam Mansar, 2005). Finally,
in our approach, the application of redesign best prac-
tices leads to one or more alternative candidates for
the existing process. The evaluation of the perfor-
mance of each alternative shows the best candidate
process which should replace the existing one.
Our approach builds on the formal representation
of a business process with Petri nets, in particular
WorkFlow nets (Aalst, 1998). Some of the features
of our model are inspired by the process modelling
tool Protos (Pallas-Athena, 2004). Current versions
of Protos are in use by thousands of organizations in
more than 25 countries. In The Netherlands alone,
more than half of all municipalities and insurance
companies use Protos for the specification of their
business processes. The focus on this real-life tool
gives our approach a practical edge, while it is still
easy to see how our approach can be generalized to
other modelling techniques and tools (e.g. ARIS).
We envision as the ultimate goal of our research
the delivery of an automated redesign tool. This tool
would support all steps of the approach in an “intelli-
gent” way. By this, we mean that the tool will not only
automate the various steps of the approach, but will
also interact with the redesigner. The redesigner will
47
Netjes M., Limam Mansar S., A. Reijers H. and M. P. van der Aalst W. (2007).
AN EVOLUTIONARY APPROACH FOR BUSINESS PROCESS REDESIGN - Towards an Intelligent System.
In Proceedings of the Ninth International Conference on Enterprise Information Systems - ISAS, pages 47-54
DOI: 10.5220/0002356900470054
Copyright
c
SciTePress
be able to indicate which performance dimensions
(time, costs, quality) should be improved, whether
certain process characteristics should perhaps not be
changed, and which promising alternatives should be
combined in constructing the best alternative. Our
approach is a solution that should primarily help re-
design novices in finding process alternatives based
on best practices. Secondly, more experienced re-
designers are supported in the creation and evalua-
tion of such alternatives in a structured and less time-
consuming manner. The structure of the paper is as
follows, section II describes the related work, section
III details the steps of our approach using an illustra-
tive example and section IV presents our conclusions
and future work.
2 RELATED WORK
Various more structured approaches to process re-
design were proposed earlier, most notably the Pro-
cessWise methodology (Calvert, 1994) and the MIT
Process Handbook (Malone et al., 2003). Also, a
variety of tools is available, e.g. MIT’s process re-
combinator tool (Bernstein et al., 1999), a number of
tools that apply case-based reasoning (Ku and Suh,
1996; Min et al., 1996), and the KOPeR tool by Nis-
sen (Nissen, 1996; Nissen, 1997; Nissen, 1998; Nis-
sen, 2000). Many existing approaches and tools are
limited in their application domain, while none of the
approaches has succeeded to gain widespread adop-
tion in industry. We have provided a more extensive
literature review on this topic in (Netjes et al., 2006).
Nissen’s work is most related to our approach
which motivates its following deeper discussion. Its
main contribution with the KOPeR tool is the con-
struction of a set of measures that, applied to pro-
cesses, would at first help to diagnose pathologies.
The pathologies are then mapped to matching trans-
formations that may be applied to the processes in
order to improve their performances. Nissen used a
process model based on nodes, attributes and edges
(Nissen, 1998).
Nissen’s work has inspired us to come up with a
similar approach that nonetheless overcomes some of
the KOPeR tool’s shortcomings:
On the process modelling side, the process model
in use by KOPeR is yet simple and the provided
examples are rather simplistic (Nissen, 1996; Nis-
sen, 1997; Nissen, 1998). Our process model is
defined as an enriched WorkFlow net allowing for
the modelling of realistic, complex business pro-
cesses.
Nissen used graph-based definitions of the mea-
sures in order to operationalize them. We have
noticed that the exact meaning of some of the
measures is unclear. We use a formal notation to
overcome this and define our measures unambigu-
ously.
Nissen adds an extra layer of indirection (process
pathologies and process transformations). We
only define a set of measures and a set of trans-
formation rules to immediately find the applicable
transformations.
Nissen’s set of presented transformation serves as
an illustration and is far from a complete coverage
of the spectrum of redesign options. We provide a
more exhaustive list of rules based on our set of 29
best practices (Reijers and Limam Mansar, 2005).
3 EVOLUTIONARY APPROACH
In our evolutionary approach towards workflow pro-
cess redesign we take an existing process model and
improve it using redesign best practices. It is evolu-
tionary, because an existing process is taken as start-
ing point instead of a clean sheet. Our approach con-
sists of six steps: 1) model an existing process (us-
ing a formal model), 2) compute process measures
(which can be seen as global process characteristics.
Values for the process measures are derived from the
existing process model and point out weaknesses in
the process), 3) evaluate condition statements to find
applicable best practices (For each best practice it is
known which process weaknesses it could solve and
the corresponding process measures are combined in
one condition statement per best practice. When a
statement evaluates to true it suggests the application
of the associated best practice. All condition state-
ments are evaluated to find the best practices which
are eligible to be applied to the process. We strive to
include as many redesign best practices as possible in
our approach and we assume that our set of best prac-
tices is complete), 4) create alternative models based
on the selected best practices (The selected best prac-
tices are used to create alternative models. A best
practice has essentially the followingparts: some kind
of construction or pattern that can be distinguished in
the existing process, an alternative to be incorporated
for the redesign and a context-sensitive justification
for this redesign), 5) evaluate the performance of the
created alternatives, and finally 6) choose the best al-
ternative and implement the new process after com-
paring the performance of the various alternatives to
find the best alternative model, cf. Figure 1.
ICEIS 2007 - International Conference on Enterprise Information Systems
48
Existing
process model
Process
measures
Condition
statements
Alternative
models
Evaluation of
alternatives
New
process model
Figure 1: Evolutionary approach towards redesign.
The following details the six steps we have de-
scribed in Figure 1. An illustrative case study is used.
3.1 Process Model
The starting point of our evolutionary approach to
workflow process redesign is the existing process
model. In order to illustrate our findings we use a
case study that describes the process of handling in-
surance claims.
Let us first describe the process: The process handles
the insurance claims of both individual and business
clients. The process starts when a claim is received.
After receipt, the claim is classified as “individual”
or “business”. Then the claim is checked for valid-
ity. Three checks, Check policy, Check amount (only
for business clients, requires the receipt of a damage
report) and Check legal are performed. A check ei-
ther results in OK (proceed with next check) or not
OK (reject claim). Claims that pass all checks are ac-
cepted and paid. Payments are authorized at the end
of each day by the finance manager. For all claims
(both rejected and accepted) a letter is written and the
claim is archived.
To model this process, we use the Worklow nets and
we define a formal process called Proto. A workflow
process is case-based, i.e. every piece of work is ex-
ecuted for a specific case, and make-to-order. A Petri
net (Reisig and Rozenberg, 1998) which models a
workflow process definition (i.e. the life-cycle of one
case in isolation) is called a WorkFlow net (WF-net).
In a WF-net, the workflow management concept task
is modelled as the Petri net concept transition, condi-
tions are modelled by places, and cases are modelled
by tokens. For a definition of WF-nets the reader is
referred to (Aalst, 1998).
The process model is created based on a formal pro-
cess definition called Proto net. The Proto net in-
cludes a process structure, related information and an
organizational model. The process structure consists
of places, transitions and flow relations just as is the
case in a WF-net. The process structure is enriched
with information related to transitions (such as exter-
nal triggers, the type of activity to be executed, XOR-
splits and -joins to model choices, responsible depart-
ments, required applications, and handled products
and services). The organizational model uses roles
at its foundation. A role is a collection of comple-
mentary skills. Allocating roles to transitions ensures
that work is performed by the relevant person. Roles
have a hierarchical relation, i.e. if two roles have the
following relation (r
, r) this means that role r is one
step higher in the hierarchy than role r
and that role
r is also able to perform the transition(s) allocated to
role r
.
The Proto net is formally defined in Definition 1.
Definition 1 (Proto net) A Proto net PN is a tuple
(P, T, F, B,C, S, J, D, DT, E, ET, G, GT, R, H,U, A, AH)
with:
P is a non empty, finite set of places;
T is a non empty, finite set of transitions (P T =
/
0). Task and activity are used as synonyms for
transition;
F (P× T) (T × P) is a set of arcs (flow rela-
tion);
B : T
P ({time, periodic, digital, mail, telephone})
relates each transition to zero, one or more trigger
types;
C : T
P ({basic, communication, check, authorize,
batch}) relates each transition to zero, one or
more activity types;
S : T {AND, XOR} relates each transition to a
split type element;
J : T {AND, XOR} relates each transition to a
join type element;
D is a non empty, finite set of departments;
DT : T
P (D) relates each transition to zero,
one or more departments;
E is a finite set of applications (i.e. software
tools);
ET : T
P (E) relates each transition to zero,
one or more applications;
G is a non empty, finite set of products and ser-
vices;
GT : T P (G) relates each transition to zero,
one or more products and services;
R is a non empty, finite set of roles;
H (R× R) is a (acyclic) set of hierarchical role
relations;
U R
N is a non empty finite bag of users. N
is the set of natural numbers {0, 1, 2, ...}. U is
a bag, i.e., for some role r R, U(r) denotes the
number of users having role r as the highest role;
A : T 6→ R relates each transition to zero or one
roles (allocation);
AH : T
P (R) relates each transition to zero,
one or more roles (hierarchical allocation) [Note
that for t dom(A): AH(t) = {r R| (A(t), r)
AN EVOLUTIONARY APPROACH FOR BUSINESS PROCESS REDESIGN - Towards an Intelligent System
49
H
} (with H
being the reflexive transitive closure
of H) and for t 6∈ dom(A): AH(t) =
/
0].
A process can be modeled according to the Proto
net definition with the modelling tool Protos (Pallas-
Athena, 2004). We have made several assumptions
regarding the creation of a process model. In our
definition of the Proto net we only take into account
the structural properties of a process model and
abstract from behavioral information.
Let us now use the Protos model for our insurance
claim process. The model of the insurance claim pro-
cess is shown in Figure 2 and is easy to understand.
individual
Claim
Damage report
End of the day Authorize pay
Manager Finance
Finance Individual & Business, FinProg
Check legal
(both XOR)
Law expert
Claim handling Individual & Business
Classify
Administrator
Distribution, Claim handling Individual & Business, Word
Receive
damage report
Postal worker
Distribution Business
Receive claim
Postal worker
Distribution Individual & Business
Pay
Assistant Finance
Finance Individual & Business, FinProg
Check policy
(XOR-split)
Administrator
Claim handling Individual & Business
Check amount
(XOR-split)
Damage expert
Claim handling Business
Archive
Administrator
Claim handling Individual & Business
Write letter
(XOR-join)
Administrator
Claim handling Individual & Business, Word
finish
class
business2
check
business3
authpay
business1
OK
acc
not OK
Figure 2: The existing insurance claim process.
The rectangular boxes in the process model are the
transitions defined in the Proto net (T), each transi-
tion has an activity type indicated by the symbol in
the box (C). The talking balloon indicates the activity
type “communication”, the trapezium indicates “ba-
sic”, the check box “check”, and the V-symbol “au-
thorize”. The three rectangles on the left side of the
process are triggers (B). A trigger indicates that a cer-
tain (external) condition has to be fulfilled before the
transition it links to can be executed. In this exam-
ple, a “Claim”, a “Damage report” and the “End of
the day” are used. The split and join types are set to
their values in the transition properties, but to show
them in the model XOR-splits and -joins are also ex-
plicitly stated in the name of a transition. On the top
right side of the transition the role allocated to the
transition is indicated (For example, the first transi-
tion “Receive claim” is executed by the role “Postal
worker”), on the bottom left side the related depart-
ment(s) (Department “Distribution” for the transition
“Receive claim”), and on the bottom right side the
products / services and the required applications (“In-
dividual and business” services for the transition “Re-
ceive claim”). The organizational model related to the
process model is depicted in Figure 3 and is also cre-
ated with Protos. It shows the roles and the number of
resources per department.
Distribution department
Finance department
Claim handling
department
Assistant
Finance
2
Administrator
2
Postal worker
2
Damage expert
3
Law expert
3
Manager
Finance
1
Manager
Distribution
1
Supervisor
Claim handling
1
Administrator
4
Manager Claim
handling
1
Figure 3: Organizational model of insurance claim process.
Looking at Figure 2 and 3, it is not easy to spot
inefficiencies in the process. Even experienced prac-
titioners will certainly need to use their expertise and
perhaps tools such as simulation to define what will
need to be improved in the process. Hence the next
step of our approach, the use of process measures.
3.2 Process Measures
The second step of our evolutionary approach is the
computation of the process measures for the existing
process model. In this section we define our set of
process measures. The starting point for the creation
of the process measures has been the work of Nissen
(Nissen, 1998). Nissen identified 19 (static) process
measures (Nissen, 1996) of which ten appeared to be
relevant in relation to the redesign best practices. The
ICEIS 2007 - International Conference on Enterprise Information Systems
50
graph-based definition Nissen presented for these
measures is replaced by our formal definition because
the formal definition provides a more precise and
unambiguous meaning for the measures.
Besides the measures included from Nissen we
developed eight new measures. These measures are
all related to the additional information incorporated
in the Proto net. In Table 1 the process measures are
defined, their range is given and it is indicated which
measures are taken from Nissen and which are new.
For Table 1 we assume the following context: PN =
(P, T, F, B,C, S, J, D, DT, E, ET, G, GT, R, H,U, A, AH)
is a Proto net.
To illustrate our results on the process measures,
let us apply them to our insurance claim process. Us-
ing the set of measures defined in Table 1 we compute
the process measures. The computed values are pre-
sented in Table 2. For instance, the measure level of
control is defined as the percentage of control tasks.
In the insurance claim process, there are 3 (control
transitions) divided by 10 (transitions) resulting in a
level of control of 0.3. Another measure, knock outs,
is defined as the percentage of knock outs in the pro-
cess. A knock out is a part of the process where cases
are checked. The result of a check can be not OK
(leading to a direct rejection of the case) or OK (lead-
ing to an acceptance of the case if all checks are OK).
The knock out can be distinguished based on the pres-
ence of a special place called the knock out place. In
our example process, there is one knock out place,
namely “not OK”, on a total of 11 places resulting in
a knock outs measure of 0.1.
Table 2: Values for process measures.
Parallelism = 0 Department share = 0.1
Process contacts = 0.2 Process hand offs = 0.4
Batch = 0 Specialization = 0.7
Periodic = 0.1 Role usage = 0.7
Level of control = 0.3 Manag. layers = 0.3
L. of authorization = 0.1 Knock outs = 0.1
IT automation = 0.2 Process size = 10
IT communication = 0 Process versions = 2
Department inv. = 0.3 User involvement = 2
In the next step, we use and combine the set of
process measures to determine condition statements
per best practice. We know which weaknesses in a
process a best practice would help to alleviate and we
derive in the next section condition statements accord-
ingly. Let us note that the complete set of measures
has been created and defined iteratively with the de-
velopment of the condition statements. Each condi-
tion statement is connected to one best practice. We
assume our set of best practices is complete and in
this work we include all best practices which require
process, data or resource information for their selec-
tion. We hypothesize that all relevant measures which
could be derived from this information and which are
necessary for the condition statements are included in
our set of measures.
3.3 Condition Statements
The third step of our evolutionary approach is the
evaluation of condition statements to select applica-
ble redesign best practices. For each best practice
we derive one condition statement which includes
one or more process measures. The application of
a certain best practice should be suggested when
the condition statement is fulfilled. The values
used in the condition statements are based on our
own redesign experience and expectation of when a
certain best practice is applicable. The validation of
these values will be future work.
We have created condition statements for 17 out of
the 29 best practices: Task elimination, Task addi-
tion, Task composition, Task automation, Knock-out,
Control relocation, Parallelism, Triage, Case man-
ager, Split responsibilities, Numerical involvement,
Specialist-generalist, Empower, Contact reduction,
Case types, Technology, Case-Based Work.
Regarding the remaining 12 best practices: 1) four
best practices appeared to have conditions similar to
other best practices and are combined, 2) four best
practices are not included because measures beyond
the process level are necessary to come to a proper
condition statement, 3) four best practices can not be
covered due to behavioral dependencies which are
not incorporated in the process definition.
For the sake of conciseness, we will only illustrate
a subset of condition statements. Moreover, the
examples are based on measures which were not
included in Nissen’s work.
For the insurance claim process 11 of the 17 condition
statements we have defined evaluate to true resulting
in the selection of a number of best practices that
appear to be applicable to redesign this process. In
what follows we describe a subset of best practices
for which the condition statements evaluated to true:
Task elimination: When applying task elimina-
tion unnecessary tasks are removed (remember: task
is a synonym for transition).
Condition statement: Apply task elimination IF Level
of control > 0.2.
Knock-out: When applying knock out tasks re-
sulting in a knock out are re-ordered.
AN EVOLUTIONARY APPROACH FOR BUSINESS PROCESS REDESIGN - Towards an Intelligent System
51
Table 1: Process measures.
Process measures with range [0, ..., 1]
Parallelism
=
|T
par
|
|T|
, perc. of parallel tasks
1
Process =
|{tT| communicationC(t)}|
|T|
, perc. of communication tasks
contacts
Batch =
|{tT| batchC(t)}|
|T|
, perc. of batch tasks
Periodic =
|{tT| periodicB(t)}|
|T|
, perc. of periodic tasks
Level of =
|{tT| checkC(t)}|
|T|
, perc. of control tasks
control
Level of =
|{tT| authorizeC(t)}|
|T|
, perc. of authorization tasks
authorization
IT automation
=
α·|{tT|ET(t)6=
/
0t6∈dom(A)}| + β·|{tT|ET(t)6=
/
0tdom(A)}|
(α+β)·|T|
, perc. of (semi-)automated tasks
IT comm.
=
|{tT |digitalB(t)communicationC(t)}|
|{tT |communicationC(t)}|
, for {tT|communicationC(t)} 6=
/
0
1 , for {tT|communicationC(t)} =
/
0
Department =
|D|
|T|
, perc. of departments
involvement
Department =
|{tT| |DT(t)|≥2}|
|T|
, perc. of tasks shared by departments
share
Process =
|{t
1
,t
2
T | t
1
•∩•t
2
6=
/
0AH(t
1
)AH(t
2
)=
/
0}|
|{t
1
,t
2
T | t
1
•∩•t
2
6=
/
0}|
, perc. of work that is handed over to another role
hand offs
Specialization
=
|{A(t)| tdom(A)}|
|T|
, specialization of roles (with a higher perc. meaning more specialists)
Role usage
=
|{A(t)| tdom(A)}|
|R|
, perc. of actively involved roles
Managerial =
|lrp|
|R|
, perc. of hierarchical layers
2
layers
Knock outs =
|{pP| |•p|>1(
t p
checkC(t)S(t)=XOR∧|t•|>1)(epEP
PN
| p⊆||ep||)}|
|P|
, perc. of k.outs
3
Range
Process size
= |T|, the number of transitions [1, 2, ..., number of transitions]
Versions = |G|, the number of products and services [1, 2, ..., number of products and services]
User =
|U|
|T|
, the average number of users per task
[0, ..., 1, ..., number of users]
involvement
* = measure taken from Nissen (Nissen, 1996)
1. A parallel transition, T
par
, is defined as T
par
T such that t T
par
if and only if there exist two elementary
paths that both start in an AND-split and end in an AND-join, t is on only one of these two paths, and the
AND-split and AND-join are the only two nodes these paths have in common.
2. A role path, rp, in PN is defined as a nonempty sequence r
k
...r
1
of roles which satisfies
(r
k
, r
k1
), ..., (r
2
, r
1
) H. Let RP
PN
be the set of all role paths in PN. Then a longest role path, lrp RP
PN
, is
defined as a role path which satisfies
rpRP
PN
|lrp| |rp|.
3. An elementary path, ep, in PN is defined as a nonempty sequence a
1
...a
k
of nodes which satisfies
(a
1
, a
2
), ..., (a
k1
, a
k
) F
1i< jk
a
i
6= a
j
. Further, let EP
PN
be the set of elementary paths in PN and let
||ep|| be the set of all nodes in the elementary path ep.
ICEIS 2007 - International Conference on Enterprise Information Systems
52
Condition statement: Apply knock-out IF Knock outs
> 0.
Split responsibilities: When applying split re-
sponsibilities the responsibility for a task will be
given to one department.
Condition statement: Apply split responsibilities IF
Department share > 0.
Case types: When applying case types new workflow
processes and product types are distinguished.
Condition statement: Apply case types IF Process
versions > 1.
Case-Based Work: When applying case-based
work each case is handled individually.
Condition statement: Apply case-based work IF
Batch > 0 OR Periodic work > 0.
It is straightforward, using Table 2, to check
that these condition statements are true for the
insurance claim process.
3.4 Alternative Models
The fourth step of our approach derives new process
models based on the selected best practices. At this
stage our approach does not support an automatic
identification of where in the process a best practice
should be applied. This is included in our future work.
However, for the sake of completeness, we here il-
lustrate the remainder of the approach for the insur-
ance claim process. Hence the following list of the 5
selected best practices for which the condition state-
ments evaluated to true, including a possible applica-
tion:
Task elimination: eliminate the control task Check
legal and ask the client to indicate whether (s)he
or some one else was responsible for causing the
damage.
Knock out: perform Check amount and Check le-
gal in a different order if Check amount requires
longer service times and / or has a lower rejection
probability than Check legal.
Split responsibilities: give the responsibility for
the task Classify solely to the Distribution depart-
ment.
Case types: distinguish one workflow process for
the individual claims and one for the business
claims.
Case-based work: remove or change the periodic
activity Authorize pay which should reduce wait-
ing times.
3.5 Evaluation of Alternatives and New
Process Model
In the final steps of our approach, the performance of
the various alternatives is evaluated and one redesign
alternative is selected for implementation. For the
evaluation, performance data (time, cost and quality
indicators) are necessary. Evaluation can be done
by simulating the model or (in simple processes)
with calculation. The alternative that provides the
best performance is selected. We have earlier found
that 11 condition statements (thus best practices)
evaluated to true for the insurance claim process (and
we only illustrated ve of them). Implementing the
best practices separately would lead to 11 redesign
alternatives for the insurance claim process. Each
redesign project has goals (for instance improvement
on throughput time or operational costs) and project
risks (Limam Mansar et al., 2006) which makes
some alternatives more promising than others. For
the insurance claim process an improvement in
throughput time will be achieved with the application
of Task elimination. An improvement on costs could
result from the use of the Knock out best practice. A
careful evaluation with performance data is necessary
to see which alternative will indeed be the best and
should replace the existing insurance claim process.
At this point in time our approach ends with the
selection of the best practices that seem fruitful for
application to the existing process. Based on the de-
scription of the best practices it is straightforward to
come up with possible applications of the best prac-
tices as we showed in our illustration. It should give a
redesign novice enough input to create an alternative
that will perform better than the existing process. In
the future our approach and the supporting redesign
tool should provide redesign alternatives which are
based on the selected best practices and evaluate the
alternatives to find the best one. This would support
both redesign novices and experts.
4 CONCLUSION AND OUTLOOK
In this paper we describe and illustrate a six steps
evolutionary approach towards workflow process re-
design. Our contribution in this paper focuses on
the first three steps leading to suggesting applicable
best practices. This suggestion will already help re-
design novices with the creation of redesign alterna-
tives. We introduced a formal process definition suit-
able for modelling realistic, complex business pro-
cesses. Our process measures have a clear and un-
AN EVOLUTIONARY APPROACH FOR BUSINESS PROCESS REDESIGN - Towards an Intelligent System
53
ambiguous meaning because of their formal notation.
Furthermore, our process measures are directly re-
lated to the redesign best practices with condition
statements.
Our current work holds limitations that we will
be addressing in the future. One direction for fu-
ture research is the extension of the current process
definition, for instance with performance information
about historic process instantiations, to be able to set
up condition statements for all redesign best practices
(steps 1, 2 and 3). Another important direction will
be the exact place in the process model where a suit-
able best practice should be applied and the deriva-
tion of the alternative model (step 4). We also aim at
automating our approach with a highly interactive re-
design tool. In addition to merely generating process
alternatives on the basis of an existing model, such
a tool will be able to process the preferences of the
redesigner for a subset of the alternatives to continue
its search for a satisfying design (steps 5 and 6). The
interaction with the redesigner and the advanced sup-
port will hopefully make our tool a truly “intelligent”
system for BPR.
ACKNOWLEDGEMENTS
This research is supported by the Technology Founda-
tion STW, applied science division of NWO and the
technology programme of the Dutch Ministry of Eco-
nomic Affairs.
REFERENCES
Aalst, W. (1998). The application of petri nets to workflow
management. The Journal of Circuits, Systems and
Computers, 8(1):21–66.
Al-Mashari, M. and Zairi, M. (1999). Bpr implementation
process: An analysis of key success and failure fac-
tors. Business Process Management Journal, 5(1):87–
112.
Bernstein, A., Klein, M., and Malone, T. (1999). The pro-
cess recombinator: a tool for generating new business
process ideas. International Conference on Informa-
tion Systems, pages 178–192.
Calvert, P. (1994). An advanced bpr methodology with
integrated, computer-based tools. In Glasson, B.,
Hawryszkiewycz, I., Underwood, B., and Weber, R.,
editors, Business Process Re-engineering: Informa-
tion Systems Opportunities and Challenges, pages
161–170. Elsevier.
Grover, V., Jeong, S., Kettinger, W., and Teng, J. (1995).
The implementation of business process reengineer-
ing. Journal of Management Information Systems,
12(1):109–144.
Kettinger, W., Teng, J., and Guha, J. (1997). Business pro-
cess change: A study of methodologies, techniques,
and tools. MIS Quarterly, 21(1):55–80.
Ku, S. and Suh, Y. (1996). An investigation of the k-
tree search algorithm for efficient case representa-
tion and retrieval. Expert Systems with Applications,
11(4):571–581.
Limam Mansar, S., Reijers, H., and Ounnar, F. (2006).
Bpr implementation: A decision-making strategy. In
Bussler, C. and Haller, A., editors, Business Process
Management Workshops: BPM 2005, volume 3812 of
Lecture Notes in Computer Science, pages 421–431.
Springer Verlag, Berlin.
Malone, T., Crowston, K., and Herman, G. (2003). Orga-
nizing Business Knowledge: The MIT Process Hand-
book. MIT Press.
Min, D., Kim, J., Kim, W., Min, D., and Ku, S. (1996).
Ibrc: Intelligent bank reengineering system. Decision
Support Systems, 18(1):97–105.
Netjes, M., Vanderfeesten, I., and Reijers, H. (2006). “in-
telligent” tools for workflow process redesign: A re-
search agenda. In Bussler, C. and Haller, A., edi-
tors, Business Process Management Workshops: BPM
2005, volume 3812 of Lecture Notes in Computer Sci-
ence, pages 444–453. Springer Verlag, Berlin.
Nissen, M. (1996). Knowledge-based organizational
process redesign: Using process flow measures to
transform procurement, phd. dissertation. Uni-
versity of Southern California, downloadable at
http://web.nps.navy.mil/ menissen/.
Nissen, M. (1997). Reengineering support through
measurement-driven inference. Intelligent Systems in
Accounting, Finance and Management, 6(2):109–120.
Nissen, M. (1998). Redesigning reengineering through
measurement-driven inference. MIS Quarterly,
22(4):509–534.
Nissen, M. (2000). An experiment to assess the perfor-
mance of a redesign knowledge system. Journal of
Management Information Systems, 17(3):25–44.
Pallas-Athena (2004). Protos User Manual. Pallas Athena
BV, Apeldoorn, The Netherlands.
Reijers, H. and Limam Mansar, S. (2005). Best prac-
tices in business process redesign: An overview and
qualitative evaluation of successful redesign heuris-
tics. Omega: The International Journal of Manage-
ment Science, 33(4):283–306.
Reisig, W. and Rozenberg, G., editors (1998). Lectures on
Petri Nets I: Basic Models, volume 1491 of Lecture
Notes in Computer Science. Springer Verlag, Berlin.
Valiris, G. and Glykas, M. (1999). Critical review of ex-
isting bpr methodologies. Business Process Manage-
ment Journal, 5(1):65–86.
ICEIS 2007 - International Conference on Enterprise Information Systems
54