Inductive Reference Model Development: Recent Results and Current Challenges

Abstract

Reference modeling offers attractive bene®ts for both research and practice. The induc-tive strategy for reference model development derivesreference models by generalizing individual enterprise models. It has recently gained attention in research, however, its practical application still faces numerous challenges. The objective of the article at hand is to identify recent contributions to the ®eld of inductive reference model development and use them to analyze the current challenges that impede the application of their results in practice. We identify at otal of 18 contributions, either scienti®c articles describing inductive methods for reference model development, or practical reports describing the concrete development of areference model for acertain domain. Theyare all analyzed by means of asix-stage-framework for reference model development. Foreach stage, we derive speci®c challenges and point out acknowledgments and potential solutions. 1I ntroduction Reference modeling offers attractive bene®ts for both research and practice [FL07]. Following the epistemologically established differentiation between rationalism and empiri-cism as twofundamental ways of cognition, reference modeling differentiates adeductive and an inductive strategy for reference model development [BS97]. Model development according to the deductive strategy employs generally accepted theories and principles, while the inductive strategy is based on the generalization of individual enterprise models. It focuses on the commonalities of the individual models and abstracts from speci®c features. Hence, deductive development proceeds from the general towards the speci®c (ªtop-down "), whereas inductive development evolves from the speci®c into the general (ªbottom-up "). Although several concepts, methods, and tools for the support of reference modeling exist by now, practical development of reference models still faces avariety of challenges. The open questions range from project preparation, overindividual steps of pre-processing and derivation, to maintenance and continuous improvement of areference model. Hence, the development process of areference model today is often barely structured, nontransparent, and only marginally justi®ed and thus hardly communicable in terms of course and characteristics. In other words, the ideal of agenerally repeatable, engineering-style approach is usually not yet reached. With that, atool support is also only possible in alimited way.

Full text

Heinrich C. Mayr,Martin Pinzger (Hrsg.): INFORMATIK 2016,
Lecture Notes in Informatics (LNI), Gesellschaft f¨
ur Informatik, Bonn 2016 739
Inductive Reference Model Development: Recent Results
and Current Challenges
Jana-Rebecca Rehse, Philip Hake, Peter Fettke, and Peter Loos1
Abstract: Reference modeling offers attractive bene®ts for both research and practice. The induc-
tive strategy for reference model development derivesreference models by generalizing individual
enterprise models. It has recently gained attention in research, however, its practical application still
faces numerous challenges. The objective of the article at hand is to identify recent contributions to
the ®eld of inductive reference model development and use them to analyze the current challenges
that impede the application of their results in practice. We identify atotal of 18 contributions, ei-
ther scienti®c articles describing inductive methods for reference model development, or practical
reports describing the concrete development of areference model for acertain domain. Theyare all
analyzed by means of asix-stage-framework for reference model development. Foreach stage, we
derive speci®c challenges and point out acknowledgments and potential solutions.
Keywords: Inductive Reference Model Development, Reference Modeling, Grand Challenges
1Introduction
Reference modeling offers attractive bene®ts for both research and practice [FL07]. Fol-
lowing the epistemologically established differentiation between rationalism and empiri-
cism as twofundamental ways of cognition, reference modeling differentiates adeductive
and an inductive strategy for reference model development [BS97]. Model development
according to the deductive strategy employs generally accepted theories and principles,
while the inductive strategy is based on the generalization of individual enterprise mod-
els. It focuses on the commonalities of the individual models and abstracts from speci®c
features. Hence, deductive development proceeds from the general towards the speci®c
(ªtop-down”), whereas inductive development evolves from the speci®c into the general
(ªbottom-up”).
Although several concepts, methods, and tools for the support of reference modeling exist
by now, practical development of reference models still faces avariety of challenges. The
open questions range from project preparation, overindividual steps of pre-processing and
derivation, to maintenance and continuous improvement of areference model. Hence, the
development process of areference model today is often barely structured, nontransparent,
and only marginally justi®ed and thus hardly communicable in terms of course and char-
acteristics. In other words, the ideal of agenerally repeatable, engineering-style approach
is usually not yet reached. With that, atool support is also only possible in alimited way.
1Institute of Information Systems at the DFKI, Campus D3.2, 66123 Saarbruecken, Jana-
Rebecca.Rehse@iwi.dfki.de, Philip.Hake@iwi.dfki.de, Peter.Fettke@iwi.dfki.de, Peter.Loos@iwi.dfki.de

740 Jana-Rebecca Rehse et al.
The article at hand intends to pave the wayfor astructured, methodical, repeatable, and
thus justi®able practical application of inductive reference modeling techniques. There-
fore, we analyze recent contributions to the ®eld of inductive reference model development
and point out the current challenges that impede the application of their results in practice.
We consider both scienti®c articles focusing on the underlying methodology and reports
describing the concrete inductive development of adomain-speci®c reference model in
practice, intending to identify as manypractical challenges as possible.
The challenges as presented in the following are comparable to the currently discussed
Grand Challenges of Mertens and Barbian [MB15]. While theydiscuss Grand Challenges
of Business and Information Systems Engineering (BISE) or ªWirtschaftsinformatik” in
general, the contribution at hand is limited to the speci®c challenges of inductive reference
modeling. Although width and relevance of the challenges discussed here are much nar-
rower in the ®rst place, we also intend to fruitfully stimulate future research. LikeMertens
and Barbian, we do not assume our list to be complete, butrather a®rst attempt towards a
more structural examination of inductive reference model development.
The article is structured as follows. After the introductory section, important foundations
of inductive reference modeling are described in Section 2. Section 3outlines the identi®-
cation of challenges by means of aliterature review. The identi®ed challenges are grouped
and described in section 4. Section 5critically assessed our ®ndings, before concluding
the article with an outlook on further research.
2Foundations of Inductive Reference Model Development
The research agenda for reference modeling can be outlined by twomain questions:
1. Questions regarding Reference Models: Howtodesign areference model for acer-
tain application domain? Areference model can capture the current organizational
context (descriptive interpretation) or makeasuggestion for an innovative organiza-
tional context (prescriptive interpretation).
2. Questions regarding Reference Modeling Methods: Howtodevelop and apply ref-
erence models? Methodological questions can examine characteristics and perfor-
mance features of known methods, re®ne existing methods, or invent and evaluate
completely newmethods.
Both research questions can be viewed regarding the inductive development strategy.An-
swers concerning the current state of research are exempli®ed by the following aspects:
1. Reference Models: Several works report on the inductive development of areference
model ([AFF11, KPR08, DM05], q.v.the attribute ªconstruction method” of the
reference models listed in the catalog at http://rmk.iwi.uni-sb.de).

Inductive Reference Model Development: Recent Results and Current Challenges 741
2. Reference Modeling Methods: Known procedure models for reference modeling in-
dicate that existing individual enterprise models as well as further sources of knowl-
edge should be identi®ed and considered within the context of reference model-
ing [Be02, Th06]. In addition, there are several contributions that suggest method-
ological approaches for different aspects of inductive reference model development
[Ar13, GvJV08, LRW09, RFL13].
Although the research agenda for inductive reference modeling has made acertain progress,
there exists avariety of challenges, which have not yet been systematically addressed.
In order to provide an extensive and systematic assessment of challenges, we followthe
method for inductive reference model development, as de®ned by Fettke[Fe14]. It con-
sists of six consecutive stages, each identifying and describing acentral task necessary for
the practical development of areference model. The stages are shown in Fig. 1and shortly
outlined in the following.
Preparation Collection Preprocessing Acquisit ion Postprocessing Evaluation &
Enhancement
Fig. 1: Stages of Inductive Reference Modeling [Fe14]
1. Preparation of Reference Model Development: Before areference model can be
developed, several preparatory decisions and assumptions have to be made regarding
the context of the reference model development. Among other things, this stage
includes identifying stakeholders’ requirements and de®ning modeling conventions.
2. Collection of Individual Models: The inductive development strategy is based on
individual enterprise models as obligatory input data. These models are collected
in this stage. Depending on the size of the reference model domain, the collection
may be split into several sub-stages, identifying classes of models and representative
organizations before collecting the actual models.
3. Preprocessing of Individual Models: After the individual models have been col-
lected, theyhavetobealigned and harmonized before theycan be used for the
derivation of areference model. This stage includes checking and establishing nec-
essary modeling conventions as well as identifying correspondences.
4. Acquisition of the Reference Model: This stage describes inductive reference model
development in anarrower sense. The preprocessed individual models are used to
derive areference model, using an arbitrary similarity measure and construction
approach. Depending on the number of input models, it may makesense to split
them into several model clusters and develop an individual reference model for each
cluster.
5. Postprocessing of the Reference Model: As the previous stage is often fully auto-
mated and, even if not, limited by the contents of the input model data, the acquired
reference model should be manually postprocessed in this stage. This may include

742 Jana-Rebecca Rehse et al.
connecting unconnected model parts, adding, deleting or renaming nodes, or com-
plementing the reference model by deductively developed model parts.
6. Evaluation and Enhancement of the Reference Model: Areference model will be
especially useful if it is regularly evaluated and replenished as necessary.New in-
dividual models have to be included, as theyarise. If the process domain changes
signi®cantly,itmight makesense to develop acompletely newreference model as
well.
3Identification and Analysis of Relevant Literature
3.1 LiteratureReview
In order to identify practical challenges to the inductive development of reference models,
the ®rst step is to identify relevant contributions to the ®eld, i.e. literature that possibly de-
scribes such challenges. Therefore, we conduct aliterature review, following the method-
ology as proposed by Fettke[Fe06]. After de®ning the objective and research question of
our literature review, we conduct the literature search by entering pre-de®ned keywords in
several research databases. The obtained results are assessed regarding their relevance to
the posed research question. Thereby,wefocus on contributions that concretely describe
an inductive method, as opposed to methodically open ones (e.g. [Th06]). While those
contributions may generally apply to the inductive strategy,theydonot report on anycon-
crete challenges, hence not providing answers relevant to our research question. In this
assessment, we followthe Model of System Design [FHL10]: The respective contribution
has to describe atechnique, i.e. areliable means to realize acertain goal, and explicate its
foundation on empirical knowledge in order to be classi®ed as inductive.
Forconducting aliterature review, aset of keywords has to be de®ned. Since the term
ªInductive Reference Model Development” is not consistently used, several different key-
word combinations assure that as much literature as possible is covered. We combine the
term ªreference model” with different verbs describing the process of automatic model
development. Each verb can appear in either its in®nitive form (ªdevelop”), as agerund
(ªdeveloping”), or as the corresponding noun (ªdevelopment”). Overall, there are eight
keyword combinations, as listed in Table 1. The search is limited to title, keywords, and
abstract of the articles. If those indicate arelation to reference modeling the full text is
assessed. We conduct the search in ®vedatabases, namely Academic Search Complete
(ASC, via EBSCO)2,AIS Electronic Library (AISeL)3,Business Source Premier (BSP,
via EBSCO)4,io-port 5,and Scopus6.
After accumulating all relevant literature, we conduct abackward search, analyzing the
citations of the identi®ed articles to account for prior contributions to consider.Finally,
2http://search.ebscohost.com/
3http://aisel.aisnet.org/
4http://search.ebscohost.com/
5http://www.io-port.net
6http://www.scopus.com

Inductive Reference Model Development: Recent Results and Current Challenges 743
we use Google Scholar to conduct aforward search, ®nding articles that cite the arti-
cles identi®ed in the previous steps. As reference modeling is traditionally atopic of the
German-speaking research on information systems, we also conduct an equivalent litera-
ture search using German keyword combinations. As the chosen databases mainly contain
English articles, this search does not yield anysigni®cant results. However, we include
German-language contributions in our forwards and backwards search to yield abetter
coverage of the literature.
Search Terms ASC AISeL BSP io-port Scopus
All Rel. All Rel. All Rel. All Rel. All Rel.
ªreference model” AND inductive 1072000011 2
ªreference model” AND mining 10 1 8 0 8 1 19 4 163 3
ªreference model” AND automatic 25 0 3 0 8 0 0 0 391 0
ªreference model” AND (generation
OR generate OR generating) 118 1 25 0 45 1 43 3 920 5
ªreference model” AND (construction
OR construct OR constructing) 72 0 52 0 30 0 45 1 692 0
ªreference model” AND (development
OR develop OR developing) 278 0 188 2 194 0 137 0 2859 4
ªreference model” AND (derivation
OR derive OR deriving) 59 0 32 1 19 0 26 1 355 0
ªreference model” AND (discovery
OR discoverORdiscovering) 38 1 2 0 20 0 18 1 142 3
Tab. 1: Quantitative Results of the Literature Review
The quantitative results of the literature search are displayed in Table 1. Foreach keyword
combination and each database, we list the overall number of search results (in column
ªAll”) and the number of relevant search results (in column ªRel.”). The relevance of an
article is assessed manually,based on its title, keywords, and abstract. While the number
of search results differs considerably,the number of relevant results is consistently very
small. The large number of insigni®cant results is caused by the fact that, although relevant
to the ®eld of reference modeling, manyofthe used search terms (e.g. ªconstruction” or
ªgeneration”) can be used in completely different contexts. In addition, manyresearch
disciplines use reference models as ascienti®c tool or methodology.
Accumulating all relevant articles and eliminating duplicates results in atotal of ten con-
tributions that we consider relevant for our research question. Performing aforwards and
backwards search yields an additional eight articles, resulting in atotal of 18 contribu-
tions, as listed in table 2. Of these 18, three [AFF11, GS14, KPR08] have the character of
apractical experience report on inductive reference modeling, whereas the remaining 15
takeonarather scienti®c point of view.

744 Jana-Rebecca Rehse et al.
3.2 Analysis of Contributions
In order to identify challenges to inductive reference model development, we analyze each
contribution for challenges that are either explicitly mentioned as such, or implicitly ad-
dressed by according measures. Forexample, apaper may require previously de®ned activ-
ity correspondences in form of identical descriptions (implicit), while others use dedicated
techniques for correspondence identi®cation (explicit). The results of this assessment are
shown in table 2. Explicit references are symbolized by afull circle ( ), implicit refer-
ences by ahalf-full circle ( ). An empty circle ( )means that the respective challenge is
not mentioned in this contribution. All challenge have to be explicitly mentioned at least
once. In order to provide abetter overviewand to point out relations, each challenge is
allocated to one of the six stages of reference modeling. An in-depth description of each
challenge is contained in the following section 4.
Challenge
[AFF11]
[Ar13]
[GvJV08]
[GS14]
[KPR08]
[LRW09]
[LRW10]
[LRW11]
[MFL14]
[MFL15]
[MG13]
[PIˇ
SK12]
[RFL13]
[RFL15]
[WFL13]
[YB11]
[Ya12]
[YWB12]
1
Indication Criteria
Modeling Language
Modeling Conventions
2
Model Character
Model Access
Collection Effort
3
Modeling Language
Modeling Conventions
Correspondences
4
Abstraction
Model Clustering
Decomposition
Model Variants
Contortion
Inductive Fallacy
Complexity
5Additions
Model Connection
6Evaluation
NewModels
Tab. 2: Foundation of Identi®ed Challenges in Literature
The results of our assessment allowfor several interesting insights. First of all, none of
the challenges is particular to one speci®c contribution. Each of them is mentioned at
least twice, although sometimes implicitly,meaning that the respective contribution de-
scribes measures to address this challenge, without explicating whyorhow it in¯uences

Inductive Reference Model Development: Recent Results and Current Challenges 745
the reference model development. Most of these implicit references appear in case studies
or empirical evaluations, such as [AFF11] or [GS14]. Rather epistemological challenges,
such as the problem of input data contortion or inductive fallacies are rarely mentioned.
Instead, most contributions focus on practical or technical challenges, such as reaching a
feasible algorithmic complexity or ways to establish correspondences between individual
models in terms of amatching. While these challenges are unquestionably important for
the practical application of the inductive strategy,epistemological aspects should also be
considered, as theyplay an important role for evaluating the resulting reference model.
Regarding the six stages of reference modeling, it is apparent that stage four,concerned
with the derivation of the reference model, contains the most challenges and is addressed
by every single contribution. As this stage contains the inductive reference model devel-
opment in anarrower sense, it is the core of most inductive development approaches and
hence poses more challenges that have to be addressed. The preparatory ®rst stage is also
well-covered, although this is mostly caused by assumptions and restrictions to the input
data. Nevertheless, this critical assessment is important for developing newapproaches to
reference modeling, which can potentially overcome these restrictions. On the contrary,
both the collection stage (two) and the postprocessing and evaluation stages (®ve and six)
are only sparsely covered by the contributions. While it is possible to develop areference
model without postprocessing and evaluating it, the collection of the input models is cen-
tral to the success of the model development itself, as the contents of the reference model is
directly in¯uenced by the contents and character of the input models. It is remarkable that
the challenges of this stage are only addressed by about half the contributions and mostly
in an implicit way. Manycontributions assume to have access to asuf®ciently large and
representative collection of digitally represented individual models, which simpli®es most
research activities, butisanon-trivial problem in practical applications.
Regarding the contributions, it is apparent that some takeonamuch broader viewon
inductive reference model development than others. Especially the practical reports on in-
ductive reference model development [AFF11, GS14, KPR08] covermanychallenges that
are not considered by other,rather methodical contributions. Among these, the degree of
examining potential challenges differs considerably.While some contributions try to take
arather hollstic viewonreference model development [RFL15, LRW11], others focus on
one particular challenge, which is analyzed in depth (for example input model abstraction
in [RFL13]). Other contributions have arather technical viewonreference model devel-
opment, focusing on algorithmic aspects and technical details (e.g. [MG13, YB11]).
4Current Challenges of Inductive Reference Model Development
4.1 Preparation of Reference Model Development
Before beginning to collect data, several preliminary questions have to be answered. Es-
sential challenges are:

746 Jana-Rebecca Rehse et al.
•Indication Criteria: It is indisputable that according individual models have to be
available to enable inductive reference model development. If, during the prepara-
tory stage, it is unclear whether an adequateamount of individual models exist
or can be easily acquired, the entire project is at risk. Hence, we require criteria
determining when to resort to an inductive development and when to avoid it. A
well-founded empirical knowledge base is acrucial success factor for the inductive
strategy [WFL13]. Moreover, for organizational reasons, the development of anew
reference model might not always be supported [LRW11].
•Choice of Modeling Language: The choice of the reference modeling language in
the context of the inductive strategy is important for tworeasons. First of all, it
makes sense to followthe individual models in order to avoid unnecessary con-
versions and transformations. Therefore, algorithmic approaches as described in
Martens et al. [MFL14] or Yahyaetal. [YWB12] require aformal mostly graph-
based representation of the individual models and the resulting reference model.
Second, due to the number and variety of the individual models to process, it is
often necessary to represent the model variants in an appropriate way. Hence, the
target language should provide concepts for variant management as described in
[GvJV08].
•Modeling Conventions: In general, modeling conventions are regarded to be impor-
tant. However, due to the plurality of modeling contexts, it cannot be assumed that
modeling conventions are always factually enforced when creating the input models.
Hence, according compromises have to be found. Especially the algorithmic solu-
tions we identi®ed address this challenge, since theyrely on characteristics such as
unique labels [LRW09] or block-structuredness of the input process models [Ar13].
4.2 Collection of Individual Models
Individual enterprise models are an obligatory input and thus anecessary requirement
for applying the inductive development strategy.Theycan either be collected originary
(primary collection)[PIˇ
SK12] or existing information can be reused (secondary collection)
[GvJV08, AFF11]. Essential challenges are:
•Representative Character of the Individual Models: If the representative character
of the individual models is questionable, there is arisk of contortion. Forinstance,
particular model variants may be overrated in the derivation of the reference model.
Meanwhile, recognizing such contortions requires to knowthe entire model domain.
In some cases, however, this domain can only reasonably be assessed after the refer-
ence model has been described. In other applications, it may makesense to purpose-
fully avoid arepresentative character,for instance to examine contrasting individual
models. In this case, the question arises howmanycontrasting cases are required.
Hence, Karowetal. [KPR08] require arich empirical database and an inductive val-
idation of the reference model. Pajk et al. [PI ˇ
SK12] address this challenge by ®rst
developing the model on the empirical data available and then assessing it according
to prede®ned requirements.

Inductive Reference Model Development: Recent Results and Current Challenges 747
•Access to Individual Models: According to Karowetal. [KPR08], accessing en-
terprise models can be dif®cult to impossible. Individual models can be subject to
con®dentiality or organizations see the risk to loose competitive advantages by al-
lowing them to circulate. Hence, cooperation cannot be expected when applying the
inductive development strategy.Therefore, the proof of concept for the method pro-
posed in [KPR08] is settled in the ®eld of public administration and it is based on
public domain models.
•Collection Effort: The effort necessary for collecting the individual models can over-
compensate the bene®ts of reference modeling. This risk is especially apparent, if
cost and bene®ts are very differently distributed among the stakeholders [RFL15].
4.3 Preprocessing of Individual Models
Before concrete methods for reference model acquisition can be applied, ageneral prepro-
cessing of the individual models appears to be useful. Essential challenges are:
•Transformation of the Modeling Language: If the individual models are not repre-
sented in auniform modeling language, it is suitable to transform and convert them
into auniform target language. This is crucial for the algorithmic approaches we
obtained, since the processing relies on auniform representation, for example as an
Event-DrivenProcess Chain [RFL15]
•Examination and Establishment of Modeling Conventions: If certain conventions
have been determined, it makes sense to examine the models for meeting them. If
deviations exist, according corrections of the individual models are required. It may
also makesense to reconstruct the factually used conventions [WFL13].
•Identi®cation of Correspondences: Single constructs in different individual models
can be equivalent, similar,ordifferent. It is essential to knowabout such corre-
spondences. The need for and the challenges of identifying these correspondences
is addressed in most of the approaches that aim at an algorithmic solution [MFL15,
Ar13, RFL15]. The correspondences can be derivedfrom solely structural charac-
teristics of the input models (for example similar graph-theoretic model features
[RFL15]) or be related to semantic aspects (for example similar process traces or
words in node labels [Ar13, MFL15, MFL14]). The fact that structural similarities
are neither necessary nor suf®cient for semantic similarities, is an additional chal-
lenge.
4.4 Acquisition of the Reference Model
The aim of the reference model acquisition stage is to derive areference model from aset
of individual models. Essential challenges are:

748 Jana-Rebecca Rehse et al.
•Abstraction and Generalization: Methods for abstracting speci®c model features are
required to generalize the contents of the individual models. If the individual models
contain commonalities, theyare to be represented in the reference model. This way,
the reference model contains the typical structures of the application domain. Hence,
the acquired reference model should abstract from speci®c features of individual
models. In analogy to the identi®cation of correspondences described above,not
only the model structure, butalso the model semantics should be considered. In
[RFL13], avariety of abstraction approaches are evaluated in terms of applying
them in the context of inductive reference model development.
•Clustering of Individual Models: Developing model clusters is important for two
reasons. First, individual models can stem from completely different contexts, so
acluster analysis can be used to identify possible domains [WFL13](for instance
for cross-sectional tasks likeaccounting or logistics). Second, Aier et al. [AFF11]
postulate acluster analysis in order to group avariety of model variants within the
same domain.
•Decomposition of Individual Models: Forthe decomposition, twoaspects should
be considered. First, individual models often consist of numerous partial models,
which should be analyzed separately [GS14]). Second, even asingle partial model
may be decomposed into fragments. This makes sense, because twomodels can be
different, while individual model fragments may be similar or even identical.
•Plurality of Model Variants: When acquiring areference model, twoopposing de-
sign objectiveshavetobebalanced. One extreme case includes all individual models
as model variants in the input model, the other extreme case only captures the com-
monalities of the individual models in the reference model and abstracts from indi-
vidual model features. Aier et al. [AFF11] describe this challenge as afundamental
con¯ict between standardization and individualization of the reference model.
•Handling Contortions: If the collected individual models do not constitute arepre-
sentative sample of the domain, it is to be checked whether possible contortions exist
and can be corrected by taking according measures. Contortions can be prevented
by selecting arepresentative or suf®ciently large set of organizations [RFL15].
•Risk of Inductive Fallacy: Philosophyingeneral and (inductive)logic in particular
often refer to the weaknesses of inductive conclusions [AFF11]. This aspect should
also be considered in the inductive development strategy.The fact that models may
contain descriptive as well as prescriptive content also has to be taken into account
[RFL13].
•Algorithmic Complexity: There exists avariety of algorithms for model analysis that
include techniques for comparing process models. These techniques include graph-
based structural comparisons [YB11, Ya12, YWB12] as well as comparisons that
deal with the processessing and comparison of natural language [Ar13, MFL15].
Due to their high runtime and memory consumption, such algorithms are unable to
cope with larger model sets as theytypically appear within the inductive develop-
ment strategy.

Inductive Reference Model Development: Recent Results and Current Challenges 749
4.5 Postprocessing of the Reference Model
Essential challenges when postprocessing areference model are:
•Additions: The inductive approach of Grger and Schumann [GS14] includes an en-
hancement of the inductively derivedreference model. Reference models can be
expected to be complemented by additional contents. The question arises, howto
combine inductive and deductive strategies into ahybrid development strategy.
•Connection of Partial Models: It should be considered to group separate reference
models. Grouping makes sense, if the individual models were accordingly decom-
posed in aprevious step [GS14, KPR08]. Adding such connections is also useful to
point out similarities between reference models of different domains (for example
warehousing processes in retail and industry).
4.6 Enhancement of the Reference Model
Reference modeling is especially useful if the reference model is constantly evaluated and
enhanced. Otherwise, there is arisk of obsolescence. Essential challenges are:
•Evaluation: In this context, it is important to assess the generated reference model
with regard to the prede®ned goals. Dependening on the speci®c stakeholder or per-
spective,the verifciation and validation varies. While Aier et al. [AFF11] conduct
interviews with process users, Pajk et al. [PIˇ
SK12] propose asimulation or even an
empirical evaluation by applying the reference model.
•NewIndividual Models: If newvariants are derivedfrom areference model, they
should be considered to be included in the reference model. Reference model stabil-
ity should be aimed for,i.e. not every variant should be included. However, it is obvi-
ously sensible to include standard adaptations, which are regularly executed during
an application, into the reference model. It might be desirable to constantly enhance
the reference model to minimize its distance to all existing variants [LRW11].
5Discussion and Conclusion
The objective of the article at hand wastoidentify acatalog of potential challenges to the
practical application of the inductive development strategy for reference modeling. This
catalog is meant to stimulate further research by providing alist of empirically grounded
challenges, which can be addressed in astructured waytoadvance the inductive strategy
both in research and practice. However, we do not claim our list of challenges to be com-
plete or exhaustive.For example, none of the contributions discusses the possibility for a
regular re-design of the reference model, which could become necessary,ifthe input mod-
els change considerably.Infuture research, we intend to validate and extend the identi®ed

750 Jana-Rebecca Rehse et al.
challenges by conducting expert interviews before dealing with each challenge individu-
ally.Interviewing reference modeling experts can offer acomplementary perspective and
lead to additional insights from apractitioner’spoint of view.
Our contribution in this article can only be astarting point for the in-depth analysis of the
challenges to inductive reference modeling. The term ªreference model” is mainly popular
in German-speaking countries and Australia, while other communities prefer terms such
as ªreusable model”. Extending our literature reviewbythose terms can further advance
our ®ndings. Meanwhile, we cannot assure that additional challenges are mentioned in
previous works on reference modeling, not focusing on the inductive strategy.However,
it is almost impossible to assess their relevance without taking explicit features of the
inductive strategy,such as the foundation on empirical knowledge, into account.
It should also be noted that none of the identi®ed contributions contains adirect reference
towards Business Process Management in particular,although it appears self-evident that
inductive reference modeling offers interesting potentials for classical BPM topics, such as
continuous process improvement, customization and implementation of standard software,
or process documentation. Traditional applications of reference modeling, such as design,
optimization, simulation, or certi®cation of business processes may also be addressed by
inductive methods. Simultaneously,manyidenti®ed challenges are related to topics that
are already addressed by BPM researchers. Forexample, advancements on naming con-
ventions, model enhancement, or variant management could be built upon to deal with the
according challenges.
Despite the numerous challenges presented in this article, the inductive development strat-
egyhas considerable advantages when compared to the deductive strategy.Analyzing in-
dividual models and their potentials for generalization is not only intellectually interesting,
butappears almost obligatory if the models are easily available. Second, adirect compari-
son does not require general principles and theories to derive areference model. Moreover,
an inductively developed reference model can be expected to have ahigher degree of de-
tailing, maturity,and acceptance. It thus appears to be especially fruitful to advance the
integration of inductive and deductive development into ahybrid strategy.Ingeneral, in-
ductive reference modeling is still at arather early stage of development and there is alot
of potential for further research.
In addition, methods of inductive reference modeling offer interesting potentials for re-
lated applications. Forexample, model comparison, model integration, or analysis of in-
ternal model variants come to mind. Besides these practical applications, there are also
interesting potentials for an application in research. Inductive methods can for example be
used for ascienti®cally well-grounded comparison of models. Theyare hence an important
foundation for one of the Grand Challenges identi®ed by Mertens and Barbian [MB15]:
Following the Human Genome Project, the ªWirtschaftsinformatik” should chart existing
IT systems in order to drawconclusions for the future employment of IT and the accu-
racyofits systems. Such activities may signi®cantly bene®t from inductive methods and
thus discovernew potentials for known works on organizational typology and the model
of shell and nucleus.

Inductive Reference Model Development: Recent Results and Current Challenges 751
References
[AFF11] Aier,Stephan; Fichter,Michael; Fischer,Christian: Referenzprozesse empirisch bestim-
men :von Vielfalt zu Standards. Wirtschaftsinformatik &Management :WuM, 3(3),
2011.
[Ar13] Ardalani, Peyman; Houy,Constantin; Fettke, Peter; Loos, Peter: Towards aMinimal Cost
of Change Approach for Inductive Reference Model Development. In: Proceedings of
the 21st European Conference on Information Systems. European Conference on Infor-
mation Systems (ECIS-13). AIS, 2013.
[Be02] Becker,J
¨
org; Delfmann, Patrick; Knackstedt, Ralf; Kuropka, Dominik: Kon®gurative
Referenzmodellierung. In (Becker,J.; Knackstedt, R., eds): Wissensmanagement mit
Referenzmodellen: Konzepte f ¨
ur die Anwendungssystem- und Organisationsgestaltung,
pp. 25±144. Springer,2002.
[BS97] Becker,J
¨
org; Sch¨
utte, Reinhard: Reference Information Systems for Retail: De®nition,
Use and Recommendations for Design and company-speci®c Adaption of Reference
Models. In: Wirtschaftsinformatik, pp. 427±448. Springer,1997. In German.
[DM05] Daun, Christine; Matheis, Thomas: Constructing areference process model for E-
government. In (Mosca, R., ed.): Proceedings of the 7th International Conference on
The Modern Information Technology in the Innovation Processes of the Industrial Enter-
prises (MITIP), Genua. pp. 10±14, 2005.
[Fe06] Fettke, Peter: State-of-the-Art des State-of-the-Art. Wirtschaftsinformatik, 48(4):257±
266, 2006.
[Fe14] Fettke, Peter: Eine Methode zur induktivenEntwicklung vonReferenzmodellen. In
(Kundisch, Dennis; Suhl, Leena, eds): Tagungsband Multikonferenz Wirtschaftsinfor-
matik 2014. Universit¨
at Paderborn, 2014.
[FHL10] Fettke, Peter; Houy,Constantin; Loos, Peter: On the relevance of design knowledge for
design-oriented business and information systems engineering. Business &Information
Systems Engineering, 2(6):347±358, 2010.
[FL07] Fettke, Peter; Loos, Peter: PerspectivesonReference Modeling. In (Fettke, Peter; Loos,
Peter,eds): Reference Modeling for Business Systems Analysis, pp. 1±20. Idea Group
Publishing, 2007.
[GS14] Gr¨
oger,Stefan; Schumann, Matthias: Entwicklung eines Referenzmodells f¨
ur die Gestal-
tung des Drittmittel-Prozesses einer Hochschule und Ableitung vonEinsatzgebieten
f¨
ur Dokumenten-und Work¯ow-Management-Systeme. Arbeitsbericht der Universit¨
at
G¨
ottingen, 1, 2014.
[GvJV08] Gottschalk, Florian; vander Aalst, Wil; Jansen-Vullers, Monique: Mining Reference Pro-
cess Models and Their Con®gurations. In: On the Move to Meaningful Internet Systems:
OTM2008 Wo rkshops. Springer,pp. 263±272, 2008.
[KPR08] Karow, Milan; Pfeiffer,Daniel; R¨
ackers, Michael: Empirical-Based Construction of Ref-
erence Models in Public Administrations. In: Multikonferenz Wirtschaftsinformatik. pp.
1613±1624, 2008.
[LRW09] Li, Chen; Reichert, Manfred; Wombacher,Andreas: Discovering Reference Models by
Mining Process Variants Using aHeuristic Approach. In (Dayal, Umeshwar; Eder,Jo-
hann; Koehler,Jana; Reijers, Hajo, eds): Business Process Management, volume 5701 of
Lecture Notes in Computer Science, pp. 344±362. Springer Berlin Heidelberg, 2009.

752 Jana-Rebecca Rehse et al.
[LRW10] Li, Chen; Reichert, Manfred; Wombacher,Andreas: The MinAdept clustering approach
for discovering reference process models out of process variants. International Journal
of Cooperative Information Systems, 19(03n04):159±203, 2010.
[LRW11] Li, Chen; Reichert, Manfred; Wombacher,Andreas: Mining business process variants:
Challenges, scenarios, algorithms. Data &Knowledge Engineering, 70(5):409±434,
2011.
[MB15] Mertens, Peter; Barbian, Dina: Researching ªGrand Challenges”. Business &Informa-
tion Systems Engineering, 57(6):391±403, 2015.
[MFL14] Martens, Alexander; Fettke, Peter; Loos, Peter: AGenetic Algorithm for the Inductive
Derivation of Reference Models Using Minimal Graph-Edit Distance Applied to Real-
World Business Process Data. In (Kundisch, Dennis; Suhl, Leena; Beckmann, Lars, eds):
Tagungsband Multikonferenz Wirtschaftsinformatik 2014. Universit¨
at Paderborn, 2014.
[MFL15] Martens, Alexander; Fettke, Peter; Loos, Peter: Inductive Development of Reference
Models Based on Factor Analysis. In (Thomas, Oliver; Teuteberg, Frank, eds): Pro-
ceedings der 12. Internationalen Tagung Wi rtschaftsinformatik (WI 2015). volume 12,
Universit¨
at Osnabr¨
uck, Osnabrck, Germany, pp. 438 ±452, 2015.
[MG13] Mejri, Asma; Ghannouchi, Sonia Ayachi: Discovering Reference Process Models in the
Context of BPM Projects. Procedia Technology,9:489±497, 2013.
[PIˇ
SK12] Pajk, Dejan; Indihar-ˇ
Stemberger,Mojca; Kovaˇ
ciˇ
c, Andrej: Reference model design: An
approach and its application. In: Information Technology Interfaces (ITI), Proceedings
of the ITI 2012 34th International Conference on. IEEE, pp. 455±460, 2012.
[RFL13] Rehse, Jana-Rebecca; Fettke, Peter; Loos, Peter: Eine Untersuchung der Potentiale au-
tomatisierter Abstraktionsans¨
atze f¨
ur Gesch¨
aftsprozessmodelle im Hinblick auf die in-
duktive Entwicklung vonReferenzprozessmodellen. In (Alt, Rainer; Franczyk, Bog-
dan, eds): Proceedings of the 11th International Conference on Wirtschaftsinformatik,
Leipzig, Germany. 2013. In German.
[RFL15] Rehse, Jana-Rebecca; Fettke, Peter; Loos, Peter: Agraph-theoretic method for the induc-
tive development of reference process models. Software &Systems Modeling, tba:1±41,
2015.
[Th06] Thomas, Oliver: Management vonReferenzmodellen: Entwurf und Realisierung eines
Informationssystems zur Entwicklung und Anwendung vonReferenzmodellen. Logos-
Verlag, 2006.
[WFL13] Walter,J
¨
urgen; Fettke, Peter; Loos, Peter: HowtoIdentify and Design Successful Busi-
ness Process Models: An Inductive Method. In (Becker,J
¨
org; Matzner,Martin, eds): Pro-
moting Business Process Management Excellence in Russia -Proceedings and Report of
the PropelleR 2012 Workshop. volume 15. European Research Center for Information
Systems, M¨
unster,pp. 89±96, 2013.
[Ya12] Yahya, Bernardo; Bae, Hyerim; Bae, Joonsoo; Kim, Dongsoo: Generating valid refer-
ence business process model using genetic algorithm. International Journal of Innovative
Computing, Information and Control, 8(2):1463±1477, 2012.
[YB11] Yahya, Bernardo; Bae, Hyerim: Generating Reference Business Process Model Using
Heuristic Approach Based on Activity Proximity.In: Intelligent Decision Te chnologies,
pp. 469±478. Springer,2011.
[YWB12] Yahya, Bernardo; Wu,Jei-Zheng; Bae, Hyerim: Generation of Business Process Refer-
ence Model Considering Multiple Objectives. Industrial Engineeering &Management
Systems, 11(3):233±240, 2012.