ArticlePDF Available

Abstract and Figures

Enterprise Architecture (EA) modeling languages can express the business-to-IT-stack for an organization, showing how changes in the IT landscape impact business aspects and vice versa. Yet EA languages provide only the final architectural design, not the rationale behind this design. In earlier work, we presented the EA Anamnesis approach for EA rationalization. We discussed how EA Anamnesis forms a complement to current EA modeling languages, showing for example design alternatives, EA artifact selection criteria and the decision making strategy that was used. In this paper, we extend EA Anamnesis with a capability for organizational learning. In particular, we present an integration of two viewpoints presented in earlier work: (1) an ex-ante decision making viewpoint for rationalizing EA during decision making, which for example captures a decision and its anticipated consequences, and (2) an ex-post decision making viewpoint, which for example captures the unanticipated decision consequences, and possible adjustments in criteria. We use a fictitious, yet realistic, case study to illustrate our approach.
Content may be subject to copyright.
EA Anamnesis: An Approach for Decision Making Analysis
in Enterprise Architecture
Georgios Plataniotis*, Public Research Centre Henri Tudor, Luxembourg & Radboud
University Nijmegen, the Netherlands & EE-Team, Luxembourg1
Sybren de Kinderen, University of Luxembourg, Luxembourg & EE-Team, Luxembourg
Henderik A. Proper, Public Research Centre Henri Tudor, Luxembourg & Radboud University
Nijmegen, the Netherlands & EE-Team, Luxembourg
Enterprise Architecture (EA) modeling languages can express the business-to-IT-stack for an
organization, showing how changes in the IT landscape impact business aspects and vice versa.
Yet EA languages provide only the final architectural design, not the rationale behind this
In earlier work, we presented the EA Anamnesis approach for EA rationalization. We discussed
how EA Anamnesis forms a complement to current EA modeling languages, showing for
example design alternatives, EA artifact selection criteria and the decision making strategy that
was used.
In this paper, we extend EA Anamnesis with a capability for organizational learning. In
particular, we present an integration of two viewpoints presented in earlier work: (1) an ex-ante
decision making viewpoint for rationalizing EA during decision making, which for example
captures a decision and its anticipated consequences, and (2) an ex-post decision making
viewpoint, which for example captures the unanticipated decision consequences, and possible
adjustments in criteria.
We use a fictitious, yet realistic, case study to illustrate our approach.
Keywords: Enterprise Architecture, Design Rationale, Organizational learning, Decision Design
Enterprise Architecture (EA) languages are considered as an instrument for expressing an
enterprise holistically (Lankhorst, 2005), linking perspectives on an organization that are usually
considered in isolation. In doing so, one can consider the organization-wide impact of a change
1 The Enterprise Engineering Team (EE-Team) is a collaboration between Public Research Centre Henri Tudor,
Radboud University Nijmegen and HAN University of Applied Sciences (
(Lankhorst, 2005; Op’t Land, Proper, Waage, Cloo and Steghuis, 2008), expressing its complete business-
to-IT-stack (Aier and Winter, 2009). For example, for a newly introduced IT application, EA modeling
languages can be used to consider implications on business processes, human resources, organizational
goals, and more.
While EA modeling languages can be used to express the holistic design of an organization, they
do not express the design decisions behind the resulting models. Although we should be careful with the
analogy, experience from the eld of software architecture shows that leaving design rationales implicit
leads to architectural knowledge vaporization” (cf. Jansen and Bosch, 2005). This means that, without
design rationale, one leaves implicit design criteria, reasons for a design, and design alternatives.
As a result of lacking rationalization architects are unable to justify their designs (Tang, Jin and
Han, 2005). Furthermore new designs are constructed in an ad hoc manner without taking into
consideration constraints implied by past design decisions (Tang et al., 2007). Here, constraints refer to
boundaries implied by the design. These boundaries can be of a business or of a technical nature, such as
the choice for a programming language implied by choosing a particular application environment.
Moreover, a survey amongst enterprise architecture practitioners (Plataniotis, 2013c) provides
indications of the usefulness of design rationalization for motivating design decisions, and for
architectural maintenance. At the same time, however, the survey shows that practitioners often forego
the use of a structured template/approach when rationalizing an architecture, relying instead on ad hoc
information capturing in tools such as Microsoft Oce.
In earlier work (Plataniotis, de Kinderen and Proper, 2012; Plataniotis, de Kinderen and Proper,
2013a; Plataniotis, de Kinderen and Proper, 2013b), we introduce the EA Anamnesis approach for archi-
tectural rationalization. EA Anamnesis derives from the ancient greek word ανάµνησις (/ ænæm"ni:sIs/),
which denotes memory and repair of forgetfulness. Our earlier work addresses two separate viewpoints
for architectural rationalization: ex-ante and ex-post. On the one hand, (Plataniotis et al., 2013a) addresses
ex-ante decision rationalization. In so doing, it focuses on concepts relevant during or just after a decision
making process, such as decision criteria and the anticipated consequences of a decision. On the other
hand, (Plataniotis et al., 2013b) focuses on a reection on past decisions. This ex-post viewpoint focuses
on concepts relevant some time after the decision making process, such as unanticipated decision
consequences and a possible subsequent change in the importance score of criteria.
In this paper, we focus on relating the ex-ante and ex-post viewpoints of EA Anamnesis. We do
so because these viewpoints naturally complement each other. For one, by a comparison of ex-post
unanticipated consequences to ex-ante decision criteria and anticipated consequences, we can inform
future decision making. In so doing, we essentially provide a good rst step in using EA Anamnesis for
“organizational learning” (cf. Conklin, 1996) in the area of decision making.
As such, the main contribution of this paper is twofold: (1) to integrate two viewpoints that we
considered previously in isolation, (2) to show how this integration can be useful for learning from past
decision making. For now, we focus our approach on a single decision maker: the enterprise architect.
At this point, we would like to note that existing rationalization approaches in software
architecture, such as (IEEE, 2005; Kruchten, 2004; Tang et al., 2007), do not consider business issues,
such as decisions related to business processes. Furthermore, while the EA language ArchiMate 2.0 (The
Open Group, 2012) has a motivational layer, it lacks concepts important for rationalization such as
considered alternatives, decision criteria, etc. As such, ArchiMate 2.0 is not a suitable language for
architectural rationalization. We discuss this in further detail in Section 2.
This paper is structured as follows. Section 2 presents the related work in design rationale,
decision strategies and challenges, Section 3 introduces our central artifact, a metamodel for capturing
decision making. Section 4 illustrates the usefulness of EA Anamnesis with an insurance industry
example. Finally Section 5 concludes.
In this section, we position our work against the existing body of knowledge. This section reports
on the stream of existing work in (1) design rationale as well as (2) decision making strategies, and (3)
factors that inuence decision making.
2.1 Design rationale
Concerning argumentation-based approaches, the Decision Representation Language (DRL)
(Lee, 1991) and Issue Based Information System (IBIS) (Kunz and Rittel, 1970) are two well-known
approaches for capturing design rationale. Both DRL and IBIS are inspired by Toulmin’s analysis of
argumentation (Toulmin, 2003) and argumentation maps. For DRL and IBIS, key rationalization concepts
are the issue, the arguments and the resolution of design argumentation. Here, for example, resolutions
are similar to Toulmin’s conclusion of an argument.
However, as pointed out by Shipman III and McCall (Shipman and McCall, 1997),
argumentation-based approaches are in practice not suitable for capturing design rationales. Mainly this is
because argumentation-based approaches require extensive documentation (Shipman and McCall, 1997).
Furthermore, argumentation-based approaches lack formality, and thus are not amenable to the computer-
based support that we aim for with EA Anamnesis. Also argumentation-based approaches do not make an
explicit relation to the design artifact under consideration, while for us it is important to focus our
rationalization on particular EA artifacts (such as elements of architectural languages).
Moreover, goal-oriented requirements engineering approaches (Elahi and Yu, 2012; Horko
and Yu, 2012; Liaskos, 2011) propose mechanisms for decision analysis and prioritization of
requirements. However, requirements engineering approaches deal with the problem space of an
architecture. Despite the fact that some concepts from goal-oriented modeling can be used to describe
design rationales, goal-oriented concepts are more generic. For one, a goal can denote a high level,
strategic, goal (for example, “make more prot”) pertaining to the problem space. However, a goal can
also denote very specic, attribute-level, criteria pertaining to the solution-space, such as the criterion
“have good usability” for a software application.
Dierently, in the words of (Savolainen, 1999), design rationale approaches such as EA
Anamnesis focus on the “solution space”. The solution space comes after the translation of high-level
goals into more specic ones (which (Savolainen, 1999) refers to as the requirements space), and before
the specic design (the design space). Since we position our work in the solution space, we opt to borrow
concepts from existing rationalization approaches. This is because they provide a set of concepts that is
specically tailored to our purposes (such as criteria, alternatives, and artifact), as opposed to the more
generic problem-oriented concepts from goal-oriented approaches.
Furthermore, since its second version the Archimate EA modeling language has a motivation
extension and an implementation and migration extension. The motivation extension is used to model the
reasons behind architectural changes, but lacks concepts common to existing rationalization approaches.
For example, the motivational extension does not capture design alternatives, the used decision making
strategy, or unanticipated consequences of decisions. Furthermore the Implementation and Migration
Extension deals with the project management and the planning of enterprise architecture changes and as
such, is not well suited for architectural rationalization.
Finally, there exist design rationale approaches for Software Architecture (Jansen and Bosch,
2005; Kruchten, 2004; Savolainen, 1999; Tang et al., 2007; Tyree and Akerman, 2005). These approaches
are template based or model based. Akin to argumentation-based rationalization approaches, template
based approaches (Savolainen, 1999; Tyree and Akerman, 2005) describe in textual format elements of
Architectural Decisions such as Rationale, Issue, Implications and etc. Dierently, model based ap-
proaches (Jansen and Bosch, 2005; Kruchten, 2004; Tang et al., 2007) provide a formal metamodel of
decision rationalization concepts, thus enabling computer-processable rationalization.
However rationalization approaches from software architecture focus on software issues (such as
code documentation). Yet these issues are dierent from those in Enterprise Architecture (Coggins and
Speigel, 2007). For one, as part of their responsibilities enterprise architects concern themselves with the
business-to-IT-stack (Aier and Winter, 2009). Here, for example, they analyze impacts of changes in IT
infrastructure to business processes and vice versa. Thus enterprise architects deal with dierent, cross-
organizational, issues compared to software architects, who deal mostly with software concerns.
2.2 Decision making strategies
Decision making strategies generally fall in two main categories: compensatory and
noncompensatory (Einhorn, 1970; Payne, 1976; Rothrock and Yin, 2008; Svenson, 1979). We briey
explain these strategies with a car buying example. In this example, a customer selects a car based upon
the criteria color of car’, carbon emission’, size of car’ and gasoline consumption’.
In compensatory decision making (Einhorn, 1970), alternatives are evaluated exhaustively, taking
all criteria and their trade-os into consideration. Criteria with high values compensate for criteria with
lower values. Finally, the alternative with the highest score is selected. For our car buying example, this
implies that a customer considers all four criteria color of car, carbon emission’, “size of carand
gasoline consumption. For example: s/he can state that color of caris of high importance, and
carbon emission, size of car and gasoline consumptionare of less. By doing so the customer then
selects a car that best complies with all these criteria.
Compensatory strategies aim to provide the best possible decision outcomes given the decision
data at hand. However, compensatory strategies require full information on how alternatives score on all
criteria, and they are time consuming (Einhorn, 1970).
Noncompensatory strategies (Einhorn, 1970), on the other hand, are consistent with the concept
of bounded rationality. This means that the rationale to make a decision is limited by factors such as hard
constraints, time constraints and the cognitive load of the decision maker. As such, noncompensatory
strategies evaluate alternatives heuristically, using only a limited number of criteria and trade-os.
Considering a noncompensatory decision strategy for our car buying example, let us now assume
that the customer lives in the city and selects between two cars: a small and a big one. Now, size of car
is a hard constraint for the customer due to the limited parking space available in the city. Therefore,
regardless of the criteria carbon emission, color of car, and gasoline consumptions/he excludes the
big car from her/his choice set.
The main characteristics of noncompensatory strategies are twofold: (1) they reduce the decision
making eort, (2) they are not demanding regarding the information needed to make the decision. As
such, it is a common practice for decision makers to use noncompensatory strategies in situations whose
limitations affect the decision making process (Payne, Bettman and Johnson, 1993). Example situations
include time stress or hard constraints. However, by denition decision makers do not take all criteria into
account when using noncompensatory decision strategies.
Last but not least, in some cases the use of a combination of compensatory and noncompensatory
decision strategies (a hybrid) is required (Elrod, Johnson and White, 2004; Jereys, 2004; Rothrock and
Yin, 2008). For example, a decision maker starts his evaluation process by excluding alternatives that do
not meet certain noncompensatory criteria, and only thereafter evaluates the remaining alternatives with a
compensatory strategy.
2.3 Decision making strategies
Decision making, in particular the choice for a decision making strategy, is inuenced by factors
such as time, information completeness, cognitive load of the decision maker, and more (Alenljung and
Persson, 2008).
We describe briey those factors and how they inuence the decision making process.
Time stress: One of the most common situations in decision making processes is time stress
(Orasanu and Connolly, 1993). Decision makers under time pressure must take critical decisions. Usually
these decisions are made last moment, in an adhoc manner, and without sucient rationalization.
Ill-structured problems: A decision problem is often complex in terms of cause-eect relations,
correlations and feedback loop between relevant factors (Orasanu and Connolly, 1993). As such, decision
problems are dicult to understand, also in terms of the impacts and outcomes that they have.
Information incompleteness: meaning that in practice information can be ambiguous or even
missing (Orasanu and Connolly, 1993).
Shifting, ill-dened, or competing goals: A decision maker can have conicting goals during
the decision making process. Decision maker should weight appropriately each of these goals in making a
decision (Ruhe, 2003).
Action and feedback loops: Decision making contains a series of loops that the decision maker
should deal with (Orasanu and Connolly, 1993). Early mistakes and poor information generate decisions
that should be reexamined. For example violations in architectural design are not disclosed early enough
and this implies that the decision making process should be repeated.
High stakes: Decision makers have also to cope with high risk decisions, especially when the
problem they are called to solve is of high importance (Orasanu and Connolly, 1993).
Multiple player situations: When multiple stakeholders are involved in a decision making
process the situation gets more complicated (Orasanu and Connolly, 1993). Stakeholders have dierent
interests, goals and expectations from a specic decision. Another common issue is the lack of shared
understanding between stakeholders for a particular problem. Multiplayer situations may result in delays
in the decision making possibly leading to a revision of the decision, with high cost impact (Regnell,
Organizational goals and norms: Organizations operate under specic goals and norms
(Orasanu and Connolly, 1993). Decision makers should make decisions in the context of these goals and
norms and should avoid making decision based only on their personal preferences. Decision makers,
regardless of their personal preferences, must evaluate alternatives with criteria that the organization sets.
Note that, for now, our approach does not provide rationalization support for decision strategies
with multiple decision makers. Currently this paper focuses on single decision maker environments.
In this section we discuss the basic motivations behind architectural rationalization with EA
Anamnesis. Thereafter we introduce an insurance case (in Section 3.2) which is used as a running case
throughout out this paper. Subsequently, we use the insurance case to illustrate our integrated EA
Anamnesis metamodel (section 3.3). Note that the insurance case study is ctitious yet realistic. This is
because it is based on the running case study used to illustrate the ArchiMate specication (The Open
Group, 2012), which in turn is based on a real insurance company.
3.1 Organizational learning with EA Anamnesis
As stated in the introduction, this paper integrates two decision making meta models: one for ex-
ante decision rationalization, and one for ex-post decision rationalization. Here, ex-ante and ex-post refer
to dierent, but complementary, perspectives on decision making.
On the one hand, ex-ante decision making concepts rationalize decision making during or just
after the decision making process. In so doing, ex-ante rationalization focuses on criteria deemed
important during decision making, and captures anticipated consequences of decisions. On the other
hand, rationalization of ex-post decision making provides a reection of unanticipated consequences of a
decision some time after the decision has been made.
We consider ex-ante and ex-post rationalization as complementary, and posit that a comparison of
the two allows for organizational learning (Conklin, 1996). For one, we foresee that an unanticipated
consequence of a decision (ex-post), when compared to decision criteria and anticipated decision
consequences (ex-ante), results in having to re-prioritize criteria and in having to revisit anticipated con-
sequences. Thus, we argue that the comparison of ex-ante to ex-post informs future decision making. This
reects the purpose of EA Anamnesis: to have an organizational memory that helps organizations to learn
from decision making.
3.2 Illustrative case
ArchiSurance is an insurance company that sells car insurance products using a direct-to-
customer sales model. It does so to reduce its operations and product costs.
Figure 1 presents the partial (business and application layers) ArchiSurance direct-to-customer
sales model, modeled with the EA modeling language Archi-Mate. ArchiMate is an Open Group2
standard EA modeling language (Lankhorst, 2005; The Open Group, 2012). Two business services
support the sales model of ArchiSurance: “Car insurance registration service” and “Car insurance
service”. Here, in accordance with the ArchiMate specication (The Open Group, 2012), a business
service refers to the functionality oered to the environment by a business process or collaboration. For
example, in this way ArchiMate separates between the functionality “Car insurance registration service,
and how this functionality is actually realized.
ArchiMate helps us understand the dependencies between dierent perspectives of an enterprise.
For example, in Figure 1 we see that the business service “Car insurance registration service” is realized
by a business process “Register customer prole”. In turn, we also see that this business process is
supported by the application service “Customer administration service”. Note here that an application
service denotes the functionality oered by an application. Thus, akin to an ArchiMate business service,
an application service denotes functionality independently of how this functionality is realized.
Although disintermediation reduces operational costs, it also increases the risk of incomplete or
faulty risk proles of customers (Cummins and Doherty, 2006). This so called “adverse selection of
proles” (cf. Cummins and Doherty, 2006) leads insurance companies to calculate unsuitable premiums
or, even worse, to wrongfully issue insurances to customers.
To reduce adverse selection of proles ArchiSurance decides to use intermediaries to sell its
insurance products. After all, compiling accurate risk proles is part of the core business of an
intermediary (Cummins and Doherty, 2006).
In our scenario John, and external architect, is hired by ArchiSurance to help guide the change to
an intermediary sales model.
2 See
John uses ArchiMate to capture the impacts that selling insurance via an intermediary has in
terms of business processes, IT infrastructure and more. For illustration purposes we will focus on the
translation of the new business process “Customer prole registration” to EA artifacts in the application
layer. The resulting ArchiMate model is depicted in Figure 2.
Figure 1. ArchiSurance direct-to-customer EA model
Here we see for example how a (new) business process “customer prole registration”, owned by
the insurance broker (cations “customer administration service intermediary” and “customer
administration service ArchiSurance”. The new business process “customer prole registration” is a
collaboration between the insurance company and the intermediary (symbolized by the double
Figure 2. ArchiSurance direct-to-customer EA model
3.3 The EA Anamnesis metamodel
Figure 3 presents the EA Anamnesis metamodel. This metamodel is an integration of two
metamodels presented in earlier work: the EA decision making metamodel (Plataniotis et al., 2013a), and
the decision relationships metamodel (Plataniotis et al., 2013b). With this integration, we allow for (1)
contextualizing the decision making process of a single decision in terms of cross cutting/intertwining
decision relationships, and (2) a comparison of decision outcomes to the original decision making
Figure 3. ArchiSurance direct-to-customer EA model
For comprehension purposes the concepts of our metamodel will be introduced in 3 sections:
decision properties (Section 3.3.1), decision making process concepts (Section 3.3.2) and decision
relationships (Section 3.3.3). Furthermore, we use the ArchiSurance case for illustration purposes.
3.3.1 Decision properties
EA decision: An EA decision represents the actual design decision that was made after an
evaluation process. (Proper and Op’t Land, 2010; Orasanu and Connolly, 1993).
Example: John makes the EA decision make customer prole registration via intermediary.
EA issue: Similar to the concept of an issue from (Tyree and Akerman, 2005), an EA issue
represents the architectural design problem that enterprise architects have to address.
Example: The EA issue create an appropriate application service to support new business
process resulting from the EA decision “introduce a new business process for customer prole
EA artifact: An EA artifact (similar to concept of an architecture element (Tang et al., 2007)) is
either the direct result produced from a set of executed EA decisions, or a representation of this result. For
now, we use an EA artifact to refer to architectural representations. Specically, we use it as a bridging
concept towards the EA modeling language ArchiMate, whereby an EA artifact allows us to link EA
decisions to concepts from ArchiMate.
Example: The EA artifact Customer prole registrationin the ArchiMate model in Figure 2 is
linked to, amongst others, the EA Decision Make customer prole registration via intermediary.
Layer: In line with the ArchiMate language (The Open Group, 2012), an enterprise is specied in
three layers: Business, Application and Technology. Using these three layers, we express an enterprise
holistically, showing not only applications and physical IT infrastructure (expressed through the
application and technology layers), but also how an enterprise’s IT impacts/is impacted by an enterprise’s
products and services and its business strategy and processes.
Example: The EA decision make customer prole registration via intermediaryis a part of the
business layer of ArchiSurance.
Unanticipated observed impact: This concept signies an unanticipated consequence of an al-
ready made decision to an EA artifact. This opposes to anticipated consequences, as indicated by
relationships such as translation or decomposition.
In current everyday practice, architects model anticipated consequences using what-if-scenarios
(Lankhorst, 2009). Unfortunately, not every possible impact of made EA decisions can be predicted. This
is especially true for enterprise architecture, where one considers impacts across the enterprise rather than
in one specic (e.g. technical) part. The outcome of EA decisions can be observed during an ex-post
analysis of the architecture (Baron and Hershey, 1988; Proper and Op’t Land, 2010). Some of the
consequences of EA decisions are revealed during the implementation phase, or during the maintenance
of the existing architecture design. These unanticipated consequences are captured exactly by the concept
of an unanticipated observed impact.
For us the main usefulness of capturing unanticipated observed impacts is that they can be used
by architects to avoid decisions with negative consequences in future designs of the architecture.
Example: The EA decision Acquisition of COTS application B has an unanticipated observed
impact Degraded user experience in the application use. This observed impact captures an
unanticipated, ex-post, side eect of acquiring COTS application B, due to unfamiliarity of users with the
new user interface that COTS application B introduces.
3.3.2 Decision making process concepts
The decision making process concepts of our metamodel focus on capturing (1) decision making
strategies that were used during the architectural design process for a specic EA decision, (2) the
rationale behind this specic decision strategy choice, and (3) available alternatives and criteria. Below
we provide the description of these concepts.
Decision-Making Strategy: This concept captures the decision making strategy used by the
enterprise architect to (1) evaluate the alternatives, and make the actual EA decision. As we mentioned in
Section 2.2, decision strategies are characterized as compensatory, noncompensatory, or as a hybrid of
these two.
In our metamodel, we specify this as follows:
Compensatory strategy
Weighted additive (WADD): In WADD strategies the criteria which evaluate the alternatives
have dierent weights. The score of each alternative is computed by multiplying each
criterion by its weight and then by taking the sum of these values. The alternative with the
highest score is chosen by the decision maker (Rothrock and Yin, 2008).
Equal weight: The score of each alternative is calculated by the same way as WADD
strategies. The dierence is that criteria have the same weight (Rothrock and Yin, 2008).
Noncompensatory strategy
Conjunctive: In conjunctive strategies, alternatives that fail to comply with one or more
criteria, are immediately excluded from the decision maker’s choice set (Rothrock and Yin,
Disjunctive: In this strategy an alternative is selected if it complies with a criterion,
irrespective on its values on other criteria (Rothrock and Yin, 2008).
Noncompensatory strategies are appropriate for evaluating alternatives in situations of
information incompleteness and absence of numerical data. For example, the “car color” criterion can not
be quantied. Alternatives which do not comply with this criterion can be eliminated from the choice set
because they do not comply with this feature.
We should also mention that there is no restriction in the use of additional decision strategies. We
include a set of common decision strategies, but we also denote in the metamodel that more decision
strategies can be supported.
“Traces to” relationship: In line with Section 2.2, a hybrid decision strategy is supported by our
metamodel. The relationship traces to signies the combination of two or more decision strategies
during the decision making process.
Example: John rejects acquisition of COTS app C because it exceeds the budget set on
beforehand by top management. Thus, here John employed a conjunctive non compensatory decision
making strategy.
Criterion: A criterion is an attribute that, to a larger or lesser extent, should be satised during
the decision making process (Elrod et al., 2004). Criteria are used for both compensatory or
noncompensatory decision making strategies. For example, if a disjunctive strategy was used, a decision
is taken on compliance with one or more criteria. Furthermore, the concepts value and weight of a
criterion are included in our viewpoint. The value concept represents the value that the decision maker
assigns to this criterion during the evaluation process, which indicates how well an artifact performs on a
particular criterion. The weight concept represents the importance of this criterion, and is typically used in
WADD strategies.
Example: After discarding acquisition of COTS app C, John considers 3 quality criteria his
evaluation, usability, interoperability and scalability. Interoperabilitywas considered as the most
important, with a weight of 10. Furthermore, “COTS application C” has the value “7” (out of 10) on the
criterion interoperability, whereas “COTS application B” has the value “3” out of 10. This dierence in
value reects that “COTS application C” performs better than “COTS application B”on the criterion
Strategy rationale: In a decision making process, the architect not only has to choose amongst
some alternatives (actual decision making process), but has also to select the decision strategy that
satises his current evaluation needs. Actually, this concept represents the rationale for the decision
strategy that was selected for the evaluation process. This is what is referred as metadecision making,
decision making about the decision process itself (Mintzberg, 1976).
As we discussed in Section 2.3, dierent factors aect the decision making process and decision
makers should adjust their decision making strategy accordingly. The concept of a strategy rationale
enables a decision maker to justify the reasons for his metadecision. We argue that time stress can be
another important factor to capture a strategy rationale. By capturing that a decision was made under time
stress, we reason that decision makers can protect themselves. For example a decision maker can justify
that a “bad” decision was made due to time constraints on the decision making process
Example: The strategy rationale “time stress” to select a non compensatory (heuristic) decision
strategy. “Time stress” is a strategy rationale, or metadecision, since it concerns decision making about
the decision making process itself, independently of specic decision criteria such as “usability”.
3.3.3 EA Decision relationships
The role of relationship concepts is to make the dierent types of relationships between EA
decisions explicit. Based on ontologies for software architecture design decisions (Kruchten, 2006; Tyree
and Akerman, 2005), we dene four types of relationships:
Translation relationship: Translation relationships illustrate relationships between decisions/EA
issues that belong to dierent layers or dierent EA artifacts. In accordance with the metamodel (Figure
3) a translation relationship relates decisions with decisions, and decisions with issues. The translation
relationship bases on the generic “is related to” relationship from (Kruchten, 2004), only we adapt it to
the basic terminology of Architects. One of the key tasks for architect is, by communication, to translate
the requirements that new EA artifacts impose (EA issue) to decisions that will support these
requirements by means of another EA artifact (Op’t Land and Proper, 2007; Strano and Rehmani, 2007).
Example: The EA decision make customer prole registration via intermediarytranslates to the
issue nd an appropriate application service. Subsequently this issue translates to a second EA decision
acquisition of COTS application B.
Decomposition relationship: The Decomposition relationship is in line with Comprises (Is
Made of, Decomposes into)of Kruchten’s ontology (Kruchten, Lago and Vliet, 2006). Decomposition
relationships signify how generic EA decisions decompose into more detailed design decisions.
Example: The EA decision acquisition of COTS application Bhas a decomposition relationship
with EA decision Application interface type 1. This is to indicate that choosing application B also
implies the more detailed choice for a particular type of user interface.
Alternative relationship: The alternative relationship type (Kruchten et al., 2006), illustrates the
EA decisions that were rejected (alternatives) in order to address a specic EA issue.
Example: Rejected EA decisions COTS application A, COTS application Cand Upgrade
existing application (inhouse) have an alternative relationship with EA issue nd an appropriate
application to interface with the intermediary. This signies that these decisions were the alternatives for
this issue.
Substitution relationship: A substitution relationship explicates how one EA decision repairs
the negative outcome of another EA Decision.
Example: The EA decision Acquisition of COTS application Bhas a negative unanticipated
observed impact on the business process Customer prole registration. This is because it leads users to
make mistakes, as we observed the concept unanticipated observed impact. As such, it is repaired by
the EA decision Application interface 2”.
We now illustrate how EA Anamnesis approach can support enterprise architects to improve their
decision making capabilities based on decision making processes and their outcomes from past cases.
Section 4.1 shows how we use the EA Anamnesis metamodel to capture EA decisions, exemplied by our
running ArchiSurance case. Subsequently Section 4.2 introduces a procedural model for using the EA
anamnesis metamodel for organizational learning, using the archisurance case for illustation purposes.
4.1 Capturing a decision making process!
For the purposes of this paper we assume that John, the enterprise architect for ArchiSurance, is a
single decision maker that is capable of identifying specic alternatives and criteria. Furthermore, we
assume that he has full information to evaluate them.
Recall (from section 3.2) that John supports ArchiSurance in changing to selling car insurance via
intermediaries. To start the decision making process based on the requirements of selling via
intermediaries, John denes the criteria that the new application should satisfy (the criteria for application
selection are grounded in (Jadhav and Sonar 2009).
For our illustrative example, John considers that the most important criteria are usability,
interoperability and scalability. Based on these criteria he identies 4 alternatives to choose from, 3
alternative Commercial O-The-Shelf (COTS) applications and 1 alternative to upgrade the existing
application in house.
Let us also assume that John receives a constraining budget limitation of 10000 for the
acquisition of new IT systems.
John is now faced with a hybrid decision strategy: on the one hand, he wants to carefully evaluate
the four alternatives on the criteria usability, interoperability and scalability (via a compensatory
strategy), but on the other hand he has to account for the hard constraint budget limitation(via a non-
compensatory strategy).
At this point John uses the EA Anamnesis approach to capture and justify his strategy selection,
as well as the alternatives and criteria of his decision problem. For the noncompensatory part, John wants
to discard all alternatives that fail to meet the cost criterion. Because of this hard constraint, he chooses a
conjunctive noncompensatory strategy (for an explanation, see Section 3.3.2) to exclude from his choice
set alternatives that exceed the maximum value of this criterion.
Table 1 summarizes the score of each alternative. COTS application C is eliminated from the
choice set because it fails to meet the maximum cost requirement.
Table 1. EA decision 13 noncompensatory conjunctive strategy
Upgrade app
As discussed in section 3.3.2, conjunctive noncompensatory strategies evaluate alternatives using
a threshold level on one or more criteria. In this example the conjunctive criterion is cost. The
alternatives COTS A”, COTS B and Upgrade applicationcomply with this criterion (Table 1) and
will be evaluated further in the next step of the decision making process. COTS C cost exceeds the
maximum limit and is eliminated from the choice set. For noncompensatory strategies, alternative scores
are boolean data types, they either comply or not with some criteria. The scores of the alternatives are also
captured by our metamodel.
For the compensatory part, John evaluates the three remaining alternatives based on the value and
the weight of each criterion. Scalabilityis the most important factor because, according to John, the ap-
plication should be able to support changes in the business processes of ArchiSurance, for example, to
support the addition of extra intermediaries.
Given the fact that the criteria that evaluate alternatives have dierent weights, John selects the
use of a weighted additive compensatory strategy. Here, John captures again his decision strategy as well
as the weights and the values of the compensatory criteria. The score of each alternative is calculated by
multiplying the value of each criterion by its weight, and then by summing up these values. Here, the
weights range from 1 not important to 10 important.
Table 2 shows us: (1) the criteria. Scalability, the most important criterion for John has a
weight of 10, while usabilityand interoperabilityhave weights 2 and 5 respectively, (2) the score on
a particular criterion for each alternative. For example: the alternative COTS B scores 9 on
scalability, whereas Upgrade app scores 4. (3) the total score of each alternative. For example:
COTS Breceives the highest score and as such, is selected by John.
Table 2. EA decision 13 compensatory weighted additive strategy
Upgrade app
Figure 4 depicts the captured decision making processes based on our approach
Figure 4. Decision making strategies for EA Decision 10
4.2 Using EA Anamnesis for organizational learning
In this section, we introduce a procedural model (Figure 5) for using EA anamnesis for
organizational learning. We illustrate its steps by means of the archisurance case and we show how the
integrated metamodel of ex-ante and ex-post perspectives assists architects in learning from past decision
making. The steps that the enterprise architect follows in order to get support from our approach are
derived from this procedural model. To visualize the metamodel for ArchiSurance we rely on Decision
Design Graphs (DDGs), a concrete syntax for the EA Anamnesis metamodel that we introduced in [8].
Finally, a textualized table (Table 3) provides summarized information for EA decisions derived from our
metamodel. Here, Table 3 as a “tooltip” for the DDGs.
Figure 5. Organizational learning procedural model
Table 3. EA decision 10 summary
Acquisition of COTS application B
EA issue:
EA Issue 05: Create an appropriate application service
to support new business process
Decision Maker:
D05: introduce application service A D13: introduce
user interface 3
COTS application A COTS application C Upgrade
existing application
usability, interoperability, scalability, cost
Observed Impact:
Observed Impact 01: Reduced performance of
customer registration service business process
Note here that Figure 6 depicts the rationalization for the nal intermediary design depicted in
Figure 2.
Case: We now move two years forward to further illustrate how EA anamnesis supports future
decision making. The customer proles of ArchiSurance’s car insurance are better calculated by using
insurance intermediaries. As a result, ArchiSurance’s management decides to also rely on intermediaries
for the remainder of its insurance products. Bob is made responsible for translating the use of
intermediaries into an appropriate enterprise architecture design. He uses ArchiMate to design the “to be”
Figure 6. EA decisions relationships visualization
Step 1: Identication of relevant EA issues/decisions. In this step, for those issues that the
architect wants to address, we identify similar issues in the rationalization information stored in EA
Case: For the sake of example, we assume that Bob wants to know how his predecessor has
supported the introduction of an intermediary with software applications. To this end, he can rely on
rationale information regarding a similar issue captured by his predecessor. For example: In the EA
ananmensis for ArchiSurance, Bob identies the “Create an appropriate application service to support
new business process” (issue 03), which addresses “Make customer prole registration via intermediary”
(decision 01).
Step 2: Ex-post analysis of decisions. In this step we analyze decision details, without
considering the unanticipated observed impacts.
Consider the issue “Create an appropriate application service to support new business
process” (issue 03), as depicted in Figure 6 and for which we provide a “tooltip” in Table 3. For the issue
“Create an appropriate application service to support new business process”, Bob can amongst others see
that the decision “Acquisition of COTS application B” was made, pertaining to the EA artifact “Customer
administration application”. Furthemore, Bob observes the alternatives “Acquistion of COTS application
A”, “Aquisition of COTS application C”, and “Upgrade existing application”, and used criteria, such as
“usability” (in line with the information capturted in Section 4.1).
Step 3: Observed unanticipated impacts. In this step, we review if there are unanticipated
observed impacts for the identied issues.
Step 4: Observed unanticipated impacts inspection. In this step we compare the unanticipated
observed impact to decision details.
Case: Bob inspects “Customer administration application” from the decision “Acquisition of
COTS application B” for Observed unanticipated impacts. He nds out that users had diculties in using
the new application, as signied by the unanticipated observed impact 01 “Degraded user experience in
the application use” (OI1 in Figure 6). As such, Bob sees that EA Decision “Acquisition of COTS
application B” (D10) had a negative observed impact on the business process “Customer prole
registration”. At this point, Bob can trace back to “Acquisition of COTS application B” (D10, as
summarized in Table 3). From the inspection of the decision making analysis in Table 2, Bob realizes that
the criterion usability, which is related with the observed unanticipated impact of user experience, was not
considered as a criterion of high importance.
Step 5: Ex-post analysis of substitution decisions. In this step we analyze how the
unanticipated observed impact was addressed.
Case: Using the DDG (Figure 6), Bob observes how John addressed the observed impact
“Degraded user experience in the application use” (OI1). First Bob nds that OI1 leads to the new EA
Issue “Find proper user interface” (IS07). Furthermore, from the DDG Bob notices that there are three
possible alternatives for “Find proper user interface” (1) “Training of users on the new application” (EA
decision 11), (2) “New User Interface 1” (EA decision 12) and, (3) “replace of existing application
interface with an interface similar to the old one” (EA decision 13). Finally, Bob notices that his
predecessor John decided for the third option (the executed EA decision 13) and rejected the other two
(i.e., that John rejected EA decisions 11, 12).
In summary, from the analysis of past EA Decisions and their implications for the enterprise
architecture, Bob can consider for his current decision making problem decision making strategies,
alternatives and criteria as well as their relative importance from past decision making processes. Based
on the outcomes of decisions, he can avoid decisions or problems that may come along. For example, if
Bob has to address the issue of an appropriate application for the new intermediary business process, he
will be aware of the possible implication “Degraded user experience in the application use”. To address
this issue he can apply the same approach as his predecessor or he can inform the interested stakeholders
regarding this concern.
In this paper we introduced a metamodel and corresponding visualization for capturing
Enterprise Architecture (EA) decision making strategies decision making strategies and a
procedural model for using the metamodel. Together, the integrated metamodel and produceral
model allow for the comparison of a decision making process with observed outcomes of a
decision. This comparison of (ex-ante) decision making with (ex-post) observed impact leads to
a better understanding of existing architectures. As such, it provides a rst step towards learning
from architectural decision making.
For future research, rst and foremost we intend to confront the illustrative examples of
our approach to enterprise architecture practitioners. From our survey amongst enterprise
architecture practitioners ( Plataniotis, de Kinderen, van der Linden, Greefhorst and Proper, 2013), we
already have indications about the perceived practical usefulness of architectural rationalization -
particularly in terms of justication and maintenance. However we require case studies to further
support those results.
In addition, we will look into how our approach deals with legacy artifacts, meaning
artifacts available prior to the introduction of a rationalization approach (such as EA anamnesis).
On the one hand, this is relevant as such legacy artifacts also imply constraints on the future
design. On the other hand we do foresee a challenge in reverse engineering rationalization for
such artifacts, as – even more so than with design rationalization of current artifacts – the
relevant design knowledge may be incomplete or even completely missing.
Furthermore, we aim to provide procedural decisional guidance for using our approach
inspired by Decision Support Systems (DSS) literature. While we feel that this paper taken a
good rst step towards capturing, representing and using a rationale, we deem further guidance
especially necessary to scale up our approach for use in real-life domains, whereby many
decisions are taken. Here, a good starting point is Silver (Silver, 1991), who introduces a
typology of dierent types of DSS and their respective characteristics. For example: a DSS that
provides support during (ex-ante) decision making has dierent characteristics than a DSS that
(ex-post) reects upon already captured information. This typology forms useful input for further
structuring procedural guidance.
Last but not least, one of our major challenges is to investigate the return of capturing
eort for our approach. Our design rationale assists architects to better understand existing EA
designs, but the eort of capturing this information might be a dissuasive factor. To address this
issue our research will focus on ways to decrease the capturing eort. One way of doing this is
by evaluating the actual practical usefulness of the concepts of the decision making strategy
viewpoint. For example we capture the strategy rationale for selecting a decision making
strategy, but whether the eort for capturing this outweighs the received benets remains to be
seen. Furthermore, by further formalization, we foresee that in a future iteration we are better
able to automatically compare factors relevant during decision making with those emerging some
time after the decision has been made. For example: to express a delta in a criterion as an explicit
result of an unanticipated consequence. This is opposed to the current manual procedure
expressed in Figure 5.
Aier, S., Winter, R. (2009). Virtual decoupling for it/business alignment -conceptual foundations,
architecture design and implementation example. Business & Information Systems Engineering
1, 150–163
Alenljung, B., Persson, A. (2008). Portraying the practice of decision-making in requirements
engineering: a case of large scale bespoke development. Requirements engineering 13, 257–279.
Baron, J., Hershey, J.C. (1988). Outcome bias in decision evaluation. Journal of personality and
social psychology 54 (569).
Burge, J., Brown, D.C. (2000). Reasoning with design rationale. In: Articial Intelligence in
Design, (pp. 611–629). Springer.
Coggins, C., Speigel, J. (2007). The methodology for business transformation v1.5: A practical
approach to segment architecture. Journal of Enterprise Architecture.
Conklin, J. (1996). Designing organizational memory: preserving intellectual assets in a
knowledge economy. Group Decision Support Systems 1, 362.
Cummins, J., Doherty, N. (2006). The economics of insurance intermediaries. Journal of Risk
and Insurance 73, 359–396.
Einhorn, H. (1970).The use of nonlinear, noncompensatory models in decision making.
Psychological bulletin 73, 221–230
Elahi, G., Yu, E. (2012). Comparing alternatives for analyzing requirements trade-offs–in the
absence of numerical data. Information and Software Technology 54, 517–530.
Elrod, T., Johnson, R., White, J. (2004). A new integrated model of noncompensatory and
compensatory decision strategies. Organizational Behavior and Human Decision Processes 95,
Horkoff, J., Yu, E. (2012). Comparison and evaluation of goal-oriented satisfaction analysis
techniques. Requirements Engineering, 1–24.
Jadhav, A., Sonar, R. (2009). Evaluating and selecting software packages: A review. Information
and software technology 51, 555–563.
Jansen, A., Bosch, J. (2005). Software architecture as a set of architectural design decisions. In
Software Architecture, 2005. WICSA 2005. 5th Working IEEE/IFIP Conference on, IEEE (pp.
Jeffreys, I. (2004). The use of compensatory and non-compensatory multi-criteria analysis for
small-scale forestry. Small-scale Forestry 3, 99–117.
Kruchten, P. (2004). An ontology of architectural design decisions in software intensive systems.
In 2nd Groningen Workshop on Software Variability (pp. 54–61).
Kruchten, P., Lago, P., Vliet, H. (2006). Building up and reasoning about architectural
knowledge. In Hofmeister, C., Crnkovic, I., Reussner, R., eds.: Quality of Software
Architectures. Volume 4214 of Lecture Notes in Computer Science. (pp. 43–58). Springer Berlin
Kunz, W., Rittel, H.W. (1970). Issues as elements of information systems. Vol. 131. Institute of
Urban and Regional Development, University of California Berkeley, California.
Lankhorst, M. (2009). Enterprise architecture at work: Modelling, communication and analysis.
Lee, J. (1991). Extending the potts and bruns model for recording design rationale. In Software
Engineering Proceedings, 13th International Conference on, IEEE (pp. 114–125).
Liaskos, S., McIlraith, S.A., Sohrabi, S., Mylopoulos, J. (2011). Representing and reasoning
about preferences in requirements engineering. Requirements Engineering 16, 227–249.
Mintzberg, H., Raisinghani, D., Theoret, A. (1976). The structure of unstructured decision
processes. Administrative science quarterly, 246–275.
Op’t Land, M., Proper, H.A. (2007). Impact of principles on enterprise engineering. ECIS 2007
Op’t Land, M., Proper, E., Waage, M., Cloo, J., Steghuis, C. (2008). Enterprise architecture:
creating value by informed governance. Springer.
Orasanu, J., Connolly, T. (1993). The reinvention of decision making. Decision making in
action: Models and methods, 3–20.
Payne, J. (1976). Task complexity and contingent processing in decision making: An information
search and protocol analysis. Organizational behavior and human performance 16, 366–387.
Payne, J., Bettman, J., Johnson, E. (1993). The adaptive decision maker. Cambridge University
Plataniotis, G., de Kinderen, S., Proper, H.A. (2012) EA Anamnesis: Towards an Approach for
Enterprise Architecture Rationalization. In Proceedings of the 2012 workshop on Domain-
specific modeling. DSM '12, New York, NY, USA, ACM (2012) 27-32
Plataniotis, G., de Kinderen, S., Proper, H.A. (2013). Capturing decision making strategies in
enterprise architecture - a viewpoint. In Nurcan, S., Proper, H., Krogstie, J., Schmidt, R., Halpin,
T., Bider, I., eds.: Enterprise, Business-Process and Information Systems Modeling. Volume 147
of Lecture Notes in Business Information Processing (pp. 339–353). Springer Berlin Heidelberg.
Plataniotis, G., Kinderen, S.d., Proper, H.A. (2013a). Relating decisions in enterprise architecture
using decision design graphs. In: Proceedings of the 17th IEEE International Enterprise
Distributed Object Computing Conference (EDOC).
Plataniotis, G., de Kinderen, S., van der Linden, D., Greefhorst, D., Proper, H.A. (2013b). An
empirical evaluation of design decision concepts in enterprise architecture. In: Proceedings of
the 6th IFIP WG 8.1 working conference on the Practice of Enterprise Modeling (PoEM 2013).
Proper, H.A., Op’t Land, M. (2010). Lines in the water. In Harmsen, F., Proper, H.A.,
Schalkwijk, F., Barjis, J., Overbeek, S., eds. Practice-Driven Research on Enterprise
Transformation. Vol. 69 of Lecture Notes in Business Information Processing (pp. 193–216).
Springer Berlin Heidelberg.
Regnell, B., Paech, B., Aurum, A., Wohlin, C., Dutoit, A., och Dag, J. (2001). Requirements
mean decisions!–research issues for understanding and supporting decision-making in
requirements engineering. In: First Swedish Conference on Software Engineering Research and
Practise: Proceedings, Citeseer.
Rothrock, L., Yin, J. (2008). Integrating compensatory and noncompensatory decision-making
strategies in dynamic task environments. Decision Modeling and Behavior in Complex and
Uncertain Environments, 125–141.
Ruhe, G. (2003). Software engineering decision support: methodology and applications.
Innovations in decision support systems 3, 143–174.
Savolainen, J. (1999). Tools for design rationale documentation in the development of a product
family. In: Position Paper Proceedings of 1st Working IFIP Conference on Software
Architecture, San Antonio, Texas.
Shipman, F.M., McCall, R.J. (1997). Integrating different perspectives on design rationale:
Supporting the emergence of design rationale from design communication. Articial Intelligence
for Engineering Design, Analysis and Manufacturing: AI EDAM, 141.
Silver, M.S. (1991). Decisional guidance for computer-based decision support. MIS Quarterly,
Strano, C., Rehmani, Q. (2007). The role of the enterprise architect. Information Systems and e-
Business Management 5, 379–396.
Svenson, O. (1979). Process descriptions of decision making. Organizational behavior and
human performance 23, 86–112.
Tang, A., Jin, Y., Han, J. (2007). A rationale-based architecture model for design traceability and
reasoning. Journal of Systems and Software 80, 918–934.
The Open Group (2012). ArchiMate 2.0 Specication. Van Haren Publishing.
Toulmin, S.E. (2003). The uses of argument. Cambridge University Press.
Tyree, J., Akerman, A. (2005). Architecture decisions: Demystifying architecture. Software,
IEEE 22, 19–27.
... • EA Decisions -This value covers approaches related to the decision-making process itself. Exemplary, it is related to the rationale behind decisions, stakeholders' influence on the decisionmaking process, or methods to evaluate alternatives (Plataniotis et al., 2013;Plataniotis et al., 2014). ...
Conference Paper
Organizations need a holistic view of their information assets in order to lead a digital transformation. Information assets can be obtained from modeling the Enterprise Architecture (EA) of the organization. However, the current EA modeling methods have been criticized for being heavy and rigid. In fact, EA modeling methods present similar problems to those faced by traditional software development (waterfall) methods. We propose that EA modeling could benefit from ideas presented in the Agile Manifesto. Previous research already pointed out the importance of finding the ”ideal” boundaries for the intersection between agile software development methods and EA practices. In this paper, we apply the Design Science Research Methodology (DSRM) to propose an Agile Enterprise Architecture Modeling Method based on agile principles and values. The proposal was demonstrated in two organizations, from which we extracted evidence such as continuous deliveries, short interactions (Sprints), proximity to customers, and systematic improvements in the process. We conclude that our proposal can improve EA modeling.
This introduces the notion of organisational subculture to ACET. During an enterprise transformation, enterprise architects interact with many different stakeholder groups coming from different departments, having different roles, functions and mindsets. We argue that enterprise architects need to pay attention to the existing organisational subcultures to get the different stakeholders on board. In this chapter, we explain that cultural differences have an indirect impact on ACET and that communication is an important intermediary factor. We reflect on potential consequences of ignoring cultural differences in the context of ACET. Finally, we suggest developing a framework to analyse the role of cultural differences in ACET. To this end, we introduce four research questions.
In this we introduce and validate a logic-based framework that serves as the underlying model for Chap. 23 The resulting logic-based framework aims to address some of the challenges identified in Chap. 15
In this chapter, we derive a reference model that provides a holistic perspective on enterprise transformations. The model includes two perspectives: on the one hand, information objects that enterprise architecture management can provide are included; and on the other hand, information requirements that enterprise transformation managers posed during interviews are integrated with those discussed in literature. Both combined perspectives lead to a comprehensive overview of information requirements and objects that are relevant during enterprise transformations.
This chapter introduces an enterprise transformation taking place in the passport issuing and registration office of the Dutch government: adding innovation to the strategic agenda of a purely on maintenance focused organisation. This transformation is divided into three projects: introduction of a formal architecture board, introduction of an innovation department, and introduction of new project types (innovation projects). During the transformation the architecture board faces different challenges linked the institutionalisation of ACET, cultural aspects, and communication defects.
In this chapter, we discuss the key ACET concepts that underpin the other components as discussed in the ensuing chapters.
This chapter introduces a design artefact for value-based componential language engineering, as a response to the research questions defined in Chap. 13 The design artefact consists of two parts: (a) two formal ontologies, representing the stakeholder perspective/language-centric perspective, respectively. These two ontologies can be used to specify catalogues of language fragments, emphasising how these are valuable to stakeholders. (b) A procedural model, inspired by situational method engineering, for creating model bundles from the fragments in the catalogues. We illustrate the artefact with an experiment on combining the modelling languages e³ value, DEMO, and ArchiMate.
In this chapter we report on the case of a globally operating insurance company that has leveraged enterprise architecture management to support business transformations. In order to do so, the company has developed enterprise architecture management capabilities that help the business structuring the business transformation particularly in the early stages before handing over respective responsibilities to more specialized corporate functions later on. This case is interesting for understanding ACET because it is one of the rare cases where enterprise architecture management truly bridges the business–IT gap.
This chapter introduces an overview on the formulation of architecture principles, guidelines for a semi-formal definition and rules for modelling the architecture principles. We give insights on analysis and impact evaluation of aforementioned principles on the design of architecture models and on the implementation of enterprise architecture.
In this chapter, we argue for a component-based approach for the construction of (visual) conceptual models, so that these models are tailored to the context-specific characteristics of a particular enterprise transformation. We offset this component-based approach against (a) “one-size-fits-all” languages, such as the enterprise architecture modelling language ArchiMate, (b) federated languages, whereby languages are related by defining (semi-)formal model transformations, and (c) domain specific language design.
Full-text available
An enterprise architecture tries to describe and control an organisation?'s structure, processes, applications, systems and techniques in an integrated way. The unambiguous specification and description of components and their relationships in such an architecture requires a coherent architecture modelling language. Lankhorst and his co-authors present such an enterprise modelling language that captures the complexity of architectural domains and their relations and allows the construction of integrated enterprise architecture models. They provide architects with concrete instruments that improve their architectural practice. As this is not enough, they additionally present techniques and heuristics for communicating with all relevant stakeholders about these architectures. Since an architecture model is useful not only for providing insight into the current or future situation but can also be used to evaluate the transition from ?as-is? to ?to-be?, the authors also describe analysis methods for assessing both the qualitative impact of changes to an architecture and the quantitative aspects of architectures, such as performance and cost issues. The modelling language and the other techniques presented have been proven in practice in many real-life case studies. So this book is an ideal companion for enterprise IT or business architects in industry as well as for computer or management science students studying the field of enterprise architecture. © Springer-Verlag Berlin Heidelberg 2005. All rights are reserved.
Conference Paper
Full-text available
Enterprise Architecture (EA) languages describe the design of an enterprise holistically, typically linking products and services to supporting business processes and, in turn, business processes to their supporting IT systems. In earlier work, we introduced EA Anamnesis, which provides an approach and corresponding meta-model for rationalizing architectural designs. EA Anamnesis captures the motivations of design decisions in enterprise architecture, alternative designs, design criteria, observed impacts of a design decision, and more. We argued that EA Anamnesis nicely complements current architectural languages by providing the capability to learn from past decision making. In this paper, we provide a first empirical grounding for the practical usefulness of EA Anamnesis. Using a survey amongst 35 enterprise architecture practitioners, we test the perceived usefulness of EA Anamnesis concepts, and compare this to their current uptake in practice. Results indicate that while many EA Anamnesis concepts are perceived as useful, the current uptake in practice is limited to a few concepts - prominently ‘rationale’ and ‘layer’. Our results go on and show that architects currently rationalize architectural decisions in an ad hoc manner, forgoing structured templates such as provided by EA Anamnesis. Finally, we interpret the survey results discussing for example possible reasons for the gap between perceived usefulness and uptake of architectural rationalization.
Conference Paper
Full-text available
Enterprise Architecture modeling languages describe an enterprise holistically, showing its business products and services and how these are realized by IT infrastructure and applications. However, these modeling languages lack the capability to capture the design rationale for decisions that lead to specific architectural designs. In our previous work we presented the EA Anamnesis approach for capturing architectural decision details. In this paper, we extend the EA Anamnesis approach with a viewpoint that captures and rationalizes decision making strategies in enterprise architecture. Such a viewpoint is useful because it helps enterprise architects reconstruct the decision making process leading up to a decision and understand how and under which circumstances this decision was made. For example, under time pressure an architect may rely on heuristics instead of examining the decision problem in depth. More specifically, we contribute: (1) a metamodel for capturing decision making strategies, which is grounded in established decision making literature, (2) an illustrative example showcasing the potential usefulness of capturing the decision making process.
Twenty years after the first publications and books on enterprise architecture, the domain is evolving from a technology-driven towards a more business-driven approach, thus empowering decision makers to adapt and transform an enterprise in order to keep up with changing business needs. At the same time the discipline of enterprise architecting has matured, leading to a better understanding of the profession of an enterprise architect. With this book, the authors aim to provide an overview of enterprise architecture including the process of creating, applying and maintaining it, thus taking into account the perspectives of CxOs, business managers, enterprise architects, solution architects, designers and engineers. They explore the results that are produced as part of an enterprise architecture, the process by which these are produced, and the role the architect plays in this process. As such, they do not describe a specific method for developing an enterprise (IT) architecture, nor do they define a specific modeling language for enterprise architecture, rather they offer the reader a fundamental way of thinking about enterprise architecture, which will enable him to select and apply the right approach, architecture framework and tools that meet the objective and context of the architecture work at hand. This approach is emphasized by discussion statements at the end of each chapter, sparking thoughts about benefits, shortcomings, and future research directions. Covering both theoretical foundations and practical use, and written in close collaboration between industry professionals and academic lecturers, Enterprise Architecture thus offers an ideal introduction for students in areas like business information systems or management science, as well as guidance and background for professionals seeking a more thorough understanding of their field of work.
"The reinvention of decision making" Naturalistic Decision Making (NDM) provides a new approach for understanding and supporting decision makers as they work in complex, time-pressured, and high-risk environments. An NDM framework provides a perspective for understanding command decision making in complex organizations such as the Combined Joint Task Force and for supporting this decision making in a variety of operations including peace operations. This paper provides a detailed description of the field of NDM. The authors begin with a description of three different decision scenarios. These scenarios facilitate a discussion of the characteristics of naturalistic decision settings and the importance of expertise for naturalistic decision making. The authors conclude the chapter with discussions of research methods and the findings from NDM research that distinguish it from more traditional decision research.
An enterprise architecture tries to describe and control an organisation's structure, processes, applications, systems and techniques in an integrated way. The unambiguous specification and description of components and their relationships in such architecture requires a coherent architecture modelling language. Lankhorst and his co-authors present such an enterprise modelling language that captures the complexity of architectural domains and their relations and allows the construction of integrated enterprise architecture models. They provide architects with concrete instruments that improve their architectural practice. As this is not enough, they additionally present techniques and heuristics for communicating with all relevant stakeholders about these architectures. Since an architecture model is useful not only for providing insight into the current or future situation but can also be used to evaluate the transition from 'as-is' to 'to-be', the authors also describe analysis methods for assessing both the qualitative impact of changes to an architecture and the quantitative aspects of architectures, such as performance and cost issues. The modelling language and the other techniques presented have been proven in practice in many real-life case studies. So this book is an ideal companion for enterprise IT or business architects in industry as well as for computer or management science students studying the field of enterprise architecture.
ContextChoosing a design solution most often involves dealing with trade-offs and conflicts among requirements and design objectives. Making such trade-offs during early stages of requirements and design is challenging because costs and benefits of alternatives are often hard to quantify.Objective The objective of this work is to develop a decision analysis method that assists in making trade-offs in the absence of quantitative data.Method In this method, stakeholders qualitatively compare consequences of alternatives on decision criteria. We propose an algorithm that generates all possible consequences of alternatives on requirements, according to the rough qualitative comparisons that stakeholders made. The possible consequences generated by the algorithm are then analyzed by the Even Swaps Multi-Criteria Decision Analysis method to determine the best solution. The Even Swaps method is a technique developed in management science to assist in multi-criteria decision making when explicit value trade-offs are not available.Results and conclusionsOur algorithm teases out the need to accurately measure or estimate costs and benefits of alternative design solutions. The algorithm automates the Even Swap process, and reuses stakeholders’ value trade-offs throughout the Even Swaps process. We applied the prototype tool in several case studies to evaluate the utility of the method. The results of case studies provide evidence that our decision aid method selects the optimum solution correctly compared to results of other similar quantitative methods, while our method does not rely on detailed numerical assessment of alternatives and importance weights of criteria.