A Case Study on
Artefact-based RE Improvement in Practice
Daniel M´endez Fern´andez1and Stefan Wagner2
1Technische Universit¨at M¨unchen, Germany
2University of Stuttgart, Germany
Abstract. Background: Most requirements engineering (RE) process
improvement approaches are solution-driven and activity-based. They fo-
cus on the assessment of the RE of a company against an external norm
of best practices. A consequence is that practitioners often have to rely
on an improvement approach that skips a profound problem analysis and
that results in an RE approach that might be alien to the organisational
needs. Objective: In recent years, we have developed an RE improve-
ment approach (called ArtREPI ) that guides a holistic RE improvement
against individual goals of a company putting primary attention to the
quality of the artefacts. In this paper, we aim at exploring ArtREPI’s
beneﬁts and limitations. Method: We contribute an industrial evaluation
of ArtREPI by relying on a case study research. Resu lts: Our results
suggest that ArtREPI is well-suited for the establishment of an RE that
reﬂects a speciﬁc organisational culture but to some extent at the cost of
eﬃciency resulting from intensive discussions on a terminology that suits
all involved stakeholders. Conclusions: Our results reveal ﬁrst beneﬁts
and limitations, but we can also conclude the need of longitudinal and
independent investigations for which we herewith lay the foundation.
Keywords: Requirements Engineering, Artefact Orientation, Software
Process Improvement, Case Study Research
Requirements engineering (RE) constitutes an important success factor for soft-
ware development projects since stakeholder-appropriate requirements are im-
portant determinants of quality. Its interdisciplinary nature, the uncertainty,
and the complexity in the process, however, make the discipline diﬃcult to in-
vestigate and to improve . For an RE improvement, process engineers have to
decide whether to opt for problem orientation or for solution orientation [2,3].
In a solution-driven improvement, the engineers assess and adapt their RE ref-
erence model, which provides a company-speciﬁc blueprint of RE practices and
artefacts, against an external norm of best practices. The latter is meant to lead
to a high quality RE based on universal, external goals (see, e.g. CMMI for
© Springer. This is the author's version of the work. It is posted here by permission of Springer for your personal use. Not for
redistribution. The deﬁnitive version was published in the conference/workshop proceeding
RE ). Solution-driven improvement approaches might thus serve the purpose
of achieving externally predeﬁned goals by implementing a set of best practices
adhered by many organisations  (e.g. as part of a certiﬁcation). They do not
necessarily consider company-speciﬁc goals, however, that dictate the notion of
RE quality within a particular socio-economic context (e.g. a company) and,
thus, may result in an RE reference model that is alien to the organisational cul-
ture. In consequence, those RE improvement approaches encounter problems and
are often rejected by practitioners [5,1]. A notion of RE quality where company-
speciﬁc goals dictate the improvement is the core of problem-driven approaches.
Besides the improvement principles, the paradigm in which the targeted RE
reference model is structured (and, thus, improved) plays an important role. A
reference model can either be activity-based or artefact-based . In short, an
activity-based improvement approach focuses on improving the quality of the
RE practices while an artefact-based one puts its focus on improving the quality
of the RE artefacts.
Most available RE improvement approaches today are solution-driven .
Yet, RE is complex by nature, and we postulate that RE quality depends on the
contribution of an RE reference model to context-speciﬁc goals. Therefore, im-
provements cannot be meaningfully implemented without a qualitative problem
investigation that reveals which goals must be achieved [3,2] and which artefacts
should be created in which way. In response to the lack of problem-driven and
artefact-based RE improvement approaches, we elaborated such an approach 
which we call ArtREPI. We further realised our approach using the EPF Com-
poser3and made ﬁrst experiences using ArtREPI in practice (see also Sect. 2).
Problem. Although we made ﬁrst conceptual and empirical contributions
to support a problem-driven and artefact-based RE improvement, we still have
little evidence on its practical beneﬁts and limitations.
Contribution. In this paper, we report on the ﬁrst industrial case study
to evaluate ArtREPI in comparison to solution-driven and activity-based REPI
approaches previously used in the same contexts. The purpose, however, is not
to evaluate only the particularities of our approach itself, but also
1. to reveal ﬁrst qualitative insights into the beneﬁts and limitations of a
problem-driven and artefact-based RE improvement in general, and
2. to lay the foundation for future independent empirical investigations.
Based on our contribution and the disclosed material , practitioners can
therefore already apply our RE improvement approach, and researchers can build
their conceptual and empirical work on our results to further explore the full
spectrum of an RE improvement.
Outline. In Sect. 2, we discuss fundamentals and related work. In Sect. 3,
we introduce the artefact-based RE improvement approach. We provide our case
study design in Sect. 4, the results in Sect. 5, and a critical reﬂection in Sect. 6,
before concluding our paper in Sect. 7.
2 Fundamentals and Related Work
Requirements engineering process improvement, as software process improve-
ment in general, is a cyclic approach to continuously analyse problems/the cur-
rent situation in RE as part of an appraisal, plann an improvement, realise the
improvement and evaluate the improvement before initiating the next iteration.
We can distinguish solution-driven approaches and problem-driven approaches
as well as artefact and activity orientation.
In literature, there exist mostly solution-driven contributions . R-CMM,
proposed by Beecham et al. , is a prominent representative of these approaches.
It is based on CMMI and an empirical investigation in twelve companies .
The investigation revealed patterns and best practices based on problems expe-
rienced by practitioners. Therefore, it aimed at a generalised, external notion of
RE quality. A technical validation using an expert panel  further illustrates
selected success criteria, such as understandability. Approaches of this category
focus on a solution-driven benchmarking of the maturity of RE according to a
speciﬁc norm of best practices and may thus lead to the problems described in
the introduction (see also [5,1] for richer investigations).
In response to their shortcoming, Pettersson et al. contributed an approach
to problem-driven RE improvement  called the iFLAP approach. Same as
in ArtREPI, they make use of qualitative methods for the problem analysis
and postulate the importance of strong stakeholder involvement. Although their
concepts are promising to conduct a problem-driven REPI, the consequential
next steps, i.e. the actual improvement realisation by crafting a new RE reference
model, was not in scope of their contribution.
To the best of our knowledge, there exists no holistic approach to a REPI
covering all improvement phases in a seamless manner, let alone considering
an improvement speciﬁcally directed at the RE artefacts. Recent work in this
direction is made by Kuhrmann et al. , but taking more a perspective on
the management of software process models. The focus is thereby set on how to
manage an artefact-based improvement rather than on how to conduct it. They
do not look at how to analyse, design, and evaluate a process in a problem-driven
manner focusing on the quality of the artefacts which is in scope of ArtREPI.
Available validation and evaluation research, which would be directly related
to the contributions of this paper, focuses on the evaluation of methods or metrics
used in isolated REPI phases, such as the analysis, on experience reports, or
on the analysis of general success factors and lessons learnt. In the case study
we present here, we therefore do not discuss the relation to existing evidence
taking into account particular approaches, but exclusively take a qualitative
view and rely on the evaluation of ArtREPI against the general perceptions and
the experiences of the participants with solution-driven, activity-based REPI in
the same context. Further details on the publication landscape on REPI can be
taken from our previously published mapping study .
Previously Published Material. In , we ﬁrst introduced the basic con-
cepts of ArtREPI and its design science principles. Since then, we realised our
approach using the EPF Composer as a means of a technical validation and
made all material (models, process documentation, document templates, and
evaluation instruments) publicly available  to support the dissemination. In a
previous short paper , we then brieﬂy reported on initial experiences from an
ongoing case study. In the paper at hands, we report on the by now completed
case study in detail including the case study design, the results containing a
second case, and the implications the results have.
3 ArtREPI: Artefact-based RE Process Improvement
Figure 1 gives an overview of the basic structure of ArtREPI, which we use for
our evaluation presented in this paper. As shown in the ﬁgure, we distinguish two
contexts important to the notion of RE quality: an external context that con-
tains norms of best practices and the socio-economic context where the notion
of quality is dictated by individual demands. External norms of best practices
are not only key to solution-driven RE improvement approaches but also impor-
tant to a problem-driven improvement as one principle is to support technology
transfer according to context-speciﬁc goals.
The socio-economic context is further characterised by a set of disciplines that
aim at managing software processes. Of particular interest is that we have process
owners who usually have the sovereignty over an RE reference model (although
often underrepresented in practice ) and a set of project participants who work
according to an (explicitly established or implicitly lived) RE reference model.
The latter is subject to an improvement in case an organisational change is
triggered. For reasons of complexity, we omit the discussion of further disciplines
or roles, such as an improvement sponsor. A change is performed as part of an RE
process improvement project which is in scope of ArtREPI. ArtREPI consists of
four phases and yields a set of mandatory and optional results. The mandatory
ones eventually serve a seamless improvement based on a uniﬁed underlying data
model of the improvement results. The full model can be taken from our online
sources  while in the following, we provide a brief overview of the single phases.
3.1 Improvement Preparation
In the preparation phase, we lay the groundwork for the improvement and aim
at getting an understanding about the application domain, such as the chemical
sector, typical constructs and rules followed therein, and the terminology used.
We discuss contemporary problems and the primary improvement goals which,
in collaboration with the process owners, we reﬁne to concrete metrics and mea-
surements to evaluate the success of the improvement after completion. Where
possible and reasonable, we reuse metrics from previously conducted ArtREPI.
The outcome of the preparation phase is an improvement plan that deﬁnes the
concrete procedures and time schedules as well as resource allocations (cases and
subjects) for subsequent steps including interviews or workshops.
1* 2* 3* 4*
Fig. 1. ArtREPI: Structuring and context.
3.2 Problem Analysis
The second stage aims at discovering problems in the use of the current RE
reference model independent of whether it is explicitly deﬁned as a company
standard or not. In case a company standard exists, we analyse it ﬁrst to get
an overview of the basic artefacts, roles, and activities. We then conduct a doc-
ument analysis in selected exemplary development projects and abstract from
the contents to build an as-is model of the RE artefacts, milestones, and roles.
We complement this syntactic analysis with a semantic analysis where we anal-
yse, for example, linguistic defects in document templates. We use the deﬁned
as-is model to conduct a gap analysis where we compare the current state of
practice with an external standard wich serves as a knowledge base. More pre-
cisely, we rely on AMDiRE (artefact model for domain-independent RE) that
deﬁnes a best practice norm for artefact-based RE (see also ). We use poten-
tial gaps from our analysis as candidates for further validation which we do via
semi-structured interviews. There, besides asking about experiences and expec-
tations, we ask the participants why certain contents identiﬁed as incomplete in
comparison to AMDiRE have or have not been speciﬁed in their projects. We
take their reasoning as our primary input to establish later on a tailoring proﬁle
for the artefact-based RE reference model. The aim of the whole analysis is not
to benchmark the process against an external standard. Instead and according to
the problem-driven nature of ArtREPI, it aims at getting a better understanding
on potential improvements which then can only be validated by project partici-
pants. To this end, we trigger a self-reﬂection via interviews by indicating only
possibilities for changes that eventually only they can judge upon.
3.3 Improvement Design
In the third stage, we conduct the actual realisation of the improvement based
on the candidates previously identiﬁed during the gap analysis. To this end, we
conduct a series of action research workshops where we build a new model can-
didate by deciding which content items to consider in the candidate (e.g. “use
case model”) before we then subsequently deﬁne the actual content model that
abstracts from concrete modelling concepts used to specify the content items
in a project. The latter serves to build a prototypical (modelling) tool or docu-
ment templates, and to establish content-related dependencies to the artefacts of
further development phases, such as of testing, serving the purpose of a process
integration into the overall software process model. We enrich the artefact model
with process elements (roles, milestones) and a tailoring proﬁle that emerges from
the interviews and that deﬁnes under which project circumstances to document
certain content items or not. Depending on the improvement project complexity
or criticality, we may perform a validation of the model candidate before enter-
ing the last stage. In this validation step, we validate the model via feedback
gathered by domain experts not involved in the improvement workshops.
3.4 Improvement Evaluation and Transfer Preparation
The success of a problem-driven improvement can eventually only be determined
by the degree to which the improvement outcome satisﬁes the improvement goals.
In the last step, we therefore conduct technical action research workshops with
project participants in pilot projects where we apply the resulting RE reference
model under realistic conditions. We specify a set of RE artefacts following the
new RE reference model. Afterwards, we compare the outcomes and the followed
process with the artefacts previously created in the same project environment by
an interview. The participants rate the new model following the criteria agreed
on in the preparation phase. In case of a positive rating, we release the new
model complemented with supporting material.
4 Case Study Design
In , we introduced in detail how we inferred our ArtREPI approach presented
in Sect. 3 from fundamental and applied research projects where we conducted
an artefact-based and problem-oriented RE improvement. We argued so far that
our approach is successful, because it emerged from successful RE improvements
leading to new RE company standards. We were aware, however, that we need
to better understand the beneﬁts and limitations we can expect from ArtREPI
and whether our approach can be used by others if we are not involved. This
motivates the design for our empirical investigations:
1. We rely on case study research with action research components in an in-
dustrial environment as we are particularly interested in elaborating qualita-
tive insights into beneﬁts and shortcomings in applying an RE improvement
approach that also depends on subjective (and social) factors. The design
follows the guideline of Runeson and H¨ost .
2. We included two cases where the second one was conducted with little direct
involvement of the ArtREPI authors. This should allow us to get an indica-
tion whether the success of an improvement depends on our inﬂuence, and
it should provide a ﬁrst step in scaling up to practice.
4.1 Objectives and Research Questions
The study has the general objective to understand the beneﬁts and limitations of
applying ArtREPI in practice. We formulate two research questions to steer the
study design structured into the evaluation of the improvement tasks (ArtREPI
stage 1-3) and of the resulting RE reference model (ArtREPI stage 4):
RQ 1 How well are process engineers supported in their RE improvement tasks?
RQ 2 How well are project participants supported by the resulting RE reference
4.2 Cases and Subjects
We applied ArtREPI as part of two improvement projects in two di↵erent con-
texts. For each project, we give a summary of the most important context in-
formation we are able to provide within the limits of existing non-disclosure
agreements in Tab. 1. We chose the contexts because of their suitability to our
research questions. We looked at two di↵erent settings (large versus small and
agile). Furthermore, REPI project 2 has, in contrast to the ﬁrst case, only little
involvement of those researchers developing ArtREPI (see also the next section).
REPI 1: Large scale RE process. The ﬁrst improvement project was
conducted as part of a research cooperation between Technische Universit¨at
M¨unchen and Wacker Chemie and was described, in parts, as the intermedi-
ate evaluation in .4Wacker Chemie is a German company that works in the
4In , we reported on ﬁrst results from the ﬁrst evaluation steps where the context-
speciﬁc evaluation (ArtREPI stage 4) was still ongoing.
Tabl e 1 . Overview of cases (REPI 1 and REPI 2).
No. Aspects Descriptions
1 Improvement goal Integration of RE into quality management.
REPI context 2 process engineers, 4 domain experts, 4 external improvement consultants.
E↵ort: 8 PM including 13 workshops.
Pilot projects 3 pilot projects, 8 participants.
2 Improvement goal Re-design of RE to support agility.
REPI context 2 main process engineers (previously coached by improvement consultants).
E↵ort: approx. 6 PM.
Pilot projects 3 project participants to rate the reference model w/o pilot projects.
chemical business and develops custom software for their operation processes
and their production sites. The improvement project aimed at deﬁning a de-
tailed artefact-based RE with a seamless integration into quality management.
REPI 2: Agile RE process. The second improvement project was con-
ducted as part of a Master’s Thesis at the company SupplyOn AG which is a
software as a service provider. The improvement project aimed at the re-design of
the RE of the Rational Uniﬁed Process to an artefact-based agile RE approach.
In contrast to project 1, where we evaluated the resulting RE reference model
explicitly via pilot projects, we relied on informal reviews and expert opinion.
4.3 Data Collection and Analysis Procedures
We answer our research questions by collecting and analysing data from the
application of ArtREPI and the assessments via questionnaires in both cases
by improvement consultants who also authored ArtREPI. To this end, the im-
provement consultants coach the process engineers on ArtREPI, its underlying
principles, and corresponding tools in a 3 hour workshop. These process engi-
neers then apply ArtREPI as described in Sect. 3 and conduct the last phase of
ArtREPI (i.e. the evaluation) by applying the resulting RE reference model in
pilot projects of the company. In REPI 1, the improvement consultants formed
part of the improvement project team after the coaching. In REPI 2, they where
not directly involved anymore after the coaching sessions.
We collect the data for both research questions using questionnaires with
open and closed questions. Each closed question is formulated as a statement
where the participants should state their agreement on a Likert scale deﬁned on
an ordinal scale from 0 (“I strongly disagree”) to a maximum of 7 (“I strongly
agree”) to avoid that they choose the middle. They have, however, always the
possibility to refuse an answer. The open questions give them the possibility to
provide a rationale for their decision. The original questionnaires can also be
found in our online material .
To answer RQ 1, we conduct an assessment by letting the involved process en-
gineers rate ArtREPI on the basis of a questionnaire where they directly compare
ArtREPI to previously followed solution-driven, activity-based improvement ap-
proaches. We summarise the questions in Tab. 2. At the end, we ﬁnally asked
the process engineers three concluding questions about the overall suitability of
the approach to cover the particularities of their context.
Tabl e 2 . Condensed RQ 1 instrument: Support of process engineers in improvement.
Criteria Statements (closed question)
Structuredness The approach was systematic.
Simplicity The approach was easy to use.
Goal orientation The approach considered problems and needs of all involved stakeholders.
Experience orientation The approach considered company culture and stakeholder experiences.
Transparency The decisions during the workshops were reproducible.
E↵ectivity The improvement has led to the desired results.
Eﬃciency I perceived the eﬃciency of the undertaking as high.
Knowledge transfer The approach actively supported knowledge transfer.
Overall suitability (1) ArtREPI was better suited than solution-driven approaches.
Overall suitability (2) I would apply ArtREPI again.
Overall suitability (3) I want to add following positive/negative aspects (open)
To answer RQ 2, we conduct an assessment by letting the involved project
participants rate the outcome of ArtREPI using again a questionnaire. This
evaluates the resulting RE reference model after application in the pilot projects
in direct comparison with the one previously used in same context to evaluate
whether the improvement goals have eventually been achieved. Table 3 sum-
marises the questions for this RQ. In the questions, we use “model” as a reference
to the RE reference model (upper part in the table) and “artefacts” as a short
reference for the artefacts created using the model (lower part in the table).
Tabl e 3 . Condensed RQ 2 instrument: Support of pro ject participants by RE ref-
erence model (resulting from the improvement tasks).
Criteria Statements (closed question)
Flexibility The model allows for ﬂexibility in the RE process.
Ease of use The model is easy to understand.
E↵ectivity The application of the model has led to the desired results.
Eﬃciency When applying the model, I perceived the eﬃciency as high.
Customisation / Tailoring The model can be tailored to pro ject-speciﬁc situation of the company.
Process integration The model is integrated into further development activities and within
the line organisation.
Structuredness of artefacts The artefacts are well structured and can be understo od by people
not involved in their creation.
Syntactic artefact quality The model supports a high syntactic artefact quality w.r.t. complete-
ness and consistency.
Traceability of artefact The model supports traceability within RE and between RE and fur-
Semantic artefact quality The mo del supports semantically consistent and complete artefacts.
Testability of artefacts The model supports the creation of testable artefacts.
For the analysis of our results, we rely on descriptive analysis and qualita-
tive interpretation of our data. We intentionally refrain from summarising visual
accumulations of the ratings (e.g. via radar charts) because of the heterogene-
ity in the data as providing answers to each criterion was not obligatory. We
therefore report on every rating given by each subject. We additionally indicate
to whether the participants experienced an improvement or deterioration of ap-
plying ArtREPI in direct comparison to previously used improvement approach
which we assume to be present if the mean values di↵er by at least one point.
5 Case Study Results
In the following, we summarise the results from the rating given in the assess-
ments, structured according to the research questions. For reasons of conﬁden-
tiality, we cannot provide details about workshop contents, the company-speciﬁc
RE reference model and the pilot projects, but we will report the rating results.
5.1 RQ 1: Support in RE Improvement Tasks
Table 4 summarises the ratings of the ArtREPI approach by the process engi-
neers in direct comparison to previously used approaches which were based on
CMMI. For each criterion, we show each subject’s ratings. We had 6 subjects
in context 1 and 2 subjects in context 2 (see Tab.1). Furthermore, we give the
mean and median as central tendencies and further show whether ArtREPI is
clearly considered better (+) or worse (). No clear comparison is indicated by
a 0. At the bottom of the table, we show the results of the rating of whether
ArtREPI was considered better suited for the respective context than previously
used REPI approaches.
Tabl e 4. Results for RQ 1: Rating of improvement procedure by subjects Sxon a
scale of 0 (“I strongly disagree”)to7(“I strongly agree”) from a process engineering
perspective. See also Tab. 2 for details on the used instrument.
REPI 1 REPI 2
S1 S2 S3 S4 S5 S6 Mean Median Comp. S1 S2 Mean Median Comp.
Structuredness Old --56534.75 5 +64 5 5 0
New 656775 6 6 65 5.5 5.5
Simplicity Old --5555 5 5 065 5.5 5.5 0
New 5565534.83 5 65 5.5 5.5
Goal Orientation Old --4644 4.5 4 044 4 4 +
New 335664 4.5 4.5 66 6 6
Experience Orientation Old --4-364.33 4
66 6 6
New 225- 52 3.2 2 43 3.5 3.5
Sustainability Old --54444.25 4 +45 4.5 4.5 0
New 6564745.33 5 45 4.5 4.5
E↵ectivity Old --55634.75 5 034 3.5 3.5 +
New 3365654.66 5 76 6.5 6.5
Eﬃciency Old --5654 5 5
34 3.5 3.5 +
New 126654 4 4.5 75 6 6
Knowledge Transfer Old --4644 4.5 4 034 3.5 3.5 +
New 555663 5 5 76 6.5 6.5
Overall Suitability New --5--5 5 5 76 6.5 6.5
Overall, ArtREPI was rated as a structured improvement approach that
tends to better support knowledge transfer than previously used approaches.
Surprising to us, however, is the result of REPI 2 conducted by people not
involved at all in the development of ArtREPI. Our assumption was that the
rating would be worse than in REPI 1, but it was better regarding the goal ori-
entation, the e↵ectivity and eﬃciency, and the support for knowledge transfer.
Qualitative statements from the open questions provide some explanations. The
subjects in REPI 1 rated that the initial preparation phase was performed in
a “too academic” fashion with “too many discussions to clarify the terminol-
ogy”, especially by those subjects with no REPI experience made before (S1
and S2). This might also be the reason for the negative comparison regarding
the experience orientation. In contrast, REPI 2 was conducted solely by process
engineers employed by the respective company and familiar with the culture and
Finally, the overall rating whether ArtREPI was better suited to achieve
the company-speciﬁc improvement goals in comparison to previously used ap-
proaches were answered positively in both cases. The process engineers stated for
both cases that they would apply ArtREPI in follow-up improvement cycles. The
answers to the open questions showed for REPI 1, however, that the engineers
expect an integration into the organisation as a prerequisite for a repetition.
Another suggestion was to check in a follow-up study whether the eﬃciency of
ArtREPI could be improved. Further informal statements included that:
1. the action research workshops, where the new artefact-based RE reference
model was crafted by the process engineers, together with the group discus-
sions, fostered discussions the engineers would otherwise not have, e.g. about
roles and responsibilities in the RE that just seemed clear to everybody.
2. independent of the results from the pilot projects (described next), they
would need additional longitudinal studies before declaring the new RE ref-
erence model as a new standard as too many changes have been made.
5.2 RQ 2: Support by resulting RE Reference Model
The success of an improvement eventually depends on whether the resulting
RE reference model achieves the improvement goals. RQ 2 therefore focuses on
evaluating the REPI outcome in pilot projects. Table 5 shows the rating of the
artefact-based RE reference model resulting from the improvement cycle from
the perspective of project participants. We distinguish criteria to rate the appli-
cation of the reference model itself and criteria to evaluate the RE artefacts pro-
duced following the reference model. In REPI 1, we applied the model in three
pilot projects (two with custom software development and one with standard
software). In REPI 2, it was not possible to fully implement the new model im-
mediately, but they included the artefacts created during the coaching sessions.
Therefore, their rating is does not reﬂect the full experience with ArtREPI.
The results indicate that ArtREPI supported the participants in achieving
their improvement goals (see Sect. 4.2). For REPI 1, we could improve, e.g.,
the traceability, the testability, and the process integration of the RE reference
model, thus, supporting the better integration into quality management which
formed the improvement goal. The negative results in the comparison to the
previously used RE reference model regarding the process quality might be ex-
plained from the complexity of the new (richer) model and the learning curve
associated with all new methods in general. In contrast, the positive rating in
REPI 2 might be explained by the new RE reference model to support agility
due to its new (light weight) simplicity.
Tabl e 5. Results for RQ 2: Rating of resulting RE reference model by subjects Sxon a
scale of 0 (“I strongly disagree”)to7(“I strongly agree”) from a project perspective.
REPI 1 REPI 2
S1 S2 S3 S4 S5 S6 S7 S8 Mean Med. Comp. S1 S2 S3 Mean Med. Comp.
Flexibility Old 770425664.63 5.5 0561 4 5 +
New 446652654.75 5 6676.33 6
Ease of Use Old 5-5464454.71 5
525 4 5 +
New 3-5352443.71 4 555 5 5
E↵ectivity Old 4-56-564 5 5 05-1 3 3 +
New 6-56-2665.16 6 6-7 6.5 6.5
Eﬃciency Old 3-45-6654.83 5
6-0 3 3 +
New 1-56-2633.83 4 5-7 6 6
Customisation Old 7-131664 4 4 05745.33 5 +
New 6-6641654.86 6 6776.67 7
Process Old 3-2336533.75 3 +5354.33 5 +
Integration New 6-4466555.14 5 5655.33 5
Structuredness Old 253532523.38 3 0342 3 3 +
New 473642534.25 4 366 5 6
Syntactic Old 05440263 3 3.5 +-22 2 2 +
Quality New 775525665.38 5.5 - 76 6.5 6.5
Traceability Old 441322633.13 3 +1301.33 2 +
New 766745645.63 6 666 6 6
Semantic Old 4423- 3733.71 3 +435 4 4 +
Quality New 6654- 5765.57 6 6655.67 6
Testability Old 123452633.25 3 +5333.67 3 +
New 545675655.38 5 6766.33 6
Overall, we observe that ArtREPI can show its strength especially in the
artefact quality as expected, because it is oriented towards artefacts. In the
process quality, the picture is, however, more mixed. Qualitative feedback to the
open questions provides explanations:
–The success of a pilot study strongly depends, beyond political factors, on
the quality of the prototypical implementation. Hence, project participants
should be involved in the technical validation, too.
–Coaching sessions before the pilot studies should not only focus on the new
RE reference models, but on the underlying principles, e.g. new levels of
abstraction in the requirements or new roles and responsibilities.
–Although artefact-based RE reference models are inherently process agnos-
tic, they should include suggestions for methods and modelling techniques
(e.g. UML-based ones) project participants are familiar with. This should
increase the organisational willingness to change.
5.3 Threats to Validity
There exist many threats to validity inherent to case study research . Threats
to the internal validity mainly arise from potential bias during the data collec-
tion, let alone because of our action research components, and because of the
general subjective nature in the ratings. We applied selected techniques to reduce
those threats, e.g. researcher and method triangulation, but to a certain extent
we were particularly interested in gathering subjective opinions by the intervie-
wees. Another mitigation strategy followed in advance was to apply ArtREPI in
the second case with little inﬂuence of the approach authors. It provides a ﬁrst
indication towards scaling up to practice , thus, it provides a ﬁrst step in
strengthening the external validity. This generalisation, however, needs further
attention in future work.
Finally, a more general problem is that many things remain (objectively)
unmeasurable. As a matter of fact, we still have a limited understanding on how
to reliably measure long-term improvement e↵ects going beyond RE, because
of the complexity of confounding variables in a software project ecosystem. In
our studies, we therefore refrained from such measurements in advance. Mea-
surements are still important to determine the success of an improvement given
that (1) RE forms part of a larger context that needs to be taken into account,
while (2) we consider problem orientation where we cannot rely on an external
reference to determine a notion of software process quality (in the sense of an
Another facet important to an improvement is ﬁnally the question how much
the notion of process quality is determinable by the quality of the artefacts
and also how much project participants eventually rely on the created artefacts.
We can observe ﬁrst empirical investigations in that direction (e.g. ). Still,
even if we can measure certain phenomena on basis of the artefacts, we still
do not fully understand to which extend the notion of RE quality eventually
manifests itself in the created artefacts. That is, the investigations are based
on the critical assumption that the success of a project and, in particular, of
an artefact-based improvement depends on the documented results on which
project participants rely. To elaborate the extent to which the application of
ArtREPI eventually leads to an improvement, and how to measure the success
of an improvement (including subjective and cognitive facets), we ﬁrst need a
better understanding on the measurability of such an improvement. In , we
provide a richer discussion on the limitations of measurements in RE.
Our results indicate that ArtREPI is well suited to cover the needs of a struc-
tured improvement where problem and artefact orientation are important, while
supporting knowledge transfer by continuous stakeholder involvement. The di-
rect comparison of the two cases further indicates that the e↵ectivity, eﬃciency,
the knowledge transfer, and even the goal orientation are strengthened when
the improvement is conducted by company members with no intervention from
outside. The results from pilot projects further suggest that the improvement
eventually achieved the local improvement goals. Our overall results therefore
strengthen our conﬁdence in the beneﬁts of ArtREPI as a self-contained and
holistic approach to a problem-driven RE improvement as long as the improve-
ment goals are in tune with the (known) beneﬁts of artefact orientation.
6.1 Limitations of ArtREPI
One beneﬁt in case study research is that it gives us the possibility to get qual-
itative feedback, i.e. explanations for the particular ratings. We are particularly
interested in revealing limitations of ArtREPI as this supports us steering the
development and the evaluation of the discussed improvement principles, and it
helps us fostering the discussions on RE improvement in general.
Limitations we could reveal by our second case concern more social aspects
of a process improvement. For example, we, as researchers involved in the de-
velopment of ArtREPI, seemed to lower the eﬃciency and e↵ectivity of an im-
provement due to long preparation phases to increase our understanding of the
domain and the terminology used. Our initial assumption that the success of
ArtREPI depends on that we ourselves should conduct the improvement thus
is wrong. However, the exact consequences of applying ArtREPI without any
involvement of the authors at all (e.g. without coaching) remain still unknown.
Also, there exists a plethora of organisational factors, such as the support by
the management, the support at project level (a champion), and general social
as well as empirical skills that constitute success factor for an improvement. It
might be possible to include some factors in the approach. For instance, one
suggestion was to involve project participants earlier during technical validation
stages to mitigate the threats arising from a missing organisational willingness
to change. Other factors, however, might not be covered at all, because models
abstracts, by nature, from desires, beliefs, experiences, and expectations, which
all are important. To fully reveal those factors, and to fully explore to which
extent they can eventually be inﬂuenced on a methodological layer, we need
more (especially longitudinal) investigations in that direction.
6.2 Success Factors for ArtREPI
Besides the general focus of problem orientation, which is to emphasise context-
speciﬁc goals over the possibility for an external certiﬁcation of an RE, we found
several factors that inﬂuence the success of an RE improvement:
–Improvement goals need to be in tune with the expected beneﬁts of the
chosen paradigms (in our case artefact orientation) known in advance.5
–Support by higher management, especially when communicating new roles
–Backup by project environments, e.g. via the early involvement of project
participants in the validation stages of a new RE reference model (before
initiating the evaluation stages).
–Reﬂection of the organisational culture.
–Social and empirical skills.
Those factors might not be surprising in themselves but together in their ex-
tent. The success of a problem-driven improvement is determined by the reﬂec-
tion of the organisational culture of a company in every facet of the improvement
approach. That is, all relevant stakeholders need to be involved in early stages to
5This implies that we need suﬃcient evidence on potential beneﬁts of artefact ori-
entation to the chosen context beforehand to justify the decision to conduct an
cover their needs, a backup by representative projects needs to be ensured (in the
sense of champions), and project participants need to identify themselves with
the resulting reference model. All those facets aim at supporting the willingness
for a change in the way of working to reshape an existing organisation .
In our current understanding, this can only be achieved by applying ex-
haustive, qualitative empirical method that foster continuous stakeholder in-
volvement but which also come themselves with limitations, e.g. regarding the
measurability of improvement e↵ects (see also our Sect. 5.3).
In this paper, we reported on our ﬁrst steps to empirically evaluate the princi-
ples of an artefact-based and problem-driven RE improvement (ArtREPI). Our
investigation should give a ﬁrst qualitative impression of the general beneﬁts
and shortcomings in direct comparison to existing principles of solution-driven
RE improvements that currently dominate the publication landscape . The
purpose was also to lay the foundation for independent empirical investigations
to eventually explore the full spectrum of RE improvement.
To this end, we reported on two industrial case studies and analysed (1)
how ArtREPI would support process engineers in achieving a problem-driven
and artefact-based RE improvement, and (2) how well project participants were
supported in their project environments by the resulting artefact-based RE ref-
erence model. We further conducted the second case with very little inﬂuence of
the approach authors to investigate, in a ﬁrst step, to what extent the results
might be inﬂuenced by us researchers being involved in the RE improvement.
Our results strengthen our conﬁdence that ArtREPI is suitable for a self-
contained, problem-driven RE improvement where the improvement goals are
in tune with the scope of artefact orientation. More important, we could also
reveal ﬁrst limitations and factors important to future work in RE improvement
research. We need to further explore the measurability of an improvement and
its long-term e↵ects, and to further support scaling up to practice. To support
the generalisation of our observations that, so far, are only valid for chosen local
improvement contexts , we need to (1) foster independent replications of our
case studies which we support by our contribution and our publicly accessible
online material , and (2) we need to take new perspectives.
Rather than further investigating the technicalities of applying ArtREPI our-
selves, we need to concentrate on exploring further social and cognitive facets
and the long-term e↵ects of an RE improvement for which our contribution has
provided one foundation and sensitisation. We thereby encourage researchers and
practitioners to join us in this endeavour to fully understand the broad spectrum
of possibilities and limitations in an artefact-based and problem-driven RE im-
Acknowledgements. We want to thank R. Wieringa and M. Broy for the
collaboration during the development of ArtREPI. We further want to thank
R. Bossek and M. Kuhrmann for their support during the systematisation of
ArtREPI, and S. Wiesi, J. Mund, J. Eckhardt, and H. Femmer for their support
in the REPI projects. Finally, we are grateful to all subjects involved during the
case studies and to B. Penzenstadler, M. Daneva, and A. Vetr`o for their valuable
feedback on previous versions of this manuscript.
1. M´endez Fern´andez, D. and Wagner, S.: Naming the Pain in Requirments Enginer-
ing: A Design for a global Family of Surveys and First Results from Germany. IST
2. Pettersson, F., Ivarsson, M., Gorschek, T., ¨
Ohman, P.: A practitioner’s Guide to
light weight Software Process Assessment and Improvement Planning. JSS (2008)
3. Napier, N., Mathiassen, L., Johnson, R.: Combining Perceptions and Prescriptions
in Requirements Engineering Process Assessment: An Industrial Case Study. TSE
4. Beecham, S., Hall, T., Rainer, A.: Deﬁning a requirements process improvement
model. Software Quality Control (2005)
5. Staples, M., Niazi, M., Je↵ery, R., Abrahams, A., Byatt, P., Murphy, R.: An
Exploratory Study of why Organizations do not adopt CMMI. JSS (2007)
6. M´endez Fern´andez, D. and Penzenstadler, B. and Kuhrmann, M. and Broy, M.:
A Meta Model for Artefact-Orientation: Fundamentals and Lessons Learned in
Requirements Engineering. In: MoDELS. (2010)
7. M´endez Fern´andez, D., Ognawala, S., Wagner, S., Daneva, M.: Where Do We Stand
in Requirements Engineering Improvement Today? First Results from a Mapping
Study. In: ESEM. (2014)
8. M´endez Fern´andez, D. and Wieringa, R.: Improving Requirements Engineering by
Artefact Orientation. In: PROFES. (2013)
9. M´endez Fern´andez, D.: ArtREPI Online Resources. https://goo.gl/8UfUee,
10. Beecham, S., Hall, T., Rainer, A.: Software Process Improvement Problems in
Twelve Software Companies: An Empirical Analysis. EMSE (2003)
11. Beecham, S., Hall, T., Britton, C., Cottee, M., Austen, R.: Using an Expert Panel
to Validate A Requirements Process Improvement Model. JSS (2005)
12. Kuhrmann, M. and Beecham, S.: Artifact-Based Software Process Improvement
and Management: A Method Proposal. In: ICSSP. (2014)
13. M´endez Fern´andez, D.: Artefact-based Requirements Engineering Improvements:
Learning to Walk in Practice. In: PROFES. (2014)
14. M´endez Fern´andez, D. and Penzenstadler, B.: Artefact-based requirements engi-
neering: The AMDiRE pproach. RE Journal (2014)
15. Runeson, P., H¨ost, M.: Guidelines for Conducting and Reporting Case Study
Research in Software Engineering. EMSE (2009)
16. Wieringa, R.: Empirical Research Methods for Technology Validation: Scaling up
to Practice. JSS (2013)
17. Liskin, O.: How Artifacts Support and Impede Requirements Communication. In:
18. M´endez Fern´andez, D. and Mund, J and Femmer, H. and Vetr`o, A.: In Quest for
Requirements Engineering Oracles: Dependent Variables and Measurements for
(good) RE. In: EASE. (2014)
19. Pﬂeeger, S.: Understanding and Improving Technology Transfer in Software Engi-
neering. JSS (1999)