Due to the increasing growth of the discipline Data Science, a partition into sub-disciplines seems appropriate. Therefore, we propose the division of Data Science in Pure Data Science, in which methods and tools are developed, and Applied Data Science, in which these methods and tools are adapted and applied to practical problems of a specific domain. This article focuses on Applied Data Science and how it should be positioned in relation to its adjoining disciplines. We also introduce the term Business Data Science as a specific form of Applied Data Science in the business domain and describe its relationship to existing terms like
Data Science (DS) is developing into an independent new subject area. Highly interdisciplinary by
nature, it takes concepts and methods from mathematics/statistics and computer science, it combines,
expands, and enhances them and applies them to new areas of application [
Although DS now possesses generic methods and algorithms that can be applied in many domains,
they often must be refined to the particular requirements of domain-specific data applications [
2022 Copyright for this paper by its authors.
To be able to address the aforementioned aspects more concretely, the focus of this article will be on the application of DS to the business domain. This seems appropriate since DS is already widely used in this domain. The authors work for different universities that have independently of each other developed teaching programs that emphasize the application of DS. Although we had never heard of each other before, we have discovered many commonalities in these programs. Through structured expert discussions over six iterations, we extracted this common view. The results given here represent a proposal for structuring the DS discipline.
As a first result, we suggest dividing DS into two sub-disciplines; a sub-discipline in which methods and tools are developed, and another sub-discipline in which these methods and tools are adapted and applied to practical problems of a specific domain. Following the conventions in Mathematics, where a similar distinction is made, we propose the terms Pure Data Science (PDS) to identify the subdisciplines focused on the further development of the scientific field of the DS and Applied Data Science (ADS) which is mainly devoted to solving application domain challenges and provide an explanation of the obtained results. This manifesto focuses on ADS and how it should be positioned in relation to its adjoining disciplines.
An overview of the sub-disciplines (i.e. PDS versus ADS) with their distinguished characteristics is shown in
Table 1. The distinguishing characteristics are the main goal of the sub-discipline, which identifies its purpose, the expected output (i.e. its expected outcome) of it, its primary beneficiary area, and the required competencies needed to develop it further.
A closer look at the degree programs named Applied Data Science shows that even these have distinctly different emphases, so that a further subdivision of the subdiscipline should be made, which will be presented in the following chapters. In this manifesto, we focus on the application of DS to business problems. There are two main reasons for this: First, a discussion of ADS is only partially possible without considering an explicit application domain. Second, the chosen application domain is particularly relevant due to its widespread use. However, it is also an area that requires differentiation from existing terms, such as business intelligence and business analytics, which are also described in this manifesto.
The paper is organized as follows. In Section 2 we present the methodology that shaped this manifesto. In Section 3 we discuss Applied Data Science in the context of its adjoining relevant disciplines by showing their respective influences. In Section 4 we focus our discussion on Business Data Science, while in Section 5 we close with a summary of our findings.
Our research shows that since 2020, more and more ADS degree programs are being founded worldwide. The simultaneous birth of many identically named study programs from different universities can hardly be a coincidence. A search in early November 2021 identified 60 of such programs. Thus, we believe that DS is an emerging discipline with multiple facets due to its multidisciplinary nature and an increasing number of application domains. The adoption of DS to application domains shapes DS into sub-disciplines having a common core but requiring different knowledge and skills tailored to the domains where it is applied. Moreover, such application induces cross-fertilization between DS and the considered application domain shaping novel competencies and, in some cases novel disciplines.
In view of the above considerations, a group of international academics and lecturers, which offer degree programs in ADS at different universities, began a discussion on emerging DS-related disciplines and started to define a taxonomy to be used to classify research directions as well as teaching projects, and to shape the emerging DS sub-disciplines.
The group worked iteratively on this manifesto, starting from the individual degree programs and their experiences. At each iteration, they focused on a specific aspect and worked on it until reaching an agreement among the authors. Finally, they concentrated on the business domain since all authors have a common interest in it.
Having discussed the division of Data Science into the sub-disciplines of PDS and ADS, we now move ADS to the center of our discussion.
In particular, we consider ADS in the context of its adjoining relevant disciplines by showing their respective influences. Figure 1 provides an overview of the relations of ADS with: (1) PDS and its related disciplines (2) Focus dependent disciplines, (3) Domain and (4) Ethics and Regulations.
The main purpose of ADS is to solve real-world problems in a domain using and possibly adapting methods and tools provided by PDS. Depending on the focus of an individual project, other disciplines such as linguistics or psychology may also be relevant.
The connections between Data Science and neighboring disciplines are often represented as a Venn diagram with Data Science as the intersection of other disciplines. This type of visualization incorrectly suggests that Data Science is merely a compilation of existing concepts from surrounding disciplines and does not adequately represent the increasing number of concepts and methods developed within Data Science – or with a more detailed view, within PDS and ADS. The type of diagram chosen in our figure is intended to highlight precisely this fact, which justifies the definition of a (sub)discipline in the first place. Self-developed methods of ADS include re-usable domain-specific strategies for applying data mining techniques, communication processes with decision makers, managing analytical projects, or quality attributes as privacy, explainability, debias, fairness, replicability, accuracy, etc. These aspects could be added to the center of the above figure. However, since a complete enumeration is not possible, we have refrained from doing so.
The domain-independent methods and tools developed by PDS rely heavily on the disciplines of mathematics/statistics, computer science and information science. Depending on the specific application domain, other specific disciplines will provide additional support. Therefore, all these disciplines play an important role in ADS as well.
We illustrate the relationships of the four neighboring areas to ADS in an example. Consider an insurance company that wants to identify fraudulent claims. For this purpose, knowledge from all four neighboring areas of ADS are relevant: From the domain (insurance) the contract-specific insurance terms and conditions, from the focus dependent disciplines (in this example linguistics) relevant methods for text analysis, from Ethics and Regulations the consideration, if social customer features should be included, and from Pure Data Science a suitable classification algorithm. Furthermore, ADSspecific concepts such as explainability and debiasing are relevant.
Having clarified that ADS is a sub-discipline of DS, a natural question is whether ADS also has subdisciplines itself. The formation of an independent sub-discipline of ADS seems appropriate to us if the following criteria are met: (a) The application domain is sufficiently large. (b) There are many questions in the application domain that can be answered with the help of ADS. (c) The questions to be answered make special adjustments and further developments of the ADS methods necessary.
Criteria (a) and (b) are intended to ensure that the application domain under consideration is significant enough to justify the creation of an independent ADS sub-discipline (with a specialized training program, special publications, etc.). Criterion (c) is to ensure that the domain questions cannot be answered using standard ADS methods alone. If that were the case, the creation of a dedicated subdiscipline would not be necessary.
There are a variety of application domains that, according to the above criteria, justify the creation of a separate sub-discipline. Besides the domain they differ in the following aspects: • focus on specific types of data (structured databases, text data, sensor data, audio/video data) • use different subsets of the “PDS toolbox” • adapt general PDS methods to the needs of the domain • may develop domain specific DS methods • may use specialized tools/software that supports domain specific DS methods • take into account domain specific requirements, e.g.
o legal and ethical concerns o data privacy o organizational standards o scientific rigor o explainability of results o fairness/bias o performance
While a complete listing of all domain-specific ADS forms is difficult, a subdivision into the two forms ADS in a scientific domain (e.g., Biological Data Science, Climate Data Science, and in more general sense, Experimental Sciences) and ADS in a non-scientific domain (e.g., Business Data Science, Public Service Data Science, Engineering Data Science) seems useful due to the differences presented below. Table 2 shows some of the differences we consider between ADS in a scientific domain and ADS in a non-scientific domain.
Applied Data Science in a scientific domain Scientific insights Often low Often deep Typically conducted by the domain experts (scientists) themselves
Applied Data Science in a nonscientific domain Pragmatic insights and their usage Often high Mostly medium Typically conducted by Applied Data Scientists
In contrast to the previous considerations, the classification of terms in the context of analytics with a business perspective has already been considered intensively in the literature, but a consensus has not yet been found. We first give an overview of the common usages of the terms Business Analytics (BA) and Business Intelligence (BI). Then we introduce the term Business Data Science and describe its relationship to BA/BI.
The literature reveals significant areas of ambiguity in relation to the demarcations that distinguish
the disciplines and subfields that concern organizational decision making. For example, some authors
still do not distinguish between the terms Business Analytics and Data Science [
A characteristic that is often used in the literature to delimit the terms is the type of statistical
methods used. It is therefore advantageous to first break down the statistics into its three sub-disciplines
[
Analytics has its origins within logic, mathematics, and science. Grammatically, the term Analytics
includes the suffix “-ics”, which refers to a body of knowledge or principles. Nelson defines Analytics
as “the scientific process or discipline of fact-based problem-solving” [
Analytics is a broad discipline that logically includes the sub-fields of BA and Data Science.
Business Analytics is primarily concerned with business relevance and actionable insights from
analyses. This concern is highlighted in a review performed by Phelps and Szabat [
Previously, and to some extent currently, BI has been described in the literature as an umbrella term
that includes the full range of available strategies and technologies that enable data driven decision
making (e.g., [
Taking a business perspective, the term Big Data Analytics has been described as a discipline that
has emerged from BI [
Analogous to the term Business Informatics in Computer Science, we propose to introduce the term
Business Data Science (BDS). In our opinion, BDS is a sub-discipline of ADS, in which the methods
and instruments of ADS are applied to issues in the business domain. Methods and instruments are
adapted to the requirements of the domain and, if necessary, expanded. BDS can therefore be defined
without reference to BI and BA. It is sufficient to concretize a general data science definition (e.g. [
Business Data Science is a field of interdisciplinary expertise in which scientific procedures are used to (semi)automatically generate business insights from conceivably complex data leveraging existing or newly developed analysis methods. The gained business insights are subsequently utilized mainly for decision support, taking into account the effects on society.
The business domain is one of the most important application areas of DS [
The relationship between BI, BA and BDS can be represented as follows. As described above, the aim of BI and BA is to support data-driven decisions in the business domain. In contrast to BI, BA also uses advanced and complex methods of statistics, information science and computer science. Historically, BI is the older discipline that was later supplemented/expanded by BA. BA continues to grow strongly as more and more questions in the business domain are being addressed with increasingly complex methods.
The goals and methods of BA and BDS are very similar and differ more in their point of view than in their content. BA represents more the internal perspective of the business domain. It focuses on the business benefits and sees the DS primarily as an auxiliary science. BDS, on the other hand, represents more of a domain-external perspective, focuses on the correct application of methods and sees the DS as a guiding discipline. BA and BDS tend to develop towards each other, so that the terms can be used synonymously in perspective.
Data Science can be divided into two sub-disciplines: Pure Date Science, in which tools and methods are developed, and Applied Data Science, in which these tools and methods are applied and adapted to practical problems of a specific domain. The focus of this manifesto is on the latter.
We have presented ADS in the context of its four neighboring disciplines: (1) PDS, (2) Focus dependent disciplines, (3) Domain and (4) Ethics & Regulations. It should be emphasized that ADS is more than the intersection of the neighboring disciplines. This aspect was illustrated by an appropriate graphical visualization in Figure 1 and by a concrete example.
Analogous to the subdivision of DS into the subdisciplines PDS and ADS, ADS can be decomposed into numerous domain-specific subdisciplines such as Business Data Science. It is useful – regarding their separating characteristics - to classify these sub-disciplines into two categories: ADS in a scientific domain and ADS in a non-scientific domain by considering the specific challenges ADS aims to solve.
An important sub-discipline of ADS focusing on a non-scientific domain is BDS. Traditionally, the terms BI and BA are often used for data-driven analytical activities in the business context. While BDS can be easily distinguished from BI, there are major methodological similarities to BA. The main difference is the perspective on the problem to be analyzed. Besides BDS, other sub-disciplines of ADS exist such as Public Service Data Science or Engineering Data Science. A closer look at these subdisciplines was not the focus of this manifesto.
The relationships between the various sub-disciplines have been summarized in Figure 2.
Data Science Pure Data Science
Applied Data Science Applied Data Science in a scientific domain
Applied Data Science in a
non-scientific domain Business Data Science …
In the future, to improve the presented classification and to enhance the developed taxonomy in order to better define emerging DS sub-disciplines with all their facets, we plan to extend the survey to a broad group of researchers and practitioners working in all areas of data science.
As a final remark, we believe DS will become even more pervasive and will have applications that we still do not imagine.
[23] Davenport, T., “Beyond unicorns: Educating, classifying, and certifying business data scientists”,
Harvard Data Science Review, 2020. [24] Miah, S. J., Solomonides, I., and Gammack, J. G., “A design-based research approach for developing data-focussed business curricula”, Education and Information Technologies 25.1 (2020): 553–581.