AI for preselection in recruting somethg

Robert Mathias Andrée, BSc
AI for Preselection in Recruiting –A
Dashboard for Recruiters
Master’s Thesis
to be awarded the degree of
Master of Science
in Business Administration
at the University of Graz, Austria
Supervised by
Uni.-Prof. Dr. Stefan Thalmann
Institute for Operations and Information Systems
Graz, December 2020

I
Table of Content
List of Abbreviations..................................................................................................III
Table of Figures ..........................................................................................................IV
List of Tables.................................................................................................................V
1 Introduction......................................................................................................... 1
1.1 Motivation of work ................................................................................... 1
1.2 Definition of Problems & Research Question .......................................... 3
1.3 Working Questions ................................................................................... 4
2 Theoretical Foundations .................................................................................... 5
2.1 Artificial Intelligence (AI) ........................................................................ 5
2.2 Conversational Agents.............................................................................. 7
2.3 Dashboard ................................................................................................. 9
3 Recruiting .......................................................................................................... 11
3.1 Differences between traditional and AI recruiting.................................. 11
3.2 The AI recruitment process..................................................................... 17
3.3 Criteria and requirements for personnel decisions.................................. 22
4 Visual Analytics ................................................................................................ 28
4.1 Design and Data Visualization................................................................ 29
4.2 Visual Concepts ...................................................................................... 34
4.3 Challenges of Visual Analytics............................................................... 44
5 Requirements for a Recruiters Dashboard .................................................... 47
5.1 Legal ....................................................................................................... 47
5.2 Recruiting................................................................................................ 50
5.3 Visual ...................................................................................................... 52
5.4 Catalogue of requirements ...................................................................... 53
6 Development and Analysis of the Dashboard Concept ................................. 55
6.1 Dashboard Design: Colour and Font....................................................... 55
6.2 Dashboard Design: Layout ..................................................................... 57

II
6.3 The Candidate List.................................................................................. 60
6.4 The Candidate Profile ............................................................................. 66
6.5 Concept Evaluation and Design Recommendations............................... 72
7 Discussion .......................................................................................................... 75
8 References.......................................................................................................... 80

III
List of Abbreviations
AI = Artificial Intelligence
CV = Curriculum Vitae
GDPR = General Data Protection Regulation
HR = Human Resources
i.e. = id est, meaning “that is.”
KPI = Key Performance Indicator
KSA = Knowledge, Skills and Abilities
KSAO = Knowledge, Skills, Abilities and Other Characteristics
NLP = Natural Language Processing
NLU = Natural Language Understanding

IV
Table of Figures
Figure 1 Components of artificial intelligence .................................................................... 6
Figure 2 Recruiting Process............................................................................................... 13
Figure 3 Recruiting process including AI and Chatbots.................................................... 21
Figure 4 Statistical Storytelling (Yau, 2013)..................................................................... 28
Figure 5 Bullet Graph (Few, 2008).................................................................................... 32
Figure 6 Advantages and Disadvantages of Visualizations............................................... 34
Figure 7 Report for performance measurements (Wexler et al.)....................................... 35
Figure 8 Ideal Dashboard - Level 1 (Ideal, 2020) ............................................................. 37
Figure 9 Ideal Dashboard - Level 2 (Ideal, 2020) ............................................................. 38
Figure 10 SmartRecruiters Dashboard - Level 1 (SmartRecruiters, 2018) ....................... 39
Figure 11 Smart Recruiters Dashboard - Level 2 (SmartRecruiters, 2018) ...................... 40
Figure 12 Examples of Colour Palettes (Bartram et al., 2017) ......................................... 55
Figure 13 Colour and Font Style of the Dashboard........................................................... 56
Figure 14 Dashboard Layout - Level 1.............................................................................. 57
Figure 15 Dashboard Layout - Layer 2.............................................................................. 58
Figure 16 Dashboard Layout - Level 3.............................................................................. 59
Figure 17 Bar Chart and Bullet Graph............................................................................... 61
Figure 18 Point System and Matching Score .................................................................... 62
Figure 19 Keywords .......................................................................................................... 64
Figure 20 Demographic Data............................................................................................. 65
Figure 21 Dashboard Mock-up - Level 1 – Candidate List............................................... 66
Figure 22 Candidate Profile – Selection and KPI Tabs..................................................... 67
Figure 23 Dashboard Mock-up - Level 2 - Overview ....................................................... 68
Figure 24 Dashboard Mock-up - Level 2 - Personality..................................................... 70
Figure 25 Dashboard Mock-up - Level 3 - Chat Protocol................................................. 71

V
List of Tables
Table 1 Catalogue of Requirements .................................................................................. 54
Table 2 Design recommendations ..................................................................................... 74
Table 3 Implementation of the catalogue of requirements ................................................ 77

Introduction
1
1 Introduction
1.1 Motivation of work
Nowadays, AI-based Information Systems are so widespread, that we probably all have been
confronted with it several times in our private or professional lives (Haenlein and Kaplan,
2019). One out of many definitions describes artificial intelligence (AI) as “[the automatization
of] activities that we associate with human thinking, activities such as decision-making,
problem solving [and] learning” (Bellman, 1978). AI systems are particularly widespread in
private use, even if this is often not apparent at first glance. Platforms like Facebook and
LinkedIn or the digital assistant Siri use different technologies in connection with AI to improve
and widen their business models (Kaplan and Haenlein, 2019; Qi Guo, 2020). If we now focus
on companies and their business functions such as production and human resources (HR), we
see a different picture regarding the usage of AI within business processes. Although a survey
confirms that 85 percent of 3000 executives believe that AI usage could create an advantage for
their companies, only a fifth have integrated AI applications. Furthermore, less than 39 percent
have a strategy regarding AI, which is especially true for small and medium sized companies
(Sam Ransbotham et al., 2017).
One of these AI applications that companies could utilize are chatbots, also called digital
assistant as already mentioned above. Chatbots are a voice or text driven software which
engages in a dialog with humans using natural language (Dale, 2016). A concrete use of a
chatbot within the company would be a digital HR assistant. In practice and theory, there are
already many different use cases in HR for this technology. Ranging from planning and
monitoring the recruitment process, to screening application documents and preselecting
suitable candidates (Nawaz and Gomes, 2020; Black and van Esch, 2020). What the above-
mentioned study already indicates is also confirmed in a further survey, with a special emphasis
on the use of AI systems, such as chatbots, in recruiting.
According to a recent study, that examined the acceptance of AI in personnel decisions, 81
percent of the interviewed recruiters see AI as a major future topic. However, it is striking and
a major finding in this research that there is a contradiction between the generally acknowledged
importance of this topic and the actual use of those systems. As reported by this work, only 37
percent of the sample are familiar with such AI systems and consequently only 10 percent
already used them in their daily work. As respondents stated in this survey, the main reason for

Introduction
2
the sparse use is the fear of losing control over personnel decisions and the lack of trust and
transparency (Hennemannm et al., 2018).
Those two studies show that practical applications of AI systems are not yet strongly integrated
into companies and their workflows, although they are described as advantageous and
promising. Therefore, we can conclude that the usage of AI applications such as chatbots has
high practical relevance and is an important subject for ongoing research. The justification for
further investigations in the field of human resources and in particular regarding personnel
decisions was already briefly mentioned in the study by Hennemannm et al. (2018). First, the
lack of transparency in decision-making and the resulting loss of trust in AI Systems. Second,
the fear of losing control over decisions. Those two issues are also highlighted in a similar way
in further literature on AI and recruiting (Eubanks, 2019; Black and van Esch, 2020;
Michaelides, 2018). In this context, challenges of biased systems and mistrust of data
collections when AI is involved, can be assigned to transparency issues. Also, the question
when to take back control of the AI system if the chatbot starts making bad decisions is
discussed frequently.
The present master thesis is embedded in the framework of AI, chatbots and recruitment, which
were just discussed above. On the one hand, the use of AI within the recruiting process, which
is supported by chatbots. In this context especially the preselection of applicants and data
visualization in a dashboard are in focus. On the other hand, we discovered two main challenges
when recruiters work with AI applications. To get those practical obstacles under control, the
central subject of the master thesis is to use the graphical user interface of the gathered chatbot
data as leverage. This interface, also known as dashboard, must be visualized in a way that
gives recruiters the necessary transparency and control to trust the system.
Although dashboards are commonly used these days, most dashboards fail to communicate with
its users effectively. This is the case when due to complicated graphics or poorly designed
navigation, it is made difficult to get the desired information from the dashboard. The
requirements for a data visualization concept therefore differ greatly with regards to the target
group of the application. Also, the person creating a dashboard must understand the power of
visual perception and design principles to implement information in a way which is aligned
with how recruiters see and think (Andrienko et al., 2020; Few, 2006).
However, visualization is not the only important factor. There is also the question of what
requirements a dashboard must meet to be a suitable tool for a recruiter to make personnel
decisions. For example, one requirement for the dashboard could be, that a recruiter has the
possibility to view the unfiltered application documents. In addition, a dashboard for decision

Introduction
3
making must also comply with the requirement coming from the current basic data protection
regulation of the European Union (GDPR), which emphasizes the aspect of transparency again.
Paragraph 22 establishes the right that individuals may not be assessed based on exclusively
automated processing systems, to protect the interests of the data subjects (Vollmer, 2020). This
regulation therefore also affects the design of a dashboard. It ultimately should lead to a
"legibility-by-design system", which means, that the user must be enabled by the design to
understand the functions, impact, consequences and background of a decision (Malgieri and
Comandé, 2017).
To conclude, the aim of the master’s thesis is to propose a dashboard design for personnel
preselection based on a literature review and selected interviews meeting the various
requirements of recruiter, GDPR and visual analytics. To make the design decision
comprehensible, the various criteria are first established from the literature and then presented
in a catalogue of requirements. The dashboard is then developed based on the formulated
criteria. During this process, different display options are compared with each other and
checked for their suitability. The dashboard was developed as a mock-up and is used in a study
in cooperation with the Institute of Psychology of the University of Graz. During this study,
expert interviews were also conducted, where a first version of the dashboard was used. The
implications of this will be included in the concluding discussion.
1.2 Definition of Problems & Research Question
The brief introduction above already highlights some issues related to creating a recruiter’s
dashboard for preselection. The main problems could be categorized into three different groups
which are also the starting point for the catalogue of requirements and therefore will guide
through the master thesis. First, what are indicators by which applicants can be compared and
which applicant data are of importance for recruiters to make a personnel decision. Second, the
different ways in which data can be visualized within a dashboard. Each of these options has
advantages and disadvantages and must be matched to the specific purpose. Third and last, the
dashboard design must comply with the GDPR regulations to ensure that recruiters can use it
in a way, that the decision-making process is reproduceable and therefore transparent. The
following main research question should answer these main issues regarding this master thesis:
What criteria should a dashboard for preselection from data gathered by a chatbot during
the recruiting process fulfil?

Introduction
4
Further, the consecutive working questions serve as a guideline throughout the master thesis,
with the goal to answer the main research question mentioned above.
1.3 Working Questions
Chapter (3) Recruiting
1. What are the differences between traditional and AI supported recruiting?
2. How does a recruiting process with the help of AI and chatbots look like?
3. Which criteria are essential for recruiters to match job profile and applicant in the
preselection stage?
4. What are key indicators to compare applicants?
5. Which amount and depth of information is necessary to make an informed decision?
Chapter (4) Visual Analytics
1. What are the advantages and disadvantages of different data visualizations in
dashboards?
2. What are visual concepts, that a recruiter is enabled to make a personnel decision based
on visual analytics in a dashboard?
3. Which challenges arise when creating a dashboard with the help of visual analytics?
Chapter (5) Requirements
1. How can a dashboard be designed so that it complies with the principle of GDPR and is
a "legibility-by-design system"?
2. Which explicit requirements can be formulated from the implications of recruiting,
visual analytics and GDPR?
Chapter (6) Development and Analysis of the Dashboard Concept
1. How could a dashboard in terms of the research question look like?
2. What are design recommendations based on the catalogue of requirements and the
conducted interview study?

Theoretical Foundations
5
2 Theoretical Foundations
Since there is the possibility that readers of the master thesis are not completely familiar with
the terms used here, this section is presented to clear up any ambiguities. Before going further,
the basic principles and definitions of artificial intelligence, chatbots, and dashboards.
2.1 Artificial Intelligence (AI)
AI has already been briefly defined in the introductory motivation of the present master thesis.
However, to better define the scope of the technology and its applications, the term AI is further
defined in this subsection. For this purpose, additional definitions are considered, as well as the
core idea and rough functionality of the technology. Then the current state of the art is described
with the help of practical examples of use.
The term AI can roughly be divided into two areas. On the one hand it is the research of
intelligent behaviour and how problems can be solved by this. On the other hand, the knowledge
gained from this is used to develop intelligent solutions, which are then converted into
automated software. The core of AI systems therefore is software, with the idea of finding
intelligent and automated solutions. However, this does not involve imitating human patterns
of action, but rather aims to find solutions outside the human sphere of action (Kreutzer and
Sirrenberg, 2019).
Despite the large number of different definitions, similarities can be observed in the message,
which are subsequently divided into two categories. The definition of Bellman chosen in the
introduction can be categorized into thinking processes and human performance (Russell et
al., 2016). Rich and Knight (1991) also describe AI in a similar way, defining it as a field of
research where computers are used to perform tasks that are currently better mastered by
humans. A second category is divided into behaviour and ideal performance or rationality.
Rational in the sense that a system "does the right thing" based on the current state of knowledge
(Russell et al., 2016). Poole et al. (1998) define AI as "[...] the study of the design of intelligent
agents". Winston (1993) goes even further into the idea of the rationality of machines by
describing AI as "computations that make it possible to perceive, reason, and act". A very
precise definition that links both categories above is as follows: Artificial intelligence is the
ability of a machine to perform cognitive tasks that we associate with the human mind. This
includes possibilities for perception as well as the ability to reason, to learn independently and
thus to find solutions to problems independently (Kreutzer and Sirrenberg, 2019).
To make a short conclusion, it can be noted that AI applications are based on software that is
used for problem solving. Furthermore, AI takes over cognitive tasks from humans, learns, and

6
can also act independently and autonomously. These diverse functionalities are ensured by the
various components of an artificial intelligence. As shown in figure 1, AI is only a collective
term for neural networks, machine learning and deep learning.
Figure 1 Components of artificial intelligence
An essential component of artificial intelligence are neural networks. Computer sciences are
trying to create intelligent networks since 1943. The neural network in the human brain, which
is a connection between neurons and also part of the nervous system, serves as a useful
foundation for mathematical models of artificial intelligence (Russell et al., 2016). The unique
feature of these neural networks is, that information is processed in parallel and therefore non-
linear dependencies can also be handled. Neural networks learn these dependencies
independently and store the generated knowledge in the individual nodes by feeding the
network with training data at the beginning. In the course of time, the network becomes
increasingly autonomous and develops further to achieve even better results. Algorithms are
used which are able to learn and improve on their own (Kreutzer and Sirrenberg, 2019). An
algorithm can basically be defined as "clerical procedure which can be applied to any of a
certain class of symbolic inputs and which will eventually yield, for each such input, a
corresponding symbolic output" (Rogers, 2002).
The whole process of learning is called machine learning. Machine learning is further described
in the literature mostly as "computational methods using experience to improve performance or
to make accurate predictions“ (Mohri et al., 2018). A special type of machine learning is the

7
so-called deep learning. The term "deep" refers to the large number of layers and links within
the neural network. Deep learning can therefore process a wider range of data, which often
leads to more accurate results than the conventional approach (Kreutzer and Sirrenberg, 2019).
Artificial intelligence is a cross-sectional technology, which means that, like the Internet, it is
not only used in one industry or specific stage within the value chain. It can be assumed that
sooner or later AI will be widely used at all stages of the value chain and in all sectors of the
economy. The best known current applications are (Russell et al., 2016; Kreutzer and
Sirrenberg, 2019; Lu, 2019):
• Natural Language Processing (NLP)
• Natural Image Processing
• Expert Systems
• Robotics
The use of NLP involves the collection and processing of text and natural language. The
application by Apple's Siri mentioned at the beginning is a good example of this. The next
subsection conversational agents will take a closer look at NLP. The keyword Natural Image
Processing describes applications that focus on the capture, processing, and storage of images.
A concrete example would be in the healthcare sector, where radiology is supported by image
recognition systems. Expert systems are applications which in turn capture, store, and process
information to derive recommendations for action. Such an expert system is used directly for
example in autonomous driving. The last area of application Robotics is already often used in
private homes. Examples of this are the vacuum cleaning or lawn mowing robots. In addition,
intelligent robots can also be found in medicine, industry and the military, where they perform
important tasks such as performing operations (Lu, 2019; Kreutzer and Sirrenberg, 2019).
2.2 Conversational Agents
This subsection is now explicitly devoted to a typical application of AI. A conversational agent,
chatbot or digital assistant is, as already mentioned at the beginning, a software that enables an
intelligent interaction between humans and computers.
The basic idea behind this technology is that a user can quickly and easily get the right answers
and information regarding his or her question. Nevertheless, a chatbot can also be used for other
applications such as entertainment, as a business tool or social factor. However, the chatbot
always acts like an intelligent creature when it communicates using text or speech. This is made
possible by the ongoing developments in AI and machine learning, which make it possible to

8
interpret and understand natural language. NLP is the basic building block, which will now be
examined more closely (Adamopoulou and Moussiades, 2020; Brandtzaeg and Følstad, 2017).
The following application forms of NLP can be distinguished (Kreutzer and Sirrenberg, 2019):
• Speech-to-Text. In this form of application, the spoken word is converted directly into
a digital text. This occurs, for example, when notes are dictated directly into the
smartphone.
• Speech-to-Speech. This form of NLP generates an answer after a voice input, which
means that the chatbot itself generates a voice output here. Applications are for example
the translation of languages or digital assistants like Amazon's Alexa.
• Text-to-Speech. The reading out of digital documents such as e-mails, short messages
and similar content is the focus of this form of application.
• Text-to-Text. The input of text causes in turn an output of text, as is the case with
translation programs such as Google Translate.
As can be clearly seen from the various forms of usage, the core application of NLP applications
consists of speech processing, or in other words, the understanding and output of human speech
in spoken and written form. The specific responsible process within NLP is called Natural
Language Understanding (NLU). The information of the input text or voice note is extracted
and assigned to specific entities to first understand single words, then sentences and finally the
whole context of the text. An entity in this context is a tool to extract the crucial parameters
from the input. Entities can be defined by the developer or the system itself. To illustrate this
in a simple example, we take the request to a virtual assistant to find out what the current
weather in Graz is like. The decisive entities in this request are Graz and weather. This way the
systems tries to fit the content meaning of the input so that the chatbot understands what
response is expected by the user. In the example just mentioned, this would be the current
weather report for Graz. Since every person has an individual written and oral form of
expression, understanding content is therefore also a central challenge for NLP applications.
They must be able to decode the intention of the author, just like the human brain would do.
Language wit, irony, sarcasm or puns are still very problematic and difficult for many AI
systems to solve (Kreutzer and Sirrenberg, 2019; Adamopoulou and Moussiades, 2020;
Chowdhury, 2003).
To classify different types of chatbots, a wider range of parameters are used in the literature.
Adamopoulou and Moussiades (2020) differentiate for example according to the service
provided, the area of knowledge, the objectives, or the construction method. However, it should
be noted that a chatbot is not exclusively belonging to one class or another, but rather the

9
different classes in each chatbot exist to a different extent. Classification by objective for
example, depends on the primary intention or goal of the chatbot, which must be achieved by
interacting with it. Here a differentiation is made between the goals of information,
conversation, and tasks. In the first case, the chatbot is constructed in such a way that
information has been saved in advance or comes from a fixed source and therefore can be passed
on to the user as effectively as possible. An example would be FAQ chatbots. The aim of the
conversation is primarily to intercommunicate with the person and to answer as correctly as
possible. The user should always be given the feeling that he is talking to another person. With
the third goal chatbots are programmed for special tasks. In this context recruiting as a task can
be mentioned. For example the onboarding and applicant communication process could be
supported by a chatbot (Adamopoulou and Moussiades, 2020).
2.3 Dashboard
The term dashboard was introduced long before the first computers were developed. Already
in the 19th century it was used in connection with carriages. At that time, a board served to
protect the driver and passengers of the carriage from dirt and mud. In the course of time, we
have come to know the term also in connection with automobiles. The driver is informed by the
dashboard with data about his speed, engine speed and the condition of the car by indicator
lights. It thus serves as a source of information to ensure that the vehicle is roadworthy. The
term dashboard is also used in a business context. It describes a system which visualizes data
to make strategic decisions based on it. The goal of both modern applications is basically the
same, namely to present data and information compressed in visuals or graphs without
distracting the viewer too much from the actual task. The dashboard therefore serves as a guide
for decision making, although it should be noted that the interpretation of the data displayed on
the dashboard is not always the same and varies depending on the user (Janes et al., 2013). Few
(2006) defines a dashboard as “[…] a visual display of the most important information needed
to achieve one or more objectives; consolidated and arranged on a single screen so the
information can be monitored at a glance”.
Furthermore, two types of features can be distinguished when designing dashboards. Functional
and visual features are defined by Yigitbasioglu and Velcu (2012). Functional features refer to
what a dashboard could do and thus also indirectly affect the visualization. Such functional
features of a dashboard could be planning, performance monitoring, communication, or
performance measurement. Visual features are about the principles of data visualization, or in
other words how to present information effectively and efficiently to a user. In the initial design

10
phase of the dashboard, it is therefore important to ensure that the functional features
correspond to its actual purpose of it. If the fit is not given here, an incomplete presentation of
information can lead to suboptimal decisions (Yigitbasioglu and Velcu, 2012).
An example of a poor fit would be if a dashboard were designed for personnel selection but
lacked the functional features to compare performance between candidates. But even if there is
a fit between the functional features and the purpose, poorly chosen visual features could cause
the dashboard user to be confused or distracted by the presentations, which in turn could lead
to poor decision-making. However, more about this in chapter 4 of this master thesis: Visual
Analytics.
Like the chatbots before, dashboards can be categorized in different ways, depending on
specific variables. According to Few (2006) some of these variables for differentiation in
dashboards could be the function, application area or interactivity. The most common form of
differentiation is certainly by function, according to which a distinction can be made between
strategic, analytical, and operational dashboards. Strategic dashboards are primarily used within
organizations by managers at different hierarchical levels. A well-known example would be the
executive dashboard, which is designed to provide a quick and easy overview of current
performance, forecasts, and goals. In contrast, analytical dashboards require a different
approach to design, as data is less aggregated and therefore more complex to present. It is
therefore necessary that the comparisons made, and graphs displayed contain more contextual
information. Furthermore, it is important to ensure that the dashboard is interactive and provides
opportunities to immerse into the underlying raw data. Dashboards that are designed for
operational use are kept simple in a similar way to strategic ones. For example they are used
for monitoring production processes, where any critical issues need to be presented quickly and
immediately (Few, 2006).
In this theoretical chapter, the term AI was defined in more depth. The central message is that
AI is based on software that is designed to solve problems. Furthermore, AI can take over tasks
that were previously performed by humans and AI is able to learn. A typical application of AI
is the conversational agent. By using NLU and NLP, the chatbot communicates with people
through text and speech. Lastly, the term dashboard was discussed, being a tool for making
decisions. The most relevant information necessary for decision making is therefore visualized
on a single screen. As we will see in the process of the next chapter, AI-supported chatbots and
dashboards find an application in modern recruiting.

Recruiting
11
3 Recruiting
In the following subchapters the focus lies on recruiting and its process. First, a general short
description is given of what is meant by recruiting and which activities and functions fall within
this scope. Afterwards, a part of the first working question of the chapter is taken up and the
changed conditions between traditional and AI recruiting are highlighted. The first subchapter
also describes the traditional recruiting process and the development of digital recruiting. The
further differences between the two processes are answered by the second working question,
by presenting and describing a typical digital recruiting process using AI and chatbots. At the
end of the second subchapter the extensive first working question is summarized and answered
again. The third subchapter will deal with questions related to the topic of personnel selection.
Formulating matching criteria and Key Performance Indicators (KPIs), whereby a comparison
between the requirement profile and the applicant can take place, will be the main objective.
The last point within the chapter Recruiting will be the question how much information is
necessary to make a well-founded personnel decision.
The concept of recruitment is, along with staffing requirements planning and personnel
deployment planning, a function within the area of human resources (HR). Human resources
are also presented as the most important capital of a company, since with the help of suitable
employees, future-oriented action can be taken. Furthermore, it is special characteristics that
organizations need from their employees to remain competitive. Innovative strength, creativity
and the ability to shape the future are just a few of them (Jung, 2017). Finding the right
personnel for these tasks, or even developing the existing ones, are some of the activities that
can be attributed to the entire HR department.
Recruiting is therefore the function of finding new personnel for the company. Recruitment can
also be understood as a process whereby a match is made between the organization and the
individual applicant (Barber, 1998). The explicit task is to find the necessary number of
employees who have the qualities and requirements for the position in demand. The recruiting
staff are then confronted with the task of making a choice between the applicants. After the
selection has been made, the recruiting process concludes with the induction of the new
employee (Jung, 2017).
3.1 Differences between traditional and AI recruiting
When we talk about traditional recruiting, we are referring to processes and approaches that
were used before mid-1990. Up to that time, mostly humans and analogue media were involved
in the function and process of recruiting (Black and van Esch, 2020). Recruiting itself was not

Recruiting
12
yet considered as an established and independent discipline in many organizations. In addition,
the operational functions of a company were not yet so developed that they were divided into
different specialist areas. For this reason, no specially trained people were used for recruiting,
as there was no demand and necessity for them as such. Recruiting was mainly done by
personnel officers or HR generalists, who also covered all other areas of the HR function. The
traditional recruiting is also often described as a “post & pray” strategy (Ullah and Witt, 2018).
In the context of analogue media, this means that companies published their job advertisements
in newspapers and other print media or on so-called job boards. The part of the pray indicates
that companies had little to no influence on the further process after the job advertisement was
published. From today's point of view, such an influence could be that one only wants to address
a specific target group which results from the job description.
One problem that arises from traditional recruiting and this mentioned “post & pray” strategy
is what Black and van Esch (2020) describe in their paper as "the analog reach and richness
frontier". It represents a trade-off between the information content and reach of a job
advertisement. In traditional recruiting, companies have had the opportunity to expand the reach
of the ad by booking additional ads in multiple print media or by increasing the print size of the
ad. However, both options involve additional costs. If a company tries to avoid these additional
costs by limiting the size of the advertisement for example, the information content of the job
advertisement will therefore suffer, which in turn means that fewer potential candidates will be
reached. A high information content is achieved when your own employees advertising job
offers to their friends and family. In this case, however, the reach is again limited, which is then
only restricted to the immediate environment of the employees (Black and van Esch, 2020).
Based on this trade-off, it is now possible to describe very well the framework conditions in
which traditional recruiting is embedded. Firstly, the discipline of recruiting was not yet
developed to that point that there are different specializations withing the HR department such
as the activity of personnel selection. Second, the limitations and trade-off between the reach
and information content of job advertisements in analogue media as discussed above. The third
point which is addressed by Black and van Esch (2020) is the fact that people themselves limit
and influence the traditional recruiting process. What the authors want to emphasize with this
last point becomes clear if one takes a brief look at the recruiting process.
It is a multi-dimensional process which in theory contains a planning and an operational
component. Planning is important because in an annual recruiting process, the rough milestones
for the year must always be set in line with the personnel requirements planning. In addition,

Recruiting
13
during the course of the year, the requirements must be adapted to the specific situation and
individually, so that recruitment can be prepared for well in advance (Ullah and Witt, 2018).
The operational component includes the actual core process of recruiting, as illustrated in the
following figure.
Figure 2 Recruiting Process
The individual steps are now briefly presented:
1. Job Posting: To prepare a job posting, detailed job requirement must first be drawn up.
Those are based on the recruitment requirements planning and needs analysis. The job
requirement is prepared jointly between the respective specialist department and the
personnel department. It is the basis on which the job description and future recruitment
is built. Both parties agree on the identification of factors critical to success, which
describe the goal and purpose of the position, as well as on requirement criteria relevant
to success. This refers in particular to the professional and technical requirement criteria,
the so-called hard skills, as well as the required personality characteristics (soft skills).
For the actual job posting, the job description must also be translated into a language
appropriate to the target group. This step is important, as it determines the success or
failure of a job advertisement (Ullah and Witt, 2018; Schulz, 2014).
2. Targeting: In this step the mix of channels through which the job advertisement should
be distributed is determined. In principle, a company internal job placement should also
be considered, and former employees should be targeted. The preferred channels today
include your own website, recruitment agencies, referral programs, job boards and
professional networking sites such as LinkedIn (Armstrong and Taylor, 2014). Now,
assuming traditional recruiting and analogue media, the possible channels are of course
not as diverse and are mainly limited to newspapers, job boards at employment agencies
and print advertising space. It is generally recommended to use the channel in which the
target group of the job advertisement is found.
3. Screening: Now the recruiter must deal with the application papers received. This means
that the information is checked and sorted based on the formulated job requirements.
The aim for the recruiter is to generate a so-called shortlist consisting of a dedicated
pre-selection of candidates (Armstrong and Taylor, 2014). This list is then used as a first

Recruiting
14
point of reference for recommendations, so that the responsible department can gain an
insight into the list of applicants. Together with the feedback from the department, the
shortlist also serves as a basis for the extended pre-selection and selection process for
interviews (Ullah and Witt, 2018).
4. Selection: “The aim of selection is to assess the suitability of candidates by predicting
the extent to which they will be able to carry out a role successfully” (Armstrong and
Taylor, 2014). In the process, the applicants are compared with the requirements for
hard and soft skills and further assessment criteria are subsequently obtained. The three
classic selection methods in this step are the job interview, references, and application
forms. In addition, further suitability tests and assessment centres can be used as further
bases for evaluation. Now other people are actually involved in the selection process
and the further procedure is not only up to the recruiter, but also to the department (Ullah
and Witt, 2018; Armstrong and Taylor, 2014).
5. Contract: In the last step, the contract negotiations take place as well as the onboarding
of the selected candidate. Furthermore, the last phase also includes the probationary
period of the new employee (Ullah and Witt, 2018).
The traditional recruiting process just described clearly shows that the recruiter is present in
every phase and further makes decisions himself. The fact that the human being influences the
process and could therefore limit the selection is particularly evident in the screening and
selection phase. Ullah and Witt (2018) speak of the so-called gut feeling, which according to
the authors is always present. This is a matter of prejudices or negative characteristics that lead
the recruiter to treat an applicant either favourably or unfavourably based on name or
demographic characteristics. In particular, the job interview popular among companies is the
subject of many studies. The possible distortion and influence in the perception of the recruiter
is often the focus of attention. Below are some of these possible influences, which were
summarized by Judge et al. (2000):
• Non-verbal communication. Smiles and eye contact can influence the interview rating.
• The external appearance. Perceived attractiveness, cosmetics and clothing as factors
influencing the recruiter's rating.
• Negative information during the interview can be weighted more heavily than positive
information.
• Information collected before the interview dominates the recruiter's evaluation. The job
interview only serves to confirm this prejudice.

Recruiting
15
• Similarity effects, with similarities in demographic characteristics between recruiter and
applicant.
An important requirement for the usefulness and validity of such an interview is the use of pre-
structured interviews upfront. According to recent studies for example demographic
characteristics then have only little influence on the recruiter's assessment during the interview
(McCarthy et al., 2010; Judge et al., 2000). Nevertheless, the overall validity of such interviews
is still around 30 percent (Huffcutt et al., 2013). Based on the above-mentioned framework
conditions of traditional recruiting and the problems where people are involved, the
circumstances in recruiting have changed due to increasing digitisation.
Since the late 1990s, three phases of digital recruiting can be identified according to Black and
van Esch (2020):
• The first phase began with the commercial use of the Internet and the emergence of the
first employment websites. The trade-off between information content and reach, as
well as the resulting restrictions, were disrupted from this point on (Black and van Esch,
2020). Through the use of employment websites, companies could now write detailed
job descriptions and make them available to every visitor on the website. The cost of a
job advertisement thus decreased, while the reach and information content increased,
compared to the use of analogue media. The internet also opened the possibility for
companies to create their own websites to present themselves and their job offers, and
to do so on any scale.
• The second phase of digital recruiting began around the turn of the century and was
mainly triggered by two developments. Firstly, the possibility to search for suitable
offers across several online job portals (Black and van Esch, 2020). Websites such as
Indeed.com, which was founded in 2004, search several job websites for the desired job
title. This gave companies and their job listings even greater reach, as the advertisement
could now be found outside their chosen platform. Secondly, the emergence of digital
professional and social networking through platforms such as LinkedIn in 2003 and
Facebook in 2004. LinkedIn is a social network for making new business contacts or
maintaining existing ones. The benefits and uses of such platforms for recruiting are
huge and will be discussed in more detail in the next sub-chapter. In a nutshell, they
offer a place to present your company and to publish information and advertisements.
On the other hand, they offer the opportunity to dive into desired target groups and
communities in order to make one's own job offers more visible within the group.

Recruiting
16
• The third phase is a result of the previous phases, which matured until around the year
2015. The outcome of this process can be described as Digital Recruiting 3.0, starting
with the use of AI within the recruiting process. One of these results is the fact that the
barrier to applying for a job is now very low and companies are almost flooded with
applications via online platforms (Black and van Esch, 2020). Surveys show that on
average around 250 people apply for an advertised position and 80 percent of them use
social media to find out more about the company in advance (Glassdoor, 2015). Johnsen
& Johnsen is a good example of what this means for international companies. According
to a study, the group received over one million applications in 2017 and this with 28,000
advertised positions (Mcilvaine, 2018). However, it is not only the high number of
applications, but also the fact that on average between 75 and 88 percent are not
qualified for the advertised position (Ideal, 2019). This is due to the low effort and entry
barrier involved for applicants in digital recruiting. They do not use up a lot of time or
money when applying to companies via an online portal. It is therefore not surprising
that 52 percent of talent acquisition leaders say that screening from a large pool of
candidates is the most difficult part of the recruiting process these days (Ideal, 2019;
Black and van Esch, 2020). The second notable outcome of the previous recruitment
phases is that the importance of the recruiter profession has increased. Compared to
traditional recruiting, the human resources function is now divided into specialised
areas. This is due to the fact that companies have recognised the importance to achieve
a fit between job and applicant (Black and van Esch, 2020).
Due to the developments of the last decades and digitalisation it can be stated that the use
of AI systems in recruiting has become a necessity. On the one hand, to cope with the mass
of applications and on the other hand, in order not to let well qualified talents from the pool
of applicants remain undiscovered. The framework conditions of traditional recruiting
mentioned at the beginning have changed during digitisation and differ greatly from current
AI supported recruiting tools. Today there is no longer a trade-off between reach and
information content, nor is there any restriction as in the case of analogue media.
Furthermore, professional recruiting is now a specialisation that can no longer simply be
taken over by HR generalists. To what extent the general conditions around the human
factor have changed and what a recruiting process supported by AI looks like will be
discussed in the next subchapter.

Recruiting
17
3.2 The AI recruitment process
Looking at the traditional recruiting process from the previous chapter, at first glance there are
many possible applications for AI-based solutions. In each of these five phases, there could be
added value for the recruiter as well as for the applicant with the help of a meaningful AI
application. Basically, it can be said that the use of AI System has changed the recruiting
industry permanently (Upadhyay and Khandelwal, 2018). Particularly repetitive activities
performed by humans are now taken over by chatbots or other AI tools. As a result, certain
steps of the traditional process have become obsolete for humans, they are now delegated by
recruiters to recruiting management systems and are thus executed fully automated (Upadhyay
and Khandelwal, 2018; Verhoeven, 2020). „AI is changing the roles that the recruiters play and
is leading to more thoughtful hiring. With AI taking care of boring and repetitive tasks,
recruiters now have more time to be creative and can focus on strategic issues” (Upadhyay and
Khandelwal, 2018). However, in order not to lose the focus of this thesis, the following
considerations will refer to the process phases of targeting, screening, and selection and to what
extent digitization, AI and chatbots are now present here.
1. Targeting
Targeting as in its traditional form is now no longer limited to the selection of suitable channels.
With the progress of digitalization and the emergence of social networks and large job
platforms, it is now possible to precisely identify target groups. This means that their data
regarding education, professional career and other qualifications is stored in huge databases and
is directly accessible to recruiters. A distinction can be made between direct and broad-based
targeting via online channels. In both options, AI systems are used, which are either supportive
or completely autonomous (Verhoeven, 2020).
External sourcing is the counterpart of the “post & pray” strategy previously mentioned and
represents direct targeting. The recruiter becomes active himself and tries to find a suitable
match for the advertised job and then presents the offer to the candidate. In this context, a
supporting AI system is entrusted with the active search in databases. The recruiter defines job
title, skills, and qualifications. The AI system then searches the database for the desired
requirements and selects suitable candidates, which the recruiter then addresses specifically.
The AI in the background learns from the actual selection of the recruiter and thus improves
the algorithm used for sourcing. An independent AI system in sourcing, on the other hand,
works completely autonomously and handles the actual pre-selection from the database, as well
as the subsequent communication with the candidate, through the use of chatbots (Verlinden,

Recruiting
18
2019). The performance of an autonomous AI sourcing system compared to a human sourcing
expert is only slightly worse, but the AI takes only 3.2 seconds, compared to the expert, who
spends between 4 and 25 hours (Eubanks, 2019).
Concerning broader targeting, the new options that have been developed because of digitisation
have already been briefly addressed. More recent AI applications generate a huge pool of
candidates by extracting data from LinkedIn, Xing, Instagram, Twitter, Facebook, job boards
and internal databases. User profiles are then created with the help of this large amount of data
(Black and van Esch, 2020). The system makes no difference, however, whether a person is
currently actively looking for a job or not. In the next step the AI matches suitable candidates
with the job profile. Based on the user profiles created, the AI system also recognizes which
channel a specific candidate should best be addressed through. There are various options such
as advertisements, banners, e-mail, or push messages to increase the chances of success. Some
AI applications go even further and even personalize the wording of the advertisement. (Black
and van Esch, 2020).
As already mentioned, chatbots are also increasingly used for targeting. In practice, for
example, this could mean that the chatbot sends a push message when a potential candidate
looks at a job ad on the company's homepage. The chatbot takes over the initial contact and the
collection of information about the potential candidate. If the candidate shows interest, the
application documents and exact user data of the applicant can be recorded during the interview.
A chatbot based on NLP could also ask questions about the application documents. The chatbot
supported by AI can therefore be seen as a front end for recruiting (Schikora et al., 2020;
Upadhyay and Khandelwal, 2018).
2. Screening and Pre-selection
In the area of screening and pre-selection of candidates, like targeting, much has changed in
recent years. The pre-selection is now already partly done during the targeting process, which
indicates in certain cases that these two phases are increasingly overlapping. As described
before, some AI systems only contact potential candidates in a specific target group. The
incoming candidates are thus already pre-selected to a certain degree from an entire pool of
people. In the case of screening, recruitment management systems play a major role, which then
sort the incoming applications using AI matching tools (Ullah and Witt, 2018). Such tools can
be chatbots that analyse and evaluate CVs and other application documents with the help of
NLU. The sentence structures and words used in the documents serve in turn as a basis for
evaluating the entire content of the application and comparing it with the requirement profile.

Recruiting
19
Furthermore, Chatbots can perform short tests and assessments, which are also included in the
evaluation of the candidate by the AI matching tool (Verhoeven, 2020; Black and van Esch,
2020). An AI and chatbot matching process could work in practice as follows (Schikora et al.,
2020; Ullah and Witt, 2018):
1. A chatbot analyses all information’s gathered from application documents and social
media profiles.
2. The AI tries to find a match between the candidate and the requirement profile.
3. In case of missing information, the chatbot will get in contact with the candidate again.
4. If all information’s are available, candidates can now be sorted by a matching score.
The chatbot then arranges further interviews or cancels people with a too low score.
5. AI analyses the language and micro expressions of the video interview and updates the
matching score based on the newly generated information.
As this exemplary screening process suggests, the human recruiter could already be completely
replaced in these phases. One reason that would support the use of AI, is the fact that algorithms
make more fair judgements in recruiting compared to humans. Human misjudgements that are
characterized by appearance, similarity effects or prejudices against race, gender and age can
therefore be minimized (Herrmann, 2016). Assuming the AI is programmed not to take these
factors into account. As with targeting, AI applications and chatbots are designed to reduce the
workload of the recruiter and provide additional features. The mentioned flood of candidates
nowadays is screened with the help of AI matching tools and pre-sorted to a compact selection.
A well-known example is Unilever, which conducted an experiment in 68 countries with a total
of 250,000 candidates. The input for the AI matching system was the candidate's LinkedIn
profile, several neuroscience games integrated into chatbots and a video interview. After the
best candidates had been pre-screened by the algorithm, the human HR management was
involved just for the final selection (Feloni, 2017).
3. Selection
In the traditional recruiting process, the selection phase is the stage where interviews and other
tests are conducted with the candidates. However, if this process is supported by AI and chatbot,
it is difficult to identify where a clear difference between the pre-selection and selection
processes can be made. Since usually all available data are already evaluated and analysed in
the screening process, this part as performed in traditional recruiting seems not necessary
anymore. Depending on the recruitment management tool used, even assessments and
interviews are carried out in advance of the actual selection, which is why this part can also be

Recruiting
20
partially included in the screening process. Meanwhile, there are also tools that create
personality profiles based on click and like behaviour on social media platforms. Theoretically,
this means that in this case it’s even no longer necessary to conduct an interview with the
candidate (Ullah and Witt, 2018; Crystalknows.com, 2020).
As the example of Unilever already pointed out, the essential steps of traditional selection are
already integrated into the screening process to create a sound matching. To what extent and at
which stage the human recruiter is then involved depends on the specific case. At Unilever, for
example, the most promising candidates were invited to spend a day with the recruiter at their
possible new workplace after the screening process (Feloni, 2017). As in the case of Unilever,
it is possible that newer approaches to getting to know a candidate personally are now used in
the selection process, since the typical process of assessment or interviews is delegated to the
AI. However, interviews, tests, and the assessment centre, which are typically conducted by
human recruiters, are still part of a recruiter’s everyday life and have their validity. In practice,
according to Ullah and Witt (2018) hardly any recruiting management tool for selection has the
necessary degree of maturity to be used effectively in the recruiting process.
After first explaining the traditional recruiting process and its framework and the role of AI and
chatbots in recruiting, the essential differences can be worked out to answer the first working
question.
• The framework conditions in which recruiting is embedded have changed
fundamentally since the 1990s. Digitalization and network effects have eliminated any
limitations in reach and information content that were previously present in analog
media. Furthermore, the increasing workload due to digitalization meant that HR had to
specialize more and more. The final point of the framework conditions was the human
factor, which could influence personnel decisions in traditional recruiting, either
consciously or unconsciously. This point is also changing more and more, as decisions
are delegated to AI algorithms, which claim to be free of prejudice and objective
judgement.
• Other differences can be seen in the recruiting process itself. In particular, the areas of
targeting, screening and selection were highlighted. In the traditional process, clear
boundaries can be drawn between the individual tasks and activities, which are also
performed by a human recruiter. In contrast to this, the recruiter's different tasks
disappear when chatbots and AI are involved. Recruiting management systems which
utilize AI and chatbots take over the part of targeting, screening, and parts of the

Recruiting
21
selection through an integrated process. Even though these are fundamentally separable
processes, the individual activities overlap so heavily when they are performed by AI
and chatbots. This is probably due to the fact that data is needed as a basis for all three
processes. Once this data is available, it is no problem for AI to approach a candidate,
screen his documents and create a matching score for the selection. Recruiting
management systems can be so advanced that the human Recruiter is only needed as the
last instance for onboarding. Or, as in the case of Unilever, recruiters adopt new methods
in the final phase of selection.
To answer the question how a recruiting process with the help of AI and chatbots could look
like, the following figure 3 is used.
Figure 3 Recruiting process including AI and Chatbots

Recruiting
22
The process shown above is intended to illustrate the various links between the individual
phases of recruiting when AI and chatbots are involved. On the one hand, the fact that different
data sources are tapped from the very beginning. This data is essential for modern recruiting
because it is the basis for pre-selecting and contacting suitable persons. The dataset is then
expanded with information provided by the candidate or generated by interviews, tests, or other
assessments. The goal is that the AI is finally able to evaluate the candidates in reference to the
requirement profile. During these phases, the chatbot has various tasks, but the central point is
the constant communication with the candidate. Establishing the first contact and collecting
application documents are tasks of the first phase of the targeting. After that, chatbots can also
ask questions, generate personality profiles with the help of AI, conduct interviews and analyse
the candidate's texts.
3.3 Criteria and requirements for personnel decisions
In order to be able to design a dashboard for personnel decisions in the further course of this
work, the most important factors that are necessary to make such a decision must first be
identified. This subchapter will first deal with the requirement profile, which was mentioned
already several times. This profile is an important part for pre-selection phase. Next, it is
essential to understand the concept of job suitability, which is closely related to the requirement
profile. Furthermore, which methods for personnel selection are frequently used in practice.
From these considerations, the matching criteria for the pre-selection can then be derived, as
well as key indicators to make candidates comparable. Finally, it will also be important for the
design of the dashboard to find out to what extent the information must be prepared.
1. Requirement Profile
A requirement profile describes which criteria a candidate should fulfil to qualify for the open
position. In the traditional recruiting process, this profile standardizes the requirements between
specialist and personnel departments. Furthermore, direct superiors, current job holders and
other employees can serve as informants to finalize the requirements. It is therefore the basis
for a fair selection process, as it contains the most important matching criteria to achieve a fit
between candidate, job requirements, team- and organizational culture. A concrete requirement
profile is also helpful for the candidate himself in order to be able to assess his personal
suitability for the position in advance, before even applying for the desired job (Weuster, 2012).

Recruiting
23
The most important factors which a requirement profile according to Schulz (2014) should
contain, are the following:
• Critical success factors - are those which are directly related to the position to be filled.
First and foremost, this would be the goal and purpose of the position. Thereby the work
goal and the main benefit of the job for the company should be explained. Secondly, the
functional tasks. Questions about the concrete tasks, task distribution, requirements and
specialist knowledge are central in this context (Schulz, 2014).
• Relevant Success factors - hard and soft skills are determined in the analysis. Hard
skills are all knowledge and skills that a candidate should bring with him/her from
his/her previous professional experience. Ideally, these characteristics are formulated in
very general terms since the expertise of the individual job descriptions is very
divergent. Examples of hard skills include leadership skills, languages or the skill of
planning and controlling. When defining soft skills, a wide range of terms are often used
to describe the requirements of the position. This is done by using lists and then putting
the terms together. However, one should be sparing here and not use more than five
terms. Examples for soft skills are patience, courage, loyalty, adaptability, or eloquence
(Schulz, 2014).
• Personality requirements - reflect the behaviours required for the job vacancy, as well
as for cooperation between colleagues or more generally, in the organization. Frequently
mentioned personality factors are the willingness to learn, ability to work in a team,
communication skills, adaptability, and the ability to assert oneself (Schulz, 2014).
2. Job suitability
Job suitability, […] also known as a recruitment or job specification, defines the knowledge,
skills, and abilities (KSAs) required to carry out the role […]”(Armstrong and Taylor, 2014).
Furthermore, job suitability also describes how high the probability of a person being suitable
for a specific professional area of work is based on the defined KSAs. In order to actually carry
out a comparison, it is first necessary to analyse the job itself. Further the role this position
plays within the company and what is needed from the candidate to fulfil it. In the following
steps, the potential candidate is then evaluated against this job and role profile (Schuler, 2013).
The job analysis is, according to Brannick et al. (2012) a process that reveals the job description
and job specification of a job. The difference between these two terms is that the job description
refers to the work performed. The specification, on the other hand, focuses on the worker. For
the analysis two descriptors can be identified. Firstly, the work activities and secondly the

Recruiting
24
worker attributes. The central task of the job analysis is to identify the tasks to be performed to
uncover the work activities. „Tasks are often grouped into meaningful collections called duties
when the tasks serve a common goal” (Brannick et al., 2012). In the case of worker attributes,
the already mentioned KSAs are the focus of the analysis. Next, the individual indicators of the
KSAs are discussed:
• Knowledge – refers, among other things, to the individual requirements for factual,
conceptual, and procedural knowledge to perform the role of the position. In this
context, domain-specific expertise or expert knowledge may also be mentioned.
Furthermore, it is generally considered an advantage, although independent of the
domain, to have a broader form of general knowledge and interests. This is why this
area of knowledge is also relevant here (Brannick et al., 2012; Armstrong and Taylor,
2014; Hunter et al., 2012).
• Skills and Abilities - are closely related to process know-how and describe what skills
and previous knowledge are required to complete a job. Furthermore, the necessary
technical skills should also be available in connection with the technical knowledge. In
more concrete terms, this means that it is necessary to have a solid understanding of the
practical and physical abilities required to apply the knowledge. Abilities include
intelligence, thinking in divergent and analytical ways, and associative skills. For
example, all of these abilities just mentioned are needed for the skill creativity
(Armstrong and Taylor, 2014; Hunter et al., 2012).
The basic concept of key indicators around the KSA could be extended by the term other
characteristics. The so-called KSAOs contain the following additional features according to
Armstrong and Taylor (2014):
• Behavioural competencies – includes the candidate's behaviour and therefore also his
or her personality to fulfil the new role in the company. Ideally, the personality and
behavioural requirements demanded from the candidate should also take values and
culture of the company into account.
• Qualifications and training – are all courses, certificates, or school qualifications that a
candidate should ideally have acquired and passed through.
• Experience – or Achievements, which someone should show to assess the probability
that the candidate is still ambitious in the future.

Recruiting
25
• Specific demand – everything a candidate should ideally achieve from a professional
point of view in the future in his future job. In the concrete case this can be an increase
in productivity or the penetration of new markets.
• Special requirements – can be a requirement for willingness to travel, mobility or
similar.
Brannick et al. (2012) describe the entire KSAOs as relevant and critical for personnel
decisions. „The logic of the psychology of personnel selection is (1) to identify those KSAOs
that are important for the performance of a job, (2) to select those KSAOs that are needed when
the new hire begins work, and which are practical and cost effective to measure, (3) to measure
applicants on the KSAOs, and (4) to use the measurements thus gathered in a systematic way
to select the best people” (Brannick et al., 2012). As (1) already points out, the selection from
the KSAO requirements should not be oversized, since this could discourage potential
candidates. Furthermore, it is often questionable how relevant the KSAOs as a whole really are
for the job in question, which varies in each specific case. Basically, it can be said that one
should select those KSAOs that are important for the job description and secondly, for which
there are suitable procedures available. Meaning that in the concrete case there the KSAOs can
be measured and evaluated (Brannick et al., 2012; Armstrong and Taylor, 2014). The result of
a job analysis should therefore consist of the most important KSAOs and tasks that best describe
the open position.
3. Methods for personnel selection
The three main approaches and methods for testing job suitability are presented now. The use
of these three selection procedures makes it possible to identify different facets of a person. In
addition, the measurement of a feature, using different methods, ensures the reliability of the
measured values (Schuler, 2013).
1. Biographical approach - is the simplest form of diagnostics. It derives information from
past performance and behaviour. This includes all school, work and training certificates,
internships, and foreign assignments. If appropriate, hobbies and relevant interests and
knowledge can also be included. The goal behind this methodology is to draw
conclusions from past experiences that are relevant for the future career path. The more
similar the past and future job are, the more likely it is that performance and behaviour
can be predicted. To the selection procedures, which are to be credited to this approach
belong the application documents such as the curriculum vitae, questionnaires or a
biographic interview (Schuler, 2013).

Recruiting
26
2. Simulations – this approach does not focus on past achievements as they could be taken
from the biographical approach. But rather those that can be provided at the present
time. When we talk about the present, this may well mean that samples of work are
provided on site. The goal is to confront the candidate with the tasks and activities of
the future job. The simulation approach therefore includes work samples, situational
interviews, and situational judgment tests. In such tests, the candidate is described a
situation with alternative actions, which is close to the future daily work routine
(Schuler, 2013).
3. Characteristics approach - at the heart of this methodology is the term potential.
Psychological tests are used with the aim of discovering possible potential that lies
untapped in the candidates. The measurement allows to identify general job-related
characteristics as well as specific skills that are often required for technical jobs.
Psychological tests are standardized and the characteristics can be formulated in
quantitative values, which allows the comparability between the individual and an ideal
value (Schuler, 2013).
Before the last question of the subchapter is examined, two further working questions can now
be answered. The question regarding the criteria for a match between job profile and candidate
can be found in the concept of the job requirements. The criteria found there was divided into
three requirement factors. Firstly, critical success factors, which look for a match in the
functional tasks and the goal and purpose of the position. Secondly, relevant success factors,
which are separated into hard and soft skills. The third factor are personality requirements. The
personality factors of the candidate should be matched with those required for the job and with
those of the organization's culture. In the pre-selection phase, it is therefore important to obtain
the necessary information using the job suitability measurement methods presented above and
then compare this information with the criteria of the requirement profile. The question of KPIs
can be best illustrated using the job suitability approach. These basically represent an
aggregated form of the requirement profile and can be divided into the so-called KSA attributes
of a candidate. Knowledge, Skills and Abilities can thus be described as the KPIs that make
candidates comparable.
At the end of this theoretical consideration of recruiting, the question arises as to how much
and in what depth data is necessary to make a well-founded personnel decision. The fact that
information is missing or wrong when making decisions is an omnipresent problem in every
day decision making. This is also true for recruiting and pre-selection. The problem behind this

Recruiting
27
is that, as Jagacinski (1991) in his study shows, candidates with missing information were
judged at a disadvantage compared to those with complete information available (Jagacinski,
1991). In addition, the vague formulation of requirement profiles or the lack of individual
criteria can lead to stereotypical views of the recruiter influencing the decision (Weuster, 2012).
It can therefore be concluded that when making a personnel decision, all desired information
from the requirement profile should be collected for the purpose of the pre-selection, as this
forms the basis for the further recruiting process. This is particularly important, as this is the
only way to ensure a fair pre-selection and assessment. The depth to which the information
must be contained depends likewise on the requirements of the position to the candidates.
Furthermore, the information of the requirement profile must also be complete, otherwise
distortions can occur. It will therefore be necessary to take steps to update the information in
the event of missing or obviously incorrect information from the candidate.
The Recruiting chapter showed the evolution of traditional recruiting to a digital process
supported by AI and chatbots. These modern recruiting management systems can take over the
targeting, screening, and parts of the selection process. This means that many activities that
were done by humans in traditional recruiting are now done by AI and chatbots. Examples of
this are conducting interviews, collecting data and constant communication with the applicant.
However, the final selection process is still within the human recruiter's area of expertise.
Therefore, the selection criteria for such a personnel decision were researched. As a first tool,
the concept of the requirement profile was introduced. Furthermore, the concept of job
suitability, according to which KPIs are defined to make candidates comparable. Based on these
two concepts, the most important data needed for personnel selection are knowledge, skills, and
abilities as well as the personality of the candidate. To make the collected data and selection
criteria visible to the human recruiter, it is essential to visualize them within a dashboard. The
next chapter therefore deals with the topic of visual analytics, the second topic in which
important requirements for the dashboard developed in this thesis can be found.

Visual Analytics
28
4 Visual Analytics
This chapter will first describe the term Visual Analytics. Afterwards the basic design concepts
of visualization will be developed. This provides the foundation for the first working question,
which deals with the various advantages and disadvantages of visualization options in
dashboards. Afterwards, visualization concepts from literature and practice will be examined.
The focus of this practical consideration is to get an insight into proven concepts. From this,
conclusions can be drawn for the conception of the dashboard developed here. Finally, the
theoretical challenges of Visual Analytics that can occur during the creation of a dashboard are
considered.
Visual Analytics is a more recent term, which is gaining importance due to the increasing
digitalization and the emergence of ever-increasing amount of data. Meanwhile, data is
collected and stored on a large scale in most areas of daily life as well as in professional
environments. The goal is to extract information that will be of beneficial to the user (Cui,
2019). If one breaks down the term visual analytics, this should give further insight to its
meaning. The term visualization can be derived from visual and is often described as a medium
for storytelling. The input for this is numerical data. The resulting output are graphs that
describe the underlying input. The visualizations tell statistical stories, which are based on a
question with a statistical concept behind, to then find the appropriate representation through a
graph (Yau, 2013). The concept of statistical stories as just described is also shown in Figure 4.
Figure 4 Statistical Storytelling (Yau, 2013)

Visual Analytics
29
The term analytics refers to data analysis, which in turn can also be represented by the concept
of data mining. „Data mining represents the work of processing, graphically or numerically,
large amounts of continuous streams of data, with the aim of extracting information useful to
those who possess them” (Azzalini and Scarpa, 2012). In summary, it can be said that „visual
analytics employs interactive visualization to integrate human judgment into algorithmic data-
analysis processes” (Cui, 2019). „To be more precise, visual analytics is an iterative process
that involves information gathering, data pre-processing, knowledge representation, interaction
and decision making. The ultimate goal is to gain insight in the problem at hand [..]” (Keim et
al., 2008).
In the further course of this discussion, however, not the whole process of visual analytics will
be considered. In particular, we will look for possible display options that are suitable for
visualizing data on a dashboard for pre-selection in recruiting. To make the dashboard
representations understandable and readable for the user, it is also essential to comply with the
most important design standards. These are generally applicable standards that are not only
used for dashboards, but also for reports and presentations.
4.1 Design and Data Visualization
The chosen design is a key challenge when designing graphs and diagrams. Decisions on fonts
and font size, colours, backgrounds, axis labels or grid lines influence the readability and
understandability of the graph. Depending on the design, the message of the representation may
vary at first glance, which can lead to misinterpretations. It is therefore important to consider
the following design standards according to Sosulski (2019) when designing diagrams and
graphs, as well as for the dashboard itself:
1. Format – The file type and resolution of the graphic varies depending on the target
media. For web-based displays, such as the recruiting dashboard, the resolution of each
element should be at least 150ppi (pixels per inch). Furthermore, it is recommended to
use vector graphics (SVG file type), which has the advantage that the graphics are
displayed without errors even on different mobile devices (Sosulski, 2019).
2. Colour – The use of colour in the visualization of data should only be applied very
sparingly. Maintaining a high data-ink ratio has a positive effect on the perception of
the visualizations. This means that colour should only be used to differentiate for
example between two categories in a bar chart or to highlight individual data points in
a line chart. A low data-ink ratio or non-data-ink, on the other hand, would be if the
above-mentioned bar chart had an additional background colour that is not backed by

Visual Analytics
30
any actual data. Therefore, a decorative design of the dashboard as well as the
representations contained within should be avoided. Furthermore, the colours to be used
should be chosen carefully, paying attention to their effect, and meaning. It would
therefore not be advantageous to display the data points in a chart with positive
development, with red, since this is generally perceived as a warning signal or warning
colour. The last important point is that the charts and the dashboard should be consistent
and repetitive, otherwise the readability of the data points will be reduced (Yigitbasioglu
and Velcu, 2012; Sosulski, 2019).
3. Text and Labels – One of the central ideas behind visual analytics is not that the display
should look particularly aesthetic. Rather, the goal is that the presented image should
convey a message or information to express a certain state of facts. For this purpose, it
is also necessary to create a context with text and labels in addition to the graphic
visualization. This ensures that the key idea of the representation reaches the viewer.
However, in the specific case it must be weighed up which type of labelling and to
which extent this is done. As with colour, the display should not be overloaded with
labels and text. With text and labels in connection with graphs and diagrams, it is more
a question of putting the information relevant to the user into context (Sosulski, 2019;
Wilke, 2019).
4. Readability – This especially applies to the font- type, -size, -direction and -colour,
which influence the readability of the visualization and therefore the dashboard. When
labelling diagrams and charts, one should refrain from placing them vertically or at a
certain angle to the axis. Furthermore, the use of italic or bold fonts within graphs should
be avoided. Good readability is the most important factor, so distracting colours and
fonts or too small/large text should not be used (Sosulski, 2019).
5. Keep it Clean – “The interior decoration of graphics generates a lot of ink that does not
tell the viewer anything new. The purpose of decoration varies—to make the graphic
appear more scientific and precise, to enliven the display, to give the designer an
opportunity to exercise artistic skills. Regardless of its cause, it is all non—data—ink
or redundant data—ink, and it is often chartjunk” (Tufte, 2001). Simply put, this means
that non-essential and non-data elements should be removed from the representations.
Some examples are the grid of a line chart or a 3D shadow of a pie chart or bar chart.
6. Density – A common error in visualization is that too much data is shown within one
display. The density in this context therefore is the amount of data shown in a single
visualization. A line chart consisting of ten different lines where the data points overlap

Visual Analytics
31
would be an example of too much data density. The visualization becomes difficult to
read and the key message is more difficult to transmit (Sosulski, 2019). Here, the right
balance must be found for the respective visualization, which may vary in a specific
case.
A recurring theme in the field of visual analytics is that the target group working with the charts
or with a dashboard is taken into account by the design. Furthermore, the purpose behind the
application needs to be considered. In the case of this master thesis, the target group consists of
recruiters, with the purpose of pre-selecting candidates. Therefore, it is necessary to choose
forms of presentation and concepts that are not completely foreign to the target group. Which
basic forms of presentation exist and what are their advantages and disadvantages are, will be
considered in the following.
To choose the right visualization, the data on which it is based must be examined more closely,
because not all forms of visualization are suitable for every data type. The data can have the
following characteristics: „categorical, univariate (a single variable), multivariate (more than
one variable), geospatial, time series, network and text“ (Sosulski, 2019). Data types suitable
for the Recruiters Dashboard and their display formats are now presented.
• Categorical Data – This data type includes the distinction between different attributes,
so it is non-numeric data. The required representations are mainly intended to illustrate
differences between attributes or categories so that comparisons can be made. For this
purpose, the various bar charts and the bullet graph are particularly suitable. The second
is a further development of the typical bar chart. Figure 5 illustrates the possibilities of
the bullet graph. How is the performance in education compared to the requirement
profile? - is a possible question in connection with recruiting, which can be answered
by this visualization. The advantage of bullet graphs in general is the easy readability
for the human eye because the ends of the bars are compared visually. Thus, the smallest
and largest items are easily recognizable, which is described as the most effective way
to compare categories. In particular, the horizontal bar chart reflects how people process
information on the screen. To take advantage of this, the labels must be positioned to
the left of each bar. One advantage of using the bullet graph is the higher information
content. Therefore, it is good for tracking performance or goals. A disadvantage with
bar charts is that labels or additional information is mandatory, otherwise it is not
readable. Furthermore, the diagram could become unclear if there are too many different
categories in one visual, i.e., bars. It is also important to make sure that a baseline (0-

Visual Analytics
32
line) is included in the diagrams, otherwise interpretation errors could occur. With bullet
charts, the amount of additional information could be distracting (Nussbaumer Knaflic,
2015; Wexler et al; Few, 2008).
Figure 5 Bullet Graph (Few, 2008)
• Simple Text and Numbers – This is not about displaying large amounts of text. Rather,
it is about extracting and displaying keywords from existing texts or documents. An
example in terms of recruiting would be to extract the most important keywords from
the CV. Furthermore, aggregated values such as the total score of a candidate could be
displayed as a whole number to set a standpoint. The representation of text and numbers
can therefore be used to describe a sentiment or situation, as well as to represent
frequencies. Possible questions in the context of the recruiting dashboard could be -
What is the highest educational level of the potential candidate? Furthermore, the
question regarding a matching score. The first question could again be answered by
displaying a keyword. The second question with the help of a number in percent. The
advantage of this method of presentation is that if keywords and individual numbers are
used sparingly, they create a clear point. Furthermore, it is a shortened and easily
readable form of communication. A last advantage is the simplicity of the creation of
this visualization. Disadvantages can occur if too many keywords or numbers are used,
as this can become very confusing - the purpose of this form of visualization is then lost.
A requirement that could become a disadvantage is the mandatory requirement of a
context. Numbers and keywords are difficult to interpret if the context of the
visualization is not clearly visible (Nussbaumer Knaflic, 2015; Sosulski, 2019; Sun et
al., 2013).

Visual Analytics
33
• Univariate Data – A variable serves as a basis for the representation of frequencies and
a value range, which represents the population. These diagrams give a deeper insight
into the data set by reading the maximum and minimum. Furthermore, the median,
frequency, and outliers. This is implemented with histograms and density charts. In
terms of recruiting, the question could be where a single candidate lies in comparison
to all candidates. The advantage of these representations is that it is very easy to
recognize around which value the data is concentrated. Further that statistical values
like the median can be visualized. A density plot also has less noise in comparison to a
histogram, as well as the possibility to display several distributions in one diagram. A
disadvantage, however, is that it is difficult to implement for smaller data sets, because
otherwise the appearance and therefore the readability suffers (Sosulski, 2019;
datavizcatalogue.com, 2019).
• Multivariate Data – These data focus on the representation of several variables. In a
diagram, comparisons can be made between the individual characteristics and a target
variable. A radar chart is the first choice for this task. A question in the context of
recruiting would be - How does the candidate perform in comparison to the requirement
profile. As highlighted in the previous chapter, several variables are crucial in this
comparison, which is why the radar chart would be an option. The advantage of a radar
chart is that you can see at a quick glance which factors perform well or poorly. This in
turn means that space can be saved when several variables are displayed in one chart.
In addition, outliers are easy to recognize and multiple radar charts can be displayed
together - which makes comparison easier. It can become however fast unclear, which
represents a disadvantage. Too many charts or too many variables within one visual, let
the diagram overflow and therefore it is difficult to read. (Nussbaumer Knaflic, 2015;
Sosulski, 2019).
The following figure shall summarize the working question assigned to this subchapter before
the next chapter deals with the visualization concepts.

Visual Analytics
34
In addition, there are some display formats that are not suitable for visual analytics and therefore
are not suitable for a dashboard. These are generally all 3D-presentations, pie charts and circle
charts. A 3D representation can lead to optical illusions. Furthermore, the human eye has
difficulty comparing angles, areas, and arc lengths. The literature therefore advises against such
representations (Sosulski, 2019; Nussbaumer Knaflic, 2015; Wilke, 2019).
4.2 Visual Concepts
Now that the basic forms of visualization and design standards for a dashboard have been
developed, three visualization concepts will be examined in detail. Therefore, two dashboards
and one report are analysed below. Here, the purpose of the visualization and the design are in
the focus. The declared goal is to identify design elements and interactions that are necessary
for the conception of a dashboard for personnel decisions to enable the recruiter to make
comparisons and evaluate the performance of a candidate. The first report is assigned to the
area of performance measurement, which makes sense in the context of recruiting, since it also
compares the performance of candidates with the requirements profile. The other two are an
Figure 6 Advantages and Disadvantages of Visualizations

Visual Analytics
35
insight into the practical design of a dashboard which is explicitly designed for personnel
decisions in the context of recruiting. At the end of this chapter, the state of the art in recruiting
dashboards is also briefly discussed.
Report - Measurement of Performance
The first example is a report (Figure 7) in which selected KPIs are compared against target
values.
Figure 7 Report for performance measurements (Wexler et al.)

Visual Analytics
36
The goal and purpose of this report is to provide a brief overview of the company's most
important KPIs. The "big picture" is to be communicated to conclude from the information
whether the company is on track with its goals. Furthermore, the dashboard shows which KPIs
are performing particularly poorly, so that any adjustments can be made. Therefore, the target
audience of such a report are decision makers who are usually part of the upper management of
a company. In Figure 7, three different forms of representation are shown. One is text and
figures under the item Key Insights. In the title of the dashboard, the most important points are
briefly listed in advance to provide a starting point, which also corresponds to the intention of
the presentation of text and figures. The other display formats are inspired by Few (2008) Bullet
Graphs, which are an effective tool for tracking performance. As seen in the figure above, the
revenue KPIs are aligned horizontally and the TV and social media ratings are displayed
vertically. An important and necessary element of the report is the colours and labels used. The
labels and the legend are necessary to correctly interpret the KPIs in the form of bullet graphs.
In the case of horizontal graphs, the dashed line is to be a first milestone, which indicates that
the goal (solid vertical line) will soon be reached. Furthermore, the lengths of the bars are
normalized with respect to the target and independent of the monetary value displayed. The
progress of the KPIs can therefore also be compared with each other. The colours used in the
horizontal bullet graph also indicate the progress. Blue indicates a very positive development;
Gray is close to being on pace; orange shows the worst performers. In the vertical bullet graphs,
there is only a dotted finish line, therefore they differ to the horizontal ones. Furthermore, there
are labels in the form of icons, text, and numbers to create a context. A critical point to note is
that the vertical bullet graphs do not use the same colour code and formatting of the bullet graph
as the horizontal ones, which shows that the design is not consistent. Furthermore, it can be
questioned whether it is necessary to separate the bars by colour, as the performance is
illustrated by the bullet graph. Reference can be made to the concept of the data-ink ratio, as
well as to the already mentioned keyword chartjunk. Another suggestion for improvement
would be to place the labels of the horizontal graphs to the left of the bar, which would increase
readability for the human eye.
Dashboard -Personnel Decision
The following screenshots are examples of dashboards from the practical everyday life of a
recruiter who already have an AI software solution integrated. The first dashboard was created
by Ideal and is an AI-powered talent screening and matching system that helps enterprise teams
make more accurate, fair, and efficient talent decisions (Ideal, 2020).

Visual Analytics
37
Figure 8 Ideal Dashboard - Level 1 (Ideal, 2020)
The figure above shows a dashboard as it is used for recruiting in practice. The target group to
be addressed is therefore clearly the recruiter. However, the purpose of the various
visualisations depends on the level at which you are currently located within the dashboard.
The first level is used to give the recruiter an initial overview of all candidates. This level of
Ideal's dashboard is shown in Figure 8. In addition, there is the possibility to filter the results.
The different candidates are displayed horizontally in a list. Only text and numbers can be
identified as the first level representation forms in this dashboard. The candidates are graded
using the American grading system (A, B, C, D, F). Further information such as the location
and the last employer of the candidate is displayed in the form of keywords. Furthermore, the
individual notes are also graded by colour. The meaning of the respective colours is added by
a legend. Here again, one can critically point out the data-ink-ratio. In this context, it is
important to check whether the colour design of the notes has an additional use, or whether this
was done for purely aesthetic reasons. With the help of the legend, it would also be possible for
the dashboard to do this without the assignment of notes and the candidates would be evaluated
only by the colours. A further critical note when awarding notes would be that the recruiter
cannot tell here to what extent two candidates with the same note differ. This is no longer
possible due to the high degree of aggregation of this form of representation. Possible
alternative forms of presentation would be a score or a bar chart as shown in the previous

Visual Analytics
38
performance report. The next figure goes one step further and shows the second level of the
dashboard, where a candidate is presented in detail.
Figure 9 Ideal Dashboard - Level 2 (Ideal, 2020)
For this purpose, four KPIs were identified here which are decisive for the grading at the first
level. The four indicators are: job fit, skills match, resume quality and screening questions.
These, in turn, are also individually evaluated here using the grading system again. In addition,
keywords are used to highlight the most important matches within those individual indicators.
A normal distribution curve (univariate data) highlights that the candidate is in the top 20
percent of all applicants - which also gives an insight into the distribution of the sample of
applicants. Furthermore, like the performance report, icons are used, which are explained by a
short text. In this case, three extraordinary characteristics are displayed, such as the fact that the
candidate has worked for a top company. On the lower half of the dashboard, the candidate's
work experience and education are then presented in the form of text. This is very similar to a
traditional CV. Another feature at this level is the selection of the candidate. The recruiter can
either give the candidate a thumbs up or a thumbs down. As with the normal distribution,
attention was paid here to the labelling to create a context. Now some points that could be
questioned regarding visual analytics. As you can see, the four KPIs, like the grades on the first
level, are coloured differently. In contrast, however, no legend has been included here. This
would not be a problem in principle if the colouring is consistent. But the combination of colour
and note on the second level does not correspond to that of the first level. It is therefore not

Visual Analytics
39
consistent, and the colours can be irritating. The keywords to the right of the KPIs are also
separated by colour. Here one could assume that they are assigned to the respective KPIs based
on the colour. However, it is not obvious. It could also be criticised that using too many
keywords in a small area can be confusing and distract from the actual purpose of this display
option. In this case, ten keywords were used which at first glance are difficult to assign to the
KPIs.
The next dashboard to be analysed is that of the hiring platform SmartRecruiters. Figure 10
shows the first level of the dashboard.
Figure 10 SmartRecruiters Dashboard - Level 1 (SmartRecruiters, 2018)
Like Ideal's dashboard, the purpose and goal of this view is to provide a rough overview of the
candidate list. There are also similarities in the presentation. For example, person-specific
information such as the place of residence and last position is again displayed as keywords
under the name. SmartRecruiters also uses a picture of the candidates and indicates the source
of the application. Each candidate is evaluated with the help of a matching score (text and
number). Furthermore, there is a kind of "circular bar" around the score, which also shows the
score as a percentage. To the right of the score there might be further information about the
candidate in the form of keywords, but unfortunately this area is greyed out in this screenshot.

Visual Analytics
40
The colour of the dashboard is very consistent and simple, but you can see a colour gradation
if the score is below 80. This could indicate quality criteria into which the candidates are divided
by the colour gradations. In summary, the first level dashboards of Ideal and SmartRecruiters
are very similar and showing the same information about the candidates. But regarding the
display option used for evaluating there are different approaches. At SmartRecruiters first level
the recruiter can already differentiate between candidates by using a score and (probably)
keywords. Compared to Ideal's dashboard, the degree of aggregation here is much lower,
despite the use of a score. The following figure shows the SmartRecruiters Dashboard in its
second level.
Figure 11 Smart Recruiters Dashboard - Level 2 (SmartRecruiters, 2018)
The visualisation of the second level in the SmartRecruiters dashboard is again for this purpose
to display the applicant profile. Compared to Ideal, only three KPIs are used here. They are
work experience, skills, and education. These are very similar to the concept of the KSA's
presented. The relevant key indicators are neither evaluated nor graded but is represented by
keywords. As the title of this section indicates, this explains the score achieved. The keywords
could have been extracted from a comparison between the requirement profile and the candidate
profile. A further representation, which is also found in the second level, is the again the
matching score. Also, with SmartRecruiters, the candidate is evaluated by the recruiter on the

Visual Analytics
41
second level. This is done here by awarding up to five stars. A point of criticism, which was
also expressed for level 2 of Ideal, is the large number of keywords that are displayed in a small
space.
Based on the three concepts presented, conclusions can now be drawn as to how a recruiter can
make a personnel decision with the help of visual analytics. First, the approaches how
performance can be presented in practice. The performance report as first example used bar
charts and the bullet graph. This allowed progress to be measured using several KPIs to then
display them in comparison to the target value. It is particularly important to put each graph in
context to make it easy to understand. As already mentioned, performance charts can of course
also be used in the context of recruiting. For instance, by using the KSAs as such indicators
with the values out of the requirement profile as a target. The two Recruiting Dashboards are
very similar in their overall conceptual design. In comparison to the performance report, those
for recruiting have several levels to click through, whereby the information density and the
degree of aggregation of the information varies. Below is a summary of the individual
characteristics according to the level of the two dashboards:
• Level 1
o Used to display all candidates who have applied for a job.
o It contains filter options and a short job description.
o Candidates are listed horizontally in rows.
o Information content on the respective candidates is low - it is an aggregated
summary for an overview.
o Candidates are assessed by means of text or number (matching score, grading
system), which is also supported by colour.
o Most important demographic data such as first and last name, place of residence
and last job title are displayed by keywords.
• Level 2
o Represents the profile of a candidate.
o Used to explain and break down the aggregated value from level 1.
o Most important requirements for the candidate are represented by KPIs.
o KPIs are either explained by keywords or again graded and separated by colour.
o Most of the information is represented by text, which leads to a higher density
of information than at the first level.
o Degree of aggregation decreases.
o Evaluation by the recruiter takes place at level 2.

Visual Analytics
42
It can be concluded that visual analytics, i.e., the presentation of aggregated data by using
graphs, is used very sparingly in a recruiting dashboard. Mainly text and numbers in the form
of keywords and aggregated scores are used. However, a clear concept of interaction can be
seen. The recruiter should click through the dashboard and the different levels to evaluate the
individual candidate. Even within the second level it is evident that even more in-depth
information can be called up. In figure 9, for example, a button to call up the candidate's report
card. In figure 10 the tab Interviews and the button "View Resume" indicate that the raw data
can also be accessed here. In the course of this work this will be an important point in the
conception of the dashboard. One may also conclude that a dashboard concept for personnel
decisions in practice is designed to be interactive and explorative. Complicated or unusual
forms of presentation and layouts are not used. By clicking through and exploring each
individual candidate, the recruiter is slowly brought closer to the personnel decision. The actual
assessment by the human recruiter takes place at level 2, i.e., on the personal applicant profile.
From these practical implications it follows that the concept should therefore be designed in
such a way that the personnel decision is outside the aggregated pre-selection and thus partially
independent of it. In concrete terms, this means that it probably should not take place at the
level with the highly aggregated data. This point becomes even clearer when considering the
legal components in chapter 5.1.
Dashboards in Recruiting – State of the Art
In this closing section of this chapter, we will now briefly discuss the state of the art of
dashboards in recruiting. As mentioned in the introduction, the use of AI applications in HR is
not yet established or common. A similar picture emerges in the area of big data and analytics.
„The HR function is lagging behind other functional areas of management in the adoption of
analytics technology and in the analysis of big data“ (Angrave et al., 2016). The fundamental
problem is that people who work in HR have little knowledge of analytics. „A different
approach to HR analytics is needed, which starts with the question of how HR data can be used
to create, capture, leverage and protect value […]. The results of this may then be used to inform
HR practice and to develop meaningful day-to-day metrics, measures and dashboards within
conventional […] analytics packages“ (Angrave et al., 2016).
Consequently, AI applications in conjunction with big data and visual analytics are areas that
are currently only very rarely found in combination within the HR function and consequently
in recruiting. This can also be deduced from the expert interview study conducted during this
master thesis. Out of the ten interviews, only one recruiting expert stated that he had worked

Visual Analytics
43
with an AI system that evaluates and displays CVs and test procedures. However, some of these
experts already work explicitly with dashboards only. These dashboards though are manually
maintained and prepared by the experts interviewed. Unfortunately, the interview did not reveal
which dashboard tool is used. It could therefore be a simple Excel table in which values are
entered and compared by the recruiter. However, criteria for personnel selection are defined
manually depending on the job advertisement. The expert spoke of "success factors" for the
evaluation - similar in content to the KPIs for professional aptitude presented in the chapter on
recruiting. Another interesting statement from these interviews was that companies have the
existing data for the use of an AI-supporting system to then also display it in dashboards.
However, from the interviews, there are two reasons why an AI system in conjunction with
dashboards has not yet been integrated. Firstly, the lack of necessity from the recruiters' point
of view, that they are not confronted with many applications. One expert spoke of around 200
applications, at which point such a system would bring actual relief to the pre-selection process.
Secondly, the lack of will on the part of the company to use an AI system, as this represents an
investment. However, two experts state that they are currently looking into this technology,
because they are of the opinion, that it will be used sooner or later (Expert Interview).
The state of art of dashboards which are powered by AI, Big Data, and Visual Analytics in
recruiting is therefore very diverse. On the one hand, there are offers from companies such as
Ideal and SmartRecruiters, which use AI and dashboards throughout the entire recruiting
process. Here, reference is made once again to the digital recruiting process as shown in Figure
3 on page 21. Especially the indication of the diverse data sources as input, regarding Big Data
and the resulting possibilities for Visual Analytics. A dashboard that is linked to such data
sources is therefore detached from the traditional recruiting process. On the other hand, the
implications from surveys and interviews (Expert Interview; Angrave et al., 2016;
Hennemannm et al., 2018), which suggest that the use of AI and Big Data is not yet far advanced
in HR and recruiting. The dashboards mentioned above by the recruiting experts can be
assigned to the traditional recruiting process with its limitations (see chapter 3.1), as neither AI
nor various data inputs are tapped here. The success factors defined by the recruiter himself and
by hand help him in the selection of personnel, with the dashboard serving as a tool for
visualisation.
It can therefore be concluded that the use of a dashboard is generally state of the art in the
context of recruiting. This is regardless of whether it is part of a traditional or digital recruiting
process. The dashboard itself always is just used as a tool. On the one hand, to visualize the

Visual Analytics
44
most important data, with the aim of aggregating it. On the other hand, it supports then the
decision making.
4.3 Challenges of Visual Analytics
The challenges of visual analytics in designing and creating a dashboard can be traced back to
the statement that visualisation is a medium for storytelling. Nussbaumer Knaflic (2015)
describes this fact as follows: „There is a story in your data. But your tools do not know what
that story is. That’s where it takes you—the analyst or communicator of the information—to
bring that story visually and contextually to life.” In this context, five challenges can be defined,
which also need to be considered when creating a dashboard. To make them understandable,
they will also be reviewed with regard to recruiting and the dashboards presented above. The
following points are based on the challenges of storytelling according to Nussbaumer Knaflic
(2015):
1. Context. As already mentioned, several times, it is particularly important in the field of
visual analytics to convey the context of the presentations. However, it is not only
necessary when creating graphics to make them understandable for the user, but the
designer himself must understand the context of the underlying data. This means that he
must understand the user and his activity, as well as identify the critical data required
for the exercise. In the case of recruiting, the dashboard designer must understand the
recruiting process, identify the most important data and decision criteria, and consider
what would be suitable forms of presentation for HR decision-makers. That is why a
not insignificant part of this thesis was dedicated to recruiting and its process
(Nussbaumer Knaflic, 2015).
2. Form of presentation. Chapter 4.1 has already described design standards and the
advantages and disadvantages of different forms of presentation. The designer must now
use this background knowledge to decide which form of presentation is best suited for
data visualisation according to the context. It is not up to the designer alone to choose a
suitable visualisation for him. The representations must serve the purpose and target
group of the dashboard. This can be an evolutionary process, which is accompanied by
feedback and requires adjustments over time. In the case of recruiting, for example, the
two dashboards considered showed slight differences in their design. The challenge is
the fact that there is no universal presentation form or dashboard concept for recruiting.
Rather, it depends on the available data and personnel decision-makers which

Visual Analytics
45
visualisations make sense and are feasible. Whether this statement also applies to the
dashboard designed here will be shown in the final chapter of the master thesis.
3. Avoid chaos. This challenge can also be derived from the design standards in chapter
4.1. "Humans' brains have a finite amount of this mental processing power. As designers
of information, we want to be smart about how we use our audience's brain power"
(Nussbaumer Knaflic, 2015). For this reason, it is necessary to pay attention to two
presented concepts. First the data-ink ratio and second the implications resulting from
the word chartjunk. As stated in all three examples, these two concepts could be used
as a basis for possible suggestions for improvement and therefore avoid chaos. If, for
example, a matching score or grade already results in a rating, the question arises to
what extent an additional colour grading improves this circumstance, or whether it is
merely for aesthetic reasons and thus falls into the chartjunk category.
4. Draw attention. For Visual Analytics to fulfil its purpose, the user must be made aware
of specific details or interactions that will guide him through the dashboard. Elements
can be highlighted in a graphic to draw the viewer's attention to a specific point. For
example, if there is a price increase in the title of a graphic, the increase in the line chart
could also be highlighted in colour to focus attention on the essential. The
SmartRecruiters dashboard's Matching Score is also displayed using a "circular bar" to
draw attention to the percentage X of 100. Level 2 of the Ideal Dashboard uses colour
to highlight the 20th percentile in the normal distribution to draw attention to the fact
that the candidate is among the top 20 percent. The challenge is to ensure that only
elements that are essential to the decision-making process are highlighted.
5. Tell a Story. The final challenge is to decide how and in what order the selected
representations are presented to the target group. More precisely, what does the click
path that the user takes through the dashboard look like and how should it be
experienced. A clear storyline is evident in the recruiting dashboards presented. It leads
through at least two levels, which have different degrees of aggregation and information
content. The different levels and tabs are made accessible to the user through interaction
by clicking on them. The main storyline starts with a rough overview of candidates. The
deeper the recruiter clicks into the dashboard, the higher the information content and
the lower the degree of aggregation of the applicant data. The recruiter is thus able to
explore the story of each individual candidate (sub storyline) to make the final personnel
decision (main storyline).

Visual Analytics
46
The chapter Visual Analytics first dealt with different possibilities of visualization, which
depend on the data type of the input. Thereby, the respective advantages and disadvantages
were identified, which will be important in the further course of the conception of the
dashboard. Next, three dashboard concepts from practice were examined. Of particular
importance were the design elements and interaction possibilities which were used to make an
informed decision. Furthermore, the most important characteristics of dashboards as they are
currently used in recruiting were identified. Finally, the challenges of visual analytics were
considered, which are closely related to storytelling. These include creating context, avoiding
chaos, and choosing appropriate forms of presentation. In the next chapter, the implications
from the chapter Recruiting and Visual Analytics are transformed into a catalogue of
requirements. This will also be the basis for answering the main research question.

Requirements for a Recruiters Dashboard
47
5 Requirements for a Recruiters Dashboard
In order to create a dashboard concept, it is first necessary to explicitly emphasise the
requirements, i.e., the criteria for such a concept. For this reason, this chapter is dedicated to
the formulation of these requirements, which are divided into three categories. As already
mentioned in the introductory motivation, one of those requirements are derived from the
implications of the of the General Data Protection Regulation (GDPR). Those will be discussed
in the following subchapter called legal. Afterwards, the requirements from the areas of
recruiting and visual analytics will be formulated. The result of this chapter should be a
catalogue of requirements which contains the central criteria to design a dashboard for
personnel selection in recruiting. These criteria form a framework and guideline on which to
base the design of the dashboard. The catalogue of requirements also provides initial answers
which can be deduced from theory to answer the main research question raised.
5.1 Legal
First of all, it is necessary to understand why it is fundamentally relevant to take the Data
Protection Regulation into account when designing the dashboard. Since 25 May 2018, the data
protection standards included in the regulation have been binding in the respective EU member
states. This applies even if they have not been incorporated into national law. The fundamental
aim of the GDPR is to protect the consumer by stricter regulation of authorities and companies
that process personal data. Without going to deep into detail about individual regulations, the
basic challenges for digital platforms can be summarised as follows: how and which personal
data is collected, stored and further processed (Datenschutz.org, 2017; Mackay, 2017). In the
context of a recruiting platform which influences the entire recruiting process, all three of the
challenges mentioned above are to be considered critical. However, since this master thesis
focuses exclusively on the conception of a dashboard, the areas of collecting and storing can be
neglected here. As a reminder, the dashboard designed here is subject to the assumption that a
chatbot supported by AI is accessing personal candidate data. This data is then analysed and
aggregated to be displayed on the dashboard. In this regard the article 22 of the data protection
regulation is of particular interest. Within this section the GDPR deals with automated
individual decision-making and profiling by formulating the following text: "The data subject
shall have the right not to be subject to a decision based solely on automated processing,
including profiling, which produces legal effects concerning him or her or similarly
significantly affects him or her" (Vollmer, 2020). The term automated processing in decision
making can be used if this takes place without human intervention. Especially if the resulting

48
decision is binding on the individual and thus affects their rights. In such a case, it must
therefore be guaranteed that the data subject is protected (Brkan, 2017). An example of this
would be the AI-supported selection of personnel in a dashboard and thus the decision about
employment for the individual. However, the protection of the individual, regarding automated
processing and decision making, according to Article 22 only needs to be taken if there is no
contract or consent between data subject and data controller. If this is not the case, the following
protective measures for the data subject can be taken according to Roig (2017): „
• Specific information to the data subject.
• The right to obtain human intervention.
• The right to express his or her point of view.
• The right to obtain an explanation of the decision reached.
• The right to challenge the decision.”
These measures can therefore also be identified as broad basic requirements for the whole
process of AI-assisted personnel selection, so that the individual is protected within the meaning
of the data protection regulation. In the context of Article 22, the literature also speaks of
readability in design when algorithms are incorporated in the decision-making process.
Readability in this context can be defined as follows: „[…] making data and analytics
algorithms both transparent and comprehensible to the people the data and processing concerns”
(Mortier et al., 2014). Malgieri and Comandé (2017) further clarify this definition of readability
and state: „legibility of data and analytics algorithms is a concept able to combine
comprehensibility of the functioning of the algorithm […] with transparency about the
commercial use of that algorithm […] in an effective way”. The present master thesis does not
deal with the full scope of the above specifications and thus does not deal with the underlying
functioning of the algorithm of personnel selection. The dashboard is only about the
representations of the output the algorithm (AI) generates. Therefore, only the implementation,
i.e., the commercial use of that algorithm is relevant here. In this specific case, one measure in
particular can be identified that has a direct impact on the design and layout of a recruiting
dashboard (Malgieri and Comandé, 2017):
• “The right to obtain human intervention” – How and through which measures can the
recruiter be authorised to influence the personnel decision in a dashboard?
The context of this requirement is based on the formulation "not to be subject to a decision
based solely on automated processing" from Article 22. For a decision not to be based solely
on automated processing systems, it is necessary that a human being is able to exert influence
(Mendoza and Bygrave, 2017). As this is formulated very vaguely, there are different views

49
and understandings of how this should be interpreted. One would be that the requirement is
sufficiently fulfilled even with minimal human intervention (Wachter et al., 2017). A concrete
example of this would be if a personnel decision is made by a recruiter based only on a score
calculated by an algorithm. Since a human being makes the actual decision, the requirement for
it not to be "solely on automated processing" would be fulfilled. However, the necessity and
usefulness of this decision made by a human being can be questioned. Based on a score, for
example a number between 1 and 100, the algorithm itself could just as well have made the
decision, picking the highest rated candidate.
According to the second view, the requirement "not to be subject to a decision based solely on
automated processing" can only be met if the intervention carried out can be described as
human. The intervention or the decision itself therefor must be considered as relevant.
Furthermore, it should represent a human function which is typically applied when a decision
is taken. In the context of recruiting, this would be the case if the recruiter's personal judgement
is used (Malgieri and Comandé, 2017).
Thus, it can be summarised that the rights and safeguards deriving from Article 22 of the Data
Protection Regulation clearly affect the design of the dashboard in one aspect. The right to
specific information, the right to challenge the decision and the right to receive an explanation
of the decision concern different phases of the AI-supported recruitment process but are outside
the purpose of the dashboard. The dashboard itself serves only as a representation of candidates,
which are collected by a chatbot and aggregated using AI algorithms. The implication from the
data protection regulation relevant for the conception of the dashboard is the selection of
personnel. As this is an act that directly affects the rights of an individual.
The following requirements must be considered when designing the dashboard to ensure that
the right to human intervention is respected and that the recruiter is able to influence the
decision:
1. The dashboard has to enable human intervention for the selection of personnel.
2. The intervention must be considered relevant and thus reflect the function of the human
recruiter.
3. The design of the dashboard should include relevant candidate information to ensure
that the recruiter's judgement and experience is used during the selection.

50
5.2 Recruiting
A closer look at the three requirements resulting from the GDPR reveals a close link to the
practical field of personnel selection. On the one hand, the dashboard must present the relevant
data for personnel decisions and on the other hand, it must also provide space for the function
of a recruiter. This means that the process of screening and selection, as it exists in the
traditional recruiting process, should also be made available to the recruiter via a dashboard.
This ensures that the personnel decision is relevant and has been influenced by human
judgement. Now it must be defined what relevant information is in the context of the personnel
decision. Furthermore, through which concept judgement and the personal experience of the
recruiter are addressed.
To illustrate the screening process in a dashboard, the recruiter must be given the opportunity
to look at the application documents. The recruiter should then be able to compare the
information from the documents with the requirements of the job. At this point reference can
be made to the concept of the requirement profile as discussed in sub-chapter 3.3. Thus, the
first requirement in the field of recruiting is that it is essential to enable a comparison between
applicant information and the requirement profile within the dashboard. This not only serves
the purpose of pre-selecting candidates but is also in line with implication from the GDPR, as
it corresponds to the function of a recruiter. What information such a requirement profile
contains and how detailed it is presented naturally depends on the job in question. However, in
order to provide a starting point for the conception of the dashboard, the factors described by
(Schulz, 2014) can be used, which have already been outlined in sub-section 3.3:
1. Critical success factors – concrete tasks, task distribution, requirements, and specialist
knowledge
2. Relevant success factors – hard skills such as languages and specialist knowledge from
previous jobs and therefore also all professional experience the candidate already has.
Soft skills, to describe the necessary personal and social skills.
3. Personality requirements – personality characteristics and behaviour patterns
The second process from traditional recruiting that can be considered relevant and should
therefore be part of the dashboard is the selection process. The stated goal here is to assess and
highlight the suitability of a candidate. In this context, the concept of job suitability can be
mentioned, as it is supposed to describe the probability of a person being suitable for a job on
the basis of the defined indicators (Schuler, 2013). Especially the evaluation and calculation of
the probability of how well a candidate fits the requirement profile is possible through the

51
collection and aggregation of data. For AI-supported recruiting dashboards, it is no problem in
practice to assign a score to candidates, as SmartRecruiters does, for example. This score is
basically nothing more than the probability of a fit between candidate and job profile. Although
the explicit presentation of a score in terms of the GDPR is questionable, the use of the concept
of job suitability is still a good approach to the selection process. For this purpose, key
indicators are first defined, such as those already presented in section 3.3. Here too, it is
important to consider which indicators are considered relevant in practice in each specific case.
However, the basic indicators consist of knowledge, skills and abilities. (Hunter et al., 2012).
The second requirement from the field of recruiting therefore is the integration and definition
of key indicators. These should empower and support the recruiter to assess the candidate
regarding job suitability. A personnel decision based on such key indicators can thus be
described as relevant, as it represents an intervention by a recruiter that is common in traditional
recruiting.
Further requirements could be defined. First the amount of information to be provided and the
choice of features which describe those indicators. Furthermore, the need to display up-to-date
and correct information to avoid distortions. Lastly, when designing the dashboard and
formulating the indicators, attention should be paid to what data can be collected from a
technical perspective. A chatbot, for example, is preferably suitable for collecting biographical
data. The chatbot is also able to collect data through standardised questionnaires, as is often the
case with personality tests. Data from simulations such as work samples or situational
interviews are therefore difficult to integrate into a dashboard powered by chatbots. The
requirements just mentioned differ in the concrete application case and are depending on
technical possibilities. Therefore they are not taken over as such in the catalogue of
requirements designed here. However, those are certainly critical issues that should be
considered in practice when designing an AI supported recruiting management tool from
scratch.
In summary, two central requirements can be formulated from the field of recruiting for the
requirement catalogue for a dashboard for the pre-selection:
1. Enable the screening process by integrating a comparison between the applicant and the
requirements profile.
2. The selection and evaluation of applicants through the integration of key indicators that
allow to draw conclusions about the applicant's suitability for the job.

52
5.3 Visual
To conclude the requirements for the dashboard, we will now look at the area of visual analytics.
Furthermore, what conclusions can be drawn from those for the design. One of the central
requirements for the visualisation are the design standards as explained in chapter 4.1. These
can be understood as basic rules. These should be applied to the different levels and
representations of the dashboard. The standards that are decisive for the following concept are
now briefly highlighted:
1. Context – To ensure that what is presented is understood by the people using the
dashboard. To guarantee this, the use of text and labels is necessary to create the context
for a presentation.
2. Colour – The basic statement of this design standard is that colour should be used
sparingly when visualising data. Colour should only be used if, for example, a
differentiation is made between different categories. Therefore, care must be taken that
colour is not used as a decorative element. Another point that should not be ignored is
the colour scheme of the dashboard layout. As this provides the framework for data
visualisation, the layout design should also be coordinated. This is necessary because
„we react emotionally as well as cognitively to visual imagery, and those emotions
influence both how we use the information presented to us and how we are affected by
its presence […]“ (Bartram et al., 2017).
3. Readability – To keep the readability of the dashboard high, it is necessary to pay
attention to the font, -size, -direction and -colour. The dashboard should be limited to
one font and should not contain more than three different font sizes (logianalytics.com,
2020). Each label, text or headline is designed to be easy to read for the user of the
dashboard.
4. Simplicity – This aspect combines the concepts of "keep it clean" and "density" as
explained in chapter 4.1. The key message from both concepts could be summarised as
removing any elements that make a presentation difficult to read or overloaded. It is
also important to ensure a certain degree of continuity in layout and presentation.
Another requirement from the field of visual analytics is the concept of statistical storytelling.
As shown in figure 4, behind every visualisation there should be a question, which is then
answered with the help of a statistical concept and the appropriate form of presentation.
Storytelling also ensures interaction with the dashboard by presenting the data exploratively on
several levels and different forms of aggregation. This is also necessary in the context of the

53
legal requirements. Storytelling therefore allows human interventions through the interaction
with the dashboard. Further the recruiter’s judgement can be encouraged by exploratory design.
In summary, two requirements for the catalogue can also be formulated from the field of visual
analytics for the recruiter’s dashboard:
1. Comply with design standards when designing the dashboard and visualisations.
1. An interactive dashboard in the context of statistical storytelling
5.4 Catalogue of requirements
Category Nr. Requirements Implications
Legal
1
The dashboard has to enable human
intervention for the selection of
personnel.
"not to be subject to a decision
based solely on automated
processing"
2
The intervention must be considered
relevant and thus reflect the function
of the human recruiter
The recruiter must be provided
with the same information about
the applicants as in the traditional
recruiting process. Furthermore,
access to the raw data must be
guaranteed.
3
The design of the dashboard should
include relevant candidate
information to ensure that the recruit’s
judgement and experience is used
during the selection
Recruiting
4
Enable the screening process by
integrating a comparison between the
applicant and the requirements profile
The following factors must be
integrated and used to formulate
key indicators: functional tasks,
expertise, hard skills, soft skills,
and personality
5
The selection and evaluation of
applicants through the integration of
key indicators that allow to draw
conclusions about the applicant's
suitability for the job
Visual
Analytics
6
Comply with design standards when
designing the dashboard and
visualisations
Context, colour, readability, and
simplicity

54
7
An interactive dashboard in the
context of statistical storytelling
Enable interactivity and data
exploration across multiple layers
Table 1 Catalogue of Requirements
Table 1 shown above serves to summarize the requirements elaborated in this chapter of the
thesis. The seven requirements found in the literature also serve as an answer to the central
research question raised in this master thesis. Further they are now a guideline for the designing
process of the recruiter’s dashboard for pre-selection. Attention should be paid to the
implications arising from the requirements. These represent the basic statement of the
respective requirement and what this means for the dashboard design. The next chapter will
design the dashboard from scratch and compare different presentation and content concepts
regarding their suitability in the context of the requirements.

Development and Analysis of the Dashboard Concept
55
6 Development and Analysis of the Dashboard Concept
In this part of the thesis, the theoretical input from visual Analytics and recruiting is transferred
step by step into a practical dashboard mock-up. The developed catalogue of requirements
serves as a guideline to check the suitability of the introduced elements. The application
InVision is used to build an interactive prototype. The graphics for the dashboard are created in
Adobe Illustrator and then added to the dashboard as an image file. As a first step, it is necessary
to define the basic layout elements. For that reason, a colour palette and font will be developed
first. Those will then be used for all the other layouts, levels, and graphics within the dashboard.
6.1 Dashboard Design: Colour and Font
“Color palettes play a central role in data visualization where they are frequently used to map
categorical attributes for effective discrimination and identification” (Bartram et al., 2017).
Furthermore, people react emotionally to colours and these emotions in turn influence how we
process information. „Affect matters in visualization for communicative intent, engagement,
and storytelling; there is evidence it supports problem solving” (Bartram et al., 2017).
According to the study by Bartram et al. (2017) colour palettes with light and unsaturated
colours have a calming effect. By using blue, green, and violet shades, trustworthiness can be
conveyed. In contrast, colour palettes with saturated and dark, brown, or red tones stand for
negative emotions. The following illustration is intended to make the differences in the colour
palettes even clearer.
Figure 12 Examples of Colour Palettes (Bartram et al., 2017)

56
As with the choice of colours, the font should also be consistent to achieve good readability on
the one hand, and continuity in the dashboard on the other. There are some basic differences in
the choice of font, which should be considered. Fonts can be divided into two categories. On
the one hand, there are Serif fonts, which are generally used for longer text passages and text
blocks, as here in this master thesis. The second category can be divided into San Serif Fonts.
Since they do not have small strokes, they are purely of the design simpler and clearer as serif
fonts. They are also known as display fonts, i.e. for advertising and for magazine and book titles
(Ali et al., 2013). According to the study conducted by Ali et al. (2013) there is no significant
difference in readability on the computer screen between serif and san serif fonts. They further
state that “from the practical standpoint, the standard practice of using serif and san serif fonts,
namely Verdana and Georgia, for computer screen reading would continue in reading long text
on websites” (Ali et al., 2013). The fonts Verdana and Georgia are also known as web-safe
fonts and belong to the Google Fonts, which are free and can be displayed well in all browsers
and applications (Mike Projkovski, 2018).
The actual choice of colour and font is of course also influenced by the personal preferences of
the designer or the corporate design of a company. In the case of the dashboard, it is the personal
preference of the creator, based on the facts just presented, that takes priority. In the following
figure the colour palette and the font used for the dashboard are presented.
Figure 13 Colour and Font Style of the Dashboard

57
The five colours presented are based on the "CalmGood" colour palette contained in Figure 12
and include blue, green, and purple shades. The first three colours will be used to differentiate
between KPI's. The colours #94FAFF and #6CC4FF are used for the layout design of the
dashboard. The selected colours should be calming and provide confidence to the dashboard
user. Verdana is used as the font for the dashboard as it is one of the most widely used web
fonts and therefore obviously has a high acceptance and readability. The consistent use of the
same colours and font serves the simplicity. Furthermore, an effort was made to implement the
necessary measures regarding colour and readability, as demanded in requirement six.
6.2 Dashboard Design: Layout
The next step is to design the basic layout of the dashboard. As can be seen from the practical
examples of SmartRecruiters and Ideal, recruiting dashboards are structured like a website.
They therefore have a header, body (content) and optionally a sidebar. The standard layout of
a website will also form the basis for the conception of this dashboard. This layout is now shown
in the next figure. The colour and font concept has already been applied.
Figure 14 Dashboard Layout - Level 1
Here the candidate list is explicitly shown, i.e., the first level of the dashboard, which is an
aggregated summary of all candidates. Therefore, an overview page on which the recruiter can
get a first impression of the individual candidates. In the header, the headline was placed
accordingly, as well as a button for filter options. In the left sidebar, buttons have been placed
to navigate between different functional areas of the recruiting software. The body consists of
a grid in which the content, i.e., the candidates, are placed for display.

58
In the context of requirement seven as well as the requirements from the legal area, it is
necessary to design several levels within the dashboard. This should enable storytelling. By
interacting and clicking on individual candidates, the recruiter can get to know each candidate
on the second level which is very similar to the dashboard used by Ideal and SmartRecruiters.
A third level will be used to display the raw data collected by the chatbot. The second and third
levels thus represent the sub storyline in terms of storytelling. The interaction and the process
of explicitly examining individual candidates more intensively represents an intervention and
function which is also carried out by personnel decision-makers in traditional recruiting. By
presenting all information and raw data on different levels, the requirement three in legal is also
fulfilled. The recruiter can thus contribute his or her experience and judgement to the decision-
making process, regardless of the aggregated first level.
As already indicated, the dashboard is therefore structured on three levels. The function of the
first level is the aggregated presentation of all candidates - the main storyline. The second level
will display the individual candidates and their relevant information – the sub storyline. On the
third level the raw data will be made accessible, i.e., in this case the chat history between chatbot
and candidate. The next two illustrations show the layout for the second and third level.
Figure 15 Dashboard Layout - Layer 2
The candidate profile consists of the same header and sidebar elements as level 1. The body
differs, however, by creating a lot of space for the communication of candidate information.

59
Four buttons have been placed to switch between the different display formats and KPI's, which
are defined in the next sub-chapter. The "Chat Protocol" button is intended to take the recruiter
to the third level, i.e., to the raw data. In the next figure you can see the raw data layer, which
is like the layout of the other layers, but in the body of the dashboard you can find a chatbot
conversation.
Figure 16 Dashboard Layout - Level 3
At the end of this sub-chapter, the associated working question of chapter 6 will now be dealt
with, i.e., what a dashboard might look like in terms of the research question. The colour
scheme, font and layout are elements that can differ in the concrete case, since, as mentioned,
personal preferences of the designer and specifications of the company play a role here. It is
therefore not in the sense of the research question to present the chosen colour tones and layout
as the best option. Rather, it is an exemplary attempt at how it could be implemented. However,
it is in the interest of the research question if the basic criteria of the catalogue of requirements
are applied. Regarding the displayed layout (header, sidebar, body) of the three levels, as well
as colours and font, it is the design standards that the dashboard wants to fulfil. The selected
layout concept should not present a challenge to the recruiter and should therefore serve
simplicity, which is why the standard website layout was used. Attention was paid to the colour
scheme and font to ensure that it has a positive effect on the readability and perception of the
viewer - i.e., in line with requirement six. A further point of the design in the context of the

60
research question is to enable interactions. This can be derived from the requirements of the
legal and visual analytics area. By explorative clicking through different levels and candidate
profiles, storytelling is made possible. The recruiter can thus pursue his or her function as a
personnel decision-maker, explore individual personal profiles and view the associated raw
data. This is in line with the GDPR, as it can be considered a relevant human intervention. The
interaction between different levels with different depth of information content thus serves the
requirements one, two, three and seven from the catalogue of requirements.
The next step in the conception of the dashboard is to design the individual levels in terms of
content and visuals.
6.3 The Candidate List
In this section, the candidate list, i.e., the first level of the dashboard, is visualised and its
contents specified. As described, the first level represents a summary of the individual
candidates to give the recruiter an initial overview. In the analysis of the SmartRecruiters and
Ideal dashboards, the following elements were displayed when presenting candidates at the first
level. These elements will also be integrated into the present dashboard concept:
1. The evaluation of the individual candidates
2. Keywords
3. the main demographic data of the candidates
1. The evaluation of the individual candidates
To integrate these elements into the dashboard, it is first necessary to define the KPI's to which
the assessment of the candidates refers. This step is part of the catalogue of requirements and
is anchored in point five. With the help of suitable indicators, the job suitability can be measured
and thus a ranking of the candidates can be achieved. The following indicators result from the
overlaps in content between the concepts of the requirement profile and KSAs, as presented in
chapter 3.3. These will be the key indicators that will be used throughout the dashboard.
• Education. In this indicator all requirements for the candidate are bundled together,
which are related to knowledge, education, schooling, and expertise. Some examples of
factors that could determine this indicator are the level of education, certificates
obtained, educational background, further trainings, languages, or technical knowledge.
• Abilities (and Skills). It contains the knowledge and skills that a candidate should bring
with him/her from his/her previous professional experience. This also complies with the

61
definition of abilities as it was done with the KSA's. Furthermore, it also includes those
points that have been identified as factors relevant and critical to success in the
requirement profile. Examples of such parameters are the concrete main and side tasks
of a job. Such tasks could be expressed in terms of hard skills, such as accounting, online
marketing, controlling, or similar. Furthermore, the professional experience in certain
functions and the associated skills required to perform them. These could be leadership
skills or creativity, i.e., again hard skills depending on the job profile. In summary, the
abilities indicator should bundle all the skills and experience needed to perform the main
and secondary tasks of the job, expressed through hard- and soft skills.
• Personality. The personality of the candidate is not an explicitly mentioned indicator in
the simple concept of the KSAs. However, personality characteristics play a role in the
so-called other characteristics and can be found in the behavioural competencies. In the
concept of the requirement profile, however, personality is one of the three main factors.
For these reasons, personality will be the final indicator for the evaluation of candidates
in this dashboard. It is necessary to assess the candidate for the suitability of his
personality characteristics, whether he fits into the existing organisational culture and
team, and whether his personality matches the requirements of the job.
Now that the three key indicators education, abilities and personality have been defined, to
which the assessment of the candidates refers, a suitable method of presentation on the first
level and the degree of aggregation must be selected. The possible ways of presentation
described in chapter 4.1 with their advantages and disadvantages are a first orientation
guide. Since the indicators are categorical data which are converted into numerical data by
an appropriate evaluation system, two different types of visualisation can be used. On the
one hand, there are different versions of bar charts (point system, classic bar chart and bullet
graph). On the other hand, the numerical data can also be summarised and presented in a
highly aggregated form, as matching scores and thus in the form of text and numbers. Below
are the figures for the options mentioned above to visualise the assessment of the candidates.
The established colour and font scheme were also used.
Figure 17 Bar Chart and Bullet Graph

62
Figure 18 Point System and Matching Score
Before choosing between these different forms of presentation, the fundamental decision on the
degree of aggregation at the first level must be made. This is not just about the visual. Through
the presentation of a numerical evaluation like a matching score, the candidate is directly
evaluated and assessed. However, this assessment is not carried out by the human recruiter. It
is done based on the assessment algorithm (AI) on which the whole recruiting dashboard is built
on. To answer this question, it is necessary to take a closer look again at the legal section of the
requirements catalogue. Malgieri and Comandé (2017) describe the decision on the basis of a
pre-calculated score as "passive human", since it involves little effort, and the first evaluation
has already been carried out automatically. They continue to explain: “[…] we can even
question about the necessary ‘human’ nature of this ‘decision’ (human or monkeys?). […] In
other words, a minimal human intervention without real influence on the outcome of the
decision cannot be sufficient to exclude the applicability of Article 22(1)” (Malgieri and
Comandé, 2017). According to this reasoning, a decision based on a matching score cannot be
considered relevant and does not reflect the function of a human recruiter. In short, the
presentation of a matching score on the first level is not in accordance with the catalogue of
requirements in the first and second point. However, this is not to imply that a matching score
should generally be excluded when designing a recruiting dashboard. If the candidates are
informed and agreed to the use of automated processing systems, a matching score can be
integrated. This assumption has not been made during the conception carried out here. A high
form of aggregation such as the matching score will therefore rejected for the first level of the
dashboard. In the next sections, the less aggregated forms such as bar chart, bullet graph and
the points system will be tested for their suitability.
The bullet graph, as shown in Figure 18, is primarily a tool that is intended to provide more
information than a conventional bar chart. Furthermore, it is often used in performance
measurement, as illustrated in the example report of chapter 4.2. However, two questions arise.
First, whether the bullet graph is suitable and familiar to recruiters. Second, whether this
presentation method meets the design standards of a recruiting dashboard. The bullet chart is

63
not a common and ordinary way of presenting data and this could lead to problems of
readability, especially among groups of people and professions where the handling and
visualisation of data is not of importance. The first question therefor can be answered with a
negative based on feedback and online research on existing recruiting dashboards. The
performance measurement report from chapter 4.2 uses the bullet graph specifically for certain
KPI's, for example to present quarterly figures in comparison to the target figure and to put
them into context. The purpose and target group of the dashboard for personnel selection is
completely different compared to a quarterly report. If ten candidates are displayed at the first
level of the dashboard, this makes a total of 30 bullet graphs (every candidate with each three
KPIs). This is not the intention of bullet graphs and this high number of complex representations
does not serve the simplicity and readability required by the design standards. The bullet graph
is therefore not used for the first level.
Finally, the simple bar chart and the point system are left to choose from. As both options are
very similar in terms of presentation, there are no major differences regarding the advantages
and disadvantages from the point of view of visual analytics. However, it should be emphasised
that the presentation in five points represents an aggregation compared to the conventional bar
chart. The evaluation is done in steps of 20 and 10 (if we assume 100 percent as the total),
whereas in the case of the bar chart the bar shows the exact value. The slight aggregation
through the evaluation in points should make the data easier to read and differentiate between
candidates. The SmartRecruiters dashboard presented also shows a similar representation. The
human recruiter has five stars at his disposal to evaluate the candidate, as it is also common
practice with online portals and shops. It can be concluded that compared to the conventional
bar chart, the representation in five points (or stars) should has a higher acceptance. This fact
could also be taken from an expert interview. He stated that in his company, candidates are
assessed with the help of 5 stars (Expert Interview). Further, due to the many applications on
the internet, this representation should also be known among professional groups that are less
statistically affine. For the above reasons, a five-point system, as shown in Figure 19, is used at
the first level to represent the KPI's and evaluate candidates.
The chosen presentation now meets the criteria of the catalogue of requirements. The five-point
system represents a medium degree of aggregation compared to the matching score and bar
chart. The judgement and experience of a recruiter can now be considered in the selection
process. By presenting a points system and integrating three indicators, it is now not possible
to make a decision based on a purely automated numerical value - as is required by the legal
requirements. By defining indicators consisting of the concepts of requirement profile and job

64
suitability (KSA's), the requirements from the recruiting process were integrated. The recruiter
is thus given aggregated information on the job suitability of the candidate at the first level. The
points can be used to draw conclusions about the extent to which the candidate matches the
requirement profile in terms of education, abilities, and personality. In the presentation itself,
attention was paid to compliance with the criteria from visual analytics. The colours of the KPI's
are those predefined for the dashboard in chapter 6.1. Furthermore, the presentation is labelled
to create a clear context. The aim of the measures taken is to create the context on the one hand,
but also to increase readability. The use of colour is purposeful and serves to differentiate
between the three indicators and will be used throughout the dashboard - this should also serve
simplicity and readability to meet the requirements of the design standards.
2. Keywords
The next element used in the analysed dashboards are keywords. This visualisation will also be
integrated into the present dashboard, although not to the same extent as it was done by Ideal
and SmartRecruiters. The keywords should rather be used in a targeted manner and within a
manageable framework. An advantage of keywords is that points of view, characteristics and
key matchings with the requirement profile can be communicated briefly and clearly. The
present dashboard concept will therefore also use keywords on the first level to integrate a
personal touch for each candidate presented. The recruiter thus receives keyword-based
information that is specifically related to the candidate. In order not to overload the dashboard,
only three keywords are assigned to each candidate. These are related to the formulated KPI's
to obtain relevant personal information about the candidate in all three categories. The
following figure shows examples of different variations for the presentation of the keywords.
Figure 19 Keywords
Of the variations presented, the third is used for the dashboard. Firstly, because the vertical
layout corresponds to the order of the KPI's point system visual. This therefore should increase
readability. Secondly, the colour coding of the keywords should provide the context to the
indicators to create a clear classification. It is necessary to pay attention to these issues to
comply with the catalogue of requirements. Finally, some examples of keywords to make this

65
more tangible for the reader. In the field of education, the catalogue of requirements could
demand a specific educational focus. If a candidate has obtained this, it could be displayed
under the green keyword for Education. The same principle could be applied to the abilities and
personality.
3. The main demographic data of the candidates
The last element of the first level is to display the most important demographic data of the
candidates. The analysed dashboards are quite consistent in their visualisation, as first name
and surname are displayed using text. In addition, the place of residence and the title of the last
job that the candidate held. These points are also adopted in the present dashboard concept.
This is also basic information, which can also be taken from the usual application documents
such as the CV. Therefore, important information which should not be withheld from the
recruiter. However, one point in which the analysed dashboards differ is the presentation of an
application picture. During the investigations, as well as interviews and feedback from experts,
which will be addressed in the final discussion, it became clear that the question of whether an
application photo should be shown is viewed differently. On the one hand, a company from
Graz, which designed a dashboard for personnel selection without picture, name, and gender to
avoid possible prejudices and influence by the recruiter. This approach is legitimate because,
as briefly discussed in this paper, it is a weakness of traditional recruiting that distortions due
to external appearance and demographic characteristics occur (Judge et al., 2000). This view
was also confirmed by an expert in personnel selection at the University of Graz, but she added
that it was not usual in the German-speaking world. While it is quite common in English-
speaking countries to conceal these applicant characteristics in the pre-selection. She therefore
advised, regarding the study to be carried out with the mock-up of the dashboard, to include
demographic characteristics as well as an application photo, as this would be more in line with
common Austrian practice among recruiters. The following figure shows how the person-
specific information is presented.
Figure 20 Demographic Data

66
Now that all three elements have been discussed and the appropriate visualisations have been
selected, these representations can be added to the first level.
Figure 21 Dashboard Mock-up - Level 1 – Candidate List
Figure 21 shows the design of the first level recruiting dashboard designed here. An effort was
made to follow the criteria of the requirements catalogue. The most important points in terms
of answering the work and research question are summarised:
1. No matching score, as otherwise the recruiter's selection of personnel would not be
relevant.
2. The KPI's education, abilities, and personality from the concepts of job suitability and
requirement profile summarised.
3. A five-point system for the presentation of these KPIs - as it is highly readable and
suitable for practical evaluation.
4. Keywords to highlight personalised matches with the requirement profile.
5. Demographic data and application picture to adapt to Austrian common recruiting
practice.
6.4 The Candidate Profile
The second level of the dashboard is the candidate profile, which can be accessed by clicking
on a candidate. At this level, the so-called sub storyline begins by giving the recruiter the
opportunity to take a closer look at the selected candidate. The analysed dashboards of Ideal
and SmartRecruiters show the aggregated values of the first level in more detail. Furthermore,
the actual selection by the recruiter was at this level. What both dashboards basically have in
common is that the degree of aggregation is significantly lower compared to the first level. For
the present dashboard concept this means that the three KPIs are being explained here at the

67
second level. This implies that data is less aggregated here and is intended to clarify the
underlying basis on which the assessment was made. Furthermore, at this level it should be
possible to carry out a comparison between the personal and requirement profiles to fulfil
requirement four. The following figure shows the provisional candidate profile, in which the
individual tabs for the KPIs were integrated.
Figure 22 Candidate Profile – Selection and KPI Tabs
Furthermore, the upper field was used to display personal data. The actual pre-selection is
carried out also in this field. And with reference to the SmartRecruiters dashboard, also in the
form of a five-point system. It is a representation with its advantages and disadvantages as it is
already used on the first level (KPIs). Thus, the reuse serves the consistency within the
dashboards. The decision that the pre-selection is made on level two can also be derived from
requirements one and three. The necessity of human intervention, as well as making the relevant
information visible. Because the actual pre-selection is made on the second level, the recruiter
is "forced" to interact. The design of the dashboard therefore ensures that the recruiter will
inevitably interact with each candidate profile. It also guarantees that the information presented
by the dashboard is accessed. The extent to which the recruiter then incorporates this
information into his or her pre-selection or how long he or she engages with the respective
profiles cannot be controlled in the context of this concept. The dashboard design can, however,
ensure that the recruiter must click on the lower level by default to make a pre-selection. To

68
make the context of the pre-selection understandable, the five-point selection visual is labelled.
Selecting by five points has another advantage. Acceptance or decline of the candidate's
invitation can take various forms. If a recruiter is not sure, he will choose the middle. This
allows the recruiter to use his or her personal judgement even more in the decision.
The next step is to create the overview tab of the candidate profile. As the name suggests, this
tab should give an idea of the most important information of the selected candidate. The
information is presented based on the three KPI's. The aggregated assessment of the first level
is to be explained this way. For example, if a candidate has received 4 points in education, the
overview tab should show which matches between the catalogue of requirements and the
candidate have led to this. The presentation form of text is used for this. The aim is to provide
the recruiter with a format which is familiar to him or her. Thus, education, abilities and
personality are presented similar to a curriculum vitae.
Figure 23 Dashboard Mock-up - Level 2 - Overview
As already defined in Chapter 6.3, information on education and further education can be found
within the education indicator. In the example above, two university degrees and two advanced
courses of further education. In this example these are also the most important matches with
the requirement profile, which is why they are listed here. Within the abilities, various tasks
and skills are presented based on work experience. The abilities and skills shown in italics font
style again represent matches with the requirement profile. In other words, skills that the

69
candidate needs and has already acquired based on his or her work experience. Furthermore,
two abilities that are required to complete the tasks of the job, which the candidate in this
example has (MS Office and SAP software). As this paper pointed out, one of the main tasks
of chatbots is to carry out personality tests or to create a personality profile. It is therefore quite
logical to present the personality of the candidate based on such a test. For this reason, a
standardised personality test was needed, which could be integrated into the dashboard. "The
Big Five framework enjoys considerable support and has become the most widely used and
extensively researched model of personality [...]" (Gosling et al., 2003). These and similar
statements, as well as the positive feedback from experts which have seen the dashboard were
reasons for choosing this test for the dashboard. However, the Big Five personality test will not
be examined or evaluated in detail here, as this is not the focus of this paper.
To evaluate the overview tab of the candidate profile about the research and work question, the
main points of the design are summarised:
1. Context and consistency in design - consistency in KPI's and selection. The KPI's of the
first level are now explained on the second level. Furthermore, again a five-point system
for the selection of the candidate.
2. Text as the primary visualisation tool to communicate KPI's clearly and in a way that
recruiters know from the traditional process.
3. Selection of the candidate on level 2 of the dashboard to allow interaction with each
profile through the design.
4. Due to the low level of aggregation and the presentation of relevant information which
lead to the ranking, the recruiter can bring in his own expertise and judgement. This is
therefore also done at the level where the recruitment decision is made. This can then
be classified as relevant. Finally, the recruiter can express his or her judgement in the
five-point scoring system to reject or invite a candidate.
The three remaining tabs of the candidate profile serve the requirement four, i.e., the need to
create a direct comparison between the requirement- and the personal profile. Each tab
represents a KPI, as shown in figure 23. The last level of the dashboard is accessible via a button
of the candidate profile. In our case it is the chat protocol, which provides access to the raw
data of the individual candidate. This is necessary to meet the implications of requirements two
and three. In the next two sub-chapters, these two elements of the dashboard are discussed.
1. KPI Tabs

70
Which form of presentation is suitable for the comparison of requirement and person profiles
can be determined by means of the data type. In the personality test based on the big-five factors,
five characteristic features can be identified. Furthermore, the KPIs education and abilities also
include several variables. On this basis, a form of presentation is required that represents
multivariate data. In the course of this work, the radar chart was found to be a suitable way to
present such data. In the context of the big-five personality test this is also a common practice
and therefore a familiar form of presentation. Each of the three KPI's will therefore be equipped
with five characteristic values to display the radar chart. These values will then be placed in
reference to the requirement profile. The radar chart will thus contain two polygons, one
representing the candidate and the other the requirements profile for comparison. The following
figure of the personality tab shows how this could look like.
Figure 24 Dashboard Mock-up - Level 2 - Personality
The above chart now shows a comparison between the personality requirements of the
candidate and the results of the fictional big-five personality test, which was conducted by a
chatbot. Attention was paid to creating the context by adding a legend and labels both on the
graphic itself and next to it. The graphic itself fits into the already existing colour scheme to
differentiate between the individual KPI's. With the help of this representation, the recruiter can
now also compare the individual variables that are more important to him personally with the
requirement profile. For the KPI's education and abilities, the variables can be freely chosen in

71
comparison to the standardised personality test. Possible variables in education can be the level
of education, educational specialisation, languages, or additional qualifications required in the
job profile. In the case of abilities, these can be individual skills or application knowledge in
MS Office or Photoshop, or tasks such as controlling, budgeting or similar. Here it is up to the
conception of the chatbot or the recruiting process to define suitable methods to collect the data
needed for the presentation. For example, the candidate could evaluate his or her knowledge
in the required abilities by asking simple questions - which then serves as input for the radar
chart. The above presentation of the radar chart should meet the requirements of the design
standard by paying attention to context and labelling. Furthermore, an attempt was made not to
overload the graphics, which of course is a challenge for a radar chart. Whether it meets the
design standards is therefore a somewhat subjective assessment.
2. Raw Data
The need to include raw data in the dashboard has already been addressed several times, which
is why, for the sake of completeness, a diagram of the last level is also shown below as the
conception of the dashboard is based on the assumption of a chatbot, a chat history is shown as
an example. CVs, certificates, school reports or letters of motivation uploaded by the candidate
during the application process could also be linked. How this looks like in a specific case may
vary, but it is important in the sense of the requirement criteria that raw data is made accessible
in order to meet the requirements of the legal area.
Figure 25 Dashboard Mock-up - Level 3 - Chat Protocol

72
6.5 Concept Evaluation and Design Recommendations
To evaluate the dashboard and to assess the applied set of requirements for their suitability as
an answer to the research question, the dashboard was used in an expert interview study with
recruiters. This study was conducted in the course of the project "The Application of Artificial
Intelligence in Personnel Selection", in which the present master thesis is embedded. A first
version of the dashboard designed here was shown to the participants of this study before they
were asked questions regarding the forms of representation, information content and
requirements of a dashboard. The interviews also contribute to answering the research question
by checking the expert’s statements regarding possible criteria for a dashboard to be used for
pre-selection. If these criteria correspond to those in the literature, this may indicate that the
formulated catalogue of requirements and thus the entire dashboard concept is a practical
prototype. Combining these two perspectives results in design recommendations for the
creation of dashboards for the pre-selection.
To provide an overview and to be able to compare individual statements, the interview excerpts
are classified into three categories. The following statements were made by 10 recruiting
experts from the practice. The first category deals with the question regarding the requirements
of a dashboard for the pre-selection. The second deals with the information content, i.e., what
information they believe such a dashboard should contain. The last category will contain
statements regarding the forms of representation. As the interview was conducted in German,
the following statements will be shortened for better understanding (Expert Interview):
1. What requirements do you think a dashboard for pre-selection should meet?
1.1. It is very important to comply with different legal requirements.
1.2. Transparency and explanation of why the candidate is proposed and displayed as
qualified or not.
1.3. Identifying the most important key figures (max. 5) of a candidate immediately
1.4. The easier to handle the better, getting to the CV with two clicks.
1.5. Clear and not overloaded, as well as intuitive handling.
1.6. Variable in the setting of characteristics.
1.7. Filter options by categories and topics.
1.8. Human intervention should be possible.
1.9. Usability, it should be self-explanatory and user-friendly.
2. What information about candidates should be presented in such a dashboard?

73
2.1. Personality and hard facts.
2.2. Level of education, previous professional experience, last companies, age, and place of
residence.
2.3. Application picture could also be left out.
2.4. Individual statistics.
2.5. Data and inputs for the visuals.
2.6. Curriculum vitae and results of any tests taken.
2.7. Application photos I would welcome.
2.8. Application photos and name are a useful tool.
2.9. A mixture of hard and soft facts.
2.10. How does the candidate's education and qualification match the job
requirements.
2.11. Relevant experience and highest level of education, skills, and expertise.
2.12. The personality in all cases.
3. What forms of representation would you prefer for the data?
3.1. I am not a big fan of network diagrams; I find pie charts and bar charts easier.
3.2. I personally prefer diagrams. With tables and text, I need longer to orientate myself.
3.3. I think that if something can be expressed in numbers, then you should do so.
3.4. Candidates are listed in tabular form, with the possibility of looking at further details.
Little information at first, based on numbers or keywords.
3.5. It all depends on the person, whether you are a person of figures, data, or facts.
The radar type, where you see different personality traits, is very, very important.
A comparison of the expert statements with those of the catalogue of requirements, which could
be derived from the literature research (see table 1 – page 51), reveals a variety of overlapping
points which strengthen the basic concept of the dashboard. The expert statements are assigned
to the three areas of legal, recruiting, and visual analytics. The statements which could be
assigned have thus been confirmed by the literature. They can therefore be understood as design
recommendations for the creation of a dashboard for recruiters.
Catalogue of requirements
(Literature)
Expert Interviews
(Statements)
Legal 1.1, 1.2, 1.4, 1.8, 2.5, 2.6

74
Recruiting 1.3, 2.1, 2.2, 2.9, 2.10, 2.11, 2.12
Visual Analytics 1.4, 1,5, 1.9, 2.4
Table 2 Design recommendations
However, there are some areas where divergent statements and opinions were expressed. In
recruiting, these were the different viewpoints regarding the presentation of the candidate's
demographic characteristics and especially the application photo. Most of the experts stated that
a photo is not important to them in principle, but that they welcome the presence of one. One
expert considered it a useful tool, but one could not rule out that it might lead to biased
decisions. Therefore, no fundamental statement can be made in this master thesis regarding the
use of application pictures, name and gender (Expert Interview). The statements about the
representation options are similarly vague as those about the application photo. No tendencies
can be identified with regard to any form of representation. Most respondents agreed that a mix
of graphics, text and figures was preferred, but no particular form of presentation was
mentioned more than once. However, the radar chart was explicitly mentioned twice and the
opinions of the two experts differed strongly. It can therefore be concluded that, as one expert
put it, the choice of presentation should depend on the people who use the dashboard daily. This
personalised view also fits in with statements 1.6 and 1.7, i.e., the requirement that a dashboard
should be variable. Variable with regard to filtering options and the characteristic values of the
KPIs (Expert Interview). From these implications a final design recommendation for the
dashboard can be formulated. A dashboard for recruiting should be adapted to the specific
persons and situation when choosing the representations and characteristics of the KPIs. Person-
specific regarding the recruiters and the company using the dashboard. The feature values of
the KPIs depend on the situation and the requirement profile of the position to be filled. The
recruiting dashboard designed in this master thesis therefore has no general validity. Rather, it
must be adapted to the circumstances to achieve high usability.
With regard to the research question raised, the catalogue of requirements as an answer, was
confirmed by the evaluation carried out. It leads to the conclusion that a large part of the
requirements mentioned by experts were met by the dashboard concept conducted in this thesis.
These statements from experts predominantly coincide with those from the catalogue of
requirements outlined in chapter five. The design recommendations from this chapter in
combination with the catalogue of requirements provide therefore a rough framework. They are
a guideline which can be used as a reference for the design of a recruiter’s dashboard.

Discussion
75
7 Discussion
The last chapter of this master thesis is dedicated to answering the research question.
Furthermore, a brief discussion of the results from the literature research and interview study is
intended, which resulted in the design recommendations.
1. Answering the research question
The research question speaks of criteria that a dashboard for the pre-selection in a recruiting
process should fulfil. The present master thesis has identified three areas that are critical when
designing a dashboard for recruiting.
First, the area of recruiting, which is moving away from an analog to a digital process in which
chatbots and artificial intelligence increasingly take on different roles. This fact also reflects
the practical necessity of a dashboard in which the collected data is visualised and made
accessible to the recruiter. Furthermore, various application examples of AI and chat bots were
outlined (video interviews, personality tests, neuroscience games). However, the actual
selection of personnel is still a domain that can be attributed to the human recruiter's field of
activity. This was also confirmed by the subsequent analysis of the expert interviews.
Formulating selection criteria and indicators, as well as creating a concept for matching the
candidate and requirement profile are initial criteria for the dashboard.
The visualisation itself is the second area that was considered in the chapter visual analytics.
The usability and suitability of display options were examined and two dashboard and a
performance report from practical experience were analysed in terms of their visualisation.
Furthermore, the concepts of statistical storytelling and design standards were explained, which
form the basis for the dashboard concept. Both concepts were adopted as important criteria
from the field of visual analytics in the requirements catalogue.
The implications resulting from the GDPR and especially article 22 are the last area that
influences the design of the dashboard. Above all, the need for the recruiter to have access to
raw data to make the aggregated visualisation explainable was emphasised. Expandability is
also one of the key requirements resulting from the interviews conducted. Furthermore, the
function and relevance of the recruiter should not be influenced by an automated process of
decision making. It is a requirement that the recruiter's judgement and expertise must be
incorporated into the decision-making process.
The table below illustrates how these criteria have been implemented and considered in the
dashboard concept presented. The seven requirements of the catalogue of requirements
represent the answer to the research question, which was identified based on a literature search.

Discussion
76
Furthermore, the interview study confirmed the relevance and approaches of the catalogue of
requirements. The most important statements of this study were compared with the
requirements and resulted in design recommendations for dashboards in recruiting. These have
also contributed to answering the research question.
Category Requirements Implementation
Recruiting The selection and evaluation of
applicants through the integration
of key indicators that allow to draw
conclusions about the applicant's
suitability for the job
Based on the overlapping concepts of the
requirement profile and KSA's, the indicators
education, abilities and personality were formed.
These reflect the hard- and soft skills, personality
features, as well as functional tasks and expertise
of an applicant.
Recruiting Enable the screening process by
integrating a comparison between
the applicant and the requirements
profile
Within the candidate profile there are three tabs
for the formulated KPIs. A radar chart shows the
individual KPIs with their feature values. The
features of a candidate are combined here with the
requirements from the requirements profile. The
recruiter can identify matches or major deviations
in the most important features at a quick glance.
Visual
Analytics
An interactive dashboard in the
context of statistical storytelling
Statistical storytelling through two storylines that
promote dashboard interaction and involve the
recruiter. On the one hand the main story, which
represents the whole activity of the pre-selection.
The first level represents the starting point. On the
other hand, the individual applicants, who each
form a side story. The recruiter navigates through
tabs on the second level up to the third level. The
pre-selection takes place on the second level,
which also represents the end of the main
storyline.
Visual
Analytics
Comply with design standards
when designing the dashboard and
visualisations
In all visuals, an attempt was made to create a
context by labelling or colour differentiation. The
use of colour was limited to what was necessary.
Five colours were chosen, which have a good and

Discussion
77
calming effect. Furthermore, a font was chosen
which is suitable for screens. In general, care has
been taken to avoid unnecessary graphics or text
to make the dashboard as simple as possible.
Legal The Dashboard hast to enable
human intervention for the
selection of personnel.
No automated or pre-calculated decision through
an overall score. Intervention by the recruiter is
therefore necessary.
Legal The intervention must be
considered relevant and thus reflect
the function of the human recruiter
The intervention takes place on the candidate
profile. Here the recruiter is provided with
information typical for an application process.
Furthermore, a comparison between the candidate
and requirement profile and raw data is available.
Judgement and personal expertise can thus be
incorporated, which is why the personnel decision
can be classified as relevant and corresponds to
the function of a recruiter.
Legal The design of the dashboard should
include relevant candidate
information to ensure that the
recruit’s judgement and experience
is used during the selection
Information on the three KPIs Education,
Abilities and Personality is presented in the form
of text on the candidate profile. These indicators
contain the most important factors for the pre-
selection based on the concept of professional
attitude and requirement profile.
Table 3 Implementation of the catalogue of requirements
2. Discussion
Finally, the results of the dashboard concept must be placed in the context of current literature.
In the process of writing this master's thesis, however, no similar scientific papers were found
that dealt with the conception or analysis of a dashboard for pre-selection. Therefore, the overall
dashboard design and concept cannot be discussed or classified based on existing literature.
Nevertheless, the dashboard was evaluated through an interview study, from which design
recommendations were derived. Based on this data, the requirements catalogue, which is the
answer to the research question, can be described as relevant. It can be concluded that this
combined topic of recruiting and visual analytics still holds possibilities for further scientific
research. This is also reflected in practice. Angrave et al. (2016) state in this context that „many

Discussion
78
in the HR profession do not understand analytics or big data, while analytics teams do not
understand HR.” Further academic work can therefore play a role in the development of suitable
visual analytics tools for AI-supported pre-selection in recruiting.
Although it is not possible to classify the dashboard as a whole, individual elements can be
discussed. In particular, the working questions that helped to formulate the catalogue of
requirements. First, the working question regarding legal competence regarding the Data
Protection Regulation. This discussion has already been held in chapter 5.1, which is why we
will now refrain from taking it up again. Furthermore, the scope of this paper shows that the
central focus of this master’s thesis does not lie in the legal area, but it is a variable that must
be considered.
In the area of recruiting, the criteria and indicators that are necessary to create a match between
the requirements profile and the applicant profile can be discussed. The working questions
around these topics represent the content of the dashboard and are therefore of central
importance for personnel selection. The requirements profile is the central building block for
defining to what extent a candidate is suitable for the advertised position. Suitability is when
the candidate's skills profile matches the requirements profile of the job as closely as possible -
it thus describes criteria that the candidate should fulfil. (Huf, 2020; Schulz, 2014; Weuster,
2012). The KPIs derived from the requirements profile are based on the concept of KSAs. This
is a common framework that is also often used in the literature, for example, to carry out
requirements analyses for emerging job trends (Chang et al., 2019). The KSAs are also a
concept of job suitability and thus a suitable tool for defining KPIs for personnel selection - and
thus also for the present dashboard concept (Armstrong and Taylor, 2014). Therefore, there is
a consensus on the pre-selection indicators formulated here: knowledge, abilities, and
personality. Both the literature and expert interviews confirmed the use of these KPIs. However,
what this master thesis does not cover in terms of content is the critical area of how the data is
obtained. Attention must be paid to proper ethical and legal implementation. In addition, the
quality of the data, which plays a crucial role in staffing decisions, has to be considered.
(Fellner, 2019). This point was also raised in the expert interview. The collection of the
personality profile through a chatbot was viewed critically, as there were concerns about its
legal feasibility (Expert Interview). Fellner (2019) therefore recommends close cooperation
between recruiters, data scientists and lawyers if one wants to benefit from the digital recruiting
process.
The next requirements for the dashboard could be taken from the work questions in the Visual
Analytics chapter. Thus, applying design standards and creating an interactive dashboard that

Discussion
79
includes elements of storytelling. It is difficult to have a discussion on this point, because as the
last design recommendation also pointed out, a dashboard must be adapted to the specific
person and situation. Also, as was mentioned earlier, direct comparison with a pre-selection
dashboard is not possible, as no scientific analysis was found for this. The designed dashboard
can therefore only be classified on the basis of generally formulated criteria. Karami et al.
(2017) have designed seven dashboard criteria in their work. These can be used to assess
whether a dashboard is effective, regardless of the application area. The seven criteria are: „user
customization, knowledge discovery, security, information delivery, alerting, visual design, and
integration and system connectivity“ (Karami et al., 2017). From these points, four common
criteria can be identified at first glance, which have also been integrated into this dashboard.
The first is the need for user customisation, as mentioned last in the design recommendations.
The second consistent criteria are knowledge discovery. By being able to dive deeper into the
data through different layers, down to the raw data. „The dashboard should meet the objectives
that are defined and understood by the users on an ongoing basis. And also, the context of the
contents being displayed in the dashboard should be in clarity“ (Karami et al., 2017). These
were also important considerations in the design of this dashboard and are part of the
information delivery criteria. The last matching criteria is visual design. Here the authors speak
of: „The dashboard should be visually appealing and engaging without overwhelming the users
but make them feel comfortable […]. It is necessary to adopt a concise and minimalist design
in order to avoid overloading the user with information, components, contents, and navigation
steps that are unnecessary“ (Karami et al., 2017). These general criteria according to Karami et
al. (2017) coincide in many points with the concept of the design standards and storytelling
presented here. How "well" or "poorly" the respective elements were implemented in the pre-
selection dashboard in the context of these criteria cannot be assessed within the scope of this
work.

Discussion
80
8 References
Adamopoulou, E. and Moussiades, L. (2020), “An Overview of Chatbot Technology”, in
Maglogiannis, I., Iliadis, L. and Pimenidis, E. (Eds.), Artificial Intelligence Applications
and Innovations, IFIP Advances in Information and Communication Technology, Vol.
584, Springer International Publishing, Cham, pp. 373–383.
Ali, A.Z.M., Wahid, R., Samsudin, K. and Idris, M.Z. (2013), “Reading on the Computer
Screen: Does Font Type has Effects on Web Text Readability?”, International Education
Studies, Vol. 6 No. 3.
Andrienko, G., Andrienko, N., Drucker, S., Fekete, J.-D., Fisher, D., Idreos, S., Kraska, T., Li,
G., Ma, K.-L., Mackinlay, J., Oulasvirta, A., Schreck, T., Schumann, H., Stonebraker, M.,
Auber, D., Bikakis, N., Chrysanthis, P., Sharaf, M. and Papastefanatos, G. (2020), Big
Data Visualization and Analytics: Future Research Challenges and Emerging
Applications.
Angrave, D., Charlwood, A., Kirkpatrick, I., Lawrence, M. and Stuart, M. (2016), “HR and
analytics: why HR is set to fail the big data challenge”, Human Resource Management
Journal, Vol. 26 No. 1, pp. 1–11.
Armstrong, M. and Taylor, S. (2014), Armstrong's handbook of human resource management
practice, 13. ed., Kogan Page, London.
Azzalini, A. and Scarpa, B. (2012), Data Analysis and Data Mining: An Introduction, Oxford
University Press USA, Oxford.
Barber, A.E. (1998), Recruiting employees: Individual and organizational perspectives,
Foundations for organizational science, Sage Publications, Thousand Oaks, Calif.
Bartram, L., Patra, A. and Stone, M. (2017), Affective Color in Visualization.
Bellman, R. (1978), An Introduction to artificial intelligence: Can computers think?, Boyd &
Fraser, San Francisco.
Black, J.S. and van Esch, P. (2020), “AI-enabled recruiting: What is it and how should a
manager use it?”, Business Horizons, Vol. 63 No. 2, pp. 215–226.
Brandtzaeg, P.B. and Følstad, A. (2017), “Why People Use Chatbots”, in Kompatsiaris, Y.,
Cave, J., Satsiou, A., Carle, G., Passani, A., Kontopoulos, E., Diplaris, S. and McMillan,
D. (Eds.), Internet Science: 4th International Conference, INSCI 2017, Thessaloniki,
Greece, November 22-24, 2017, Proceedings, 2017, Cham, Springer International
Publishing, Cham, pp. 377–392.
Brannick, M.T., Cadle, A. and Levine, E.L. (2012), “Job Analysis for Knowledge, Skills,
Abilities, and Other Characteristics, Predictor Measures, and Performance Outcomes”, in

Discussion
81
Schmitt, N. (Ed.), The Oxford handbook of personnel assessment and selection, Oxford
library of psychology, Oxford University Press, New York, pp. 119–146.
Brkan, M. (2017), “Do Algorithms Rule the World? Algorithmic Decision-Making in the
Framework of the GDPR and Beyond”, SSRN Electronic Journal.
Chang, H.-C., Wang, C.-Y. and Hawamdeh, S. (2019), “Emerging trends in data analytics and
knowledge management job market: extending KSA framework”, Journal of Knowledge
Management, Vol. 23 No. 4, pp. 664–686.
Chowdhury, G.G. (2003), “Natural language processing”, Annual Review of Information
Science and Technology, Vol. 37 No. 1, pp. 51–89.
Crystalknows.com (2020), “Hiring”, available at: https://www.crystalknows.com/hiring
(accessed 22 July 2020).
Cui, W. (2019), “Visual Analytics: A Comprehensive Overview”, IEEE Access, Vol. 7, pp.
81555–81573.
Dale, R. (2016), “The return of the chatbots”, Natural Language Engineering, Vol. 22 No. 5,
pp. 811–817.
datavizcatalogue.com (2019), “Density Plot - Learn about this chart and tools to create it”,
available at: https://datavizcatalogue.com/methods/density_plot.html (accessed 31 July
2020).
Datenschutz.org (2017), “EU-Datenschutzgrundverordnung (DSGVO): Verbindliches
Datenschutzrecht für alle!”, Datenschutz, 11 May, available at:
https://www.datenschutz.org/eu-datenschutzgrundverordnung/ (accessed 19 August 2020).
Eubanks, B. (2019), Artificial intelligence for HR: Use AI to support and develop a successful
workforce, Kogan Page, London, United Kingdom.
Expert Interview, Die Anwendung künstlicher Intelligenz in der Personalauswahl:
Anforderungen zur Nachvollziehbarkeit von Entscheidungen und.
Fellner, K. (2019), Moderne Personalauswahl: Renommierte Experten über Trends, neue
Technologien, Chancen und Risiken in der Eignungsdiagnostik, essentials, Springer
Fachmedien Wiesbaden, Wiesbaden.
Feloni, R. (2017), “Consumer goods giant Unilever has been hiring employees using brain
games and artificial intelligence -- and it's a huge success”, Business Insider Australia, 29
June, available at: https://www.businessinsider.com.au/unilever-artificial-intelligence-
hiring-process-2017-6 (accessed 21 July 2020).
Few, S. (2006), Information dashboard design: The effective visual communication of data, 1.
ed., O'Reilly, Beijing.

Discussion
82
Few, S. (2008), “Visual Business Intelligence – Bullet Graphs for Not-to-Exceed Targets”,
available at: https://www.perceptualedge.com/blog/?p=217 (accessed 31 July 2020).
Glassdoor (2015), “50 HR & Recruiting Stats That Make You Think|50 HR & Recruiting
Stats That Make You Think”, Glassdoor, 20 January, available at:
https://www.glassdoor.com/employers/blog/50-hr-recruiting-stats-make-think/ (accessed
20 July 2020).
Gosling, S.D., Rentfrow, P.J. and Swann, W.B. (2003), “A very brief measure of the Big-Five
personality domains”, Journal of Research in Personality, Vol. 37 No. 6, pp. 504–528.
Haenlein, M. and Kaplan, A. (2019), “A Brief History of Artificial Intelligence: On the Past,
Present, and Future of Artificial Intelligence”, California Management Review, Vol. 61
No. 4, pp. 5–14.
Hennemannm, L., Schlegel, L. and Hülskötter, A. (2018), “Künstliche Intelligenz in der
Personalvorauswahl”, available at: https://raven51.de/kuenstliche-intelligenz-in-der-
personalvorauswahl/ (accessed 3 April 2020).
Herrmann, B. (2016), Die Auswahl: Wie eine neue starke Recruiting-Kultur den
Unternehmenserfolg bestimmt, Wiley, s.l.
Huf, S. (2020), Personalmanagement, Studienwissen kompakt, 1st ed. 2020.
Huffcutt, A.I., Culbertson, S.S. and Weyhrauch, W.S. (2013), “Employment Interview
Reliability: New meta-analytic estimates by structure and format”, International Journal
of Selection and Assessment, Vol. 21 No. 3, pp. 264–276.
Hunter, S.T., Cushenbery, L. and Friedrich, T. (2012), “Hiring an innovative workforce: A
necessary yet uniquely challenging endeavor”, Human Resource Management Review,
Vol. 22 No. 4, pp. 303–322.
Ideal (2019), “Resume Screening: A How-To Guide For Recruiters | Ideal”, available at:
https://ideal.com/resume-screening/ (accessed 20 July 2020).
Ideal (2020), “Ideal | AI Powered Screening and Matching”, available at: https://ideal.com/
(accessed 5 August 2020).
Jagacinski, C.M. (1991), “Personnel decision making: The impact of missing information”,
Journal of Applied Psychology, Vol. 76 No. 1, pp. 19–30.
Janes, A., Sillitti, A. and Succi, G. (2013), “Effective dashboard design”, Cutter IT Journal,
Vol. 26, pp. 17–24.
Judge, T.A., Cable, D.M. and Higgins, C.A. (2000), “The Employment Interview: A Review
of Recent Research and Recommendations for Future Research”, Human Resource
Management Review, Vol. 10 No. 4, pp. 383–406.

Discussion
83
Jung, H. (2017), Personalwirtschaft, De Gruyter Studium, 10., aktualisierte Auflage, De
Gruyter Oldenbourg, Berlin, Boston.
Kaplan, A. and Haenlein, M. (2019), “Siri, Siri, in my hand: Who’s the fairest in the land? On
the interpretations, illustrations, and implications of artificial intelligence”, Business
Horizons, Vol. 62 No. 1, pp. 15–25.
Karami, M., Langarizadeh, M. and Fatehi, M. (2017), “Evaluation of Effective Dashboards:
Key Concepts and Criteria”, The open medical informatics journal, Vol. 11, pp. 52–57.
Keim, D.A., Mansmann, F., Schneidewind, J., Thomas, J. and Ziegler, H. (2008), “Visual
Analytics: Scope and Challenges”, in Simoff, S.J., Böhlen, M.H. and Mazeika, A. (Eds.),
Visual Data Mining, Lecture Notes in Computer Science, Vol. 4404, Springer Berlin
Heidelberg, Berlin, Heidelberg, pp. 76–90.
Kreutzer, R.T. and Sirrenberg, M. (2019), Künstliche Intelligenz verstehen: Grundlagen –
Use-Cases – unternehmenseigene KI-Journey.
logianalytics.com (2020), “Dashboard Design Fundamentals - The Definitive Guide to
Dashboard Design”, available at:
https://www.logianalytics.com/dashboarddesignguide/dashboard-design-fundamentals/
(accessed 9 September 2020).
Lu, Y. (2019), “Artificial intelligence: a survey on evolution, models, applications and future
trends”, Journal of Management Analytics, Vol. 6 No. 1, pp. 1–29.
Mackay, D. (2017), “Technology and GDPR: Is your platform ready?”, ITProPortal, 24
October, available at: https://www.itproportal.com/features/technology-and-gdpr-is-your-
platform-ready/ (accessed 19 August 2020).
Malgieri, G. and Comandé, G. (2017), “Why a Right to Legibility of Automated Decision-
Making Exists in the General Data Protection Regulation”, International Data Privacy
Law, Vol. 7 No. 3.
McCarthy, J.M., van Iddekinge, C.H. and Campion, M.A. (2010), “ARE HIGHLY
STRUCTURED JOB INTERVIEWS RESISTANT TO DEMOGRAPHIC SIMILARITY
EFFECTS?”, Personnel Psychology, Vol. 63 No. 2, pp. 325–359.
Mcilvaine, A.R. (2018), “Data in the Driver’s Seat”, HRExecutive.com, 5 June, available at:
https://hrexecutive.com/talent-acquisitions-leaders-use-ai-to-improve-hiring/ (accessed 20
July 2020).
Mendoza, I. and Bygrave, L.A. (2017), “The Right Not to be Subject to Automated Decisions
Based on Profiling”, in Synodinou, T.-E., Jougleux, P., Markou, C. and Prastitou, T.
(Eds.), EU Internet Law, Springer International Publishing, Cham, pp. 77–98.

Discussion
84
Michaelides, M. (2018), “The Challenges of AI and Blockchain on HR Recruiting Practices”,
The Cyprus Review, Vol. 30 No. 2.
Mike Projkovski (2018), “Welche Schriftart für dein Webdesign?”, available at:
https://99designs.de/blog/web-digitales-design/schriftarten-webdesign-auswaehlen/
(accessed 28 October 2020).
Mohri, M., Rostamizadeh, A. and Talwalkar, A. (2018), Foundations of machine learning,
Adaptive computation and machine learning, Second edition.
Mortier, R., Haddadi, H., Henderson, T., McAuley, D. and Crowcroft, J. (2014), Human-Data
Interaction: The Human Face of the Data-Driven Society, available at:
https://arxiv.org/pdf/1412.6159.
Nawaz, N. and Gomes, A.M. (2020), “Artificial Intelligence Chatbots are New Recruiters”,
SSRN Electronic Journal.
Nussbaumer Knaflic, C. (2015), Storytelling with data: A data visualization guide for
business professionals, John Wiley & Sons, Hoboken, New Jersey.
Poole, D.L., Mackworth, A.K. and Goebel, R. (1998), Computational intelligence: A logical
approach, Oxford Univ. Press, New York, NY.
Qi Guo (2020), “The AI Behind LinkedIn Recruiter search and recommendation systems”,
available at: https://engineering.linkedin.com/blog/2019/04/ai-behind-linkedin-recruiter-
search-and-recommendation-systems (accessed 11 June 2020).
Rich, E. and Knight, K. (1991), Artificial intelligence, 2. ed., McGraw-Hill, New York.
Rogers, H. (2002), Theory of recursive functions and effective computability, 5. print, MIT
Press, Cambridge, Mass.
Roig, A. (2017), “Safeguards for the right not to be subject to a decision based solely on
automated processing (Article 22 GDPR)”, European Journal of Law and Technology,
Vol. 8 No. 3.
Russell, S.J., Norvig, P., Davis, E. and Edwards, D. (2016), Artificial intelligence: A modern
approach, Always learning, Third edition, Global edition, Pearson, London.
Sam Ransbotham, David Kiron, Philipp Gerbert and Martin Reeves (2017), “Reshaping
Business With Artificial Intelligence: Closing the Gap Between Ambition and Action”,
MIT Sloan management review, Vol. 59 No. <b>ISSN: </b>1532-9194.
Schikora, C., Galster, S. and Högerl, D. (2020), “Digitalisierung im Recruiting: Chatbots”, in
Harwardt, M., Niermann, P.F.-J., Schmutte, A.M. and Steuernagel, A. (Eds.), Führen und
Managen in der digitalen Transformation, Springer Fachmedien Wiesbaden, Wiesbaden,
pp. 265–283.

Discussion
85
Schuler, H. (2013), “Personalauswahl”, in Stock-Homburg, R. (Ed.), Handbuch Strategisches
Personalmanagement, Springer Fachmedien Wiesbaden, Wiesbaden, pp. 29–58.
Schulz, L.M. (2014), Das Geheimnis erfolgreicher Personalbeschaffung: Von der
Bedarfsidentifikation bis zum Arbeitsvertrag, Springer Gabler, Wiesbaden.
SmartRecruiters (2018), “AI for Recruiting Software | SmartAssistant”, available at:
https://www.smartrecruiters.com/recruiting-software/artificial-intelligence-recruitment-
tools/ (accessed 5 August 2020).
Sosulski, K. (2019), Data visualization made simple: Insights into becoming visual,
Routledge Taylor & Francis Group, New York.
Sun, G.-D., Wu, Y.-C., Liang, R.-H. and Liu, S.-X. (2013), “A Survey of Visual Analytics
Techniques and Applications: State-of-the-Art Research and Future Challenges”, Journal
of Computer Science and Technology, Vol. 28 No. 5, pp. 852–867.
Tufte, E.R. (2001), The visual display of quantitative information, 2. ed., 9. printing, Graphics
Press, Cheshire, Conn.
Ullah, R. and Witt, M. (2018), Praxishandbuch Recruiting: Grundlagenwissen - Prozess-
Know-how - Social Recruiting, 2nd ed., Schäffer-Poeschel Verlag für Wirtschaft Steuern
Recht GmbH, Stuttgart.
Upadhyay, A.K. and Khandelwal, K. (2018), “Applying artificial intelligence: implications
for recruitment”, Strategic HR Review, Vol. 17 No. 5, pp. 255–258.
Verhoeven, T. (Ed.) (2020), Digitalisierung im Recruiting, Springer Fachmedien Wiesbaden,
Wiesbaden.
Verlinden, N. (2019), “9 Intriguing Uses of AI in Recruitment in 2019 - Harver”, available at:
https://harver.com/blog/ai-in-recruitment-2019/#CandidateSourcing (accessed 21 July
2020).
Vollmer, N. (2020), “Article 22 EU General Data Protection Regulation (EU-GDPR)”,
available at: https://www.privacy-regulation.eu/en/article-22-automated-individual-
decision-making-including-profiling-GDPR.htm (accessed 12 June 2020).
Wachter, S., Mittelstadt, B. and Floridi, L. (2017), “Why a Right to Explanation of
Automated Decision-Making Does Not Exist in the General Data Protection Regulation”,
International Data Privacy Law, Vol. 7 No. 2, pp. 76–99.
Weuster, A. (2012), Personalauswahl I: Internationale Forschungsergebnisse zu
Anforderungsprofil, Bewerbersuche, Vorauswahl, Vorstellungsgespräch und Referenzen,
3., akt. u. überarb. Aufl. 2012, Gabler Verlag, Wiesbaden.
Wexler, S., Shaffer, J. and Cotgreave, A., “The Big Book of Dashboards”.

Discussion
86
Wilke, C.O. (2019), Fundamentals of Data Visualization, O’Reilly Media Inc, Farnham.
Winston, P.H. (1993), Artificial intelligence, 3. ed., reprinted with corr, Addison-Wesley,
Reading, Mass.
Yau, N. (2013), Data points: Visualization that means something, 1. Aufl., Wiley,
Indianapolis, Ind.
Yigitbasioglu, O.M. and Velcu, O. (2012), “A review of dashboards in performance
management: Implications for design and research”, International Journal of Accounting
Information Systems, Vol. 13 No. 1, pp. 41–59.

AI for preselection in recruting somethg

More Related Content

Similar to AI for preselection in recruting somethg (20)

Recently uploaded (20)

AI for preselection in recruting somethg