ArticlePDF Available

Acceptance Acceptance conditions of algorithmic decision support in management

December 2023

Authors:

Ruhr University Bochum

This thesis explores the acceptance of decision-aiding technologies in management, which is a challenging component in their use. To address the lack of research on algorithmic decision support at the managerial level, the thesis conducted a vignette study with two scenarios, varying the degree of anthropomorphizing features in the system interface. Results from the study, which included 281 participants randomly assigned to one of the scenarios, showed that the presence of anthropomorphized features did not significantly affect acceptance. However, results showed that trust in the system was a crucial factor for acceptance and that trust was influenced by users' understanding of the system. Participants blindly trusted the system when it was anthropomorphized, but the study emphasized that system design should not focus on the benefits of blind trust. Instead, comprehensibility of the system results is more effective in creating acceptance. This thesis provided practical implications for managers on system design and proposed a structural model to fill a research gap on acceptance at the managerial level. Overall, the findings may assist companies in developing decision support systems that are more acceptable to users.

Content uploaded by Kiram Iqbal

Content may be subject to copyright.

Junior Management Science 8(4) (2023) 887-925

Junior Management Science

journal homepage: www.jums.academy

Editor:

DOMINIK VAN AAKEN

Advisory Editorial Board:

FREDERIK AHLEMANN

JAN-PHILIPP AHRENS

BASTIAN AMBERG

THOMAS BAHLINGER

MARKUS BECKMANN

CHRISTOPH BODE

SULEIKA BORT

ROLF BRÜHL

KATRIN BURMEISTER-LAMP

JOACHIM BÜSCHKEN

CATHERINE CLEOPHAS

NILS CRASSELT

BENEDIKT DOWNAR

RALF ELSAS

KERSTIN FEHRE

MATTHIAS FINK

DAVID FLORYSIAK

GUNTHER FRIEDL

MARTIN FRIESL

FRANZ FUERST

WOLFGANG GÜTTEL

NINA KATRIN HANSEN

ANNE KATARINA HEIDER

CHRISTIAN HOFMANN

SVEN HÖRNER

KATJA HUTTER

LUTZ JOHANNING

STEPHAN KAISER

NADINE KAMMERLANDER

ALFRED KIESER

NATALIA KLIEWER

DODO ZU KNYPHAUSEN-AUFSESS

SABINE T. KÖSZEGI

ARJAN KOZICA

CHRISTIAN KOZIOL

MARTIN KREEB

TOBIAS KRETSCHMER

WERNER KUNZ

HANS-ULRICH KÜPPER

MICHAEL MEYER

JÜRGEN MÜHLBACHER

GORDON MÜLLER-SEITZ

J. PETER MURMANN

ANDREAS OSTERMAIER

BURKHARD PEDELL

MARCEL PROKOPCZUK

TANJA RABL

SASCHA RAITHEL

NICOLE RATZINGER-SAKEL

ASTRID REICHEL

KATJA ROST

THOMAS RUSSACK

FLORIAN SAHLING

MARKO SARSTEDT

ANDREAS G. SCHERER

STEFAN SCHMID

UTE SCHMIEL

CHRISTIAN SCHMITZ

MARTIN SCHNEIDER

MARKUS SCHOLZ

LARS SCHWEIZER

DAVID SEIDL

THORSTEN SELLHORN

STEFAN SEURING

ANDREAS SUCHANEK

TILL TALAULICAR

ANN TANK

ORESTIS TERZIDIS

ANJA TUSCHKE

MATTHIAS UHL

CHRISTINE VALLASTER

PATRICK VELTE

CHRISTIAN VÖGTLIN

STEPHAN WAGNER

BARBARA E. WEISSENBERGER

ISABELL M. WELPE

HANNES WINNER

THOMAS WRONA

THOMAS ZWICK

Volume 8, Issue 4, December 2023

JUNIOR

MANAGEMENT

SCIENCE

Maryam Hammad,

Power to the CEO? Sources of CEO

Power and Its Influences on Strategic Choic es and

Firm Performance

MohidFarooq Butt,

Logical Reasoning in Management:

From “Philosopher Kings” to Logical Managers?

Georg Streicher,

Digital Transformation in Family Businesses

KiramIqbal,

Acceptance conditions of algorithmic decision

support in management

Max-Gerrit Meinel,

Transparenz und organisationale

Legitimität: Eine experimentelleStudieam Beispiel

einesfiktivenUnternehmens

Philip Christoph Häberle,

Discussionofautomotivetrends

and implicationsforGerman OEMs

Regina Maria Martha Förg,

Chances and challenges for the

members of the Fairtrade-supply chain: a case study

of Chile and Switzerland

Stefano Mattana,

Berufseinstiegin Zeitenmobiler Arbeit:

Eine empirischeStudieüberden Einfluss mobilen

Arbeitensauf das psychischeWohlbefin den

kaufmännischerAuszubildenderin Hambu rg

Jan Philipp Natter,

Enabling E-Mobility: How Electric Grids

Can Support High EV Adoption with Residential PV

and Battery Energy Storage Systems

Hannah Fernsebner,

Virtual Reality Transforming the Digital

Learning Environment: An Analysis of Students’

Acceptance

827

845

865

887

926

955

993

1010

1040

1081

PublishedbyJuniorManagement Science e.V.

Acceptance conditions of algorithmic decision support in management

Kiram Iqbal

Ruhr-Universität Bochum

Abstract

This thesis explores the acceptance of decision-aiding technologies in management, which is a challenging component in their

use. To address the lack of research on algorithmic decision support at the managerial level, the thesis conducted a vignette

study with two scenarios, varying the degree of anthropomorphizing features in the system interface. Results from the study,

which included 281 participants randomly assigned to one of the scenarios, showed that the presence of anthropomorphized

features did not signiﬁcantly affect acceptance. However, results showed that trust in the system was a crucial factor for

acceptance and that trust was inﬂuenced by users’ understanding of the system. Participants blindly trusted the system when

it was anthropomorphized, but the study emphasized that system design should not focus on the beneﬁts of blind trust. Instead,

comprehensibility of the system results is more effective in creating acceptance. This thesis provided practical implications

for managers on system design and proposed a structural model to ﬁll a research gap on acceptance at the managerial level.

Overall, the ﬁndings may assist companies in developing decision support systems that are more acceptable to users.

Keywords: Decision support systems; Algorithmic management; Artiﬁcial intelligence; Anthropomorphizing; Technology

acceptance.

1. Introduction and area of problem

Recent advances in technology enable aid for business

in the context of problem-solving (J. R. Evans & Lindner,

2012). In practice, the usage of systems aiding decisions is

low. Therefore it is necessary to research on acceptance con-

ditions. This introduction outlines the practical and theoret-

ical necessity of deriving acceptance conditions for research.

Furthermore, the structure of the thesis is outlined.

1.1. Objective and research question

The scientiﬁc ﬁeld of business analytics and business in-

telligence has gained high importance in strategic manage-

ment. In this context, it is important to differentiate between

these terms. Business analytics is deﬁned as a process where

data is converted to actions through an analysis of this data

in the context of organizational problem solving or decision-

making (J. R. Evans & Lindner,2012). Business intelligence

is deﬁned as the use of various technologies like information

technology to help managers to gain insights about their busi-

ness and to improve decision-making (Gluchowski,2016).

Since analytic procedures are based on algorithms the term

business analytics can be used as a synonym for algorithmic

decision support.

Despite the rise of opportunities for algorithmic decision

support, arising challenges should not be neglected. Chal-

lenges are legal issues like ownership and privacy of data and

technical obstacles like analysis of complex data and scaling

of algorithms (Mishra & Silakari,2012).

One of the most challenging components due to the use of

algorithmic decision support in business is the acceptance of

these systems by users. From a user’s perspective, one major

problem is that precise algorithms generate the perception of

authoritative correctness therefore human beings can feel in-

ferior toward algorithms. Especially the introduction of deep

learning algorithms in artiﬁcial intelligence (Linardatos, Pa-

pastefanopoulos, & Kotsiantis,2020) and the scaling of algo-

rithms (Mishra & Silakari,2012) lead to higher accuracy and

precision which in turn makes the human being feel inferior

to algorithms. In this regard, it is necessary to do further re-

search on the acceptance condition of algorithmic decision

support.

Therefore, this paper conducts an analysis on the follow-

ing research question: Which conditions lead to an accep-

DOI: https://doi.org/10.5282/jums/v8i4pp887-925

K. Iqbal /Junior Management Science 8(4) (2023) 887-925888

tance of algorithmic decision support in management?

1.2. Theoretical and practical research gap

To answer the research question, it is necessary to make

an analysis of the state of the art in research and elucidate

the research gap. Various studies do research on the topic of

acceptance of artiﬁcial intelligence-based technologies. Has-

tenteufel and Ganster (2021) apply this topic to the digital

transformation in banking. Therefore, they use the tech-

nology acceptance model by Davis, Bagozzi, and Warshaw

(1989). Hastenteufel and Ganster (2021) identify the trust-

worthiness, perceived usability and social inﬂuence as accep-

tance conditions for algorithmic decision support. Gersch et

al. (2021) do research about the challenges particular in trust

in collaborative service delivery with artiﬁcial intelligence in

the ﬁeld of radiology. Therefore they conduct interviews

with various stakeholders in radiology. They identify trust

as an indicator to cope with uncertainties. Furthermore,

they identify that cognitive trust is built in the ﬁrst contact

with the user. With repeated experience, the user develops

affective trust. Understandability and comprehensibility are

important for users. Further challenges are the change of

own position in the workplace due to the introduction of

support through artiﬁcial intelligence and arising of new

duties and prerequisites in the design of the socio-technical

system. Therefore, explainable artiﬁcial intelligence should

take into account the perspective of different stakeholders.

Rathje, Laschet, and Kenning (2021) do research about trust

in banking. Therefore, they develop their own research

model based on the models by Mayer, Davis, and Schoorman

(1995), Gefen, Karahanna, and Straub (2003) and Davis

(1989). They conducted a survey with 119 participants

where the afﬁnity to technology is high. Rathje et al. (2021)

identify that trust has a relationship to the intention to use

the technology. Pütz, Düppre, Roth, and Weiss (2021) do re-

search on the topic of acceptance of voice and chatbots. They

use the technology acceptance model (TAM) of Davis (1989)

and the extended version of Venkatesh and Davis (2000)

and Venkatesh and Bala (2008) to analyze the acceptance of

this technology. The approach used by Pütz et al. (2021) is

literature-based. They identify a relation between perceived

usability and perceived user-friendliness. Further results are

a relation between perceived user-friendliness and intention

to use the technology.

Scheuer (2020) develops an acceptance model for the use

of artiﬁcial intelligence. The model developed by Scheuer is

called the KIAM model. The KIAM model is an extension of

the TAM model and is considered the Artiﬁcial Intelligence

Acceptance Model. Whereas KI is referred to as the german

term for AI. The AI acceptance model (KIAM) consists of a

holistic acceptance model that addresses the characteristics

of the theoretical properties of an AI compared to a classi-

cal computer system. Scheuer (2020) assumes that an AI is

accessible via a technology (e.g., a smartphone application)

and enriching it with Narrow AI services (e.g., a chatbot in-

tegration, Speech-To-Text, or Text-To-Speech) through which

a user can interact with the AI in natural language. Based

on this, two essential components emerge ﬁrst, the classical

technology in the form of a software application, and sec-

ond, the dialog component for interacting with the AI in the

background. For the classical technology and the investiga-

tion of its acceptance, Scheuer uses the existing TAM model

by Venkatesh and Bala (2008) TAM 3. However, for the di-

alog component and the resulting interaction between the

AI and the user, Scheuer (2020) differentiates to what ex-

tent the user accepts the AI as a personality or even as a

complete person. For this, he considers that psychological

models for measuring sympathy and affection apply as per-

sonality acceptance takes precedence over pure technology

acceptance. In this regard, Scheuer highlights that if the ﬁl-

ter of the perception of the system as a personality is taken

into account an AI is recognized as a personality. This re-

lationship with technology can be described as interpersonal

acceptance. According to inter-parental acceptance-rejection

theory (IPART) (Rohner & Khaleque,2002), interpersonal ac-

ceptance is generated by warmth and affection in the rela-

tionship and is based on sympathy. Sympathy, in turn, is de-

pendent on reciprocity in human behavior of communication

and sameness of character traits. Reciprocity of behavior is

subsequently inﬂuenced by a perceived and radiated attrac-

tiveness of and to the other person and positive external per-

ception. Interpersonal acceptance in decision support is a

new component for analyzing acceptance conditions. There-

fore this thesis considers interpersonal acceptance for deriv-

ing acceptance conditions. Due to a lack of research ﬁndings

of algorithmic decision support on managerial-level, this the-

sis aims to identify acceptance conditions, in order to con-

tribute to research and practice. This section aimed to em-

phasize the research gap and underline what has been al-

ready used in the context of academic literature. Summing

up, the section shows that there is a need for investigating the

conditions of accepting algorithmic decision support systems

from a managerial perspective.

1.3. Outline of the thesis

This thesis aims to answer the following research ques-

tion: which conditions lead to an acceptance of algorithmic

decision support in management? In order to answer the re-

search questions and derive the conditions that lead to an

acceptance of algorithmic decision support in management,

it is necessary to provide a better understanding of the theo-

retical foundation regarding algorithmic decision support in

management and explain how this takes place in practice.

This will be presented in section two where the relevance of

algorithmic decision support is outlined. Hereby, the advan-

tages of the integration of business analytics into business are

examined. Necessary technological foundations are given in

order to understand the underlying technology behind algo-

rithmic decision support and understand the rapid develop-

ment in performance of computing architecture.

Furthermore, acceptance conditions are derived from the

literature. At ﬁrst theories for an increase usage of technol-

ogy are examined. In addition, the term acceptance plays

an important role in the context of the research question, as

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 889

the conditions that lead to an acceptance of algorithmic de-

cision support in management are investigated. To further

elaborate on the role of acceptance from a theoretical point

of view, different acceptance models that exist in the liter-

ature are presented. Findings from literature from of non-

managerial-levels are used to derive hypotheses for accep-

tance conditions.

Afterward, a structural equation model will be derived

based on the thoughts of the TAM for conducting a quantita-

tive study (vignette study) to provide empirical evidence to

answer the research question. The target group for the em-

pirical study will be managers and students in future man-

agement positions as the research question focuses on the

acceptance of algorithmic decision support in management.

The items are derived from Scheuer (2020) who introduced

the KIAM model which contains the TAM of Venkatesh and

Bala (2008). The items are used in a vignette study (Wason,

Polonsky, & Hyman,2002). The results are analyzed empiri-

cally and descriptive statistics are provided.

Before estimating the structural equation model, the

quality indicators for the measurement models and struc-

tural models are examined.

In the next section, the survey data is analyzed by esti-

mating a structural equation model. The results of the anal-

ysis are discussed in a further section and contextualized to

ﬁndings in literature. This section puts emphasis on the in-

terpretation of the results where the quantitative results are

transferred into qualitative measures and reﬂected in the the-

oretical foundations. In addition to this, the ﬁndings will be

applied and compared to the results of the state of the art in

literature. Afterward, the theoretical and practical implica-

tions are presented along with the limitations of the study.

The conclusion summarizes the ﬁndings of the thesis.

2. Understanding acceptance of algorithmic decision

support

In order to answer the research question, it is necessary

to outline theoretical foundations. The following section will

emphasize the importance of algorithmic decision support for

strategic management. At ﬁrst, the relevance of algorithmic

decision support is derived on a general level. Further, al-

gorithmic decision support is applied to the business context

where advantages of the application of this technology are

outlined. Afterward, the underlying technological compo-

nents or related technologies are addressed for a sufﬁcient

technological foundation.

2.1. Relevance of algorithmic decision support

In order to understand the relevance of algorithmic deci-

sion support, it is important to understand what decisions are

and when they occur. According to Mallach (1994), decisions

are part of the problem-solving process and are deﬁned as a

reasoned choice between available alternatives. The litera-

ture identiﬁes two types of decision-making processes. The

intuitive decision-making approach and the rational decision-

making approach (Alvarez, Barney, & Young,2010). These

approaches are based on the two types of cognitive processes

of Stanovich and West (2000) and are deﬁned as System 1

(based on intuition) and System 2 (based on reasoning). An

intuitive decision-making approach is deﬁned as a decision

based on biases and heuristics (Alvarez et al.,2010). Indi-

viduals tend to use various kinds of heuristics in judgmental

decisions (Tversky & Kahneman,1974).

Managers tend more toward the intuitive decision-

making approach than the rational decision-making ap-

proach (Anderson,2015). Anderson (2015) identiﬁed that

only 29% of senior executives of 1135 surveyed base their

decision on data and analysis, where 30 % of them use their

intuition or experience and 28 % of them use advice or ex-

perience of others as a source of decision. The majority of

the surveyed managers use availability heuristics to make

decisions which implies that most managers tend to use

the intuitive decision-making approach. The use of heuris-

tics and biases may lead to efﬁcient decision-making or to

decreased decision quality. Various studies show that the oc-

currence of biases lowers the quality of decisions (Camerer

& Lovallo,1999;Carr & Blettner,2010;Everett & Fairchild,

2015;Forbes,2005;Kahneman & Tversky,1996;Koellinger,

Minniti, & Schade,2007;La Hayward, Forster, Sarasvathy, &

Fredrickson,2010). According to Carr and Blettner (2010)

especially the quality of hot decisions1is strongly related to

the success or survival of companies. This literature shows

those wrong decisions by an intuitive decision-making ap-

proach can lead to the failure of the company. On the other

hand, the advantage of intuitive decision-making is that it

may be faster than rational decision-making. Intuitive de-

cision making is based on System 1 which is faster than

System 2 (Kahneman,2003). The rational decision-making

approach is based on System 2.

In the literature, there is no mutual agreement on an ex-

act description of the process of the rational decision-making

approach. Bazerman and Moore (2012) specify the rational

decision-making approach as a rational model of decision-

making assuming that people follow a certain process. The

rational decision-making process by them is segmented into

six phases: (1) perfectly deﬁne the problem (2) identify all cri-

teria (3) accurately weigh all of the criteria according to pref-

erences (4) know all relevant alternatives (5) accurately assess

each alternative based on each criterion (6) accurately calcu-

late and choose an alternative with the highest perceived value

(Bazerman & Moore,2012).

The main problem by the rational decision-making ap-

proach is that human-beings do not have complete informa-

tion (Biswas,2015).

The Prospect Theory addresses the problem of bounded

rationality and gives the advice to use biases and heuristics

when rational decision-making is not applicable (Kahneman,

Slovic, Slovic, & Tversky,1982). Despite incomplete infor-

mation, a manager may use the rational decision-making ap-

proach for problem-solving process. It is tautologic to imply

1Hot decisions are deﬁned as decisions who are critical for companies’

success Janis and Mann (1977)

K. Iqbal /Junior Management Science 8(4) (2023) 887-925890

that decisions based on incomplete information lead to a de-

creased decision quality because the use of incomplete infor-

mation is referred to as the availability heuristic. The effects

of heuristics and biases on decision quality are mentioned

above.

To overcome this vicious cycle the literature suggests a

different kind of decision aids. Decision support systems

(DSS) are a particular technological form of a decision aid.

First DSS help decision-makers by giving them more informa-

tion and extending their decision-relevant knowledge (Mal-

lach,1994). Referring to previous thoughts extended infor-

mation would increase decision quality. Huber (1990) iden-

tiﬁes that managers using computer-assisted decision aiding

would make better decisions. Mcafee, Brynjolfsson, Dav-

enport, Patil, and Barton (2012) consider data-driven deci-

sions better than intuitive decisions because they are based

on evidence. Despite the dynamic development of technol-

ogy computer-aided decision support is not new. In fact, it is

more than 50 years old. The First DSS application was built

in 1970 (Watson & Wixom,2007). The usage of DSS has

various advantages.

Carlson (1977) identiﬁes that DSS can be used in all

decision-making phases. DSS can help to make the rational

decision-making process better by partially reducing previ-

ous incomplete information. Nevertheless, the past 50 years

led to an increased computing power by the factor of ap-

proximately 67.41 Million2according to Moore’s law (Moore,

1965). A better example to understand the increased com-

puting power is given in the following. Assuming no change

in algorithms, operations that needed approximately 2.13

years of calculation to give decision aid in 1970 can now be

processed within one second. Considering the rise of new

and better algorithms which differ in performance since they

are evaluated by runtime (Güting & Dieker,1992;Mcafee et

al.,2012) the performance of algorithmic decision support

is increased. In fact, new algorithmic technologies like ar-

tiﬁcial intelligence, big data analytics, neural network, etc.,

leverage the performance of decision support systems. This

increase in the performance of decision support systems may

lead theoretically to an extensive improvement of a rational

decision-making process by reducing time and incomplete in-

formation in theory. At the practical level, necessary data

for information processing should be available since infor-

mation is processed out of data by analytics. The analysis of

data to support decision-making is considered business ana-

lytics (Shanks & Bekmamedova,2012). Besides supporting

decisions, business analytics has a wide range of impacts on

business. Therefore it is necessary to understand the impact

of algorithmic decision support on business and the underly-

ing technologies of algorithmic decision support.

2Meaning the computing power is doubled every second year due to con-

stant costs of transistors. The necessary mathematical operation is 226. 52

years were passed. These years are divided by two results in the power of

26.

2.1.1. Advantage of business analytics in management

In order to understand the impact of business analytics

on management, it is necessary to understand the role and

tasks of management.

Management is deﬁned as leadership in the efﬁcient, in-

formed, purposeful and planned conduct of complex orga-

nized activity (Andrews,1980). The activity is characterized

by high complexity and the desirability to increase the intu-

itive competence of the executing manager. Andrews (1980)

suggests the need for a unitary concept for reducing the com-

plexity of the manager’s job and identiﬁes strategy as a pos-

sible solution to reduce complexity. Therefore it is important

to distinguish between operational and strategic activities.

According to Porter (1996), operational activities are about

performing similar activities. They differ only if they are per-

formed in a more efﬁcient way than rivals. Porter (1996)

deﬁnes strategy as the creation of a unique and valuable po-

sition, involving various sets of actions. Therefore Andrews

(1980) delivers the approach of a schematic development of

an economic strategy. According to Andrews (1980), it is

necessary to identify external opportunities and risks and get

insights into the corporate capabilities and resources in terms

of strengths and weaknesses and consider all combinations

of internal and external analysis to evaluate and determine

the best match for opportunity and resources. In the end, a

choice is derived which is called an economic strategy. This

schematic development of an economic strategy is relevant

in theory and practice because the SWOT-Analysis is based

on this scheme (Kotler, Berger, & Bickhoff,2010). Andrews’s

(1980) approach shows that strategy is all about the evalu-

ation and selection of choices – similar to the deﬁnition of

making decisions. Porter (1996) conﬁrms that strategy is the

deliberate disregard of other alternatives by purposefully lim-

iting what a company should do. Strategic management can

be considered as the reasoned choice or decision between

the combination of strategies from the internal and external

analysis. As mentioned before there are two decision-making

approaches.

The highest valuated companies in the world can be con-

sidered as successful in competition due to the ﬁnancial in-

dicator. The top ﬁve companies with the highest valuation

in May 2022 are Apple, Saudi Aramco, Microsoft, Alphabet

and Amazon (Companiesmarketcap,2022). Except for Saudi

Aramco, the highest valuated companies could establish their

market position due to the use of algorithmic support, ex-

plicitly through the use of artiﬁcial intelligence (Rainsberger,

2021). Rainsberger (2021) shows four dimensions where al-

gorithmic aid (artiﬁcial intelligence) revolutionizes business

activities. The four dimensions are strategy, performance, ef-

fectiveness and competence. In the following, the wide range

of impacts on business analytics is outlined. Especially strate-

gic management is affected by business analytics.

The assumption that an alternative future can be derived

from certain past events (Luhmann,1990) is essential for

algorithmic aid. This assumption is essential, since analyt-

ics is based on historical data (descriptive analytics), esti-

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 891

mates future outcomes (predictive analytics) and determines

actions for optimizing business outcomes (prescriptive ana-

lytics) (Apté, Dietrich, & Fleming,2012). Descriptive ana-

lytics enable organizations to calibrate opportunities by pro-

viding insights into what happened previously in their inter-

nal and external environment (van Rijmenam, Erekhinskaya,

Schweitzer, & Williams,2019). Anticipating a possible future

leads to a competitive advantage (Koch,2015). Côrte-Real,

Oliveira, and Ruivo (2017) specify that algorithmic aid al-

lows effective internal and external knowledge management

enhancing organizational agility. Côrte-Real et al. (2017) ad-

dress the scheme of economic strategy by Andrews (1980)

for sensing opportunities and threats and seizing possible

chances.

The implementation of algorithmic aid in the internal and

external analysis of a company can gain insights into the in-

ternal processes and external events (Benaben et al.,2019)

with the possibility to analyze this data and make predictions

of future internal processes and external events. Consider-

ing all combinations of internal and external analysis a more

precise evaluation and determination for the best match of

opportunity and resources is possible. The literature sug-

gests that algorithmic aid (predictive analytics) leads to bet-

ter decision-making by improving business value and com-

petitive performance (LaValle, Lesser, Shockley, Hopkins, &

Kruschwitz,2011;Shanks & Bekmamedova,2012). A pos-

sible explanation for this relation is that predictive analyt-

ics helps companies to remain competitive by anticipating

changing environments and adapting to these changes (Ha-

jkowicz et al.,2016).

The distinction between strategic and operational activi-

ties was outlined. We showed that business analytics can en-

hance strategic activities. Other dimensions of Rainsberger

(2021) address operational activities. In the following, a

detailed description of enhanced operational activities is de-

rived.

The second dimension of Rainsberger (2021) is perfor-

mance. Operational activities can be enhanced by business

analytics since we showed that algorithmic aid (Big Data An-

alytics) improves business performance (Mcafee et al.,2012).

Furthermore, Chen, Preston, and Swink (2015) and Apté et

al. (2012) show that algorithmic aid enhances operational

efﬁciency. An example of operational efﬁciency is improved

workforce planning and reduced need for new hires and a re-

duction in overtime (D. Barton & Court,2012). Further ben-

eﬁts are decreased cost for IT-Infrastructure and efﬁcient data

delivery resulting in saving time (Watson & Wixom,2007).

An example of the reduction of costs is preventing and mon-

itoring fraud in organizations. Analytics enable fraud detec-

tion at reasonable costs (Mishra & Silakari,2012). All in all

algorithmic aid helps to make effective decisions faster (Reid,

McClean, Petley, Jones, & Ruck,2015) even enabling to au-

tomate operational workﬂows (Iansiti & Lakhani,2020) re-

sulting in greater performance.

The third dimension of Rainsberger (2021) is compe-

tence. Gartz (2004) shows that business intelligence can

enhance the representation and evaluation of companies’

knowledge using knowledge-based systems. Therefore al-

gorithmic aid can help to preserve knowledge within the

company and make information ﬂow more efﬁcient (Watson

& Wixom,2007).

The fourth dimension of Rainsberger (2021) is effective-

ness. The literature suggests explicit effectiveness of the

use of algorithmic aid in sales and marketing (Halper,2014;

Mishra & Silakari,2012;Rainsberger,2021). The effective-

ness is shown by the term pervasive business intelligence.

Pervasive business intelligence is providing users with infor-

mation for better job performance (Watson & Wixom,2007).

The advantage of pervasive business intelligence is that data

is delivered to the certain user who needs the data to take an

effective decision (Rainsberger,2021). Furthermore, algo-

rithmic aid can provide insights into customer habits & pat-

terns by analyzing customer data (Hamilton & Koch,2015).

Therefore the use of algorithms enables personalized con-

textual interaction with customers (Brahm, Cheris, & Sherer,

2016). Customization to customers’ needs is a very effective

form of gaining competitive advantage at operational level

since data-based customization to customers’ needs brings

value (Davenport,2013). On the other hand customer pri-

oritization by analyzing customer proﬁtability through digi-

tal devices can increase effectiveness of business (Davenport,

2013). The effectiveness of a business can be measured by

ﬁnancial indicators. J. R. Evans and Lindner (2012) suggest

that algorithmic aid can increase proﬁtability, revenue and

shareholder return. Furthermore, companies’ goals can be

reached faster with the use of analytics (Rainsberger,2021).

2.1.2. Technological foundations for algorithmic decision

support

The main goal of algorithmic decision support is to gain

value-creating information (Mikalef, Pappas, Krogstie, &

Pavlou,2020). The information is derived from data (Azvine,

Cui, Nauck, & Majeed,2006;Benaben et al.,2019). The ab-

straction levels of data, information, decision and knowledge

are shown in Figure 1.

According to Benaben et al. (2019) data is a formalized

observation of the reality. Information is deﬁned as the result

of the interpretation of data through algorithmic methods

(Benaben et al.,2019). The process of applying data analysis

and discovery algorithms over the data is described as Data

Mining (Fayyad, Piatetsky-Shapiro, & Smyth,1996). Ben-

aben et al. (2019) deﬁne the exploitation of information gen-

erated by data mining as a decision. The deﬁnition of Ben-

aben is not contradictory to the deﬁnition of Mallach (1994)

mentioned earlier in this paper due to the fact that the infor-

mation provides the ability for reasoning in choice-settings.

The last distinction by Benaben et al. (2019) is knowledge.

Knowledge is a capitalized static information about ex-

tracted abstract concepts or previous experience (Benaben

et al.,2019). As described earlier the interpretation of data

is executed by algorithms. Therefore, it is necessary to de-

ﬁne algorithms. The literature has a broad deﬁnition of algo-

rithms. Moschovakis (2001) outlines the necessity to deﬁne

algorithms precisely. According to Moaschavakis, a rigorous

K. Iqbal /Junior Management Science 8(4) (2023) 887-925892

Figure 1: K-DID framework presenting the abstraction levels of data, information, decision and knowledge (Source: Benaben

et al. (2019))

deﬁnition can lead to a wrong identiﬁcation of abstract ma-

chines or mathematical models of computers.

According to Güting and Dieker (1992), an algorithm is

deﬁned as a speciﬁc process of tasks with a clear order of

tasks run by mechanical or technical devices to receive an

expected output for a task. Furthermore, they characterize

that every task has to be described clearly and is executable

with ﬁnite effort in ﬁnite time leading to a termination of an

algorithm. Therefore algorithms can metaphorically be seen

as a recipe for a problem-solving process. The recipes for the

problem-solving process can vary in tasks. In the end the best

performance of a recipe matters.

The implementation of a certain data type on algorithmic-

level is characterized by data structures (Güting & Dieker,

1992). Algorithms differ in performance if they are used

in other data structures than intended. Short runtimes are

performance measures for algorithms. The selection of algo-

rithms is based on a runtime analysis (Güting & Dieker,1992;

Knebl,2019). The runtime analysis does not contain com-

puting power of the underlying hardware run on algorithms.

In the evaluation of algorithms, it is necessary to distinguish

between runtime and computing time since the computing

time involves the performance of hardware and algorithm

combined. In practice, computing time is a desirable perfor-

mance measure for algorithms. Computing time can be re-

duced by aiming for a low runtime of an algorithm or using

performant hardware. Therefore decision support can per-

fectly aid in the decision-making process since the quality of

algorithms is evaluated by time.

Recent advances in hardware show a leveraging effect on

computing power. Besides Moore’s law, other advances in

hardware can be seen in Butters or Kryder’s law. Butter’s

law indicates that the amount of data transmitted by ﬁber-

glass doubles every 9th month (Rainsberger,2021). Fur-

thermore, Kryder’s law states that storage capacity doubles

every 13th month proportional to one square-centimeters of

a hard drive (Rainsberger,2021). These technological ad-

vances have exponential growth by deﬁnition leading to rad-

ical advances of exploitation in the business context. Despite

the rapid development of technology, the conception of com-

puting hardware exhibits weakness in performance due to

architectural issues. Computing architecture nowadays is di-

vided into Central Processing Unit (CPU) and Random Ac-

cess Memory (RAM) deﬁned as Von-Neumann-architecture

(Leimeister,2019). The CPU interprets and executes com-

mands in sequential order and the RAM saves necessary data

for the necessary point of time for processing (Leimeister,

2019). Shi (2021) and Rosenberg (2017) show several prob-

lems in Von-Neumann architectures. Shi (2021) states that

Moore’s law will reach its physical limit in the coming 10

- 15 years. Furthermore, the sequential processing of com-

mands leads to an inefﬁciency in comparison to the actual

brain. The human brain has advances against the computer

in coping with novelty, complexity and ambiguity (Furber,

2016). The calculating speed and precision of a computer

is higher than that of human brains but the level of intel-

ligence of computers is low (Furber,2016;Shi,2021). In

fact, various research ﬁelds of computer science are inspired

by the human brain. Therefore neuromorphic computer ar-

chitecture is a solution toward the challenges faced by Von-

Neumann architecture. Neuromorphic computing is inspired

by the research ﬁndings of the structure and operation of

the brain (Furber,2016). Neuromorphic computing aims

to extract the formidable complexity of the biological brain

and apply this knowledge to practical engineering systems

(Furber,2016). An example of neuromorphic computer ar-

chitecture is the product of the german startup from Bochum

called GEMESYS Technologies. This startup develops a neu-

romorphic chip that substitutes Von-Neumann architectures

(GEMESYS Technologies,2022). Recent breakthroughs in

neuromorphic computing research show that computing ar-

chitecture can become intelligent. Kagan et al. (2021) intro-

duce a new system architecture called DishBrain. The Dish-

Brain integrates neurons into digital systems to leverage their

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 893

innate intelligence. Kagan et al. create a synthetic biolog-

ical intelligence by harnessing the computational power of

living neurons. Therefore the DishBrain can exhibit natural

intelligence and create a new computing architectures by po-

tentially substituting Von-Neumann architectures (Kagan et

al.,2021). Future developments in computing architecture

are to use the human brain as a processing unit by creating

an interface between the human brain and the computing

system developing an interface to human brain (Kreutzer &

Sirrenberg,2019). These interfaces are called Brain Machine

Interfaces (Kreutzer & Sirrenberg,2019).

Since software is a leverage for enhancing computing per-

formance (Rosenberg,2017) it is necessary to put emphasis

on recent advances in algorithmic developments.

AI is an expression of form of algorithms. The term AI has

a wide range of deﬁnitions, and the selection of a speciﬁc def-

inition results in path dependency for research (Wang,2019).

Considering the path dependency interpreted by Wang, a

more general deﬁnition of AI is used. According to Rich

(1985), AI is the science of enabling computers to do things

that humans are currently better at. According to Kreutzer

and Sirrenberg (2019), AI is the ability of a machine to per-

form cognitive tasks. This includes reasoning skills, learning,

and ﬁnding solutions to problems independently. The idea of

an AI is ﬁrst introduced by Turing. Turing (1950) investigates

the ability of machines to think. Therefore he argues that a

successful imitation game by a computer can lead to the sug-

gestion that machine can think. In the imitation game, the

computer is programmed for doing realistic conversations.

The imitation game is successful if a human being can differ-

entiate whether the subject in the conversation is a computer

or another human. According to Turing (1950), the percep-

tion of the interrogator plays a role in the evaluation of the

question of whether machines can think or not. Therefore,

if a machine is perceived as a human, Turing considers that

this machine can think. The test for considering a system as

intelligent according to Turing’s deﬁnition of intelligence is

deﬁned as the Turing test.

Dellermann, Ebel, Söllner, and Leimeister (2019) deﬁne

intelligence as the ability to achieve complex goals, reason,

learn and adaptively perform effective activities within an

environment. Moreover, they extend the concept of intelli-

gence by dividing it into human intelligence and machine in-

telligence to gain complementary capabilities and augment

each other (Dellermann et al.,2019). As mentioned before,

computer architectures are in terms of performance in intel-

ligence lower than human brains. Therefore computer archi-

tectures are inspired by the human. Dellermann et al. (2019)

introduce the term of hybrid intelligence in terms to combine

the advances of human brain and computer systems. The

same applies to algorithms. The research ﬁeld of compu-

tational intelligence aims to develop algorithms devised to

imitate human information processing and reasoning mech-

anisms for processing complex and uncertain data sources

(Iqbal, Doctor, More, Mahmud, & Yousuf,2020). Further

technologies inspired by humans are neural networks (Iqbal

et al.,2020;Kreutzer & Sirrenberg,2019).

A neural network is a computer system containing hard-

ware and software inspired by human brain (Kreutzer & Sir-

renberg,2019). A neural network has multiple CPUs in or-

der to approximate simultaneous information processing. In

Figure 2the structure of the neural network is shown. The

ﬁrst layer is the Input Layer where data is stored as input for

further processing by the following Hidden Layer. Following

layers are deﬁned as Hidden Layer. A Hidden Layer can take

the outputs of previous Hidden Layers and do further pro-

cessing generating a new output which is processed by the

following Hidden Layer. The last layer is deﬁned as the Out-

put Layer. The Output layer generates a new output of the

previously generated outputs by the previous Hidden Layer.

Each processing algorithm of a neural network can vary from

the other. In Hidden Layers machine learning algorithms are

also used (Kreutzer & Sirrenberg,2019).

Machine learning (ML) on general level is deﬁned as a set

of methods that can automatically detect patterns in data and

use uncovered patterns to predict future data or to support

other kinds of decision-making under uncertainty (Murphy,

2012). Murphy (2012) states that ML provides automation in

data analysis. Therefore he suggests three types of learning

algorithms supervised learning, unsupervised learning and

reinforcement learning (Murphy,2012). Supervised learning

is when results the machine should process are given a pri-

ori. The machine is trained to process the right results (Rains-

berger,2021). When results are not given a priori a ML algo-

rithm is deﬁned as unsupervised learning. Here the machine

identiﬁes automatically patterns in data and creates results

(Rainsberger,2021). Reinforcement learning is inspired by

human learning the machine gets rewards for right results

and punishments for wrong results (Buxmann & Schmidt,

2021;Murphy,2012;Rainsberger,2021). Punishments and

rewards are normally associated with the teaching process

(Turing,1950). Reinforcement learning is inspired by the

ﬁndings of Turing (1950) who suggests instead of program-

ming a simulation of an adult mind programming a simula-

tion of a child’s brain can lead to a simulation of an adult

brain in future. If ML is applied in neural networks the term

Deep Learning is used in literature (Kreutzer & Sirrenberg,

2019).

As mentioned before data is necessary for data mining.

Besides the technological advances in hardware and algo-

rithms, data itself is developing in a broad way. It is nec-

essary to deﬁne the term Big Data due to the fact that Big

Data is essential for the previously mentioned technologies.

Mashingaidze and Backhouse (2017) show various deﬁni-

tions of Big Data in literature and practice. Considering the

broad range of deﬁnitions for Big Data, Mashingaidze and

Backhouse (2017) synthesize all deﬁnitions into a new one.

According to them, Big Data is data that is high in volume

gathered from a variety of sources or data formats and is

generated at high velocity. Conventional technologies are in-

sufﬁcient for the management of Big Data due to the high

level of complexity of Big Data. Therefore new advanced

technologies and techniques for storage and analysis of data

are required (Mashingaidze & Backhouse,2017). Data Ware-

K. Iqbal /Junior Management Science 8(4) (2023) 887-925894

Figure 2: Structure of a neural network (Source: Own illustration according to Kreutzer and Sirrenberg (2019))

houses are the necessary technology for coping with the high

level of complexity of Big Data (Leimeister,2019). Accord-

ing to Leimeister (2019), Data Warehouses are a specialized

technology for the storage of data and information. Further-

more, they are the underlying technology for business intelli-

gence. Their function is to gather data and ensure data qual-

ity considering uniformity, consistency and freedom from er-

ror of data (Leimeister,2019). They provide data for other

systems using interfaces (Leimeister,2019).The mentioned

technologies can be used to assist humans in their work. The

collaboration of technology and humans is called human-

machine interface (Blutner et al.,2009). Human-machine

interfaces are used to execute hybrid intelligence. Sheridan

and Verplank (1978) develop ten automation levels for assist-

ing human. In Figure 3the ten automation levels are shown

the terms used in the illustration are developed by Blutner et

al. (2009).

The ﬁrst level of automation is the manual control where

the human executes tasks without the aid of the computer.

In the second level (Action Support) the computer suggests

choices for task execution. The Batch processing reduces the

number of choices and takes a preselection of choices. The

selection of the choices is done by the human. In the next

level (Shared Control), the computer reduces the degree of

choices and processes a result for one alternative which is

suggested to the human. The human has the possibility to

select the processed choice. The next level is Decision Sup-

port where the conﬁrmation of the human is necessary for

the computer-aided execution of the processed result. In the

Blended Decision Making level, the human only has the right

for interventions. The computer does the tasks automati-

cally. The next level is Rigid system where the task execu-

tion is automated through computer aid. Human role here is

to monitor task execution. At the Automated Decision Mak-

ing level, the monitoring by human is only provided by re-

quest. The level of control by the computer is extended in

the next level where the computer executes the tasks auto-

matically. Monitoring possibility is only provided after a de-

cision. Furthermore, the last level of automation is full au-

tomation where the computer has full control over the task

and the human is ignored (Blutner et al.,2009;Sheridan &

Verplank,1978). Systems, where various technologies and

concepts are combined to aid in management, are deﬁned

as business intelligence systems (Nedelcu,2013). Therefore,

a decision support system can be seen as an assistant sys-

tem for management tasks. The use of various technologies

to aid managers in decision-making is deﬁned as business

intelligence (Baars & Kemper,2021;Gluchowski,2016). In

fact, the term business intelligence has no uniform deﬁnition.

Furthermore, it is important to distinguish between business

analytics and business intelligence. Mashingaidze and Back-

house (2017) show various deﬁnitions of business intelli-

gence (BI) and business analytics (BA) in literature and prac-

tice. BI is deﬁned as a set of integrated strategies, applica-

tions, technologies, architectures, processes and methodolo-

gies in order to gather, store, retrieve and analyze data to sup-

port decision-making (Mashingaidze & Backhouse,2017).

According to Mashingaidze and Backhouse (2017), BA is de-

ﬁned as a set of skills, applications, technologies, architec-

tures, processes and methodologies used to collect, store and

retrieve data for the purpose of data mining in order to sup-

port decision-making, inform business strategy and drive per-

formance. The data mining techniques can be descriptive,

predictive and prescriptive from scientiﬁc disciplines such as

mathematics and statistics. Since this paper outlined the

technological foundations and beneﬁts of algorithmic deci-

sion support it is necessary for summarization to illustrate

the components of decision support system. The ﬁgure 4

shows the framework of Delen and Demirkan (2013). Busi-

ness processes create transaction data which are gathered

in data warehouses. These data warehouses include various

data sources for data mining. Interfaces with other systems

process new data, information or knowledge. The processed

ﬁndings are used for consultation of decision makers or to

identify opportunities and risks (environmental monitoring).

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 895

Figure 3: Illustration of ten automation levels for assisting humans (Source: Own illustration according to Sheridan and

Verplank (1978) and Blutner et al. (2009))

Figure 4: A conceptual framework for service-oriented decision support systems (Source: Delen and Demirkan (2013))

2.2. Deriving the need for acceptance conditions: technol-

ogy acceptance models

Since this paper showed a wide range of beneﬁts of algo-

rithmic decision support on business activities, one can sug-

gest to easily implement algorithmic aid to outperform the

competition. In fact, the beneﬁts of algorithmic aid are not

from the application themselves. Moreover, the integration

of algorithmic aid into business where business processes are

transformed and redesigned deliver these kinds of beneﬁts

(Apté et al.,2012). Further literature shows that the appli-

cation of algorithmic aid can even lead in a failure to real-

ize expected performance gains (Mikalef, Boura, Lekakos, &

Krogstie,2019). Mikalef et al. (2019) outline that organi-

zational aspects and managerial skills are more important in

an uncertain environment than application of technology it-

self. Laudon, Laudon, and Schoder (2016) outline factors

that determine the success or failure of an information sys-

tem. They state that four factors inﬂuence the result of the

implementation in terms of design, cost, usage and data. The

factors of Laudon are the involvement of users and considera-

tion of their inﬂuence which is also supported by (Korsgaard,

Schweiger, & Sapienza,1995). The next factor of Laudon et

al. (2016) is the support from the management. Due to the

fact that this paper focuses on decision support systems used

K. Iqbal /Junior Management Science 8(4) (2023) 887-925896

by managers, managers and users are the same people. How-

ever, the need of support from management is a crucial fac-

tor for decision support systems (Mcafee et al.,2012;Rains-

berger,2021). A further factor of Laudon et al. (2016) is the

degree of complexity and the risk of the implementation pro-

cess. The next factor of Laudon et al. (2016) is the manage-

ment of the implementation process. Laudon et al. (2016)

refer here to the change management process.

In general, the technological change management pro-

cess is related to challenges (Orlikowski, 1992). In speciﬁc,

Larson and Chang (2016) show that the adoption of BI appli-

cations and services is challenging for organizations. Accord-

ing to Savolainen (2016), the commitment to a change pro-

cess can be predicted by acceptance. Scheuer (2020) shows

that the aim of acceptance research is the explaining of be-

havior of users in terms of rejection or afﬁrmation due to

the use of material or non-material (artiﬁcial) technologies.

Therefore he deﬁnes acceptance as the willingness of some-

one to voluntarily accept, acknowledge, approve or agree

with a subject. In fact, a decision suggested by system in-

hibits quality if the decision fulﬁlls the goals and is accepted

by users (Sharma, Mithas, & Kankanhalli,2014).

Scheuer (2020) shows various kinds of acceptance mod-

els in the literature. Acceptance research varies in the point

of technology usage. Various research ﬁelds focus on ac-

ceptance: information system research, marketing research,

behavioral consumer research, psychology and philosophy

(Königstorfer,2008). The information system research is

characterized by the technology acceptance model. The

technology acceptance model (TAM) was ﬁrst introduced

by Davis (1989) and states that the attitude toward using

a technology is inﬂuenced by the perceived usefulness or

perceived ease of use. Furthermore, TAM by Davis (1989)

was extended by social inﬂuence mechanisms and cognitive

instrumental processes who inﬂuence decision making or

an attitude towards using a technology (Venkatesh & Davis,

2000). This extension is deﬁned as TAM 2 (Venkatesh &

Davis,2000). Moreover, a further extension of TAM 2 was

introduced and deﬁned as TAM 3 (Venkatesh & Bala,2008).

TAM 2 by Venkatesh and Davis (2000) is extended by de-

terminants of perceived ease of use by Venkatesh (2000).

The determinants of perceived ease of use are deﬁned in the

following. Perceived Enjoyment is deﬁned as the perception

of joy resulting from system use (Venkatesh,2000). The

second determinant of perceived ease of use is Computer

Self-Efﬁcacy which is deﬁned as the subjective perception to

have the necessary capabilities to perform a speciﬁc task us-

ing a computer. A further determinant by Venkatesh (2000)

is the Computer Playfulness deﬁned as the degree of cognitive

spontaneity while interacting with the computer (Venkatesh

& Bala,2008). Perception of External Control is deﬁned as

the perception of support for the computer from organiza-

tional and technological resources. A further determinant

is Computer Anxiety which stands for the perception of fear

while using the computer. The last determinant of Venkatesh

(2000) is the Objective Usability which is deﬁned as a com-

parison of the system in terms of task completion.

Scheuer (2020) develops an acceptance model for the use

of artiﬁcial intelligence. The model developed by Scheuer is

called KIAM model (Scheuer,2020). The KIAM model is an

extension of the TAM model and is considered as the Artiﬁcial

Intelligence Acceptance Model. Whereas KI is referred to the

german term for AI. The AI acceptance model (KIAM) con-

sists of a holistic acceptance model that addresses the char-

acteristics of the theoretical properties of an AI compared to

a classical computer system. Scheuer assumes that an AI is

accessible via a technology (e.g., a smartphone application)

and enriching it with Narrow AI services (e.g., a chatbot in-

tegration, Speech-To-Text, or Text-To-Speech) through which

a user can interact with the AI in natural language. Based

on this, two essential components emerge ﬁrst, the classical

technology in the form of a software application, and second,

the dialog component for interacting with the AI in the back-

ground. For the classical technology and the investigation of

its acceptance, Scheuer (2020) uses the existing TAM model

by Venkatesh and Bala (2008) TAM 3. However, for the dia-

log component and the resulting interaction between the AI

and the user, Scheuer (2020) differentiates to what extent

the user accepts the AI as a personality and sees the system

as a complete person or as a technology. Furthermore, he

shows that the perception of the system as technology or as

person determines suitability of acceptance models.

For this, he considers that psychological models for mea-

suring sympathy and affection apply as personality accep-

tance takes precedence over pure technology acceptance if

the system is seen as a person. In this regard, Scheuer (2020)

highlights that if the ﬁlter of the perception of the system as a

personality is considered and an AI is recognized as a person-

ality. This relationship with the technology can be described

by interpersonal acceptance models.

On the other hand, if a system is perceived as a technol-

ogy TAM is suitable (Scheuer,2020). He shows that the per-

ception of a system as a technology has an inﬂuence on ac-

ceptance.

Since Scheuer (2020) shows that the perception of a sys-

tem as a technology or a person determines how acceptance

is created it is necessary to consider the perception of the

system for acceptance conditions. Therefore it is necessary

to derive further acceptance conditions for decision support

systems.

Since decision support systems may inﬂuence the deci-

sion of the user it is important to understand the concept of

persuasive technologies in order to derive acceptance con-

ditions. A persuasive computer is an interactive technol-

ogy that changes the attitude or behavior of the user (Fogg,

1998). The intention for technology usage is key here. If

a person is using the interactive technology with the intent

to extend or change his or her attitude or behavior the type

of intent is deﬁned as autogenous by Fogg (1998). The in-

tent is deﬁned as exogenous if the access to the interactive

technology is given by others (Fogg,1998). Furthermore,

the intent of creation or production of interactive technology

is deﬁned as endogenous by Fogg. Additionally, Fogg (1998)

differentiates three types of computer functions. Fogg (1998)

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 897

sees the computer as a tool for reducing barriers, increasing

self-efﬁcacy, decision support and change of mental models.

Moreover, the computer is seen as a medium for providing

experience by insights and visualization, promoting under-

standing of causal relationships and motivating through ex-

perience. At least Fogg (1998) sees the computer as a social

actor who creates relationships by establishing social norms,

invoking social rules and dynamics and providing social sup-

port or sanction. Fogg (1998) demonstrates that it is impor-

tant to see decision support systems as persuasive technolo-

gies with various kinds of functions for usage.

The structuration approach from DeSanctis and Poole

(1994) focuses on the social structures for human activity

provided by technology. Further, they differentiate between

two social structures. Firstly the features of the technol-

ogy called the structural and secondly the general experi-

ence towards the feature of the technology, deﬁned as spirit.

The adaptive structuration approach of DeSanctis and Poole

(1994) can be seen as a structuration approach that high-

lights the user-centricity in terms of user experience (UX) and

user interface (UI). The spirit of the social structure can be

referred to UX due the deﬁnition by Hassenzahl and Tractin-

sky (2006). They deﬁne UX as a consequence of a user’s

internal state while interacting with the technology. UI is

deﬁned as including all aspects of system design that affect a

user’s participation in handling the system (Smith & Aucella,

1983). If the features of the system are not comprehensible

a mismatch between system and user is given leading to a

decreased effectiveness of the information system (Barbosa

& Hirko,1980).

UX and UI optimization may be a fundamental part of

acceptance research for information technologies since Gong

(2008) shows that an anthropomorphized interface leads to

higher social responses from users. Scheuer (2020) identi-

ﬁed that anthropomorphizing a system interface has a posi-

tive inﬂuence on the perception of the system as technology

and as a person. Therefore the hypothesis should be tested

that an anthropomorphizing of a system leads to an ac-

ceptance of the system (H1).

In the following, the necessity of user-centricity due to

the process of implementation of algorithmic decision sup-

port is outlined. Makarius, Mukherjee, Fox, and Fox (2020)

research how to successfully integrate AI into an organiza-

tion. They outline that comprehension of employees is cru-

cial for a successful integration of AI. Furthermore, Makar-

ius et al. (2020) formulate research necessity in the ﬁeld of

trust of decision-makers in the output of decision support sys-

tems, fostering team identiﬁcation between AI systems and

users. Scheuer (2020) identiﬁed a positive inﬂuence of trust

on acceptance. Trust may be an acceptance condition for

decision-support systems. Therefore the hypothesis should

be tested that trust leads to acceptance (H2).

Rainsberger (2021) identiﬁes major challenges in the

adoption of AI systems. According to Rainsberger insufﬁ-

cient knowledge about the beneﬁts of the technology, lacking

trust towards the technology and insufﬁcient resources for

technology implementation hinder AI adoption in an orga-

nization. He summarizes that these fallacies arise due to

ignorance of the decision-makers. The TAM by Venkatesh

and Bala (2008) considers the result demonstrability as the

transparency of the information system in result process-

ing. Since Scheuer identiﬁes that trust is inﬂuenced by the

transparency of the system the hypothesis higher trans-

parency/comprehensibility of a system leads to more

trust is tested (H3).

Gartz (2004) also sees challenges in implementation of

technologies for decision support due to missing awareness

or lack of interest and motivation of management. It is nec-

essary for senior leaders to recognize the importance of deci-

sion support (Grossman & Siegel,2014;Mcafee et al.,2012).

One major challenge here is the change in the decision-

making culture. In important decisions companies often rely

on “HIPPO” (highest paid person’s opinion) (Mcafee et al.,

2012). If a decision support system is introduced which is

by deﬁnition a persuasive technology does the senior leader

or decision maker allow themselves to be overruled by data

(Mcafee et al.,2012)? Orlikowski and Robey (1991) as-

sumes that more information in the decision-making process

leads to a higher power of the decision maker.Does more in-

formation lead to a shift in the decision-making process from

the HIPPO as an expert to HIPPO as an interrogator (Mcafee

et al.,2012) leading to a decreased power? Does the senior

leader accept a decision support system that may lead to a

shift in their role in form of decreased power? Therefore the

hypothesis more transparency leads to higher perceived

participation of the user in the decision-making process

(H4) is tested.

Participation in decision-making leads to a higher percep-

tion of fairness (Korsgaard et al.,1995). Newman, Fast, and

Harmon (2020) show that participation possibilities in the

decision-making process increase trust. Do participation pos-

sibilities in the decision-making process lead to an increased

perception of trust if the output is not comprehensible due

to the black box of algorithmic aid? In order to answer these

questions, the hypothesis the higher the perceived partici-

pation of the system-user in the decision-making process

the higher the perceived trust towards the system (H5) is

tested.

The TAM model considers the perceived usefulness as an

indicator for acceptance. Therefore it is necessary to con-

sider the perceived intelligence of the system as an accep-

tance condition. Furthermore, Scheuer (2020) tests the ef-

fect of perceived intelligence and trust. Despite no evidence

from Scheuer for a relationship between trust and the per-

ceived intelligence, trust may be a mediator for the relation-

ship between perceived intelligence and acceptance. There-

fore, the hypothesis the higher the perceived intelligence

of the system is the higher the trust (H6) should be tested.

As mentioned earlier Scheuer (2020) identiﬁes that the

perception of a system as a technology or as a person creates

acceptance. The following hypotheses should be tested. The

higher the perception of the system as technology is, the

higher acceptance (H7).Furthermore the higher the per-

ception of the system as a person, the higher acceptance

K. Iqbal /Junior Management Science 8(4) (2023) 887-925898

(H8).

2.3. State of the art in literature: acceptance conditions of

algorithmic decision support

In order to answer the research question, it is necessary

to make an analysis of the state of the art in research. Vari-

ous studies do research on the topic of acceptance of artiﬁcial

intelligence-based technologies which is one of the technolo-

gies for algorithmic decision support.

Hastenteufel and Ganster (2021) apply this topic to the

digital transformation in banking. They analyze the accep-

tance of Robo Advisors using a modiﬁed TAM model. There-

fore they use the technology acceptance model by Davis et

al. (1989) as a foundation. Hastenteufel and Ganster (2021)

identify trustworthiness, perceived usability and social inﬂu-

ence as acceptance conditions for algorithmic decision sup-

port. Similarly to Hastenteufel and Ganster (2021), Rathje

et al. (2021) do research about trust in banking. Therefore

they develop their own research model based on the models

by Mayer et al. (1995), Gefen et al. (2003) and Davis (1989).

They conducted a survey with 119 participants where the

afﬁnity to technology is high. Rathje et al. (2021) identify

that trust has a relationship with the intention to use the tech-

nology. Despite relevant ﬁndings for this thesis, these papers

analyze the acceptance on consumer-level.

Gersch et al. (2021) do research about the challenges par-

ticular in trust in a collaborative service delivery with artiﬁ-

cial intelligence in the ﬁeld of radiology. Therefore they con-

duct interviews with various stakeholders in the radiology.

They identify trust as an indicator to cope with uncertain-

ties. Furthermore, Gersch et al. (2021) identify that cogni-

tive trust is built in the ﬁrst contact with the user. With re-

peated experience, the user develops affective trust. Under-

standability and comprehensibility are important for users.

Further challenges are the change of own position in the

workplace due to the introduction of support through arti-

ﬁcial intelligence and arising of new duties and prerequi-

sites in the design of the socio-technical system (Gersch et

al.,2021). Therefore they suggest that explainable artiﬁ-

cial intelligence should consider the perspective of different

stakeholders. Since the objective of research in this paper is

applied to the health industry these results are partially ap-

plicable for this thesis.

Pütz et al. (2021) do research on the topic of accep-

tance of voice and chatbots. They use the technology ac-

ceptance model of Davis (1989) and the extended version of

Venkatesh and Davis (2000) and Venkatesh and Bala (2008)

to analyze the acceptance of this technology. The approach

used by Pütz et al. (2021) is literature-based. They identify

a relation between perceived usability and perceived user-

friendliness. Further results are a relation between perceived

user-friendliness and intention to use the technology. Since

the research of Pütz et al. (2021) focuses on the acceptance

of voice and chatbots these results are applicable to the re-

search question of this thesis if these kinds of technology is

considered.

Lee’s (2018) study shows how people perceive decisions

made by algorithms compared with decisions made by hu-

mans. He made an online experiment by using four manage-

rial decisions that required human or mechanical skills. By

manipulating the decision maker in terms of algorithm and

human he measured the perceived fairness, trust and emo-

tional response. In mechanical tasks, decisions made by al-

gorithms and humans were perceived as equally in fairness,

trustworthiness and evoked similar emotions. Decision made

by humans for mechanical tasks differ in terms of trustworthi-

ness due to the attribution to managers’ authority. Decisions

made by algorithms were perceived as fair and trustworthy

due to attribution perceived efﬁciency and objectivity. In hu-

man tasks, decisions made by humans evoke positive emo-

tions which can be attributed to social recognition. Further

Lee (2018) identiﬁes that human task made by algorithms are

perceived as less fair and trustworthy. Furthermore, decision

made by algorithms in human task evoke negative emotions

due to the perception of a dehumanizing experience of be-

ing tracked and evaluated by machines. The perceived lack

of intuition and subjective decision capabilities caused lower

perception of fairness and trustworthiness. Newman et al.

(2020) analyze the perceived fairness of decision-making by

algorithms in human resource management. They assume

that algorithms increase procedural fairness. Further they

assume that decisions made by algorithms are less accurate

than identical decisions made by humans. Newman et al.

(2020) prove that individuals perceive decisions made by

algorithms as less fair than comparable decisions made by

humans. Further they outline that algorithms are perceived

as reductionistic leading to a decreased perception of fair-

ness. Newman et al. (2020) show that organizational com-

mitment is affected in a negative way by decisions made by

algorithms where the perception of fairness has a mediat-

ing effect. However, the negative effect of decisions made

by algorithms is mitigated in decisions made by hybrid in-

telligence where the human has more involvement. Further-

more, high transparency in algorithmic decisions has a neg-

ative effect on perceived fairness and leads to decontextual-

ization. On the other hand transparency in human decision

leads to an increase in the perception of fairness and cause

less decontextualization.

Lee (2018) shows that the perception of algorithms lies

in the decision context and characteristics of the decision.

Newman et al. (2020) show that the perception of fairness

is human-centered. Despite the strong consideration of Lee

(2018) on acceptance or Newman et al. (2020) on perceived

fairness at worker-level, these results are applicable for this

thesis. Newman et al. (2020) ﬁndings about the role of trans-

parency will be considered for this thesis.

Panagiotarou, Stamatiou, Pierrakeas, and Kameas (2020)

conﬁrm Lee’s (2018) results since they reveal that task char-

acteristics matter in order to understand people’s experi-

ences with algorithmic technologies. Furthermore, they ﬁnd

to prove that participants with different levels of techni-

cal skills have statistical differences in perceived usefulness

of the technology, perceived ease of use, intention to use

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 899

the technology and actual use of the technology. Sagnier,

Loup-Escande, Lourdeaux, Thouvenin, and Valléry (2020)

analyzes the acceptance of Virtual Reality (VR). They iden-

tiﬁed an indirect effect of personal innovativeness on the

intention to use due to the fact that people with high per-

sonal innovativeness have an interest in new technologies

and are more likely to perceive a higher usefulness which

leads to an intention to use Virtual Reality. Besides task char-

acteristics, the literature suggests that the characteristics of

the person who is going to use the technology matter. The

results from (Panagiotarou et al.,2020;Sagnier et al.,2020)

can be considered for further analysis.

Uysal, Alavi, and Bezençon (2022) analyze potential

harmful and beneﬁcial effects while using artiﬁcial intelli-

gent assistants (AIA) such as Alexa. They identify that an-

thropomorphism of artiﬁcial intelligent agents increase con-

sumer satisfaction through increased trust but also threatens

user identity by undermining their comfort by a high degree

of anthropomorphism of the technology. Further, the per-

ception of threat to user-identity increases if the consumer

relationship is closer and the relationship is longer (Uysal

et al., 2021). The perception of threat to human identity

can be mitigated when consumers are aware of data se-

curity solutions and adopt them in relationship with AIA.

The hypothesis by Uysal et al. (2022) that higher anthro-

pomorphism reduces consumer satisfaction and consumer

well-being was not supported. Further ﬁndings of Uysal et

al. (2022) indicate that higher threat to human identity re-

duces consumer comfort through decreasing consumer’s AI

empowerment. This effect is attenuated when consumers

with a long relationship to AIA are aware of data issues of

their usage (Uysal et al.,2022). Scheuer’s (2020) ﬁndings

imply a new acceptance model based on the dual-process

theory (J. S. B. T. Evans & Stanovich,2013;Kahneman &

Schmidt,2012). He distinguishes between acceptance based

on System 1 (IPART) and System 2 (TAM). Scheuer (2020)

identiﬁes that the acceptance for AI systems is dependent

on the acceptance of the speciﬁc technology medium, ac-

ceptance of AI as a technology and the interpersonal accep-

tance. If the degree of anthropomorphism is high the AI is

considered as a personality. The use of AI systems which

are considered as a personality is emotion-driven. Therefore

Scheuer (2020) states that TAMs are not suitable to measure

acceptance of an AI system if an AI system is considered as

personality. Interpersonal acceptance models should be con-

sidered to describe acceptance of AI systems. On the other

hand, TAMs are suitable if the user perceives the AI system as

a technology (Scheuer,2020). Furthermore, Scheuer iden-

tiﬁes that users seek to use an AI system where the degree

of machine learning is controllable and transparent. Since

Panagiotarou et al. (2020) and Sagnier et al. (2020) showed

the relevance of task characteristics, Uysal et al. (2022) and

Scheuer (2020) put emphasis on the relevance of system

features by anthropomorphizing the interface. The ﬁndings

of Uysal et al. (2022) and Scheuer (2020) have a high ap-

plicability for this thesis. The specialty of Scheuer’s research

approach is that he refers to the Turing’s idea of intelligence.

Considering the computer developed to the point where it is

perceived as a human being by the user (Turing,1950). Since

no literature except Scheuer (2020) focuses on interpersonal

acceptance Scheuer’s ﬁndings contribute to this thesis.

Bader and Kaiser (2019) research on the assessment

of the role of artiﬁcial intelligence in workplace decisions.

Bader and Kaiser (2019) outline the spatial and temporal de-

tachment of decision-making. They explore how users deal

with algorithmic decision-making and how user interfaces

inﬂuence the involvement of decision-making. Bader and

Kaiser (2019) argue due to sociomateriality the detachment

to decision-making gets reduced. They outline that AI has

a dual role in workplace decisions. On the one hand, AI

creates human attachment due to emotion driven affective

entanglement. On the other hand, AI facilitates detachment

due to deferred decisions and manipulation in data. The dual

role of AI results in high and low involvement in interactions

with the interface. The involvement of interfaces in research

will be necessary for this paper.

Merendino et al. (2018) explore whether Big Data has

changed strategic decision-making processes on board-level.

They identify a lack in cognitive capabilities in order to cope

with Big Data. Furthermore, they outline a friction in group

cohesion on board-level which has consequences on the

decision-making process. Merendino et al. (2018) show that

boards seek new ways of working in order to avoid informa-

tion silos and relying on capabilities of third parties such as

consultants in order to handle Big Data. Merendino et al.

(2018) ﬁndings are applicable to this thesis due to the fo-

cus on managers. However, Merendino et al. (2018) results

address decision making of managers in a group.

Abhari, Vomero, and Davidson (2020) analyze the psy-

chological motivation behind the use of BI tools. Therefore

they use the Needs-Affordances-Features framework by Kara-

hanna, Xin Xu, Xu, and Zhang (2018). At ﬁrst, they identify

that the need for autonomy and competence in business en-

vironment motivates the use of BI tools where psychological

affordance features of autonomy, collaboration and commu-

nication are addressed. Further, they outline that the need

for relatedness, having a place and self-realization motivates

the use of BI that afford the psychological features of col-

laboration and communication. Since Abhari et al. (2020)

researches the adoption of BI on a voluntary user-level the

ﬁndings of them are applicable for this thesis.

Meske, Bunde, Schneider, and Gersch (2022) show that

explainability is a prerequisite for fair AI. Therefore Meske

et al. deﬁne explainable AI (XAI) by distinguishing it from

interpretable AI. If humans can directly make sense of a ma-

chine’s decision without additional explanation interpretable

AI is given (Guidotti et al.,2018). Giving additional informa-

tion for an explanation as a proxy to comprehend the arguing

process is deﬁned as XAI (Adadi & Berrada,2018).

The TAM by Venkatesh and Bala (2008) considers per-

ceived comprehensibility of the system as an acceptance con-

dition. An explanation as proxy tries to create acceptance by

having a higher result demonstrability. XAI tries to overcome

the boundary between artiﬁcial and material by approaching

K. Iqbal /Junior Management Science 8(4) (2023) 887-925900

the cognitive System 2 of the human. By deﬁnition, it reduces

simultaneously the amount of information approaching the

cognitive system System 1 of the human by using availability

heuristics. System 1 may lead to misinterpretation which can

be reduced through a personalized XAI according to Meske et

al. (2020). As the literature showed high lack in comprehen-

sibility of the decision support systems, it may be challenging

to create an acceptance through XAI.

The state of the art in literature shows that the research

on acceptance in algorithmic decision support is new as the

oldest literature is from 2018. Furthermore, few studies such

as Merendino et al. (2018) analyze the acceptance of algo-

rithmic decision support on managerial-level. However, they

do not consider the acceptance of a single manager since they

focus on board-level decision-making. Other studies show

various results in the acceptance of algorithmic decision sup-

port. These studies do not focus on managerial-level. The

study of Uysal et al. (2022) shows that anthropomorphism

can lead to an increased consumer satisfaction and increase

in trust. Since trust has a relationship to the intention to use

the technology Rathje et al. (2021), it is necessary to consider

anthropomorphizing of decision support systems for an anal-

ysis of acceptance. Therefore anthropomorphizing an inter-

face can be seen as the structuration approach by DeSanctis

and Poole (1994).

No literature except Scheuer (2020) and Uysal et al.

(2022) focuses on the acceptance of anthropomorphized

systems. Furthermore Scheuer (2020) showed that anthro-

pomorphizing AI systems leads to an use that is emotion-

driven. Since Bader and Kaiser (2019) show a dual role of

AI in the workplace, it is necessary to analyze empirically

whether anthropomorphizing the AI system may mitigate

the detachment from AI system. The answer to this ques-

tion may lead to conditions for an acceptance of algorithmic

support on managerial level.

3. Analyzing acceptance conditions: methodological ap-

proach

This thesis aims to answer the following research ques-

tion: which conditions lead to an acceptance of algorithmic

decision support in management? To answer the research

question two approaches are chosen. A vignette study is

conducted along with a quantitative survey. The results are

empirically analyzed and afterwards, a structural equation

model is derived to illustrate the conditions that may lead to

the acceptance of algorithmic decision support. In this re-

gard, it is ﬁrst important to explain why in the context of

the research question it was important to select the innova-

tive approach of a vignette study along with a quantitative

survey. According to this, in the following section the ratio-

nale behind the methodological approach of this thesis will

be presented.

3.1. Theoretical foundation of a vignette study

One of the most frequent tools to investigate the beliefs,

attitudes and judgments of respondents is the combination

of quantitative research and vignette study. Vignette stud-

ies are particularly helpful when research is designed to as-

sess judgment from respondents about speciﬁc scenarios. In

academic literature quantitative surveys along with vignette

analysis were innovational breakthroughs as they allowed a

new way of assessing public opinion in form of a survey while

keeping the element of integrating contextual perception of

speciﬁc situations. In the past, quantitative vignette studies

have been used in different disciplines such as psychology

by Barrera and Buskens (2007), Dülmer (2001)orWalster

(1966) and marketing by Wason et al. (2002) or sociology

by Alves and Rossi (1978), Beck and Opp (2001)orJasso

and Webster Jr (1999). Atzmüller and Steiner (2010) deﬁne

a vignette as a “carefully constructed description of a person,

object, or a situation representing a systematic combination

of characteristics.” (Atzmüller & Steiner,2010). According

to Dubinsky, Jolson, Kotabe, and Lim (1991), vignettes help

identify management decisions. They especially outline that

“Vignettes can be particularly illuminating with respect to

managerial implications; an appropriately constructed and

relevant [vignette]can help management discern where spe-

ciﬁc action is necessary” (Dubinsky et al.,1991). Another

beneﬁt of vignette studies is that the design of vignettes al-

lows to simultaneously present several explanatory as well as

context-dependent factors through which more realistic sce-

narios are possible (Atzmüller & Steiner,2010). Moreover,

vignettes can be presented in different forms such as text, di-

alog, cartoons, pictures, audios, or videos. Depending on the

research setting and research question a vignette can inhibit

an experimental design feature.

Several researchers argue that vignette surveys are supe-

rior to normal question-based surveys. In this regard, Wason

and Cox (1996) support this statement by outlining that vi-

gnette surveys provide greater realism. Robertson (1993)

underlines that vignette studies offer a greater range of sit-

uational and contextual factors. Similarly, Barnett, Bass,

and Brown (1994) state that vignette studies “approximate

real-life decision-making situations”. Alexander and Becker

(1978) further explain that vignette studies supply standard-

ized stimuli to all respondents which makes a replication of

the study easier and enhances the measurement reliability.

On the other hand, Cavanagh and Fritzsche (1985) argue

that vignette studies also improve construct validity (Wason

et al.,2002). Furthermore, they outline that vignette studies

increase the involvement of the respondents and decrease the

potential of errors through not paying attention to questions

or answering the same throughout the survey.

Researchers claim that in the context of vignette stud-

ies the target group plays an essential role and an appro-

priate population should be selected. Stevenson and Bodkin

(1998) argue that with regard to the decision-making process

vignette studies can be targeted toward students as the stu-

dents are tomorrow’s business professionals. Regarding the

design of vignette studies, researchers suggest that vignettes

should be designed adequately and not much detailed. Hy-

man and Steiner (1996) argue that vignettes should not be

so detailed that they overburden respondents.

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 901

Grant and Wall (2009) highlight that especially in the

context of management research it is important to under-

stand causal relationships which in turn requires the use

of experimental or quasi-experimental designs. Through

vignette studies, exactly this aspect is addressed as this re-

search design improves our knowledge about causal rela-

tionships (Aguinis & Bradley,2014). The vignette survey

methodology tackles participants with carefully constructed

and realistic scenarios to assess dependent variables includ-

ing intentions, attitudes, and behaviors. An example of pro-

viding insights on the causal relationships through vignette

surveys is illustrated by McKelvie, Haynie, and Gustavsson

(2011) where they addressed the impact of uncertainty in

the decision-making process of entrepreneurs. In particular,

they provided an evidence on which type of uncertainty had

an effect on whether entrepreneurs choose to exploit or not

to exploit opportunities (Aguinis & Bradley,2014;McKelvie

et al.,2011). Aguinis and Bradley (2014) conducted a re-

view on 30 management journals from 1994 to 2003 and

provided evidence that vignette surveys are a way to address

the problem of internal and external validity.

In this regard, vignette surveys have been used in sev-

eral contexts and formats. Cook (1979) investigate whether

Americans support programs for social groups in need of aid

or not. For this, they used text vignettes. On the other hand,

Atzmüller, Kromer, and Elisabeth (2014) took a closer look

at peer violence among adolescents. For their approach, they

used short video vignettes. Also, audio vignettes have been

used for example by Atzmüller et al. (2014) to investigate

radio news on crimes. Several scholars claim that vignette

surveys are ﬂexible and have a wide range as they allow par-

ticipants to come out of their comfort zone and perceiving

different experimental settings in form of videos, audios, text

etc. Moreover, vignette surveys allow participants to move

away from socially desirable answers or politically correct an-

swers which in turn reduces biases (Steiner, Atzmüller, & Su,

2016).

Based on the aforementioned aspects a vignette survey

seems to be an appropriate tool to ﬁrst, identify management

decisions (Dubinsky et al.,1991). Secondly, to include dif-

ferent experimental settings in a survey such as videos, au-

dios, text, etc. (Steiner et al.,2016). Thirdly, to construct

realistic scenarios and consider context-dependent aspects

(Atzmüller & Steiner,2010). Because of this, the approach of

a vignette survey was selected to answer the research ques-

tion. In the next section it will be explained why structural

equation modeling is relevant in the context of the research

question and why is it used for the empirical approach in

this thesis. In particular, why is structural equation modeling

used to illustrate the results of the survey.

3.2. Structural equation modeling

With the method of structural equation modeling (SEM)

it is possible to simultaneously model complex relationships

among multiple dependent and independent variables (Hair

Jr. et al.,2021). Moreover, there are two options in SEM

which are common factor-based-SEM (CB-SEM) and partial

least squares SEM (PLS-SEM). In this regard, the option of

common factor-based SEM is mostly used in the context of

accepting or rejecting hypotheses, which serves as an indi-

cator to conﬁrm or reject theories. In a practical manner,

this approach of common factor-based-SEM investigates how

closely a proposed theoretical model is able to reproduce a

covariance matrix for the considered dataset. On the other

hand, we have the PLS-SEM. For this thesis, a PLS-SEM was

used. In the following, it will be explained why PLS-SEM

is more appropriate than a common factor-based SEM for

this thesis. PLS-SEM should be conducted if the objective of

research is an exploratory research for theory development

(Hair Jr. et al.,2021). The objective of research of this thesis

is to identify acceptance conditions. Therefore, the approach

of PLS-SEM ﬁts to the objective of research of this thesis.

According to Jöreskog and Wold (1982), PLS-SEM is a

“causal predictive” approach and aims at explaining the vari-

ance of the dependent variable. Basically, a partial least

square (PLS) path consists of two essential elements.

One element is the inner model whereas the other one is

the outer model. The inner model is referred as a structural

model which links together constructs. The outer model,

on the other hand, is referred as the measurement model.

These measurement model shows the relationships between

the constructs and the indicator variables as rectangles. The

ﬁgure 5demonstrates the inner and outer model. Another

beneﬁt of PLS-SEM is that there is a high efﬁciency in pa-

rameter estimation, and it is ﬂexible in terms of its model-

ing properties. According to Hair Jr., Matthews, Matthews,

and Sarstedt (2017), PLS-SEM is a prediction-oriented ap-

proach and is mostly used in exploratory research. PLS-SEM

maximizes the amount of explained variance of endogenous

constructs in a path model and provides a better understand-

ing of the underlying causes and predictions (Shmueli et al.,

2019).

In addition to this through PLS-SEM, it is also possible

to include control variable to account for the target con-

struct’s variation. Furthermore, PLS-SEM allows the assess-

ment of not only reﬂective but also formative measurement

models along with single-item constructs, with no identiﬁca-

tion problems. Regarding the single items, it can be said that

they have the advantage of being not complicated in terms

of the scales and result in higher response rates where the

questions are easily understood and answered (Fuchs & Dia-

mantopoulos,2009;Sarstedt & Wilczynski,2009). Hair Jr. et

al. (2021) further point out that a global single item is sufﬁ-

cient and captures the essence of the construct, especially in

the context of executing a redundancy analysis.

As explained in the aforementioned section, the path

model for PLS-SEM will be presented. This thesis aims to an-

swer the following research question: which conditions lead

to an acceptance of algorithmic decision support in manage-

ment? In the theoretical foundation, acceptance conditions

for algorithmic decision support were derived and formu-

lated as hypotheses. The state-of-the-art shows that the

degree of anthropomorphizing an AI system may lead to an

acceptance. Therefore, a new model (ﬁgure 6) is derived

K. Iqbal /Junior Management Science 8(4) (2023) 887-925902

Figure 5: A simple path model (Source: Hair Jr. et al. (2021))

from the previously formulated hypotheses.

Due to hypotheses tests the path model is tested for va-

lidity using PLS-SEM. The hypotheses tests are executed in

a two-step process according to Hair Jr. et al. (2021). Hair

Jr. et al. (2021) suggest to ﬁrst conﬁrm reliability and valid-

ity of the measurement model and then testing the structural

equation model for validity.

In order to create a measurement model, a survey is

conducted where the degree of anthropomorphizing of the

system-interface is manipulated. The items from the survey

are used for creating a reﬂective measurement model.

The variables in the path model are measured as latent

constructs in a reﬂective measurement model. The deﬁnition

of constructs is described in table 1. In table 2the hypotheses

are summarized.

In an experimental setting, these hypotheses are tested

for validity. These interactions with a system is simulated by

a vignette study.

3.3. Design and parameters of the survey

A vignette study is conducted in german language where

the scenario is described. In this scenario the survey partic-

ipant is in the situation of a manager in a dynamic market

environment where he has to make a hot decision accord-

ing to Janis and Mann (1977). Furthermore, a decision sup-

port system aids in the decision-making. The decision sup-

port system is introduced in the scenario with high prediction

capabilities. Moreover, the decision support system is imple-

mented as an interface. The survey participants simulate an

interaction with a decision support system. Due to the inter-

action, the survey participants are involved in the problem-

ﬁnding process. Furthermore, the system suggests a solution

to the problem without further explanation. The survey par-

ticipants have to make the choice to accept the suggestion or

to reject the suggestion and make their own decision. The

structure of the survey is shown in ﬁgure 7.

There are two interfaces with a different degrees of an-

thropomorphizing features. The interface with low anthro-

pomorphizing features is created by embedding HTML and

javascript code into the survey tool. The interface is named

Lisa, shown in ﬁgure 8.

Furthermore, the interface implemented as an interac-

tive video with high anthropomorphizing features is named

Maria, shown in ﬁgure 9.

The degree of anthropomorphizing features is relatively

high due to the use of professional tools. In order to cre-

ate the interface Maria, an AI actor is created with the tool

Colossyan (Colossyan,2022) and saved as a video. Further-

more, the tool Tolstoy (Tolstoy,2022) is used to make the

interface more interactive. Therefore the videos created by

Colossyan (2022) are ordered through the use of various con-

ditions leading to a high degree of anthropomorphizing fea-

tures. The degree of anthropomorphizing could have been

maximized through voice inputs. Despite the substitution of

voice input for the interaction of the user via textual or but-

ton components, voice outputs could be implemented in the

interface. Furthermore, the AI actress uses gestures while

speaking.

The participants are randomly distributed to one inter-

face. After the interaction with the interface, the participant

has to make a decision, where he can approve the suggested

decision by the system or reject the suggestion and choose an

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 903

Figure 6: Path model for hypotheses testing (Source: Own illustration)

Table 1: Construct deﬁnition

Constructs Description

Perception as technol-

ogy is abbreviated as

“TEC”

The extent to which the system-user perceives the an-

thropomorphized system as technology. A low value (1)

for perception as technology indicates that the system is

not perceived as a technology.

Perception as person is

abbreviated as “PER”

The extent to which the system-user perceives the an-

thropomorphized system as a person. A low value (1)

for perception as person indicates that the system is not

perceived as a person.

Trust is abbreviated as

“TRU”

The extent to which the user has trust towards the sys-

tem. A low value (1) for trust indicates that the user

does not trust the system.

Transparency is abbre-

viated as “TRAN”

The perceived comprehensibility of the system results. A

low value (1) for transparency indicates that the results

processed by the system were not perceived as trans-

parent in terms of comprehensibility of the decision-

making.

Computer Power is ab-

breviated as “CPOW”

The perceived control of the system within the decision-

making process by the user. A low value (1) for computer

power indicates that the system-user perceives his par-

ticipation in decision-making process as high. A high

value (5) indicates that the system-user perceives the

participation of the system in decision-making process

as high.

Intelligence is abbrevi-

ated as “INT”

The perceived intelligence of the system. A low value

(1) for intelligence indicates that the user perceives the

system as not intelligent.

Acceptance is abbrevi-

ated as “AIACC”

The willingness to voluntarily approve the system. A

low value (1) for Acceptance indicates that the user is

not willing to use the presented system in the future.

Decision is abbreviated

as “ACC”

The ﬁnal decision after receiving aid from the system.

A low value (0) indicates that the user has rejected the

suggested decision from the system. A high value (1)

indicates that the user has accepted the suggested deci-

sion from the system.

K. Iqbal /Junior Management Science 8(4) (2023) 887-925904

Table 2: Formulation of hypothesis

Construct No. Hypothesis

Acceptance

(ACC; AIACC)

H1 An anthropomorphizing of a system leads to an ac-

ceptance of the system.

Acceptance

(ACC; AIACC)

H2 A higher trust in a system leads to higher acceptance.

Trust

(TRU)

H3 A higher transparency/comprehensibility of a sys-

tem has a positive effect on the trust towards the

system.

Computer

Power

(CPOW)

H4 A higher comprehensibility of a system has a positive

effect on the perceived participation of the user in

the decision-making process.

Trust

(TRU)

H5 The higher the perceived participation of the system-

user in the decision-making process is, the higher the

perceived trust towards the system.

Trust

(TRU)

H6 The higher the perceived intelligence of the system

is, the higher the trust.

Acceptance

(ACC; AIACC)

H7 The higher the perception of the system as technol-

ogy is, the higher the acceptance.

Acceptance

(ACC; AIACC)

H8 The higher the perception of the system as person is,

the higher the acceptance.

Figure 7: Structure of the survey (Source: Own illustration)

Figure 8: System interface of Lisa (Source: Own illustration)

alternative decision. Furthermore, the participants are sur-

veyed for their experience while interacting with the system.

In the end demographical data were surveyed.

The survey was created with the survey tool Unipark.

This tool saves cookies on the devices of the participants and

prevents multiple participations in the survey from the same

user. The survey was online from 25.07.2022 to 07.08.2022

and distributed via various channels. Despite no speciﬁc tar-

get group, the target group was varied across the distribu-

tion channels. The survey was shared on social networks

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 905

Figure 9: System interface of Maria (Source: Own illustration)

like Linkedin, Instagram and Whatsapp. The target group

on Linkedin was speciﬁed as managers or people in lead-

ership positions. The Linkedin post where the survey was

shared had impressions of 6824 (07.08.2022) meaning the

call for participation has reached 6824 people. Furthermore,

students and researchers were targeted due to the distribu-

tion of ﬂyers. 1000 Flyers were printed. People in the uni-

versity were asked to participate in the survey while giving

them the ﬂyer. Furthermore, a ﬂyer could be used multiple

times meaning that minimum of 1000 students or researchers

could be approached by ﬂyers. The survey was shared multi-

ple times on multiple Whatsapp and Instagram accounts with

daily views of approximately 200 people. Leading to a dis-

tribution of approximately 3000 people as a non-speciﬁable

target group. The sharing activities lead to a distribution of

the survey to 11000 people who could be accounted multiple

times.

746 people clicked on the survey and 253 people can-

celed their survey participation or did not give their consent

to the survey leading to 493 people who started the survey.

212 people canceled their survey participation after starting

the survey leading to 281 people who have fully participated

in the survey. Due to the cancel activities, an equal distri-

bution of the interfaces among the participants could not be

guaranteed.

3.4. Assessment of measurement model

The latent constructs were measured by the previous de-

scribed survey. The operationalization of constructs was

derived from the study of the KIAM-Model by Scheuer

(2020). Since Scheuer considers various items from TAM

by Venkatesh and Bala (2008) in his research, the items from

TAM are used for this study. The operationalization of the

items is executed in german language since the target group

of the study are german speaking students and employees.

All operationalized items for the interface Lisa are shown in

table 3. Items for the interface Maria are mostly identical to

the interface Lisa. The items for Maria only differ from Lisa

when the item consists the name of the interface. Similar

to the study of Scheuer (2020) the measurement of items is

executed on a 5-point-likert scale. Scheuer argues that the

use of this scale minimizes time for the survey participants

and delivers a higher precision than on a 7-point-likert scale

due to more intuitive responses on perception.

On the 5-point-likert scale the rejection of the statement

is coded as one and is increased by one each for a lower re-

jection or higher afﬁrmation of the statement where the max-

imum afﬁrmation of the statement is coded with a ﬁve.

The construct of acceptance was measured with a further

construct deﬁned as “ACC” where the variable is named as

“acc” and is measured on a binary scale. The binary scale

is used to assess the acceptance due the interaction with the

system. A zero in this binary variable reﬂects rejection of the

suggested decision. Moreover, one reﬂects an afﬁrmation of

the suggested decision.

The construct of acceptance could have been measured as

a higher order construct, where the constructs “AIACC” and

“ACC” are reﬂective measures for the higher order construct.

This study composites the constructs “AIACC” and “ACC” as

no single constructs because the estimation of the SEM is con-

ducted partially for each acceptance construct leading to a

higher accountability of acceptance conditions. A composi-

tion of both constructs to one higher order construct may dis-

tort the results due to the different scaling of both constructs.

Therefore two SEM are estimated where the construct “ACC”

is used to validate the results from the estimation with the

construct “AIACC”.

As shown previously Scheuer (2020) states that the per-

ception of a system as a technology or a person determines

how acceptance is created. Therefore it is necessary to sep-

arate the measurement of both systems into individual mea-

surement models. A separation of measurement models in-

K. Iqbal /Junior Management Science 8(4) (2023) 887-925906

Table 3: Operationalization of items

Construct Item Question Reference

INT int_1 “Das System wirkt intelligent” “Das System wirkt intelligent”

(Scheuer,2020)

TEC tec_1 “Ich habe das System als Technologie

wahrgenommen”

“Ich habe das System als Technologie

wahrgenommen“ (Scheuer,2020)

TEC tec_3 “Ich habe Lisa als Technologie

wahrgenommen”

“Ich habe das System als Technologie

wahrgenommen“ (Scheuer,2020)

PER per_1 “Ich habe das System als Persön-

lichkeit wahrgenommen“

“Ich habe das System als Persön-

lichkeit wahrgenommen“ (Scheuer,

2020)

PER per_2 “Ich habe Lisa als Persönlichkeit

wahrgenommen“

“Ich habe das System als Persön-

lichkeit wahrgenommen“ (Scheuer,

2020)

PER per_3 “Ich habe in Lisa Menschlichkeit

wahrgenommen“

CPOW part_1 „Ich habe keine Kontrolle über die

Nutzung des Systems“

“Ich habe Kontrolle über meine

Nutzung des Systems”–TAM

(Scheuer,2020)

TRAN tran_1 “Die Entscheidung des Systems ist

transparent”

TRAN tran_2 “Ich kann die Entscheidung des Sys-

tems nachvollziehen“

TRAN tran_3 “Für mich ist hinreichend transpar-

ent, wie das System funktioniert“

“Für mich ist hinreichend trans-

parent, wie das System funktion-

iert“(Scheuer,2020)

TRU tru_1 Ich vertraue dem System “Ich vertraue dem System” (Scheuer,

2020)

TRU tru_2 Das System wirkt vertrauensvoll -

TRU tru_3 „Ich vertraue auf die Ergebnisse des

Systems“

“Ich vertraue auf die Ergebnisse

des Systems” (Scheuer,2020)

AIACC aiacc_1 „Die zuvor vorgestellte künstlichen

Intelligenz würde ich aktiv verwen-

den, wenn ich Zugriff auf dieses Sys-

tem habe und die Rahmenbedingun-

gen gegeben sind“

“Eine künstliche Intelligenz wie

dieses System würde ich aktiv ver-

wenden, wenn ich Zugriff auf diese

habe und die Rahmenbedingungen

gegeben sind” (Scheuer,2020)

AIACC aiacc_2 „Angenommen ich hätte Zugriff auf

das System, würde ich es nutzen

wollen“

“Angenommen ich hätte Zugriff auf

das System, würde ich es nutzen

wollen”-TAM (Scheuer,2020)

AIACC aiacc_3 „Ich würde das System freiwillig

nutzen, wenn die Rahmenbedingun-

gen gegeben wären“

“Ich würde das System, wenn

die Rahmenbedingungen gegeben

wären, freiwillig nutzen”-TAM

(Scheuer,2020)

ACC acc Decision of the user -

creases the accountability of estimation for the certain inter-

face. A separation of measurement leads to major challenges

in the minimum sample size required.

For path coefﬁcients of minimum 0.11 a minimum sample

size of 113 is required to have signiﬁcant path coefﬁcients

on a 10 % signiﬁcance level. Furthermore, a sample size of

minimum 155 is required to have signiﬁcant path coefﬁcients

on a 5 % signiﬁcance level (Hair Jr. et al.,2021).

Since the measurement model of Maria has a sample size

of 127 and the measurement model of Lisa has a sample size

of 154 the requirements for signiﬁcant path coefﬁcients on

a 10 % signiﬁcance level are fulﬁlled. The minimum sam-

ple size required for path coefﬁcient with a minimum value

of 0.21 is 112 with a signiﬁcance level of 1% (Hair Jr. et al.,

2021). Both measurement models exceed the minimum sam-

ple size required for signiﬁcant path coefﬁcients with a min-

imum value of 0.21.

In the following, the sample of the survey is described.

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 907

4. Analysis and ﬁndings of the survey

In the next section, the descriptive statistics of both mea-

surement models are shown for describing the underlying

sample. Furthermore, the data from the survey is analyzed

according to Hair Jr. et al. (2021). At ﬁrst, the quality indi-

cators for both measurement models are assessed. Moreover,

the quality indicators of the structural models are evaluated.

At the end of the section, the results from the analysis are

discussed.

4.1. Descriptive statistics: ﬁrst ﬁndings

The sample size is 281, where eight people did not re-

spond to sociodemographic questions. Furthermore, the

average age of the participants is 25,66 years where the

youngest participant being 18 years old and the oldest par-

ticipant being 66 years. The distribution of ages is shown in

the appendix. In addition to this 57.14 % of the participants

were male and 41.03% were female. 1.47% of survey respon-

dents were not speciﬁable and 0.37% of survey participants

classiﬁed their gender as diverse.

More than 50% of the survey participants were students.

The second largest group of the survey is classiﬁed as Man-

agers. Further job descriptions of the survey participants are

shown in ﬁgure 10.

Furthermore, 23.13% of survey participants stated that

they have already gathered management or leadership ex-

perience with a duration of more than two years. Stan-

dard Industrial Classiﬁcation (SIC) was used to determine

the branch classiﬁcations of the companies of survey partic-

ipants. The majority of branches were not speciﬁable. 62

survey respondents classiﬁed their branch in “Services”. The

branch of “Agriculture, Forestry, Fishing” and “Mining” was

not present in the sample. Furthermore, the survey partic-

ipants were asked to classify their organization. 32.6% of

survey participants responded that the classiﬁcation is not

speciﬁable. 17.95% of survey participants stated that they

are working in public institutions. Further classiﬁcations of

job institutions are shown in ﬁgure 11.

Since Panagiotarou et al. (2020) show the relevance of

personal innovativeness on acceptance, it is necessary to have

insights into the afﬁnity to technologies or personal innova-

tiveness of survey participants. At ﬁrst, 91.21% of survey par-

ticipants evaluated that they have the necessary capabilities

for handling Ofﬁce-software. Furthermore, 68,5% of survey

participants responded that they have no experience in pro-

gramming. 15.02% of participants stated that they have re-

cently gathered experience in programming (less than years).

16.48% of participants reported that they have more than

years of experience in programming. Since programming ca-

pabilities afford a high level of technical afﬁnity 31.5% of

survey participants can be classiﬁed as the minimum share

of survey participant with a high level of technical afﬁnity.

People who spend time for gaming have the need to inform

their self about the latest hardware. Therefore the survey

measured the technical afﬁnity by asking survey participants

whether they like to spend their free time gaming. 40.66%

of survey participants have answered with „Yes“ to this ques-

tion. Besides the gaming experience, it is important to mea-

sure the experience with VR-technology for the assessment

of personal innovativeness. Since Sagnier et al. (2020) ar-

gue that the use of new technologies like VR is related to

personal innovativeness, the survey participants were ques-

tioned whether they have used VR-technology before. 51.28

% of survey participants have answered with „Yes“ to this

question. Overall there is a high afﬁnity to technology and

personal innovativeness among the survey participants. The

experience in handling Ofﬁce-software may not be a measure

for technical afﬁnity because these are relevant job capabili-

ties and are often expected as general knowledge in practice.

The high share of participants who have Ofﬁce-capabilities

shows a representativeness of the sample since these capabili-

ties are expected as general knowledge. Furthermore, 31.5%

to 51.28% of survey participants can be classiﬁed as partic-

ipants with a higher level of technical afﬁnity leading to an

overall high technical afﬁnity of the sample.

4.1.1. Descriptive statistics: measurement model of Lisa

The summary of descriptive statistics of the measurement

model Lisa are shown in table 4.

The maximum variance for a ﬁve-point likert scale on

mathematical foundation is 2.00 3. Therefore the measure-

ment model of Lisa shows a moderate to high variance within

the variables. The acceptance variables show a low to mod-

erate variance. As anticipated the Interface Lisa is perceived

more as a technology (mean =3.9870) than a person (mean

=2,4156).

Furthermore, the correlation matrix of constructs (table

5) shows that “CPOW” is negatively correlated to other con-

structs which is expected. The correlation of the perception

of the system as technology is negatively correlated with the

perception of the system as person. These negative correla-

tions imply a validity of the measurement.

4.1.2. Descriptive statistics: measurement model of Maria

The descriptive statistics measurement model of Maria

are shown in table 6. As shown previously the maximum

variance for a ﬁve-point likert scale is 2.00. Therefore the

measurement model of Maria shows a moderate to high vari-

ance within the variables. Similar to measurement model of

Lisa, acceptance variables show a low to moderate variance.

Furthermore, the interface Maria is perceived more as

a technology (mean =4.1417) than a person (mean =

2,5669). Since Maria is an anthropomorphized interface

the perception of the system as technology should be lower

3High dispersion on ﬁve-point likert scale means that every number

should be distributed equally among the scale. Therefore a quantity of one

number at each point of scale can be considered for further calculations due

to the reduction of complexity. The average among the scale is equal to the

median. The average is three. This average is considered for the variance

calculation. The sum of squares of the difference between the observation

and the mean is equal to 10. 10 is divided by the number of observations,

leading to a variance of 2.

K. Iqbal /Junior Management Science 8(4) (2023) 887-925908

Figure 10: Job description of survey participants (Source: Own illustration)

Figure 11: Classiﬁcation of job institutions (Source: Own illustration)

than the perception of the system as technology. In fact, the

system Lisa has a lower mean value for the perception as

technology than the system Maria, which may indicate that

the anthropomorphized system has failed the Turing test

(Turing,1950). Furthermore, the mean value for acceptance

parameters of the system Lisa are slightly higher than the

system Maria. The correlation matrix in table 7shows that

“CPOW” is no more negatively correlated to all other con-

structs which was not expected. Similar to the measurement

model of Lisa the correlation of the perception of the system

as technology with the perception of the system as person is

negative which indicates a validity of the measurement.

4.2. Quality indicators for measurement and structural

model

Hair Jr. et al. (2021) show that the ﬁrst step in the as-

sessment of measurement models is the examination of indi-

cator reliability. Hair Jr., Risher, Sarstedt, and Ringle (2019)

recommend indicator loadings, above 0.708 as reliable indi-

cators. Indicators under this threshold should be considered

for a removal. Indicator loading below 0.4 should always

be eliminated from the measurement model (Hair Jr. et al.,

2021).

Therefore, a factor analysis was conducted in R using the

package “seminr” by (Hair Jr. et al.,2021). The results of

the factor analysis are shown in the appendix. The measure-

ment model was adjusted by deleting indicators with low

values for indicator loadings. The ﬁnal stage of the factor

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 909

Table 4: Descriptive summary of the measurement model of Lisa

Descriptive statistics

n min q25 mean median q75 max sd var

l_per_1 154 1 2 2.4091 2 3 5 1.1411 1.3021

l_per_2 154 1 1 2.3312 2 3 5 1.1719 1.3733

l_per_3 154 1 2 2.4156 2 3 5 1.0646 1.1334

l_int_1 154 1 3 3.3766 4 4 5 1.1265 1.2690

l_tran_1 154 1 1 2.0974 2 3 5 1.1130 1.2388

l_tran_2 154 1 2 2.6299 3 4 5 1.1713 1.3719

l_tran_3 154 1 1 2.2338 2 3 5 1.2250 1.5006

l_part_1 154 1 2 2.9481 3 4 5 1.1013 1.2130

l_tru_1 154 1 2 2.9870 3 4 5 1.0846 1.1763

l_tru_2 154 1 2 3.1039 3 4 5 1.0427 1.0872

l_tru_3 154 1 2 3.1494 3 4 5 1.0833 1.1736

l_tec_1 154 1 4 3.9870 4 5 5 0.9768 0.9541

l_tec_3 154 1 3 3.8571 4 5 5 1.0125 1.0252

l_aiacc_1 154 1 3 3.5974 4 4 5 0.9533 0.9088

l_aiacc_2 154 1 3 3.5649 4 4 5 0.9283 0.8618

l_aiacc_3 154 1 3 3.5779 4 4 5 0.9754 0.9514

l_acc 154 0 1 0.7987 1 1 1 0.4023 0.1618

Table 5: Correlation matrix of Lisa

Correlations of constructs of Lisa

TEC PER TRAN INT CPOW TRU AIACC

TEC 1 -0.312 0.109 0.064 -0.095 0.056 0.177

PER -0.312 1 0.347 0.469 -0.169 0.381 0.258

TRAN 0.109 0.347 1 0.557 -0.274 0.536 0.268

INT 0.064 0.469 0.557 1 -0.353 0.581 0.414

CPOW -0.095 -0.169 -0.274 -0.353 1 -0.492 -0.385

TRU 0.056 0.381 0.536 0.581 -0.492 1 0.514

AIACC 0.177 0.258 0.268 0.414 -0.385 0.514 1

analysis is shown in table 8. Indicator loadings from the ini-

tial measurement model are shown in the appendix. Due

to low loadings the item “tran_4” and “part_2” had to be

removed from the measurement model. Further measure-

ment errors in construct validity were identiﬁed. Therefore

“tec_2”, “int_2”,”int_3”,”part_4”, ”part_5” were eliminated in

the measurement model.

After removing the previously stated indicators the mea-

surement models were tested for ﬁnal indicator loadings.

The item “part_3” had an indicator loading of 0.589 for

the measurement model of Maria. Therefore Hair Jr. et al.

(2021) suggest to examine internal consistency. The neces-

sary threshold for internal consistency could not be reached

by both measurement models. Deleting an indicator should

be considered when a removal leads to an increase in relia-

bility (Hair Jr. et al.,2021). Therefore the item “part_3” was

removed from the measurement model.

The second step for evaluating reﬂective measurement

models according to Hair Jr. et al. (2021) is the examination

of internal consistency reliability. Hair Jr. et al. state that the

use of Cronbach’s alpha is a very conservative reliability mea-

sure. Further, they assume that composite reliability (rhoC)

may be a too liberal measure. Therefore they suggest to use

the reliability measure of rhoA. The reliability summary is

shown in table 9.

The results from the reliability summary show high reli-

ability of the measurement model. Since the rhoA-value for

the construct of “TEC” in the measurement model of Maria

is higher than 1, which may imply measurement errors. The

correlation of items (Appendix 12) for measurement model

of Maria show no anomalies since these items are correlated

0.716. The composite reliability shows reliability values ex-

ceeding the threshold of 0.7 to 0.9 suggested by Hair Jr. et

al. (2021).

They state value above 0.9, especially above 0.95 imply a

redundancy of indicators. As stated by Hair Jr. et al. (2021)

the composite reliability measure may be too liberal measure

for internal consistency. The results on the measure of Cron-

bach’s alpha, which is a conservative measure for reliability,

show that the constructs “TEC”, “ PER” and “TRAN” can be

K. Iqbal /Junior Management Science 8(4) (2023) 887-925910

Table 6: Descriptive summary of measurement model of Maria

Descriptive statistics

n min q25 mean median q75 max sd var

m_per_1 127 1 2 2.4961 2 3 5 1.0902 1.1885

m_per_2 127 1 2 2.5118 2 3 5 1.1117 1.2360

m_per_3 127 1 2 2.5669 3 3 5 1.1026 1.2157

m_int_1 127 1 3 3.4720 4 4 5 0.9048 0.8187

m_tran_1 127 1 1 2.3150 2 3 5 1.1866 1.4079

m_tran_2 127 1 2 2.8413 3 4 5 1.1298 1.2764

m_tran_3 127 1 1 2.2598 2 3 5 1.1071 1.2256

m_part_1 127 1 2 2.8898 3 4 5 1.0633 1.1306

m_tru_1 127 1 2.5 2.9921 3 4 5 0.9128 0.8333

m_tru_2 127 1 3 3.1969 3 4 5 0.9001 0.8101

m_tru_3 127 1 3 3.0709 3 4 5 0.9101 0.8283

m_tec_1 127 1 4 4.1417 4 5 5 0.8704 0.7575

m_tec_3 127 2 3.5 4 4 5 5 0.8909 0.7937

m_aiacc_1 127 1 3 3.3780 4 4 5 0.9994 0.9989

m_aiacc_2 127 1 3 3.4016 4 4 5 1.0333 1.0676

m_aiacc_3 127 1 3 3.4488 4 4 5 1.0213 1.0430

m_acc 127 0 1 0.7953 1 1 1 0.4051 0.1641

Table 7: Correlation matrix of Maria

Correlations of constructs of Maria

TEC PER TRAN INT CPOW TRU AIACC

TEC 1 -0.386 -0.154 -0.049 0.028 -0.077 -0.128

PER -0.386 1 0.311 0.373 -0.060 0.349 0.418

TRAN -0.154 0.311 1 0.424 -0.185 0.428 0.276

INT -0.049 0.373 0.424 1 -0.102 0.591 0.531

CPOW 0.028 -0.060 -0.185 -0.102 1 0.045 -0.057

TRU -0.077 0.349 0.428 0.591 0.045 1 0.657

AIACC -0.128 0.418 0.276 0.531 -0.057 0.657 1

considered as “good” in terms of reliability. Furthermore, the

constructs of “TRU” and “AIACC” slightly exceed the thresh-

old of 0.9. The constructs of “INT” and “CPOW” have a alpha-

value of 1.0 because they are single item constructs.

The third step for the assessment of reﬂective measure-

ment model according to Hair Jr. et al. (2021) is convergent

validity. Therefore they suggest to examine the measure of

average variance extracted (AVE). Furthermore, they state

that the AVE should exceed the value of 0.5. The results from

the examination of convergent validity are shown in table 9.

All constructs exceed the threshold suggested by Hair Jr. et

al. (2021).

In order to evaluate reﬂective measurement models, the

fourth step of Hair Jr. et al. (2021) is the assessment of dis-

criminant validity. They suggest to avoid the Fornell-Larcker

Criterion by Fornell and Larcker (1981) due to an inability of

the criterion to identify discriminant validities issues. There-

fore Hair Jr. et al. recommend to examine the heterotrait-

monotrait ratio of correlations (HTMT) (Henseler, Ringle, &

Sarstedt,2015). The results of the examination of the dis-

criminant validity are shown in table 10.

Hair Jr. et al. (2021) suggest that the values for HTMT

should be signiﬁcantly lower than the threshold of 0.85.

The values for HTMT shown in table 10 are below the sug-

gested threshold. Furthermore, a signiﬁcance test is con-

ducted where the structural equation model is bootstrapped

by 10000 samples for generating standard errors and conﬁ-

dence intervals. The signiﬁcance test shows that the upper

bound of the 95% conﬁdence interval is signiﬁcantly lower

than the suggested threshold of 0.85 (Hair Jr. et al.,2021).

The results of the bootstrapped values for HTMT are shown

in the appendix. All bootstrapped paths are signiﬁcantly

lower than the suggested threshold leading to discriminant

validity.

4.3. Analyzing acceptance conditions and robustness checks

of study

Since the previous tests show that the measurement mod-

els fulﬁll reliability and validity measures, the structural

model can be evaluated for testing the hypotheses. Before

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 911

Table 8: Loadings summary for Lisa and Maria

Loadings summary of Lisa

TEC PER TRAN INT CPOW TRU AIACC

l_per_1 0 0.929 0 0 0 0 0

l_per_2 0 0.908 0 0 0 0 0

l_per_3 0 0.870 0 0 0 0 0

l_int_1 0 0 0 1 0 0 0

l_tran_1 0 0 0.890 0 0 0 0

l_tran_2 0 0 0.881 0 0 0 0

l_tran_3 0 0 0.880 0 0 0 0

l_part_1 0 0 0 0 1 0 0

l_tru_1 0 0 0 0 0 0.946 0

l_tru_2 0 0 0 0 0 0.883 0

l_tru_3 0 0 0 0 0 0.929 0

l_tec_1 0.946 0 0 0 0 0 0

l_tec_3 0.954 0 0 0 0 0 0

l_aiacc_1 0 0 0 0 0 0 0.919

l_aiacc_2 0 0 0 0 0 0 0.908

l_aiacc_3 0 0 0 0 0 0 0.916

Loadings summary of Maria

TEC PER TRAN INT CPOW TRU AIACC

m_per_1 0 0.895 0 0 0 0 0

m_per_2 0 0.872 0 0 0 0 0

m_per_3 0 0.865 0 0 0 0 0

m_int_1 0 0 0 1 0 0 0

m_tran_1 0 0 0.902 0 0 0 0

m_tran_2 0 0 0.879 0 0 0 0

m_tran_3 0 0 0.918 0 0 0 0

m_part_1 0 0 0 0 1 0 0

m_tru_1 0 0 0 0 0 0.923 0

m_tru_2 0 0 0 0 0 0.888 0

m_tru_3 0 0 0 0 0 0.909 0

m_tec_1 0.851 0 0 0 0 0 0

m_tec_3 0.976 0 0 0 0 0 0

m_aiacc_1 0 0 0 0 0 0 0.934

m_aiacc_2 0 0 0 0 0 0 0.929

m_aiacc_3 0 0 0 0 0 0 0.954

Table 9: Internal consistency reliability and convergent validity

Summary of internal consistency reliability and convergent validity

Lisa alpha rhoC AVE rhoA Maria alpha rhoC AVE rhoA

TEC 0.892 0.949 0.902 0.896 TEC 0.835 0.912 0.838 1.349

PER 0.890 0.930 0.815 0.967 PER 0.850 0.909 0.770 0.854

TRAN 0.863 0.914 0.781 0.892 TRAN 0.883 0.927 0.810 0.888

INT 1 1 1 1 INT 1 1 1 1

CPOW 1 1 1 1 CPOW 1 1 1 1

TRU 0.908 0.943 0.846 0.911 TRU 0.892 0.933 0.822 0.897

AIACC 0.902 0.938 0.835 0.902 AIACC 0.933 0.957 0.882 0.935

assessing the structural model, the hypothesis that an an-

thropomorphizing of a system leads to an acceptance of the

system (H1) is tested for validity. Therefore the mean of the

constructs of acceptance in the measurement model of Maria

K. Iqbal /Junior Management Science 8(4) (2023) 887-925912

Table 10: Summary of dicriminant validity

HTMT table of Lisa

TEC PER TRAN INT CPOW TRU AIACC

TEC

PER 0.365

TRAN 0.125 0.392

INT 0.069 0.490 0.592

CPOW 0.101 0.176 0.278 0.353

TRU 0.065 0.419 0.593 0.611 0.516

AIACC 0.198 0.272 0.289 0.435 0.405 0.566

HTMT table of Maria

TEC PER TRAN INT CPOW TRU AIACC

TEC

PER 0.464

TRAN 0.183 0.366

INT 0.043 0.403 0.446

CPOW 0.025 0.067 0.191 0.102

TRU 0.100 0.394 0.476 0.619 0.047

AIACC 0.127 0.468 0.297 0.549 0.059 0.717

has to be signiﬁcantly higher than the mean of the constructs

of acceptance in the measurement model of Lisa. Therefore

a T-test is conducted in R (R Core Team,2013), shown in

table 11.

The T-test shows that the mean of the anthropomorphized

system is not signiﬁcantly higher than the mean of the tex-

tual interface. Therefore the hypothesis that an anthropo-

morphizing of a system leads to an acceptance of the system

(H1) is not supported.

In fact, the mean-value of the textual interface for the

construct of AIACC is higher than the mean-value of the an-

thropomorphized system. This difference in AIACC is not sig-

niﬁcant. Furthermore, the means of the constructs of ACC

have equal means for both observation groups. Moreover, the

mean-values for other constructs do not differ signiﬁcantly,

leading to the assumption that an anthropomorphizing of the

system has no signiﬁcant effect on acceptance measures. To

conﬁrm this assumption the structural models are assessed

in order to understand how acceptance is created. Further-

more, equal ﬁndings in both structural models support the

assumption that there is no signiﬁcant effect of anthropomor-

phizing the system on acceptance.

In order to evaluate the structural model Hair Jr. et al.

(2021) suggest to ﬁrst examine potential collinearity issues

in the constructs. Therefore the structural model is tested for

variance inﬂation factors. According to Becker, Ringle, Sarst-

edt, and Völckner (2015) variance inﬂation factors above the

threshold of 3.0 assume issues with collinearity. The results

shown in table 12 indicate no issues for potential collinearity.

In the second step Hair Jr. et al. (2021) suggest to exam-

ine the signiﬁcance of the structural model. Before evaluat-

ing the signiﬁcance of the structural model it is important to

outline the approach in order to estimate the model.

As mentioned earlier the dataset is divided into two mea-

surement models. Furthermore, the construct of AIACC is

considered as the main indicator for measuring acceptance.

For conﬁrming the results from PLS-SEM estimation, the con-

struct of ACC is considered in a second structural model. The

models only differ in the path coefﬁcients from the predictors

of acceptance to acceptance across both structural models.

For reducing illustrative complexity the construct of ACC is

added to the illustration of the SEM-estimation in ﬁgure 12

and ﬁgure 13. It is important to outline that both acceptance

constructs were not estimated in a single SE. All in all two

structural models with two measurement models were esti-

mated by using R, speciﬁcally the “seminr”-package by (Hair

Jr. et al.,2021), leading to four estimations of PLS-SEM.

In order to examine the signiﬁcance of the path coefﬁ-

cients, it is necessary to perform bootstrapping standard er-

rors for calculating conﬁdence intervals (Hair Jr. et al.,2021).

The summary of the bootstrapped paths is shown in the ap-

pendix. Figure 12 and ﬁgure 13 show the path coefﬁcients af-

ter bootstrapping. Further green paths indicate positive path

coefﬁcients whereas red paths indicate negative path coefﬁ-

cients. The signiﬁcance of paths is marked by stars. A p-value

smaller than 0.01 is marked with three stars, p-value greater

than 0.01 and smaller than 0,05 is marked with two stars and

a p-value greater than 0.05 and smaller than 0.1 is marked

with one star. The signiﬁcance of paths aids to support the

previously formulated hypothesis.

The path coefﬁcients from Trust to Acceptance show a

positive inﬂuence of trust on acceptance. This path is sig-

niﬁcant for both acceptance measures and by both measure-

ment models. The SEM for Lisa shows a path coefﬁcient of

β=0.502 (p <0.001; 5% CI =0.372; 95% CI =0.625)

for the construct Decision (ACC) and 0.445 (p <0.001; 5%

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 913

Table 11: T-test summary for antropomorphizing interfaces

System PER INT CPOW TRAN TRU TEC AIACC ACC

Lisa 2.3852 3.3762 2.9480 2.3203 3.0800 3.3766 3.5800 0.7952

Maria 2.5249 3.4720 2.8897 2.4682 3.0866 3.4645 3.4094 0.7952

p-value 0.2407 0.4352 0.6531 0.2343 0.9518 0.4997 0.1223 1.0000

Table 12: VIF-values for structural model evaluation

Structural model of Lisa

AIACC TRU CPOW

TEC PER TRU TRAN INT CPOW TRAN

1.153 1.344 1.217 1.464 1.546 1.154 .

Structural model of Maria

AIACC TRU CPOW

TEC PER TRU TRAN INT CPOW TRAN

1.180 1.336 1.144 1.250 1.220 1.036 .

CI =0.308; 95% CI =0.571) for the construct Acceptance

(AIACC).

Furthermore, the SEM for Maria shows a path coefﬁcient

of β=0.223 (p =0.008; 5% CI =0.068; 95% CI =0.371) for

the construct Decision and 0.582 (p <0.001; 5% CI =0.372;

95% CI =0.625) for the construct Acceptance (AIACC).

The path coefﬁcients from Transparency to Trust show

a positive inﬂuence of transparency on trust. This path is

signiﬁcant for both measurement models. The SEM for Lisa

shows a path coefﬁcient of β=0.277 (p <0.001; 5% CI =

0.155; 95% CI =0.399). Furthermore, the SEM for Maria

shows a path coefﬁcient of β=0.237 (p =0.005; 5% CI =

0.084; 95% CI =0.388).

The path coefﬁcient from Transparency to Computer

Power shows a negative inﬂuence of transparency on trust.

This path is signiﬁcant for both measurement models. The

SEM for Lisa shows a path coefﬁcient of β=- 0.277 (p <

0.001; 5% CI =- 0.396; 95% CI =- 0.149). Furthermore,

the SEM for Maria shows a path coefﬁcient of β=- 0.186 (p

=0.027; 5% CI =- 0.336; 95% CI =-0.028).

The path coefﬁcient from Computer Power to Trust show

different signiﬁcant result of the trust towards the system.

This path is signiﬁcant for both measurement models. The

SEM for Lisa shows a path coefﬁcient of β=- 0.301 (p <

0.001; 5% CI =- 0.395; 95% CI =- 0.205). Furthermore,

the SEM for Maria shows a path coefﬁcient of β=0.142 (p

=0.038; 5% CI =0.011; 95% CI =0.272).

The path coefﬁcient from Intelligence to Trust shows a

positive inﬂuence of Intelligence on Trust. This path is signif-

icant for both measurement models. The SEM for Lisa shows

a path coefﬁcient of β=0.321 (p <0.001; 5% CI =0.172;

95% CI =0.460). Furthermore, the SEM for Maria shows a

path coefﬁcient of β=0.504 (p <0.001; 5% CI =0.374;

95% CI =0.633).

The path coefﬁcients from Perception as technology to Ac-

ceptance show a positive inﬂuence of trust on acceptance.

The result on path signiﬁcance differs for both measurement

models. The SEM for Lisa shows a path coefﬁcient of β

=0.096 (p =0.160; 5% CI =- 0.004 ; 95% CI =0.225)

for the construct Decision (ACC) and 0.205 (p <0.001; 5%

CI =0.073; 95% CI =0.336) for the construct Acceptance

(AIACC). Furthermore, the SEM for Maria shows a path co-

efﬁcient of β=- 0.084 (p =0.17; 5% CI =- 0.220; 95% CI

=0.057) for the construct Decision and 0.003 (p =0.5; 5%

CI =- 0.145; 95% CI =0.175) for the construct Acceptance

(AIACC).

The path coefﬁcients from Perception as person to accep-

tance show a positive inﬂuence of trust on acceptance. This

path is signiﬁcant for both measurement models. Further, the

path to the acceptance measure Decision (ACC) is not signif-

icant. The SEM for Lisa shows a path coefﬁcient of β=-

0.069 (p =0.160; 5% CI =- 0.239; 95% CI =0.225) for

the construct Decision (ACC) and 0.159 (p =0.025; 5% CI

=0.033; 95% CI =0.283). Furthermore, the SEM for Maria

shows a path coefﬁcient of β=- 0.091 (p =0.170; 5% CI

=- 0.035; 95% CI =0.261) for the construct Decision and

0.214 (p =0.003; 5% CI =0.091; 95% CI =0.339) for the

construct Acceptance (AIACC).

After assessing the signiﬁcance of the path coefﬁcients it

is necessary to evaluate the explanatory power of the model

(Hair Jr. et al.,2021). Therefore the measure of R2explains

how much of the variance of the construct is explained by the

model. R2-values of 0.75 indicate a substantial, R2-values

of 0.5 show a moderate and R2values of 0.25 state a low

explanatory power (Hair Jr., Ringle, & Sarstedt,2011). The

measure of R2may increase by the number of explanatory

variables. Therefore it is important to consider the measure

of Adjusted R2. The limitation of Adjusted R2-measure is,

that it may consider the number of explanatory variables but

this measure is not a precise indicator as R2(Hair Jr. et al.,

2021).

The results shown in table 13 state a moderate to low

K. Iqbal /Junior Management Science 8(4) (2023) 887-925914

Figure 12: SEM for Lisa after Bootstrapping (Source: Own illustration)

Figure 13: SEM for Maria after Bootstrapping (Source: Own illustration)

Table 13: Explanatory Power of model: R2

R-squared

Lisa AIACC TRU CPOW ACC Maria AIACC TRU CPOW ACC

R20.303 0.483 0.075 0.257 0.473 0.407 0.034 0.090

Adj. R20.289 0.473 0.069 0.242 0.460 0.393 0.027 0.068

explanatory power of both models. Furthermore, the model

of Lisa and Maria only describes 7.5% and 3.4% of the vari-

ance of the construct of Computer Power. This may indicate

that Computer Power is inﬂuenced by unobserved variables

which are not measured by the model. The effect size f2may

be a further measure to evaluate the explanatory power of

the model (Hair Jr. et al.,2021). Results of f2-measure are

shown in the appendix.

Furthermore, the assessment of the predictive power of

the model is the next step in order to evaluate the structural

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 915

model (Hair Jr. et al.,2021). Predictive power is deﬁned as

the ability of a model to predict new observations (Hair Jr.

et al.,2021). Therefore the sample is divided into a hold-

out and multiple training samples. The training sample are

is estimated and evaluated by predicting performance while

comparing the results to the holdout sample (Hair Jr. et al.,

2021). In order to perform cross validation, this process is

repeated by the number of subsamples where the holdout

sample is changed to a training sample and a further training

sample is changed to a holdout sample. Therefore the mea-

sure of root-mean-square error (RMSE) and mean absolute

error (MAE) is calculated. Furthermore, prediction errors

from a linear regression model for each indicator are calcu-

lated. The structural model needs to show lower prediction

errors than the benchmark of prediction errors generated by

the linear model.

In order to perform the prediction, the sample was di-

vided into 10 subsamples. Furthermore the process calcu-

lation of prediction errors is repeated 10 times. The results

generated by the “seminr”-package in R (Hair Jr. et al.,2021)

are shown in table 14.

The results show that six out of seven indicators of the

model Lisa have lower prediction errors (RMSE) than the

benchmark of the linear model. Furthermore, ﬁve out of

seven indicators of the model Maria have lower prediction

errors than the benchmark of the linear model. According to

Hair Jr. et al. (2021) a majority of indicators under the bench-

mark of the linear model imply medium predictive power.

The evaluation of the structural model showed no issues

with collinearity in constructs, signiﬁcant path coefﬁcients,

moderate explanatory power and medium predictive power

for new observations.

4.4. Discussion of study results

After evaluating the structural model, the next section

will interpret the results. Therefore the previously formu-

lated hypotheses are evaluated for validity. Furthermore,

the ﬁndings of the study are reﬂected in prior ﬁndings in

research. Possible explanations for the ﬁndings are given,

derived from previous research.

4.4.1. Interpreting study results

The hypothesis that an anthropomorphizing of a system

leads to an acceptance (H1) is not supported by the study.

The descriptive statistics show higher mean values for the

perception of the system as technology for the anthropomor-

phized system than for the textual system. On the other

hand, the perception of the system as a person has higher

mean values of the anthropomorphized system than the tex-

tual system which may indicate appropriate system design in

terms of anthropomorphizing. Furthermore, the results from

the t-test indicate that there is no signiﬁcant inﬂuence of an-

thropomorphizing the system on acceptance conditions. De-

spite these ﬁndings, this study shows signiﬁcant differences

in acceptance conditions which may be referred to an inﬂu-

ence of anthropomorphizing features of the system on the

acceptance. The results are shown in table 15.

Furthermore, the study shows empirical evidence that

trust towards the system is the main indicator for creating ac-

ceptance by users since the path coefﬁcients from Trust to AI

acceptance are above 0.5. The hypothesis that higher trust in

a system leads to higher acceptance (H2) is supported by this

study. An anthropomorphizing of the system has higher path

coefﬁcients from Trust to AI acceptance which may indicate

that higher trust towards the system has a higher inﬂuence

on acceptance in more anthropomorphized systems.

The results show that trust towards the system is inﬂu-

enced by the transparency or the comprehensibility of the

system. This study shows a signiﬁcant inﬂuence of Trans-

parency on Trust with a path coefﬁcient of 0.277 for the tex-

tual system and a path coefﬁcient of 0.237 for the anthro-

pomorphized system. Therefore the hypothesis that higher

transparency of a system has a positive effect on the trust

towards the system (H3) is supported. Furthermore, the to-

tal effects shown in the appendix imply that the mediator

variables which are inﬂuenced by transparency increase the

inﬂuence of transparency of the system on trust towards the

system for textual interface with a signiﬁcant path coefﬁcient

of 0.362. The total effects statistic shows that transparency is

a signiﬁcant predictor for acceptance with a path coefﬁcient

of 0.161 and the total effects statistic for the anthropomor-

phized system implies that the mediating variable which is

inﬂuenced by transparency decreases the inﬂuence of trans-

parency on trust. The path from Transparency to Trust re-

mains signiﬁcant with a path coefﬁcient of 0.123. Similar to

the textual system, transparency is a signiﬁcant predictor of

acceptance with a path coefﬁcient of 0.125 in anthropomor-

phized systems.

The tautologic relation between Transparency to Com-

puter Power could be proven empirically. This study shows

a signiﬁcant effect of Transparency on Computer Power with

a path coefﬁcient of – 0.276 for the textual interface and a

path coefﬁcient of – 0.186 for the anthropomorphized sys-

tem. Therefore the hypothesis that a higher comprehensi-

bility of a system has a positive effect on the perceived par-

ticipation of the user in the decision-making process (H4) is

supported. Since the construct of Computer Power has low

explanatory power the suggested predictor may not be sufﬁ-

cient and other unobserved variables would be more suitable

for predictors. The anthropomorphized system has lower

path coefﬁcients than the textual interface which may in-

dicate that anthropomorphizing features decrease the effect

that transparency leads to a lower perception of a higher role

in decision-making by the system (Computer Power).

Further ﬁndings of the study are that Computer Power

has a signiﬁcant effect on Trust with a path coefﬁcient of -

0.301 for the textual system and a path coefﬁcient of 0.141

for the anthropomorphized system. Therefore the hypothesis

that the higher the perceived participation of the system-user

in the decision-making process is, the higher the perceived

trust towards the system (H5), is supported partially. These

ﬁndings were not expected since the literature showed that

higher participation possibilities lead to an increase in trust.

Similar to the literature the results suggest that a perception

K. Iqbal /Junior Management Science 8(4) (2023) 887-925916

Table 14: Summary of prediction errors

Prediction error measures for PLS of Lisa

l_aiacc_1 l_aiacc_2 l_aiacc_3 l_tru_1 l_tru_2 l_tru_3 l_part_1

RMSE 0.865 0.832 0.874 0.842 0.825 0.866 1.068

MAE 0.677 0.626 0.683 0.660 0.663 0.680 0.906

Prediction error measures for LM of Lisa

l_aiacc_1 l_aiacc_2 l_aiacc_3 l_tru_1 l_tru_2 l_tru_3 l_part_1

RMSE 0.913 0.863 0.930 0.847 0.856 0.885 1.027

MAE 0.696 0.632 0.720 0.651 0.657 0.709 0.866

Prediction error measures for PLS of Maria

m_aiacc_1 m_aiacc_2 m_aiacc_3 m_tru_1 m_tru_2 m_tru_3 m_part_1

RMSE 0.811 0.789 0.797 0.801 0.694 0.801 1.060

MAE 0.650 0.637 0.673 0.634 0.568 0.632 0.867

Prediction error measures for LM of Maria

m_aiacc_1 m_aiacc_2 m_aiacc_3 m_tru_1 m_tru_2 m_tru_3 m_part_1

RMSE 0.871 0.816 0.854 0.813 0.646 0.778 1.171

MAE 0.692 0.653 0.700 0.639 0.502 0.579 0.975

of higher participation of the system in decision-making de-

creases trust towards the system for the textual interface. For

the anthropomorphized interface the perception of higher

participation of the system in the decision-making process in-

creases the trust in the system. This effect may be explained

due a lower perception of high participation of the system

in decision-making shown in the comparison of means in

descriptive statistics. Furthermore, the total effects statistic

shows that Computer Power is a signiﬁcant predictor of ac-

ceptance with a path coefﬁcient of – 0.082 for anthropomor-

phized interfaces. On the other hand, the path coefﬁcient of –

0.135 is not signiﬁcant for textual interfaces. This total effect

statistics show that acceptance is increased if the system has

higher power in decision making for anthropomorphized in-

terfaces. On the other hand, there is no signiﬁcant inﬂuence

of the perceived power of the system in decision-making on

acceptance for textual interfaces.

The results show that the perceived intelligence of the

system has an inﬂuence on the trust in the system. This

study shows a signiﬁcant inﬂuence of Intelligence on Trust

with a path coefﬁcient of 0.321 for the textual interface and

a path coefﬁcient of 0.521 for the anthropomorphized inter-

face. Therefore the hypothesis that the higher the perceived

intelligence of the system is, the higher the trust (H6) is sup-

ported. The path coefﬁcients of the anthropomorphized sys-

tem show that the perceived intelligence has a greater role

in predicting trust than the path coefﬁcient of the textual in-

terface. Furthermore, the total effects statistic shows that In-

telligence is a signiﬁcant predictor of acceptance with a path

coefﬁcient of 0.142 for the textual interface and a path co-

efﬁcient of 0.294 for the anthropomorphized interface. This

total effect statistic shows that a higher perceived intelligence

of the system should be considered in order to create accep-

tance of the users.

The results show that the perception of the system as a

technology has an inﬂuence on acceptance. This study shows

a signiﬁcant inﬂuence of the perception of the system as tech-

nology on the acceptance by users with a path coefﬁcient of

0.205 for the textual interface. For anthropomorphized inter-

faces the path coefﬁcient of – 0.003 is not signiﬁcant. There-

fore hypothesis that the higher the perception of the system

as technology is, the higher the acceptance (H7), is supported

partially.

Furthermore, the study results show that the perception

of the system as a person has an inﬂuence on the acceptance

of users. A Signiﬁcant inﬂuence of the perception of the sys-

tem as a person with a path coefﬁcient of 0.159 for textual

interface and path coefﬁcient of 0.214 for an anthropomor-

phized system is identiﬁed. Therefore hypothesis that the

higher the perception of the system as person is, the higher

the acceptance (H8) is supported. The results may indicate

that adding anthropomorphizing features for textual inter-

faces may not be necessary in order to create acceptance since

the path coefﬁcients for the perception of the system as a

technology is higher than the path coefﬁcient of the percep-

tion of the system as a person.

4.4.2. Theoretical relevance of study results

The acceptance research on the non-managerial level

showed that trust is a major condition to create acceptance

by users (Hastenteufel & Ganster,2021;Rathje et al.,2021;

Scheuer,2020;Uysal et al.,2022). This study shows that

ﬁndings from literature are applicable on the managerial

level. Speciﬁcally, trust is identiﬁed as a major condition for

creating acceptance by users. Furthermore, this study shows

how trust is inﬂuenced. Since Makarius et al. (2020) iden-

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 917

Table 15: Results on acceptance conditions for algorithmic decision support

From INT CPOW TRAN TRAN PER TEC TRU CPOW TRAN INT

To TRU TRU CPOW TRU AIACC AIACC AIACC AIACC AIACC AIACC

Path coefﬁcients: 0.321 -0.301 -0.276 0.277 0.159 0.205 0.445 -0.135 0.161 0.142

Lisa

T-Statistics 3.678 -5.276 -3.606 3.691 1.973 2.490 5.574 -3.532 3.625 3.097

5% CI 0.172 -0.395 -0.396 0.155 0.033 0.073 0.308 -0.201 0.094 0.094

95 % CI 0.460 -0.205 -0.149 0.399 0.283 0.336 0.571 -0.076 0.238 0.238

Signiﬁcance ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Path coefﬁcients: 0.504 0.142 -0.186 0.237 0.214 -0.003 0.582 0.082 0.123 0.294

Maria

T-Statistics 6.420 1.790 -1.985 2.607 2.834 -0.009 9.048 1.739 2.182 4.599

5 % CI 0.374 0.011 -0.336 0.084 0.091 -0.145 0.472 0.006 0.028 0.193

95 % CI 0.633 0.272 -0.028 0.388 0.339 0.175 0.683 0.161 0.217 0.402

Signiﬁcance ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

K. Iqbal /Junior Management Science 8(4) (2023) 887-925918

tiﬁed cognitive issues in terms of strategic decision making

where they identiﬁed the necessity to do further research in

how decision-makers trust the output received from AI sys-

tems. This study shows how trust is inﬂuenced. Furthermore,

Venkatesh and Davis (2000) show that perceived usefulness

is an indicator of acceptance. The study results show that the

perceived intelligence of the system is an acceptance condi-

tion for decision support. Furthermore, the results suggest

that the system should be perceived as useful and exhibit

a certain intelligence conﬁrming the research on TAM by

Venkatesh and Davis (2000). Despite no ﬁndings on the in-

ﬂuence of perceived intelligence on trust by Scheuer (2020),

this study showed empirical evidence that the perceived in-

telligence of system has an inﬂuence on trust. Research on

acceptance states that it is necessary to have a transparent

system in terms of the comprehensibility of a decision pro-

cess (Gersch et al.,2021;Meske et al.,2022;Scheuer,2020;

Venkatesh & Davis,2000). On the other hand, Newman et

al. (2020) state that an increase in transparency may lead

to a decontextualization of workers. Therefore the results of

this study show that the ﬁndings from acceptance research

are applicable on the managerial level, speciﬁcally that trans-

parency of a result-generating process inﬂuences the trust in

the system in a positive way. Transparency of the decision-

making process of the system is identiﬁed as a necessary

condition in order to create acceptance by users.

Further results of this study show evidence for the tauto-

logic relationship between the transparency of result process-

ing and the perceived participation of the system in decision-

making. Orlikowski and Robey (1991) assume that more in-

formation in the decision-making process leads to a higher

power of the decision-maker. Considering the manager as

a decision-maker the study showed that more information

for the manager increases his perceived power in decision-

making. Furthermore, the perceived power of the system

in decision-making decreases. This effect can be explained

by the perception of authoritative correctness of algorithms.

Precise algorithms may generate the perception of correct-

ness therefore human beings can feel inferior to algorithms

(Martini,2019).

On the other hand, a higher perceived power of the sys-

tem in the decision-making process may leads to an increase

in trust towards the system if the system is anthropomor-

phized. Mcafee et al. (2012) questioned whether managers

would accept a decision support system which may lead to

a shift in their role in form of decreased power. The re-

sult showed that higher perceived power by the system in

decision-making is an acceptance condition for anthropo-

morphized systems. These results are contradictory to prior

research and to attribution theory by Kelley and Michela

(1980).

Attribution theory states that individuals seek to under-

stand the cause of their own behavior (Kelley & Michela,

1980). Since this study shows that the power of the sys-

tem in decision-making process can be achieved by reducing

the transparency of the system, it is assumable that users can

have an increased trust in the system even if the system is not

comprehensible. Furthermore, a low comprehensibility of a

system may not lead to an understanding of the cause of own

contribution on the success caused by the decision. Speciﬁ-

cally, anthropomorphized systems cause blind trust. This as-

sumption may be irrational in terms of research ﬁndings in

the necessity of explainable systems regarding result process-

ing (Gersch et al.,2021;Meske et al.,2022;Scheuer,2020;

Venkatesh & Bala,2008). This irrational assumption may be

explained by interpersonal acceptance. As mentioned before

insufﬁcient knowledge and a lack of trust hinder the adop-

tion of decision support systems (Rainsberger,2021). An an-

thropomorphizing of systems may build up a personal rela-

tionship by generating sympathy and affection towards the

system which results in interpersonal acceptance (Rohner &

Khaleque,2002). Furthermore, an anthropomorphized sys-

tem may lead to higher perception of effectiveness of the sys-

tem causing automation bias. Anthropomorphized systems

may be perceived as more effective leading to the tendency

to over-rely on decisions made by algorithms (Meske et al.,

2022). Goddard, Roudsari, and Wyatt (2012) shows that au-

tomation bias leads to a potential failure to detect mistakes

made by algorithms. Expectancy theory by Isaac, Zerbe, and

Pitt (2001) may explain this irrational assumption. Isaac et

al. (2001) state that individuals choose a decision based on

the expected outcome of a decision. Therefore a high per-

ception of intelligence may lead to greater expectancy in the

outcome of the decision. Further possible explanation for

the positive effect of higher perceived power of the system

on trust towards the system may be a perception of fairness.

Korsgaard et al. (1995) show that participation possibilities

as the consideration of an input brought for decision-making

or the inﬂuence of the input brought for decision making on

the outcome of a decision create procedural justice which is

a prerequisite for fairness. As Lee (2018) and Newman et al.

(2020) show the perception of a fair or trustworthy decision

depends on whom the decision is made. Decisions made by

the anthropomorphized system may be perceived fairer due

to the anthropomorphizing features. Since the anthropomor-

phized system was more perceived as a technology than a

person this explanation may be partially valid. Since the per-

ception of the system as technology was higher in the anthro-

pomorphized system than in the textual system, the Turing

test (Turing,1950) failed. The failure of the Turing test may

purpose that the system was not perceived as intelligent by

the users. In fact, the perceived intelligence of the system was

higher in the anthropomorphized systems than in the textual

system which may propose that Turing’s deﬁnition of intelli-

gence is outdated. Intelligence may be connected to the per-

ception of anthropomorphizing features according to Waytz,

Cacioppo, and Epley (2010) like cognition, emotions or inter-

activity. Furthermore, Lee (2018) shows that decisions made

by humans evoke positive emotions due to the possibility

of social recognition. Since anthropomorphized systems are

characterized through the perception of cognitive capabilities

in technology like emotions Waytz et al. (2010), users may

see a psychological pleasure or social gain while interacting

with the technology. Therefore, the social exchange theory

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 919

(Emerson,1976) may be applicable in order to conﬁrm ﬁnd-

ings on acceptance conditions. The social exchange theory

states that the interaction between two humans is character-

ized by an exchange of costs and utilities. Utilities may be the

effectiveness of the system (Goddard et al.,2012;Lee,2018;

Martini,2019), psychological pleasure or the enjoyment of

system usage (Waytz et al.,2010) and costs may be the per-

ception of inferiority (Baumann-Habersack,2021;Lee,2018;

Newman et al.,2020), possible detachment of decision mak-

ing (Bader & Kaiser,2019) in terms of involvement or the risk

of failure due to data discrimination (Newman et al.,2020).

Lawler and Thye (1999) show that emotion deepen the na-

ture of the relationship between humans. Furthermore, they

show that due to the rise of emotions, humans tend to fo-

cus on the decision rather than on the decision process in a

group. Therefore social exchange theory may be a possible

explanation for blind trust.

The theory of Uncanny Valley by Mori, MacDorman, and

Kageki (2012) shows that anthropomorphizing features lead

to an increase of acceptance inﬂuenced by trust (Scheuer,

2020). They state that an increase of anthropomorphizing

features to a certain point lead to a radical reduction of accep-

tance. Furthermore, Mori et al. (2012) outline that after the

critical point of reduction a certain high degree of anthropo-

morphizing leads to increasing effects on acceptance. Since

the anthropomorphized system had high features of anthro-

pomorphizing like gestures, human embodiment and voice

output. The survey participants may felt an imperfection of

anthropomorphizing leading to higher perception of the sys-

tem as technology.

5. Acceptance conditions of algorithmic decision support

for practice and research

Since the literature shows that research on acceptance

conditions for management is critical in order to enhance

the potential of algorithmic decision support (Grossman

& Siegel,2014;Laudon et al.,2016;Mcafee et al.,2012;

Mikalef et al.,2019;Rainsberger,2021;Reid et al.,2015).

This paper identiﬁed plenty of acceptance conditions. There-

fore it is necessary to categorize ﬁndings for practice and

identify limitations for further research.

5.1. System design implications

This paper showed that an optimization of an interface

in terms of anthropomorphizing has no effect on the accep-

tance. Despite no ﬁnding, the way acceptance is created dif-

fers in optimized interfaces. Therefore practitioners should

ﬁrst deﬁne their goal in terms of algorithmic decision support

where they have to specify the role of the user. It is necessary

to adjust the optimization of the interface to the intention to

use the system. If a user should question the output of the

decision support system, the decision-processing of the sys-

tem should be transparent leading to a higher power of the

user during the usage. Therefore anthropomorphized system

would not be suitable.

On the other hand, if a user should rely on the output

of the decision support system, the system should exhibit a

higher perceived intelligence leading to a higher trust. Fur-

thermore, the system should be perceived as a person in order

to create acceptance. Therefore anthropomorphized systems

would be suitable. The research showed that the expected

outcomes of an anthropomorphizing is dependent on the sys-

tem design. Therefore practitioners should pay attention to

a suitable degree of an anthropomorphized system in order

to avoid the Uncanny Valley proposed by Mori et al. (2012).

Practitioners should examine which degree of anthropomor-

phized system is beneﬁcial in order to fulﬁll their goals. The

implications show that system design is key in order to opti-

mize the interface to create acceptance.

Furthermore, the decision support system should be

trustworthy since trust is identiﬁed as the main indicator

for creating acceptance. In order to create trust according

to Lemke, Monett, and Mikoleit (2021) ethical principles

should be considered while designing the system. Speciﬁ-

cally, beneﬁcence, transparency, nonmaleﬁcence, autonomy,

justice, and privacy are principles for an ethical usage of AI

according to M. C. Barton and Pöppelbuß (2022). The deci-

sion processing of the system should be transparent leading

to a higher power of the user. Further performance measure-

ment of the decision may lead to the realization of a positive

impact of own contribution on the decision (attribution the-

ory). Systems with high power in the decision making pro-

cess should be avoided since they have a negative effect on

trust. On the other hand anthropomorphized systems with

high power in decision-making lead to an increase in trust

(blind trust). Future technology advances in hardware like

neuromorphic computer architecture, DishBrain and Brain

Machine Interface or advances in algorithms like Computa-

tional Intelligence or Super Artiﬁcial Intelligence may lead

to an affordance of blind trust. Since anthropomorphized

systems should be introduced when a reliance on the system

is afforded, practitioners have the possibility to avoid ethical

principles by designing a system with low transparency lead-

ing to blind trust. They should carefully evaluate whether

they want to beneﬁt from blind trust. It may be beneﬁcial

in order to create acceptance. The research showed that

the beneﬁts (total effects) from blind trust are smaller than

the beneﬁts from trust created by comprehensibility (total

effects). Therefore systems with high effectiveness due to

technological advances should be transparent for the user

because they lead to higher trust and make it possible to

identify their own contribution to the outcome of the deci-

sion (attribution theory). Furthermore, advanced systems

have to exhibit intelligence. The manager should rely on the

system knowing that the system processes decision aid with

high precision. Therefore the effectiveness of the system

should be communicated properly in order to beneﬁt from

expectancy theory (Isaac et al.,2001).

This study showed that an anthropomorphizing may not

have a direct effect on the acceptance. One of the ﬁrst an-

thropomorphized system was introduced by Microsoft called

Clippy (Swartz,2003). The rejection of this assistant was

K. Iqbal /Junior Management Science 8(4) (2023) 887-925920

high due to malfunction and a low effectiveness of the sys-

tem (Swartz,2003). Due to claims, Microsoft has removed

the function of Clippy in Ofﬁce (Swartz,2003). Despite the

failure of Clippy an interface with a similar degree of anthro-

pomorphizing may be beneﬁcial for advanced decision sup-

port systems in order to avoid the Uncanny Valley.

5.2. Limitations and future research

Further precise implications for practice could be derived

if the study did not have limitations. The study results were

based on an interaction of users with the system. Therefore

vignettes have to be designed carefully which could imitate a

realistic scenario. Vignettes may distort the perception of the

user through the framing of information. This study carefully

examined framing of information. The vignette was framed

in terms of transparency of the system. The system used in

the vignette was not comprehensible. Therefore the descrip-

tive statistics conﬁrm that on average the users do not under-

stand the decision processing of the results of both systems.

This distortion was necessary in order to examine whether

users would accept the system even if it is not comprehen-

sible. Therefore the vignette described an interaction with

hybrid intelligence where the level can be classiﬁed as De-

cision Support System. Further levels of hybrid intelligence

were not speciﬁed. Since this study showed that the partici-

pation of the user in the decision-making process is important

for building a trustworthy system, further levels of hybrid in-

telligence should be considered for future research.

Due to measurement errors, constructs of the study con-

tain single-items which may be not beneﬁcial since exoge-

nous variables are not measurable directly. Nevertheless, the

literature shows that single item constructs are appropriate

measures for an exploratory research. Since the research

question focused on the exploration of acceptance conditions

this study examined valid results through the use of PLS-

SEM. In order to validate the constructs on theoretical level,

a further study should be conducted where the data is ana-

lyzed by a common factor-based structural equation model

(CB-SEM).

One major problem of the study is that the perception of

the system as technology of the system of the anthropomor-

phized system was higher than that of the textual system.

This may indicate that the system design of the vignette was

affected by the Uncanny Valley by Mori et al. (2012). Since

this study has aimed to maximize the level of anthropomor-

phizing a speciﬁc high degree of anthropomorphizing was

reached. The degree of anthropomorphizing is not speciﬁ-

able uniformly. Therefore the research has to develop a scale

for identifying the degree of anthropomorphizing where fea-

tures of system design are speciﬁed in order to derive the

degree of anthropomorphizing. Due to the non-existence of

a certain scale of anthropomorphizing the degree of anthro-

pomorphizing was chosen arbitrarily, which may distort the

results. Further research can focus on the acceptance of an-

thropomorphized systems with different scales of anthropo-

morphizing. The effect of Uncanny Valley is identiﬁed in the

cancellation statistics of the survey. Most cancellations of the

survey were done on the page of the introduction of the an-

thropomorphized system (75 survey participants). The re-

sults may be distorted since the users were annoyed by the

presence of a human-like system which resulted in the can-

cellation of the survey. This group would have provided other

results. Further research could examine whether a maxi-

mization of the anthropomorphizing features may lead to a

perception of the system as a person and examine the effect

of interpersonal acceptance on acceptance?

The T-test showed that anthropomorphizing has no effect

on the acceptance or acceptance-creating variables. Further-

more, two PLS-SEM were estimated to identify how accep-

tance is created. This approach could be optimized by us-

ing anthropomorphizing as moderating variable. Since both

models show similar effects except for the aspect of blind

trust in anthropomorphized systems, a lack of explanatory

power exists in the difference in the results of both systems.

R2is low in the construct of system power for both SEM

models highlighting the affordance of a research setting with

moderating effect of anthropomorphizing. Furthermore, the

method of PLS-SEM maximizes the explanatory power of the

model R2(Hair Jr. et al.,2021). Low R2values indicate

that variables were omitted in research, which may lead to

a problem of causal identiﬁcation (endogeneity). Since the

research focuses on an exploration of acceptance conditions

a validation of endogeneity was not necessary. Therefore fu-

ture research should examine endogeneity to identify causal-

ities for the acceptance conditions.

This study examined the acceptance conditions for a sin-

gle decision-maker. In practice, decision situations may be

more complex. Merendino et al. (2018) show that algorith-

mic decision support can create tension in boards. Therefore

it is necessary to examine acceptance conditions for further

decision scenarios. Future research should identify whether

acceptance conditions for single managers are applicable for

more complex decision scenarios, like group decisions.

6. Conclusion

The aim of this thesis was to investigate the conditions

that lead to the acceptance of algorithmic decision support

systems. In this study, it was especially important to con-

sider the decision-making process of managers. According

to this the target group of this study was german speaking

students and employees including managers. To analyze dif-

ferent conditions, that may lead to the acceptance of algo-

rithmic decision support systems it was necessary to choose

a methodological approach that considers different scenarios

but also provides insights on the perceptions, beliefs, and at-

titudes of the target group. Based on this, a vignette study

along with a quantitative survey was used for the data col-

lection for the thesis. In total 281 german speaking student

and employees including managers participated in the study

during the period from 25.07.22 -07.08.22.

Furthermore, to analyze the conditions of acceptance an

estimation of a PLS-SEM model was conducted.

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 921

In the theoretical section, it was assumed that anthropo-

morphizing features may lead to a situation where the user

perceives the system as a person and accordingly shows more

trust and acceptance towards it. But the result show, an ex-

act opposite behavior of the users. As in the vignette study,

two scenarios were presented a textual scenario and a sce-

nario considering anthropomorphizing features. The users

perceive the anthropomorphized scenario as a technology

and show more trust and acceptance towards a scenario that

is not anthropomorphized. Accordingly, the results indicate

that there is no signiﬁcant inﬂuence of anthropomorphizing

the system on acceptance.

On the other hand, this thesis shows how acceptance dif-

fers across both distinct system. This study conﬁrms that

higher trust in a system leads to higher acceptance. In ad-

dition to this, the results show that trust in the system is in-

ﬂuenced by the transparency or the comprehensibility of the

system. In this regard it might be interesting to investigate

how a system can be designed to receive more trust. In other

words, how can the variable transparency or comprehensi-

bility be further elucidated to generate more trust which in

the end leads to a situation where the user accepts a system?

In this regard different vignette settings might be helpful to

investigate scenarios that lead to more transparency and in

turn to more trust and acceptance.

Moreover, this study presented several implications for

managers and academics. It needs to be mentioned that ex-

ponential development in technology can help to aid strategic

and operational decisions in management and can be crucial

in order to be competitive in dynamic markets. Nevertheless,

decision support systems are not used in practice which has

many reasons. The literature shows that major challenges

arise in the domain of management. Studies show that only

few decision-makers understand data concepts well. There-

fore the acceptance of algorithmic decision support is not

given in the practice. Research on acceptance has identiﬁed

many conditions in order to foster acceptance of information

systems. Nevertheless, the research focuses on the accep-

tance on worker- or user-level. This study focuses on the gap

in the existing literature on management-level. The research

question is which conditions lead to an acceptance of algo-

rithmic decision support in management.

Summing up, the literature on persuasive technology

shows that an optimization of interfaces leads to more inter-

action with the technology. Anthropomorphizing is identiﬁed

as an appropriate way to optimize interfaces. Therefore a

vignette study design is conducted, where the survey partici-

pants simulate an interaction with a decision support system

where the anthropomorphizing is manipulated due to two

alternating degrees of anthropomorphizing (low and high).

The data for both systems were measured on distinct mea-

surement models. Initially, the results show that there is no

effect of anthropomorphizing on acceptance, which may be

biased by Uncanny Valley.

Practitioners should ﬁrst deﬁne the level of hybrid in-

telligence in order to design the system. The system de-

sign should consider effects from the study. Beneﬁts from

blind trust are not recommendable since the creation through

transparency has higher total effects than the total effect of

the perceived power of the system in decision-making pro-

cess. Furthermore, the system has to be effective which may

be realized by technological advances. The effectiveness of

the system has to be communicated in an appropriate level

to enhance the perceived intelligence of the system.

This study showed which conditions lead to an accep-

tance of algorithmic decision support in management in an

explorative study design. These conditions of acceptance

could be conﬁrmed by further research through a CB-SEM.

All in all, it needs to be mentioned that this study ﬁrstly,

provided a theoretical contribution by deriving a Structural

model and based on the thoughts of the TAM. Secondly, this

study provided an empirical contribution at a managerial

level as 281 survey respondents participated in this study and

shared their perceptions and attitudes towards two scenarios

constituting two systems.

Finally, this study provided a practical contribution by

showing how companies can use this model as an indicator

to design systems and which conditions are necessary in or-

der to create acceptance for users. All in all, this study con-

tributes to the research gap on acceptance on managerial-

level.

K. Iqbal /Junior Management Science 8(4) (2023) 887-925922

References

Abhari, K., Vomero, A., & Davidson, E. (2020). Psychology of Business

Intelligence Tools: Needs-Affordances-Features Perspective. In Pro-

ceedings of the Annual Hawaii International Conference on System Sci-

ences. Hawaii International Conference on System Sciences.

Adadi, A., & Berrada, M. (2018). Peeking inside the black-box: a survey

on explainable artiﬁcial intelligence (XAI). IEEE Access,6, 52138–

52160.

Aguinis, H., & Bradley, K. J. (2014). Best Practice Recommendations for De-

signing and Implementing Experimental Vignette Methodology Stud-

ies. Organizational Research Methods,17(4), 351–371.

Alexander, C. S., & Becker, H. J. (1978). The Use of Vignettes in Survey

Research. Public Opinion Quarterly,42(1), 93–104.

Alvarez, S. A., Barney, J. B., & Young, S. L. (2010). Debates in entrepreneur-

ship: Opportunity formation and implications for the ﬁeld of en-

trepreneurship. In Handbook of entrepreneurship research (pp. 23–

45). Springer.

Alves, W. M., & Rossi, P. H. (1978). Who should get what? fairness judg-

ments of the distribution of earnings. American Journal of Sociology,

84(3), 541–564.

Anderson, C. (2015). Creating a data-driven organization: Practical advice

from the trenches (1st ed.). O’Reilly Media Inc.

Andrews, K. R. (1980). The concept of corporate strategy.

Apté, C., Dietrich, B., & Fleming, M. (2012). Business leadership through

analytics. IBM Journal of Research and Development,56(6), 7: 1–7:

Atzmüller, C., Kromer, I., & Elisabeth, R. (2014). Peer Delinquency:

Wahrnehmung und Bewertung typischer Jugenddelikte aus der Sicht

Jugendlicher als Grundlage für Präventionsmaßnahmen. Innovation

und Technologie (BMVIT), Wien, Österreich.

Atzmüller, C., & Steiner, P. M. (2010). Experimental Vignette Studies in

Survey Research. Methodology,6(3), 128–138.

Azvine, B., Cui, Z., Nauck, D. D., & Majeed, B. (2006). Real Time Busi-

ness Intelligence for the Adaptive Enterprise. In The 8th IEEE In-

ternational Conference on E-Commerce Technology and The 3rd IEEE

International Conference on Enterprise Computing, E-Commerce, and

E-Services (CEC/EEE’06). IEEE.

Baars, H., & Kemper, H.-G. (2021). Business Intelligence & Analytics–

Grundlagen und praktische Anwendungen. Auﬂ., Wiesbaden in

Druck. Retrieved from https://link.springer.com/content/

pdf/10.1007/978-3-8348-2344-1.pdf

Bader, V., & Kaiser, S. (2019). Algorithmic decision-making? the user inter-

face and its role for human involvement in decisions supported by

artiﬁcial intelligence. Organization,26(5), 655–672.

Barbosa, L. C., & Hirko, R. G. (1980). Integration of Algorithmic Aids into

Decision Support Systems. MIS Quarterly,4(1), 1.

Barnett, T., Bass, K., & Brown, G. (1994). Ethical ideology and ethical

judgment regarding ethical issues in business. Journal of Business

Ethics,13(6), 469–480.

Barrera, D., & Buskens, V. (2007). Imitation and learning under uncertainty:

a vignette experiment. International sociology,22(3), 367–396.

Barton, D., & Court, D. (2012). Making advanced analytics work for you.

Harvard Business Review,90(10), 78–83.

Barton, M. C., & Pöppelbuß, J. (2022). Prinzipien für die ethische Nutzung

künstlicher Intelligenz. HMD Praxis der Wirtschaftsinformatik,59(2),

468–481.

Baumann-Habersack, F. H. (2021). Autorität, Algorithmen und Konﬂikte

– Die digitalisierte Renaissance autoritärer Führungsprinzipien. In

Kooperation in der digitalen Arbeitswelt (pp. 279–291). Wiesbaden,

Springer Gabler.

Bazerman, M. H., & Moore, D. A. (2012). Judgment in managerial decision

making. John Wiley & Sons.

Beck, M., & Opp, K. (2001). Der faktorielle Survey und die Messung von

Normen. KZfSS Kölner Zeitschrift für Soziologie und Sozialpsychologie,

53(2), 283–306.

Becker, J.-M., Ringle, C. M., Sarstedt, M., & Völckner, F. (2015). How

collinearity affects mixture regression results. Marketing Letters,

26(4), 643–659.

Benaben, F., Lauras, M., Montreuil, B., Faugere, L., Gou, J., & Mu, W. (2019).

Physics of Organization Dynamics: An AI Framework for opportunity

and risk management. In 2019 International Conference on Industrial

Engineering and Systems Management (IESM). IEEE.

Biswas, T. T. (Ed.). (2015). Measuring Intrinsic Quality of Human Decisions

(Vol. 9346). Cham, Springer.

Blutner, D., Cramer, S., Krause, S., Mönks, T., Nagel, L., Reinholz, A., &

Witthaut, M. (2009). Assistenzsysteme für die Entscheidungsunter-

stützung. In Große Netze der Logistik (pp. 241–270). Berlin, Heidel-

berg, Springer.

Brahm, C., Cheris, A., & Sherer, L. (2016). What Big Data Means for Cus-

tomer Loyalty. Brief, Bain and Company, August, 7.

Buxmann, P., & Schmidt, H. (2021). Grundlagen der Künstlichen Intelligenz

und des Maschinellen Lernens. In Künstliche Intelligenz (pp. 3–25).

Berlin, Heidelberg, Springer Gabler.

Camerer, C., & Lovallo, D. (1999). Overconﬁdence and excess entry: An

experimental approach. American economic review,89(1), 306–318.

Carlson, E. D. (1977). Decision support systems: personal computing ser-

vices for managers. Management Review,66(1), 4–11.

Carr, J. C., & Blettner, D. P. (2010). Cognitive control bias and decision-

making in context: Implications for entrepreneurial founders of

small ﬁrms. Frontiers of Entrepreneurship Research,30(6), 2.

Cavanagh, G. F., & Fritzsche, D. J. (1985). Using vignettes in business ethics

research. Research in Corporate Social Performance and Policy,7, 279–

293.

Chen, D. Q., Preston, D. S., & Swink, M. (2015). How the use of big data

analytics affects value creation in supply chain management. Journal

of management information systems,32(4), 4–39.

Colossyan. (2022). Create videos with AI actors, real easy. https://www

.colossyan.com/.

Companiesmarketcap. (2022). Largest Companies by Market Cap. https://

companiesmarketcap.com/.

Cook, F. L. (1979). Who should be helped? Public support for social services:

Public Support for Social Services.

Côrte-Real, N., Oliveira, T., & Ruivo, P. (2017). Assessing business value of

Big Data Analytics in European ﬁrms. Journal of Business Research,

70, 379–390.

Davenport, T. H. (2013). Analytics 3.0. Harvard Business Review,91(12),

64–72.

Davis, F. D. (1989). Perceived Usefulness, Perceived Ease of Use, and User

Acceptance of Information Technology. MIS Quarterly,13(3), 319–

340.

Davis, F. D., Bagozzi, R. P., & Warshaw, P. R. (1989). User acceptance

of computer technology: A comparison of two theoretical models.

Management Science,35(8), 982–1003.

Delen, D., & Demirkan, H. (2013). Data, information and analytics as ser-

vices. Decision Support Systems,55(1), 359–363.

Dellermann, D., Ebel, P., Söllner, M., & Leimeister, J. M. (2019). Hybrid In-

telligence. Business & Information Systems Engineering,61(5), 637–

643.

DeSanctis, G., & Poole, M. S. (1994). Capturing the Complexity in Advanced

Technology Use: Adaptive Structuration Theory. Organization Sci-

ence,5(2), 121–147.

Dubinsky, A. J., Jolson, M. A., Kotabe, M., & Lim, C. U. (1991). A

Cross-National Investigation of Industrial Salespeople’s Ethical Per-

ceptions. Journal of International Business Studies,22(4), 651–670.

Dülmer, H. (2001). Bildung und der Einﬂuss von Argumenten auf

das Moralische Urteil. KZfSS Kölner Zeitschrift für Soziologie und

Sozialpsychologie,53(1), 1–27.

Emerson, R. M. (1976). Social Exchange Theory. Annual Review of Sociology,

2(1), 335–362.

Evans, J. R., & Lindner, C. H. (2012). Business Analytics: The Next Frontier

for Decision Sciences. College of Business, University of Cincinnati.

Decision Science Institute,21(12). Retrieved from http://www.cbpp

.uaa.alaska.edu/afef/business{_}analytics.htm

Evans, J. S. B. T., & Stanovich, K. E. (2013). Dual-process theories of higher

cognition: Advancing the debate. Perspectives on Psychological Sci-

ence,8(3), 223–241.

Everett, C. R., & Fairchild, R. J. (2015). A theory of entrepreneurial over-

conﬁdence, effort, and ﬁrm outcomes. Journal of Entrepreneurial

Finance,17(1), 1–27.

Fayyad, U., Piatetsky-Shapiro, G., & Smyth, P. (1996). From data mining to

knowledge discovery in databases. AI magazine,17(3), 37–54.

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 923

Fogg, B. J. (1998). Persuasive computers: perspectives and research di-

rections. Proceedings of the SIGCHI conference on Human factors in

computing systems, 225–232.

Forbes, D. P. (2005). Are some entrepreneurs more overconﬁdent than oth-

ers? Journal of business venturing,20(5), 623–640.

Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation mod-

els with unobservable variables and measure-ment error. Journal of

marketing research,18(1), 39–50.

Fuchs, C., & Diamantopoulos, A. (2009). Using single-item measures for

construct measurement in management research: Conceptual issues

and application guidelines. Die Betriebswirtschaft,69(2), 195.

Furber, S. (2016). Large-scale neuromorphic computing systems. Journal of

Neural Engineering,13(5), 051001.

Gartz, U. (2004). Enterprise information management. In Business intelli-

gence in the digital economy: opportunities, limitations and risks (pp.

48–75). IGI Global.

Gefen, Karahanna, & Straub. (2003). Trust and TAM in Online Shopping:

An Integrated Model. MIS Quarterly,27(1), 51.

GEMESYS Technologies. (2022). Wir bauen einen vom menschlichen Gehirn

inspirierten Computer. https://gemesys.tech/.

Gersch, M., Meske, C., Bunde, E., Aldoj, N., Wesche, J. S., Wilkens, U.,

& Dewey, M. (2021). Vertrauen in KI-basierte Radiologie – Erste

Erkenntnisse durch eine explorative Stakeholder-Konsultation. In

Künstliche intelligenz im dienstleistungsmanagement (pp. 309–335).

Wiesbaden, Springer Gabler.

Gluchowski, P. (2016). Business Analytics–Grundlagen, Methoden und Ein-

satzpotenziale. HMD Praxis der Wirtschaftsinformatik,53(3), 273–

286.

Goddard, K., Roudsari, A., & Wyatt, J. C. (2012). Automation bias: a system-

atic review of frequency, effect mediators, and mitigators. Journal of

the American Medical Informatics Association,19(1), 121–127.

Gong, L. (2008). How social is social responses to computers? the function

of the degree of anthropomorphism in computer representations.

Computers in Human Behavior,24(4), 1494–1509.

Grant, A. M., & Wall, T. D. (2009). The neglected science and art of quasi-

experimentation: Why-to, when-to, and how-to advice for organi-

zational researchers. Organizational Research Methods,12(4), 653–

686.

Grossman, R., & Siegel, K. (2014). Organizational models for big data and

analytics. Journal of Organization Design,3(1), 20–25.

Guidotti, R., Monreale, A., Ruggieri, S., Turini, F., Giannotti, F., & Pedreschi,

D. (2018). A survey of methods for explaining black box models.

ACM computing surveys (CSUR),51(5), 1–42.

Güting, R. H., & Dieker, S. (1992). Datenstrukturen und Algorithmen.

Springer.

Hair Jr., J. F., Hult, G. T. M., Ringle, C. M., Sarstedt, M., Danks, N. P., &

Ray, S. (2021). Partial Least Squares Structural Equation Modeling

(PLS-SEM) Using R: A Workbook. Springer International Publishing

AG.

Hair Jr., J. F., Risher, J. J., Sarstedt, M., & Ringle, C. M. (2019). When to use

and how to report the results of PLS-SEM. European Business Review,

31(1), 2–24.

Hair Jr., J. F., Matthews, L. M., Matthews, R. L., & Sarstedt, M. (2017).

PLS-SEM or CB-SEM: updated guidelines on which method to use.

International Journal of Multivariate Data Analysis,1(2), 107–123.

Hair Jr., J. F., Ringle, C. M., & Sarstedt, M. (2011). PLS-SEM: Indeed a Silver

Bullet. Journal of Marketing Theory and Practice,19(2), 139–152.

Hajkowicz, S., Reeson, A., Rudd, L., Bratanova, A., Hodgers, L., Mason, C.,

& Boughen, N. (2016). Tomorrow’s digitally enabled workforce:

Megatrends and scenarios for jobs and employment in Australia over

the coming twenty years.

Halper, F. (2014). Predictive analytics for business advantage. TDWI Re-

search, 1–32.

Hamilton, B., & Koch, R. (2015). From predictive to prescriptive analytics.

Strategic Finance,96(12), 62.

Hassenzahl, M., & Tractinsky, N. (2006). User experience - a research

agenda. Behaviour & Information Technology,25(2), 91–97.

Hastenteufel, J., & Ganster, F. (2021). Einﬂussfaktoren auf die Akzeptanz

von Robo Advisors: Digitale Kommunikation in der Anlageberatung.

Springer Fachmedien Wiesbaden.

Henseler, J., Ringle, C. M., & Sarstedt, M. (2015). A new criterion for

assessing discriminant validity in variance-based structural equation

modeling. Journal of the Academy of Marketing Science,43(1), 115–

135.

Huber, G. P. (1990). A theory of the effects of advanced information tech-

nologies on organizational design, intelligence, and decision making.

Academy of Management Review,15(1), 47–71.

Hyman, M. R., & Steiner, S. D. (1996). The vignette method in business

ethics research: Current uses and recommendations. Marketing:

Moving Toward the 21st Century, 261–265.

Iansiti, M., & Lakhani, K. R. (2020). Competing in the age of AI: How

machine intelligence changes the rules of business. Harvard Business

Review,98(1), 60–67.

Iqbal, R., Doctor, F., More, B., Mahmud, S., & Yousuf, U. (2020). Big Data an-

alytics and Computational Intelligence for Cyber–Physical Systems:

Recent trends and state of the art applications. Future Generation

Computer Systems,105, 766–778.

Isaac, R. G., Zerbe, W. J., & Pitt, D. C. (2001). Leadership and motivation:

The effective application of expectancy theory. Journal of managerial

issues, 212–226.

Janis, I. L., & Mann, L. (1977). Decision making: A psychological analysis of

conﬂict, choice, and commitment. Free press.

Jasso, G., & Webster Jr, M. (1999). Assessing the gender gap in just earnings

and its underlying mechanisms. Social Psychology Quarterly, 367–

380.

Jöreskog, K. G., & Wold, H. O. A. (1982). Systems under indirect observation:

Causality, structure, prediction. North Holland.

Kagan, B. J., Kitchen, A. C., Tran, N. T., Parker, B. J., Bhat, A., Rollo,

B., . . . Friston, K. J. (2021). In vitro neurons learn and exhibit

sentience when embodied in a simulated game-world. bioRxiv.

Retrieved from https://www.biorxiv.org/content/biorxiv/

early/2021/12/03/2021.12.02.471005.full.pdf

Kahneman, D. (2003). A perspective on judgment and choice: mapping

bounded rationality. American psychologist,58(9), 697.

Kahneman, D., & Schmidt, T. (2012). Schnelles Denken, langsames Denken.

Siedler Verlag.

Kahneman, D., Slovic, S. P., Slovic, P., & Tversky, A. (1982). Judgment under

uncertainty: Heuristics and biases. Cambridge university press.

Kahneman, D., & Tversky, A. (1996). On the reality of cognitive illusions.

1939-1471.

Karahanna, E., Xin Xu, S., Xu, Y., & Zhang, N. (2018). The Needs–

Affordances–Features Perspective for the Use of Social Media. MIS

Quarterly,42(3), 737–756.

Kelley, H. H., & Michela, J. L. (1980). Attribution theory and research.

Annual review of psychology,31(1), 457–501.

Knebl, H. (2019). Algorithmen und Datenstrukturen: Grundlagen und

probabilistische Methoden für den Entwurf und die Analyse. Springer

Vieweg.

Koch, R. (2015). From business intelligence to predictive analytics. Strategic

Finance,96(7), 56–58.

Koellinger, P., Minniti, M., & Schade, C. (2007). “I think I can, I think I can”:

Overconﬁdence and entrepreneurial behavior. Journal of economic

psychology,28(4), 502–527.

Königstorfer, J. (2008). Akzeptanz von technologischen Innovationen:

Nutzungsentscheidungen von Konsumenten dargestellt am Beispiel von

mobilen Internetdiensten.

Korsgaard, M. A., Schweiger, D. M., & Sapienza, H. J. (1995). Building Com-

mitment, Attachment, and Trust in Strategic Decision-Making Teams:

The Role of Procedural Justice. Academy of Management Journal,

38(1), 60–84.

Kotler, P., Berger, R., & Bickhoff, N. (2010). The quintessence of strategic

management. What You Really Need to Know to Survive in Business,

Berlin.

Kreutzer, R. T., & Sirrenberg, M. (2019). Künstliche Intelligenz verstehen ([1.

Auﬂage]ed.). Wiesbaden, Springer Fachmedien.

La Hayward, M., Forster, W. R., Sarasvathy, S. D., & Fredrickson, B. L.

(2010). Beyond hubris: How highly conﬁdent entrepreneurs re-

bound to venture again. Journal of business venturing,25(6), 569–

578.

Larson, D., & Chang, V. (2016). A review and future direction of agile, busi-

ness intelligence, analytics and data science. International Journal of

Information Management,36(5), 700–710.

K. Iqbal /Junior Management Science 8(4) (2023) 887-925924

Laudon, K. C., Laudon, J. P., & Schoder, D. (2016). Wirtschaftsinformatik:

Eine Einführung (3rd ed.). Pearson Studium.

LaValle, S., Lesser, E., Shockley, R., Hopkins, M. S., & Kruschwitz, N. (2011).

Big data, analytics and the path from insights to value. MIT sloan

management review,52(2), 21–32.

Lawler, E. J., & Thye, S. R. (1999). Bringing Emotions into Social Exchange

Theory. Annual Review of Sociology,25(1), 217–244.

Lee, M. K. (2018). Understanding perception of algorithmic decisions: Fair-

ness, trust, and emotion in response to algorithmic management. Big

Data & Society,5(1), 205395171875668.

Leimeister, J. M. (2019). Dienstleistungsengineering und-management: Data-

driven service innovation. Springer.

Lemke, C., Monett, D., & Mikoleit, M. (2021). Digitale Ethik

in datengetriebenen Organisationen und deren Anwendung am

Beispiel von KI-Ethik. In Data Science anwenden (pp. 33–52).

Springer.

Linardatos, P., Papastefanopoulos, V., & Kotsiantis, S. (2020). Explainable

AI: a review of machine learning interpretability methods. Entropy,

23(1), 18.

Luhmann, N. (1990). Risiko und Gefahr. In Soziologische aufklärung 5 (pp.

131–169). Springer.

Makarius, E. E., Mukherjee, D., Fox, J. D., & Fox, A. K. (2020). Rising

with the machines: A sociotechnical framework for bringing artiﬁcial

intelligence into the organization. Journal of Business Research,120,

262–273.

Mallach, E. G. (1994). Understanding Decision Support Systems and Expert

Systems. Richard D. Irwin. Inc., USA.

Martini, M. (2019). Blackbox Algorithmus. Grundfragen einer Regulierung

künstlicher Intelligenz, Berlin.

Mashingaidze, K., & Backhouse, J. (2017). The relationships between def-

initions of big data, business intelligence and business analytics: a

literature review. International Journal of Business Information Sys-

tems,26(4), 488–505.

Mayer, R. C., Davis, J. H., & Schoorman, F. D. (1995). An Integrative Model

Of Organizational Trust. Academy of Management Review,20(3),

709–734.

Mcafee, A., Brynjolfsson, E., Davenport, T. H., Patil, D. J., & Barton, D.

(2012). Big data: the management revolution. Harvard Business

Review,90(10), 60–68.

McKelvie, A., Haynie, J. M., & Gustavsson, V. (2011). Unpacking the uncer-

tainty construct: Implications for entrepreneurial action. Journal of

business venturing,26(3), 273–292.

Merendino, A., Dibb, S., Meadows, M., Quinn, L., Wilson, D., Simkin, L.,

& Canhoto, A. (2018). Big data, big decisions: The impact of big

data on board level decision-making. Journal of Business Research,

93, 67–78.

Meske, C., Bunde, E., Schneider, J., & Gersch, M. (2022). Explainable Ar-

tiﬁcial Intelligence: Objectives, Stakeholders, and Future Research

Opportunities. Information Systems Management,39(1), 53–63.

Mikalef, P., Boura, M., Lekakos, G., & Krogstie, J. (2019). Big data analyt-

ics and ﬁrm performance: Findings from a mixed-method approach.

Journal of Business Research,98, 261–276.

Mikalef, P., Pappas, I., Krogstie, J., & Pavlou, P. A. (2020). Big data and

business analytics: A research agenda for realizing business value.

0378-7206.

Mishra, N., & Silakari, S. (2012). Predictive analytics: a survey, trends, ap-

plications, oppurtunities & challenges. International Journal of Com-

puter Science and Information Technologies,3(3), 4434–4438.

Moore, G. E. (1965). Cramming more components onto integrated circuits.

McGraw-Hill New York.

Mori, M., MacDorman, K. F., & Kageki, N. (2012). The Uncanny Valley [From

the Field].IEEE Robotics & Automation Magazine,19(2), 98–100.

Moschovakis, Y. N. (2001). What Is an Algorithm? In Mathematics unlimited

— 2001 and beyond (pp. 919–936). Berlin, Heidelberg, Springer.

Murphy, K. P. (2012). Machine learning: A probabilistic perspective. MIT

Press.

Nedelcu, B. (2013). Business intelligence systems. Database Systems Jour-

nal,4(4), 12–20.

Newman, D. T., Fast, N. J., & Harmon, D. J. (2020). When eliminating bias

isn’t fair: Algorithmic reductionism and procedural justice in human

resource decisions. Organizational Behavior and Human Decision Pro-

cesses,160, 149–167.

Orlikowski, W. J., & Robey, D. (1991). Information Technology and the

Structuring of Organizations. Information Systems Research,2(2),

143–169.

Panagiotarou, A., Stamatiou, Y. C., Pierrakeas, C., & Kameas, A. (2020).

Gamiﬁcation Acceptance for Learners with Different E-Skills. In-

ternational Journal of Learning, Teaching and Educational Research,

19(2), 263–278.

Porter, M. E. (1996). What is strategy? Harvard Business Review,74(6),

61–78.

Pütz, C., Düppre, S., Roth, S., & Weiss, W. (2021). Akzeptanz und Nutzung

von Chat-/Voicebots. In Künstliche intelligenz im dienstleistungsman-

agement (pp. 361–383). Wiesbaden, Springer Gabler.

R Core Team. (2013). R: R: A language and environment for statistical comput-

ing : reference index. http://www.R-project.org/. R Foundation

for Statistical Computing.

Rainsberger, L. (2021). KI–die neue Intelligenz im Vertrieb. Springer Books.

Rathje, R., Laschet, F.-Y., & Kenning, P. (2021). Künstliche Intelligenz in der

Finanzdienstleistungsbranche – Welche Bedeutung hat das Kunden-

vertrauen? In Künstliche Intelligenz im Dienstleistungsmanagement

(pp. 265–286). Wiesbaden, Springer Gabler.

Reid, C., McClean, J., Petley, R., Jones, K., & Ruck, P. (2015). Seiz-

ing the information advantage: How organisations can un-

lock value and insight from the information they hold: A

PwC report in conjunction with Iron Mountain. https://

www.pwc.es/es/publicaciones/tecnologia/assets/

Seizing-The-Information-Advantage.pdf.

Rich, E. (1985). Artiﬁcial intelligence and the humanities. Computers and

the Humanities,19(2), 117–122.

Robertson, D. C. (1993). Empiricism in business ethics: Suggested research

directions. Journal of Business Ethics,12(8), 585–599.

Rohner, R. P., & Khaleque, A. (2002). Parental acceptance-rejection and

life-span development: A universalist perspective. Online readings in

psychology and culture,6(1), 1–10.

Rosenberg, J. (2017). Security in embedded systems: Important Security

Concepts, Security And Network Architecture, Software Vulnerabil-

ity And Cyber Attacks, Security And Operating System Architecture.

In A. Vega, P. Bose, & A. Buyuktosunoglu (Eds.), Rugged Embedded

Systems: Computing in Harsh Environments (pp. 149–205). Else-

vier/Morgan Kaufmann.

Sagnier, C., Loup-Escande, E., Lourdeaux, D., Thouvenin, I., & Valléry, G.

(2020). User Acceptance of Virtual Reality: An Extended Technol-

ogy Acceptance Model. International Journal of Human–Computer

Interaction,36(11), 993–1007.

Sarstedt, M., & Wilczynski, P. (2009). More for less? a comparison of single-

item and multi-item measures. Die Betriebswirtschaft,69(2), 211.

Savolainen, S. (2016). Could Acceptance Predict Commitment in Organisa-

tional Change? Impact of Changes Caused by Succession From the

Viewpoint of Non-family Employees in Small Family Firms. Manage-

ment,4(5), 197–215.

Scheuer, D. S. (2020). Akzeptanz von Künstlicher Intelligenz. Springer.

Shanks, G., & Bekmamedova, N. (2012). Achieving beneﬁts with business

analytics systems: An evolutionary process perspective. Journal of

Decision Systems,21(3), 231–244.

Sharma, R., Mithas, S., & Kankanhalli, A. (2014). Transforming decision-

making processes: a research agenda for understanding the impact

of business analytics on organisations. European Journal of Informa-

tion Systems,23(4), 433–441.

Sheridan, T. B., & Verplank, W. L. (1978). Human and Computer Control of

Undersea Teleoperators. Defense Technical Information Center.

Shi, Z. (2021). Intelligence Science: Leading the Age of Intelligence. Elsevier

and Tsinghua University Press.

Shmueli, G., Sarstedt, M., Hair, J. F., Cheah, J., Ting, H., Vaithilingam,

S., & Ringle, C. M. (2019). Predictive model assessment in PLS-

SEM: guidelines for using PLSpredict. European Journal of Market-

ing,53(11), 2322–2347.

Smith, S. L., & Aucella, A. F. (1983). Design guidelines for the user interface to

computer-based information systems. MITRE CORP BEDFORD MA.

Stanovich, K. E., & West, R. F. (2000). Individual differences in reasoning:

Implications for the rationality debate? Behavioral and brain sciences,

23(5), 645–665.

K. Iqbal /Junior Management Science 8(4) (2023) 887-925 925

Steiner, P. M., Atzmüller, C., & Su, D. (2016). Designing valid and reliable

vignette experiments for survey research: A case study on the fair

gender income gap. Journal of Methods and Measurement in the Social

Sciences,7(2), 52–94.

Stevenson, T. H., & Bodkin, C. D. (1998). A Cross-National Comparison

of University Students’ Perceptions Regarding the Ethics and Accept-

ability of Sales Practices. Journal of Business Ethics,17(1), 45–55.

Swartz, L. (2003). Why people hate the paperclip: Labels, appearance,

behavior, and social re-sponses to user interface agents.

Tolstoy. (2022). Tolstoy: A new way to communicate, with interactive video.

https://www.gotolstoy.com/.

Turing, A. M. (1950). Mind. Mind,59(236), 433–460.

Tversky, A., & Kahneman, D. (1974). Judgment under Uncertainty: Heuris-

tics and Biases: Biases in judgments reveal some heuristics of think-

ing under uncertainty. Science,185(4157), 1124–1131.

Uysal, E., Alavi, S., & Bezençon, V. (2022). Trojan horse or useful helper?

a relationship perspective on artiﬁcial intelligence assistants with

humanlike features. Journal of the Academy of Marketing Science,

1–23. Retrieved from https://link.springer.com/article/

10.1007/s11747-022-00856-9

van Rijmenam, M., Erekhinskaya, T., Schweitzer, J., & Williams, M.-A.

(2019). Avoid being the Turkey: How big data analytics changes

the game of strategy in times of ambiguity and uncertainty. Long

Range Planning,52(5), 101841.

Venkatesh, V. (2000). Determinants of Perceived Ease of Use: Integrating

Control, Intrinsic Motivation, and Emotion into the Technology Ac-

ceptance Model. Information Systems Research,11(4), 342–365.

Venkatesh, V., & Bala, H. (2008). Technology acceptance model 3 and a

research agenda on interventions. Decision sciences,39(2), 273–315.

Venkatesh, V., & Davis, F. D. (2000). A Theoretical Extension of the Tech-

nology Acceptance Model: Four Longitudinal Field Studies. Manage-

ment Science,46(2), 186–204.

Walster, E. (1966). Assignment of responsibility for an accident. Journal of

personality and social psychology,3(1), 73.

Wang, P. (2019). On Deﬁning Artiﬁcial Intelligence. Journal of Artiﬁcial

General Intelligence,10(2), 1–37.

Wason, K. D., & Cox, K. C. (1996). Scenario utilization in marketing re-

search. Advances in Marketing. Texas: Southwestern Marketing Asso-

ciation, 155–162.

Wason, K. D., Polonsky, M. J., & Hyman, M. R. (2002). Designing vignette

studies in marketing. Australasian Marketing Journal,10(3), 41–58.

Watson, H. J., & Wixom, B. H. (2007). The current state of business intelli-

gence. Computer,40(9), 96–99.

Waytz, A., Cacioppo, J., & Epley, N. (2010). Who Sees Human? The Stabil-

ity and Importance of Individual Differences in Anthropomorphism.

Perspectives on psychological science : a journal of the Association for

Psychological Science,5(3), 219–232.

ResearchGate has not been able to resolve any citations for this publication.

On the Reality of Cognitive Illusions

Article

Full-text available

Jul 1996

The study of heuristics and biases in judgment has been criticized in several publications by G. Gigerenzer, who argues that “biases are not biases” and “heuristics are meant to explain what does not exist” (1991, p. 102). This article responds to Gigerenzer's critique and shows that it misrepresents the authors' theoretical position and ignores critical evidence. Contrary to Gigerenzer's central empirical claim, judgments of frequency—not only subjective probabilities—are susceptible to large and systematic biases. A postscript responds to Gigerenzer's (1996) reply.

Partial Least Squares Structural Equation Modeling (PLS-SEM) Using R: A workbook

Book

Full-text available

Nov 2021

Partial least squares structural equation modeling (PLS-SEM) has become a standard approach for analyzing complex inter-relationships between observed and latent variables. Researchers appreciate the many advantages of PLS-SEM such as the possibility to estimate very complex models and the method’s flexibility in terms of data requirements and measurement specification. This practical guide provides a step-by-step treatment of the major choices facing researchers when analyzing PLS models using R, a free software environment for statistical computing, which runs on Windows, MacOS, and UNIX computer platforms (https://www.r-project.org/). Adopting the R software’s SEMinR package, which brings a friendly syntax to creating and estimating structural equation models, each chapter offers a concise overview of relevant topics and metrics, followed by an in-depth description of a case study. Simple instructions give readers the “how-tos” of using SEMinR to obtain solutions and document their results. Rules of thumb in every chapter provide guidelines on best practices in the application and interpretation of PLS-SEM.

The Concept of Corporate Strategy

Chapter

Dec 1997

Kenneth R Andrews

Strategic management has been increasingly characterized by an emphasis on core competences. Firms are advised to divest unrelated businesses and return to core business. Moreover, competitive advantage is now increasingly seen as a matter of efficiently deploying scarce knowledge resources to product markets. Much of this change in emphasis has occurred because of the emergence of a unified and rigorous approach to strategy, often called the resource-based approach. This Reader brings together extracts from the seminal articles that created this dominant perspective in strategic management. It includes the pioneering work of Selznick, Penrose, and Chandler and more recent writing by Wernerfelt, Barney, Teece, and Prahalad and Hamel.

Beobachtungsstudie zur Akzeptanz Künstlicher Intelligenz

Chapter

Feb 2020

Dennis Scheuer

Entlang der zuvorgegangenen Analyse der spezifischen Differenzierungsmerkmale Künstlicher Intelligenz in Kapitel 2.3, der vergleichenden Betrachtung der bestehenden Akzeptanzmodelle für innovative Technologien in Kapitel 3.6 und der Weiterentwicklung des Technology Acceptence Model 3 in Kapitel 4 ergibt sich das Referenzmodell für KI-Akzeptanz aus Kapitel 4.3.

A Theory of Entrepreneurial Overconfidence, Effort, and Firm Outcomes

Article

Mar 2015

Determinants of Perceived Ease of Use: Integrating Control, Intrinsic Motivation, and Emotion into the Technology Acceptance Model

Article

Jan 2000

Viswanath Venkatesh

Autorität, Algorithmen und Konflikte – Die digitalisierte Renaissance autoritärer Führungsprinzipien

Chapter

Oct 2021

Frank Baumann-Habersack

Algorithmische Systeme sind heute schon ein wichtiger Bestandteil der Entscheidungsfindung in Unternehmen. Etwas ist aber erst wenigen bewusst: Menschen selbst führen bereits seit Längerem Algorithmen im gesellschaftlichen Kontext aus, was nun aber immer mehr auf der Managementebene digitalisiert wirksamer wird – eine Tatsache, die noch nicht genug beleuchtet wurde. Dieser Beitrag beschäftigt sich daher einführend mit der asymmetrischen Wirkmacht von Algorithmen, deren Wechselwirkung mit Führungsautorität und der drohenden Reinszenierung autoritärer Führungspraktiken. Die in diesem Zusammenhang zwangsläufig entstehenden Konflikte zwischen Mitarbeiter:innen und Führungskräften gilt es als Chance zu verstehen und für die Zukunftsfähigkeit zu nutzen. Noch ist die aktive Mitgestaltung an diesem Transformationsprozess möglich. Autorität ist für unsere Zukunft zu wichtig, um ihre Gestaltung, auch im digitalen Raum, (anti-)autoritären Kräften zu überlassen.

Einflussfaktoren auf die Akzeptanz von Robo Advisors, Digitale Kommunikation in der Anlageberatung

Article

Jan 2021

In diesem essential werden die Faktoren untersucht, die die Nutzungsabsicht eines Robo Advisors beeinflussen. Hierbei werden zunächst die Digitalisierung in der Finanzindustrie sowie die Entstehung und Funktionsweise von Robo Advisorn thematisiert. Anschließend werden mithilfe einer auf dem Technologieakzeptanzmodell beruhenden empirischen Untersuchung die Akzeptanz von Robo Advisory gemessen und auf dieser Basis Handlungsempfehlungen formuliert. Der Inhalt • Digitaler Wandel in der Bankenbranche • Definition und Funktionsweise von Robo Advisory • Technologieakzeptanzmodell • Konzeption und Auswertung einer Studie zur Akzeptanz von Robo Advisory Die Zielgruppen • Wissenschaftler/Forschung Die Autor*innen Prof. PD Dr. Jessica Hastenteufel ist Professorin für Betriebswirtschaftslehre an der IU Internationale Hochschule und Privatdozentin an der Universität des Saarlandes. Felix Ganster ist Wertpapierspezialist bei der UmweltBank AG Nürnberg.

Akzeptanz und Nutzung von Chat-/Voicebots

Chapter

Jul 2021

Dieser Beitrag beschäftigt sich mit der Akzeptanz und Nutzung von Technologien im Allgemeinen und von Chat-/Voicebots im Speziellen. Dazu wird insbesondere untersucht, welche Einflussfaktoren der Akzeptanz im Kontext von Chat-/Voicebots eine Rolle spielen. Auf Basis einer Literaturanalyse werden die individuellen Unterschiede der Nutzer, die Systemcharakteristika, der soziale Einfluss und erleichternde Bedingungen diskutiert. Anschließend werden weitere Einflussfaktoren aus der Perspektive der Praxis ergänzt. Der Beitrag schließt mit einem Fazit und Ausblick.

Vertrauen in KI-basierte Radiologie – Erste Erkenntnisse durch eine explorative Stakeholder-Konsultation

Chapter

Jul 2021

Künstliche Intelligenz verspricht im Bereich Radiologie große Verbesserungspotenziale. Wenige Vorarbeiten weisen auf die überragende Bedeutung von Vertrauen hin, damit die verschiedenen Stakeholder bereit sind, ein entsprechendes KI-basiertes Service System in Anspruch zu nehmen. Im Rahmen einer explorativen Stakeholder-Konsultation konnten wertvolle Hinweise für zukünftige Forschung abgeleitet werden. Hierzu zählen u.a. eine notwendige Stakeholderdifferenzierung relevanter Vertrauensaspekte, die Identifikation besonders wichtiger „Moments of (Dis-) Trust“, die Bedeutung von Surrogat-Informationen sowie der Erklärbarkeit KI-gestützter Entscheidungen.

Acceptance Acceptance conditions of algorithmic decision support in management

Abstract

Recommended publications

Explainable Artificial Intelligence for Human Decision Support System in the Medical Domain

AI-based Spacecraft Operations and the Issue of Lacking Trust - First Results of an AI Trustability...

Unveiling the Evolution of Big Data Analytics Capability: A Comprehensive Analysis

A Meta Survey of Quality Evaluation Criteria in Explanation Methods