A User Acceptance Model for Robotic Process Automation [original]

A User Acceptance Model for Robotic Process

Automation

Judith Wewerka

Institute of Databases and

Information Systems

Ulm University

Ulm, Germany

0000-0002-4809-2480

Sebastian Dax

Chair of Information Systems III

University Regensburg

Regensburg, Germany

[email protected]urg.de

Manfred Reichert

Institute of Databases and

Information Systems

Ulm University

Ulm, Germany

0000-0003-2536-4153

Abstract—Robotic Process Automation (RPA) is the rule-based

automation of business processes by software bots mimicking

human interactions to relieve employees from tedious work. How-

ever, any RPA initiative will not be successful if user acceptance

is poor. So far, variables influencing RPA user acceptance have

not been systematically investigated. The objective of this paper

is to develop a model for assessing RPA user acceptance as

well as variables influencing it. We derive this model using the

Technology Acceptance Model (TAM) and extend TAM by RPA-

specific variables. Our empirical validation indicates that the

most important variables, which significantly influence perceived

usefulness and perceived ease of use are facilitating conditions,

result demonstrability,innovation joy, and social influence. These

findings can be used to derive concrete recommendations for the

design and implementation of RPA bots increasing acceptance of

employees using the bots during their daily work. For the first

time, an RPA user acceptance model is presented and validated

contributing to an increased maturity of RPA projects.

Index Terms—Robotic Process Automation, Technology Accep-

tance Model, User Acceptance

I. INTRODUCTION

Robotic Process Automation (RPA) aims to automate busi-

ness processes or parts of them with software bots (bots for

short) by mimicking human interactions with the graphical

user interface [6]. Recently, many RPA approaches were

implemented and the RPA software market grew by 60%

in 2018 [17]. On one hand, RPA shall relieve employees

from tedious works [22]. On the other, it shall save full time

equivalents [53]. Employees might, therefore, refuse the use

of bots, fearing that they loose their job otherwise [14]. To

cope with these fears, the basic variables determining RPA

user acceptance need to be understood.

This paper explores these variables by developing an RPA

user acceptance model and evaluating it in the automotive

industry. A widely used model for predicting technology adop-

tion behavior is the Technology Acceptance Model (TAM)

[32], [47]. TAM has been applied in areas like, e.g., mobile

payment [32], mobile library applications [55], and search-

based software engineering [36]. TAM represents beliefs about

the consequences of a behavior by internal variables describ-

ing individual perceptions. These internal variables, in turn,

are influenced by external variables. Internal variables include

perceived usefulness (PU), perceived ease of use (PEOU),

Fig. 1. TAM 1 model.

attitude toward use, and behavioral intention to use (BI)

(cf. Figure 1). In particular, PU, i.e., “the degree to which a

person believes that using a particular system would enhance

his or her job performance” [12] and PEOU, i.e., “the degree

to which a person believes that using a particular system

would be free of effort” [12], directly influence attitude

toward use, which has an impact on BI, and, finally, on the

actual use. PU is hypothesized to influence BI directly. These

hypothesized interconnections are represented by directed ar-

rows (cf. Figure 1), which indicate a positive influence of

the source variable on the target variable. Though there is a

variety of technology adoption and acceptance models, TAM is

considered as the by far most widely used and most influential

predictive interpretation model in the analysis and prediction

of information technology adoption behavior [32]. Thus, our

RPA user acceptance model is based on TAM 1 [11] and its

extensions TAM 2 and TAM 3.

The RPA user acceptance model can be used to develop

an understanding of the variables of RPA adoption and use.

Consequently, underutilization of future RPA bots shall be

prevented and the adoption of already existing RPA bots

shall be increased. Overall, the RPA user acceptance model

presented in this paper shall help companies to actually realize

the economic and technological benefits of RPA.

II. METHODOLOGY

This section introduces the methodology for deriving and

validating the RPA user acceptance model (cf. Figure 2). In

general, we distinguish two major steps, which are presented

in Sections III and IV. Step 1 reviews literature on TAM

to collect external variables. Moreover, it studies literature

Fig. 2. Methodology.

on RPA to identify inclusion and exclusion criteria, which

are then applied on the collected external variables. These

variables, the basic model from TAM literature, and additional

RPA-specific variables form our RPA user acceptance model.

Step 2 validates this RPA user acceptance model. A survey

is conducted and the obtained data is used to validate the

RPA user acceptance model. The analysis of the obtained data

allows for optimizing the RPA user acceptance model.

III. DERIVING AN RPA USER ACCEPTANCE MODEL

We apply the first step of our method (cf. Figure 2, left) to

obtain an initial RPA user acceptance model, i.e., the collection

of external variables from TAM literature (cf. Section III-A),

the identification of inclusion and exclusion criteria from RPA

literature (cf. Section III-B), and the application of these

criteria on the variables (cf. Section III-C).

A. Collecting External Variables from TAM Literature

We review TAM literature to collect external variables.

TAM 1 has two extensions: TAM 2 [48] and TAM 3 [47]. The

former investigates external variables influencing PU, while

the latter determines external variables affecting PEOU. Note

that these extensions exclude attitude toward use, originally

covered by TAM 1. On one hand, it has proven to be non-

significant in practice. On the other, as the connection between

PEOU and BI is significant, it is added to the model.

Table I summarizes external variables introduced by the

models TAM 1 - TAM 3. To cover a variety of external

variables and to explain user acceptance in the best possible

way, we additionally consider the Unified Theory of Accep-

tance and Use of Technology 1 and 2 (UTAUT 1, UTAUT 2)

[49], [50]. The latter is based on eight existing user accep-

tance models and is a good summary of possible variables

influencing user acceptance. We omit the interconnections

between the variables and refer to [10], [11], [13], [33],

[37], [44], [46]–[51] instead. In addition to external variables,

moderating variables may be added to a technology accep-

tance model. The latter moderate the influence of external

variables on internal ones like, e.g., experience [23], [48]–

[50], voluntariness [2], [23], [48], [49], age [49], [50], and

gender [49], [50]. Figure 3 visualizes the difference between

external,internal, and moderating variables.

TABLE I

External variables DERIVED FROM TECHNOLOGY ACCEPTANCE

LITERATURE.

External variable Definition Model

Subjective norm, so-

cial influence

Social influence of people being

important to the user

TAM 2,

UTAUT 1

Image Use of the technology enhances the

reputation or status of the user

TAM 2

Job relevance New technology is applicable to

the job of the user

TAM 2

Result demonstrabil-

ity

Effects from the use of technology

are tangible for the user

TAM 2

Output quality The technology performs job-

related tasks

TAM 2

Computer self-

efficacy

Users’ perception of their ability to

use computers

TAM 3

Perceptions of exter-

nal control, facilitat-

ing conditions

Availability of resources and sup-

port structure to facilitate the use

of the technology

TAM 3,

UTAUT 1

Computer anxiety Users’ fear or apprehension when

using computers

TAM 3

Computer playful-

ness

Intrinsic motivation of users for

applying emerging technologies

TAM 3

Perceived enjoyment,

hedonic motivation

System-specific fun and pleasure

for using the technology

TAM 3,

UTAUT 2

Objective usability Actual effort required to complete

specific tasks using the technology

TAM 3

Price value Trade off between perceived bene-

fits of the technology and the costs

caused by its use

UTAUT 2

Habit Automatic behavior due to learning UTAUT 2

Fig. 3. Relation between external,internal, and moderating variables.

B. Identifying Inclusion/Exclusion Criteria in RPA Literature

This section reviews RPA literature in order to derive

inclusion and exclusion criteria for deciding whether or not

external variables (cf. Table I) shall be considered in the

RPA user acceptance model. As RPA is a rather new research

area, we additionally consider critical success factors dis-

cussed in Business Process Management [38], Process-aware

Information Systems [39], and Workflow Implementation [8],

[9], [41] to obtain more rigor criteria. Table II shows the

derived inclusion and exclusion criteria. Note that all criteria

are initially derived from RPA literature.

C. Applying the Criteria on the Variables

We apply the inclusion and exclusion criteria (cf. Table II) to

the external variables (cf. Table I), resulting in those variables

that might affect PU and PEOU in the RPA context.

A variable is included in the RPA user acceptance model if

it fulfills at least one inclusion criterion and is excluded if any

exclusion criterion is met. Besides the RPA-specific exclusion

criteria, we introduce two additional ones:

TABLE II

INCLUSION/EXCLUSION CRITERIA DERIVED FROM RPA LITERATURE.

Inclusion criterion Reference

RPA-INCL-1 IT security requirements [8], [28], [29], [53]

RPA-INCL-2 Robust and reliable oper-

ation of RPA bots

[9], [28], [29]

RPA-INCL-3 High response speed of

RPA bots

[28]

RPA-INCL-4 Stable and secure infras-

tructure

[29]

RPA-INCL-5 Competent support struc-

ture in the Center of Ex-

cellence

[27], [28]

RPA-INCL-6 Detailed documentation [29], [38], [39]

RPA-INCL-7 Management commitment [27], [28], [38], [39], [41]

RPA-INCL-8 User involvement [27]–[30], [41], [52]

RPA-INCL-9 Removal of fear of job

losses

[8], [38], [39]

RPA-INCL-10 Communication of the

RPA advantages for

employees

[28]–[30], [52], [53]

RPA-INCL-11 Characteristics and func-

tionality of RPA

[3], [26], [28], [52], [53]

Exclusion criterion Reference

RPA-EXCL-1 Characteristics and func-

tionality of RPA

[3], [26], [28], [52], [53]

RPA-EXCL-2 RPA price model in indus-

try

[6], [28]

- CTX-EXCL-1: Variable is out of scope for our research,

i.e., the external variable does not contribute to a basic

understanding of RPA user acceptance.

- CTX-EXCL-2: Variable is outdated. Note that the first

technology acceptance models are around 40 years old

and not every variable is valid in its original version

anymore.

Table III summarizes the external variables we included as

well as the reason for inclusion. The criteria are elaborated in

detail in the following:

-Social influence. According to literature on RPA use

cases, management commitment is important to adopt

RPA bots (RPA-INCL-7).

-Job relevance. To be job-relevant, the bots need to be

robust and reliable (RPA-INCL-2). Users should state

relevant bot features to be relieved from repetitive tasks

(RPA-INCL-8, RPA-INCL-11).

-Result demonstrability. This variable tests whether the bot

is reliable, documented, and results are tangible (RPA-

INCL-2, RPA-INCL-6).

-Computer self-efficacy. According to RPA literature, users

should be involved in RPA projects and their fear of

losing their job should be reduced (RPA-INCL-8, RPA-

INCL-9). This variable varies from user to user.

-Facilitating conditions. To investigate whether support by

experts (RPA-INCL-5) or documentation (RPA-INCL-6)

is available to the user. As reported by RPA literature,

both aspects constitute important success factors for RPA.

-Innovation joy. This variable is derived as a positive form

of computer anxiety and is included to investigate whether

TABLE III

External variables INCLUDED IN THE RPA USER ACCEPTANCE MODEL.

External variable Inclusion criteria

Social influence RPA-INCL-7

Job relevance RPA-INCL-2, RPA-INCL-8, RPA-INCL-11

Result demonstrability RPA-INCL-2, RPA-INCL-6

Computer self-efficacy RPA-INCL-8, RPA-INCL-9

Facilitating conditions RPA-INCL-5, RPA-INCL-6

Innovation joy RPA-INCL-9

Hedonic motivation RPA-INCL-11

the attitude of the user toward new technology (RPA-

INCL-9) has an influence on acceptance.

-Hedonic motivation. RPA should be used to redeploy

employees from boring routine work to more interesting,

decision-making tasks (RPA-INCL-11). Thus, hedonic

motivation might influence RPA bot acceptance.

Having a closer look at the inclusion criteria, we notice that

RPA-INCL-1 and RPA-INCL-10 have never been used to

include an external variable from Table I. Thus, it is rea-

sonable to include additional RPA-specific external variables

not listed in this table. [42] integrates trust with TAM, as

“trust is essential for understanding interpersonal behavior”

[42]. In an RPA-specific context, trust can be observed from

both a reliability and a risk perspective. Therefore, trust meets

RPA-INCL-1, RPA-INCL-2, RPA-INCL-4, and RPA-INCL-

6 and is included in our RPA user acceptance model. We

further add variable user involvement (RPA-INCL-8, RPA-

INCL-10) to the RPA user acceptance model as it refers to

user involvement during the design and implementation of

RPA bots. Additionally, it checks whether the user has been

informed about the opportunities provided by automation.

Table IV depicts external variables excluded from the RPA

user acceptance model and corresponding exclusion criteria:

-Image. This variable is excluded as in a working context,

reputation is nowadays improved through hard work and

good results, but not by using a specific technology

(CTX-EXCL-2).

-Output quality. The output of RPA bots is either correct

or incorrect. Therefore, output quality is not measurable

on a scale and thus it is excluded (RPA-EXCL-1).

-Computer playfulness. RPA bots only work in the defined

way and the user cannot play with them or try out

different functions (RPA-EXCL-1).

-Objective usability. The execution time of the bot cannot

be influenced by the user once the bot is triggered.

Therefore, this variable is excluded (RPA-EXCL-1).

-Price value. In practice, users do not pay to use RPA

bots. Hence, price value is excluded (RPA-EXCL-2).

-Habit. To measure this variable, it is required to conduct

the survey several times. Our goal is to obtain a basic

understanding of RPA user acceptance. Thus, we exclude

this variable (CTX-EXCL-1).

The moderating variables experience, voluntariness, age, and

gender are excluded due to CTX-EXCL-1. Remember that the

TABLE IV

External variables EXCLUDED FROM THE RPA USER ACCEPTANCE MODEL.

External variable Exclusion criteria

Image CTX-EXCL-2

Output quality RPA-EXCL-1

Computer playfulness RPA-EXCL-1

Objective usability RPA-EXCL-1

Price value RPA-EXCL-2

Habit CTX-EXCL-1

Fig. 4. RPA user acceptance model.

objective of this study is to get a basic understanding of RPA

user acceptance as well as the variables influencing it.

Internal variables as well as all interconnections are

strongly oriented to the original TAM. We exclude actual use

as TAM is more likely to predict BI than actual use [45],

which is in line with our research goal. Figure 4 shows the

RPA user acceptance model including all variables (external

and internal), interconnections, and resulting hypotheses (H1-

H14). Remember that an interconnection hypothesizes a pos-

itive influence of the source variable on the target variable of

the arrow.

IV. VALIDATING THE RPA USER ACCEPTANCE MODEL

To validate the derived RPA user acceptance model em-

pirically, a quantitative survey is developed, reviewed, and

sent to actual and potential RPA users (cf. Fig 2, right). The

Appendix summarizes all external and internal variables of

the RPA user acceptance model, the corresponding question

items, and references (cf. Table VI). All variables are queried

by three question items each (cf. [11]). Most question items

are taken from the questionnaires of TAM literature. Any

adjustments of the question items to the RPA context are

documented in Table VI. Each question item is tokenized

to facilitate the reference to specific ones. The user answers

the question items by a 7-point Likert scale ranging from 1

(“I totally disagree”) to 7 (“I totally agree”). Additionally,

questions to gather general information, i.e., gender, age, and

experience with RPA bots, are included at the beginning of

the questionnaire. After reviewing the questionnaire with some

Fig. 5. Descriptive analysis.

trail subjects, we sent it to 150 RPA users in a large automotive

company.

In the following, the data gathered with this questionnaire

are analyzed using the partial least squares-structural equation

modeling with the SmartPLS V.3.2.9 software. Firstly, we

assess the minimum sample size and conduct a descriptive

analysis (cf. Section IV-A). We adhere to the guidelines set

out for the use of partial least squares-structural equation

modeling in information systems [4], [5], [21]. Therefore,

secondly, the quality of the measurement model is checked

(cf. Section IV-B). This analysis considers the interconnections

between the variables and the questions in the questionnaire.

Thirdly, the structural model, i.e., the interconnection between

the variables themselves, is analysed (cf. Section IV-C). Hence,

we compare the theoretical model with reality as characterized

by the gathered data.

A. Descriptive Analysis

The minimum sample size for partial least squares-structural

equation modeling should be equal to the maximum of: ten

times the largest number of question items used to measure

a variable, i.e., 30, or ten times the largest number of arrows

directed at a variable, i.e., 50 [19]. Therefore, the minimal

sample size for our model is 50. Out of the 150 employees

to whom the questionnaire had been sent, we received 50

responses, resulting in a response rate of 33.33% and fulfilling

the minimal sample size.

Of the 50 respondents, 58% (N=29) are male and 42%

(N=21) are female. The majority is 30 years or younger (46%,

N=23); 34% (N=17) of the respondents have an age between

31 and 40 years, 10% (N=5) are between 41 and 50 years old,

and 10% (N=5) are 50 years or older (cf. Figure 5, left).

Note that 56% (N=28) of the respondents have been using

RPA bots for more than six months, 16% (N=8) between three

and six months, 18% (N=9) between one and three months,

and 10% (N=5) less than a month (cf. Figure 5, right).

B. Measure Model Assessment

Before testing the hypotheses in the structural model, the

quality of the measurement model is assessed in terms of

reliability (Cronbach’s Alpha, Composite Reliability), conver-

gent validity (Average Variance Extracted, Factor Loadings),

and discriminant validity (Fornell-Larcker Criterion, Cross

Loadings). The respective tables can be found in the Appendix.

Reliability refers to the degree to which a measure is free of

variable error [15]. To test reliability, Cronbach’s Alpha and

Composite Reliability are evaluated. As minimum threshold

value for Cronbach’s Alpha we use 0.600 as we are still

in an exploratory research phase [4]. This threshold is not

reached by variables user involvement (0.519) and facilitating

conditions (0.523) (cf. Table VII). However, for early stages of

research, Cronbach’s Alpha values between 0.500 and 0.600

are recommended [40] and are fulfilled by all variables. The

values of Composite Reliability lie between 0.744 (facilitat-

ing conditions) and 0.906 (PEOU), exceeding the suggested

value of 0.700 [4] (cf. Table VII). As Composite Reliability is

regarded as more suitable to assess reliability [4], the model

holds the reliability constraints.

Convergent validity refers to whether the items comprising

a scale behave as if they are measuring a common underlying

variable [12]. Convergent validity is assessed by Average

Variance Extracted and Factor Loadings. An average vari-

ance extracted of 0.500 or higher is satisfactory meaning that

on average a variable explains (more than) 50% of the variance

of its items [43]. All variables fulfill this constraint with values

ranging from 0.509 for user involvement to 0.764 for PEOU

(cf. Table VII). Factor Loadings, in turn, correspond to the

correlation coefficients of each question item with its variable

[5]. Factor Loadings of 0.500 or greater are necessary for

achieving practical significance [18]. Note that this constraint

is not met by REL3 (0.413), FAC3 (0.422), and HED3 (0.406).

However, values between 0.300 and 0.400 are still acceptable.

Therefore, all question items fulfill the requirements and

convergent validity is ascertained (cf. Table VII).

Discriminant validity corresponds to the degree to which

a variable is truly distinct from others [18]. The Fornell-

Larcker Criterion and Cross Loadings are used to assess

discriminant validity. The former is fulfilled if each variable

shares more variance with its assigned question items than

with other variables in the model [4]. This constraint is met

by all variables (cf. Table VIII). The latter criterion, i.e.,

Cross Loadings, demands that the loadings of a question item

with its variable should be higher than the loadings with all

remaining variables [19]. This constraint is not met by FAC3,

which is, therefore, omitted from the empirical evaluation of

the structural model (cf. Table IX).

Furthermore, the RPA user acceptance model is checked for

high levels of multicollinarity by evaluating the Variance In-

flation Factor, which should not exceed 5 [19]. The question

items of our RPA user acceptance model show no signs of

multicollinearity with a maximum variance inflation factor of

2.805 for PEOU3 (cf. Table VII).

In summary, the quality of the measurement model is ascer-

tained, and the constraints for reliability as well as convergent

and discriminant validity are met.

C. Structural Model Assessment

The structural model is assessed in terms of hypotheses

testing and coefficients of determination using the partial least

squares algorithm. The maximum number of iterations is set

to 500 and the stop criterion to 10−7.Path coefficients are

standardized regression coefficients used in assessing causal

Fig. 6. Structural model with path coefficients, significance levels, and R2

values.

linkages and relative effects between statistical variables in

partial least squares-structural equation modeling [19], [24].

Regarding the assessment of path coefficients in partial least

squares models, a minimum threshold of 0.100 is considered to

be significant [34]. The coefficients of determination are de-

noted as R2values measuring the proportion of the variance of

the internal variables explained by the external ones [20]. The

R2values should be judged in relation to studies investigating

the same internal variable. R2values can be high for well

understood phenomena or low for less understood phenomena

[54]. The bootstrapping algorithm (number of Bootstrap sub

samples = 5000) is used to compute the significance level of

the path coefficients, i.e., p-values. If the p-value exceeds 5%,

there is no empirical support to retain a question item or a

variable’s interconnection, and, thus, its theoretical relevance

should be questioned [19].

Figure 6 depicts R2values of each internal variable and

the path coefficients as well as p-values of each hypothesis.

Non-significant variables and interconnections are printed in

grey. The following external variables have positive and

significant influence on PU:social influence (b= 0.242,

p < 0.001), job relevance (b= 0.108,p < 0.05), and result

demonstrability (b= 0.282,p < 0.001). PU has a positive and

significant influence on BI (b= 0.665,p < 0.001). In turn,

PEOU is significantly influenced in positive direction by trust

(b= 0.109,p < 0.05), innovation joy (b= 0.276,p < 0.001),

and facilitating conditions (b= 0.489,p < 0.001). Moreover,

PEOU influences PU (b= 0.182,p < 0.05) and BI (b= 0.096,

p < 0.1) positively and significantly. Thus, Hypotheses H1-

H3, H7, H9, H10, and H12-H14 are supported.

V. DISCUSSION

The presented results are discussed in the following. First,

statistically significant interconnections are explained (cf. Sec-

tion V-A). Second, statistically non-significant interconnec-

tions are analysed and possible reasons for non-significance

are provided (cf. Section V-B). Third, the R2values of the

RPA user acceptance model are compared with R2values

in TAM 1 - TAM 3 (cf. Section V-C). Fourth, we derive

recommendations for increasing RPA user acceptance based

on the statistically significant interconnections in our RPA user

acceptance model (cf. Section V-D). Fifth, threats to validity

are discussed (cf. Section V-E).

A. Statistically Significant Interconnections

Hypothesis H1 is in line with TAM 2, TAM 3, UTAUT

1, and UTAUT 2, describing the leveraging effect of social

influence on PU. The interconnection shows a comparatively

high path coefficient of 0.242. On one hand, this indicates the

importance of management commitment and support. On the

other, word-of-mouth propaganda is revealed as a driver for

RPA adoption. Employees perceive RPA as useful, increasing

their productivity, if colleagues consider the technology as

useful. Although 72% of the respondents have been using RPA

bots over at least three months, the impact of social influence

on PU is high. For the early phases of using RPA bots, we

assume that social influence has an even greater importance.

The hypothesized influence of job relevance on PU is em-

pirically supported (H2) although the path coefficient (0.108)

is only slightly above the threshold of 0.100. H2 emphasizes

that PU of RPA becomes higher if the RPA bot takes over

frequently recurring or time-consuming tasks. Question item

REL3 checks whether the workload of a respondent can

hardly be handled without RPA bots (cf. Table VI). The low

factor loading of 0.413 (cf. Table VII) indicates that RPA

is a supporting technology not being capable of replacing

workforce.

Hypothesis H3, i.e., “result demonstrability positively influ-

ences PU”, is supported. Note that this finding is in line with

the observations made in TAM 2 and TAM 3. Employees need

to understand what RPA bots can do for them and why their

usage is beneficial. If users cannot trace efficiency gains back

to RPA, they do not perceive RPA usefulness and eventually do

not accept it, although RPA bots produce the desired results.

Result demonstrability is the strongest driver for PU showing

a path coefficient of 0.282.

Regarding the predictors of PEOU,trust (H7) has a signif-

icant, but low impact (0.109). Nevertheless, for employees it

is important to trust the RPA bots to perform as designed and

deliver the desired results. If trust is low, the user monitors

and verifies all results produced by RPA bots. Thus, process

time and effort increase and negatively influence PEOU.

Hypothesis H9 describes the influence of innovation joy

on PEOU and is supported with a high path coefficient of

0.276. Innovation joy follows a more general approach than

computer anxiety (in TAM 3). In particular, it examines the

attitudes of users concerning new technologies and their fear

of being replaced by them. Question item JOY3 states that

the respondent is not afraid of being replaced by computers

in near future (cf. Table VI). As opposed to JOY1 and JOY2,

JOY3 is characterized by a low mean of 5.300 and a factor

loading of 0.600 (cf. Table VII). Altogether, respondents do

not fully agree with this statement. The individual joy of a

user to explore new technologies plays a major role for the

PEOU of RPA.

Facilitating conditions correspond to the external variable

with the strongest effect on PEOU (H10). The path coefficient

of 0.489 shows the highest influence for all external variables

of the RPA user acceptance model. The significance of the

interconnection is in line with the empirical investigations

made in TAM 3 (perceptions of external control), UTAUT 1,

and UTAUT 2. Although question item FAC3 is omitted from

the RPA user acceptance model due to discriminant validity

violations (cf. Table IX), it is essential for the employees to be

provided with the resources and knowledge necessary to use

the RPA bots. Literature on technology acceptance emphasizes

that in early stages older employees attach more importance

to receiving assistance with new technology. Although 80% of

the respondents are 40 years or younger and 72% have been

using RPA for more than 3 months, facilitating conditions have

a strong influence on PEOU.

PEOU significantly influences PU with a path coefficient

of 0.182 (H12). This finding is in line with TAM 1, TAM 2,

and TAM 3. The easier RPA bots are to operate, the more

useful they might be. Thus, the use of RPA bots should not

cause user efforts. According to TAM 3, the effect of PEOU on

PU becomes stronger with increasing experience. 72% of the

respondents have been using RPA bots for more than 3 months

and have gained enough experience to judge the difficulty of

using RPA bots. This results in the strong influence, PEOU

has on PU in the RPA context.

The interconnection between PEOU and BI (H13) is par-

tially supported. The path coefficient of 0.096 is slightly below

the threshold of 0.100, and the p-value of 0.093 is statistically

significant if a 90% confidence interval is acceptable [19].

In TAM 1, the influence of PEOU on BI is justified by

the fact that the performance advantages of a system can

be outweighed by the effort of using it. In TAM 3, the

effect of PEOU on BI weakens with increasing experience.

Note that an experienced user is accustomed to the way a

systems works and knows how to use it. In UTAUT 1, the

effect of effort expectancy on BI is only significant after

initial training; it becomes non-significant with increasing

experience. Therefore, the weak influence of PEOU on BI is

due to the experience the respondents have with RPA.

PU is the key driver for BI with a path coefficient of

0.665 (H14). PU, in turn, is significantly impacted by social

influence,job relevance,result demonstrability, and PEOU.

The interconnection shows that RPA bots should contribute to

a substantial increase in job performance and productivity to

be adopted by the employees. UTAUT 1 empirically validates

that men and particularly younger men place more importance

on task accomplishment. These findings are in line with

our empirical investigation in the RPA context: 80% of the

respondents are 40 years or younger and 58% are male.

B. Statistically Non-Significant Interconnections

Regarding RPA use cases, we include a number of external

variables in the RPA user acceptance model. Though not all

are statistically significant, they still provide valuable insights.

The influence of user involvement on PU (H4) is statistically

not significant. As a potential explanation, its question items

INV1 and INV2 are distinct from INV3 (cf. Table VI). The

latter checks whether the user was informed about the benefits

RPA provides for him. Theoretically, RPA can lead to higher

job satisfaction. INV1 and INV2 ask whether the respondent

is involved in clarifying the automation needs and the testing

of RPA bots. We hypothesize that users involved in designing

and testing the requirements of RPA bots better understand

how useful the technology is. However, this effect has not

been proven true for our sample group.

The hypothesized effects of trust on PU (H5) and BI (H6)

are statistically not significant. Theoretically, the intercon-

nection between trust and PU is justified by the fact that

the user is dependent on the programmer of the RPA bots

[42]. Empirically, this reasoning does not apply, as RPA is

a supporting technology and users still have the possibility

to complete a certain task manually, if they do not trust the

RPA bots. Moreover, trust is hypothesized to create positive

attitudes toward a certain technology, reducing the uncertainty

associated with its use, i.e. increasing BI [42]. This reasoning

is statistically not confirmed by the survey. PU is the key driver

for BI, whereas trust becomes less important. Finally, privacy

risk concerns seem to be non-relevant as well.

In the RPA context, computer self-efficacy has no significant

influence on PEOU (H8). RPA requires only basic computer

skills, which does not pose a problem concerning our sample

group. The assumed effect of different computer-related skills

of users on their PEOU does not vindicate.

The hypothesized interconnection between hedonic moti-

vation and BI is statistically not significant (H11). Question

item HED3 states that the use of RPA bots is entertaining

(cf. Table VI). Its factor loading is extremely low compared

to the other question items of hedonic motivation (cf. Table

VII). RPA bots do not claim to be enjoyable, their main

purpose is to increase productivity. According to UTAUT 2,

hedonic motivation becomes less important in predicting BI

with increasing experience and can be a reason to explore

emerging technologies. In the long term, employees use RPA

for more pragmatic purposes, e.g., to be relieved from non-

value adding tasks.

C. Comparing Coefficients of Determination

The R2values are compared with existing technology

acceptance literature to evaluate their quality. Note that the

values in Table V are non-deterministic as they depend on,

e.g., the respective study participants and their experience. Our

RPA user acceptance model explains 67% of the variance in

BI. This value is higher than the R2values of BI in TAM 1,

TAM 2, and TAM 3. Variables social influence,job relevance,

and result demonstrability explain 62% of the variance in PU.

Compared to TAM 2, the R2value of the RPA user acceptance

model exceeds the value of TAM 2 and ranks in the middle

compared to TAM 3. External variables of PU are not modeled

in TAM 1 and the R2value cannot be compared. Variables

TABLE V

R2VALUES IN COMPARISON.

TAM 1 TAM 2 TAM 3 RPA user accep-

tance model

BI 0.35-0.40 0.34-0.52 0.40-0.53 0.67

PU N/A 0.40-0.60 0.52-0.67 0.62

PEOU N/A N/A 0.40-0.60 0.57

trust,innovation joy, and facilitating conditions explain 57%

of the variance in PEOU. Note that this value is highly

acceptable compared to TAM 3 whose R2value lies between

40% and 60%. PEOU is not modeled in TAM 1 and TAM 2

and, therefore, a comparison is impossible.

D. Recommendations for Increasing RPA User Acceptance

Based on the statistically significant interconnections of the

RPA user acceptance model, recommendations to increase

RPA user acceptance are derived. We recommend ensuring

management support and establishing RPA opinion multi-

pliers, which both serve as a kind of technology ambassador

advertising RPA use in personal contact with their colleagues.

This finding is in line with literature on RPA use cases, which

additionally observes that the number of RPA projects grows

exponentially, if management is promoting the technology

[27]. Based on the influence of job relevance on PU, the

potential use cases must be examined thoroughly. Any

automation must introduce real benefits, otherwise it is not

worth for the employee to learn how to use RPA bots, e.g.,

the predicted savings of full time equivalents due to RPA

automation should be carefully chosen. This is in line with the

literature on RPA use cases, e.g., [53] proposes to automate

processes saving three full time equivalents and more.

Intensive communication and practical demonstration of

RPA advantages, e.g., through live demonstrations of existing

RPA bots, are crucial as well. These tasks could be accom-

plished by the RPA opinion multipliers, which can convince

their colleagues that RPA relieves them from repetitive and

boring tasks, and enables them to work on more sophisticated

and value-adding activities. The recommendations are based

on the high influence of result demonstrability and innovation

joy on PU and PEOU respectively. Our empirical data show

that employees are generally not hesitant to try out new

technologies, but have concerns that RPA eventually leads to

job losses. RPA literature considers the removal of fear of job

losses and the clarification of new opportunities the automation

offers as crucial success factor for RPA projects [28].

Areliable and trustworthy operation of RPA bots needs

to be ensured although the influence of trust on PEOU is low.

Literature on RPA use cases suggests that RPA bots should

have built-in controls and checkpoints to verify that the process

ran correctly, e.g., regular updates on the current processing

status may increase confidence in the technology [29].

Facilitating conditions are the most influential variable of

the RPA user acceptance model as well as key success driver

of RPA projects. On one hand, extensive training sessions

should be offered to users, e.g., which information and in

which manner information must be provided to the RPA bots.

Detailed documentation and user manuals help employees to

look up the way RPA bots work. On the other, users need to be

equipped with the resources necessary to use RPA bots. This

includes the infrastructure as well as the access rights required.

Our empirical investigation shows that facilitating conditions

are one of the key variables of RPA for our rather young and

experienced sample group. This indicates that support should

be continuously provided to employees of all ages during the

complete life cycle of the RPA bots.

Special attention should be paid to a user-friendly de-

sign. This includes, the communication between RPA bots

and users. Moreover, RPA bots could use optical character

recognition instead of rigid selectors to find information on

dynamically changing platforms like a website.

The statistically non-significant interconnections enable fur-

ther recommendations. Regarding our sample group, user

involvement does not significantly influence PU and is, there-

fore, not necessary. Computer self-efficacy has no influence

on PEOU. RPA requires only basic computer skills. The

interconnection between hedonic motivation and BI is not

significant. Thus, the enhancement of hedonic motivation does

not contribute to a sustainable use of RPA bots.

E. Threats to Validity

As thread to validity, the RPA user acceptance model is only

validated in one company in the automotive domain. Note

that carrying out the survey in other companies or industry

sectors might lead to different results. Moreover, our work is

limited to the concepts of the chosen literature on technology

acceptance and not all possible variables influencing RPA user

acceptance might have been considered. Finally, our sample

group is young and experienced with RPA. An increased

number and a more balanced distribution of respondents could

further enhance the explanatory power of the model.

VI. SUMMARY AND OUTLOOK

Our goal was to gain a basic understanding of key variables

influencing RPA user acceptance. This goal has been achieved

by expanding TAM by external variables. The developed

RPA user acceptance model is empirically validated. Our

results confirm that social influence,job relevance, and result

demonstrability positively influence PU.Trust,innovation joy,

and facilitating conditions are significant as well as positive

influencing variables of PEOU. Note that the goodness of

fit of our RPA user acceptance model is highly satisfying

compared to other TAMs. Finally, concrete recommendations

for developing new RPA bots and improving existing RPA

bots are developed. Future work can focus on incorporating

additional external variables into the developed RPA user

acceptance model, testing our results with diverse user groups,

and modeling moderating variables, e.g., gender, age, or ex-

perience to deepen the understanding of RPA user acceptance.

APPENDIX

REFERENCES

[1] R. Agarwal and E. Karahanna, “Time Flies When You’re Having Fun:

Cognitive Absorption and Beliefs about Information Technology Usage,”

MIS Q, vol. 24, no. 4, pp. 665–694, 2000.

[2] R. Agarwal and J. Prasad, “The Role of Innovation Characteristics and

Perceived Voluntariness in the Acceptance of Information Technologies,”

Decis Sci, vol. 28, no. 3, pp. 557–582, 1997.

[3] S. Aguirre and A. Rodriguez, “Automation of a Business Process Using

Robotic Process Automation (RPA): A Case Study,” in Workshop on

Eng Appl. Springer, 2017, pp. 65–71.

[4] M. Al-Emran, V. Mezhuyev, and A. Kamaludin, “PLS-SEM in Informa-

tion Systems Research: A Comprehensive Methodological Reference,”

in Int Conf on Advanced Intelligent Systems and Informatics. Springer,

Cham, 2018, pp. 644–653.

[5] J. C. Anderson and D. W. Gerbing, “Structural equation modeling in

practice: A review and recommended two-step approach,” Psychol Bull,

vol. 103, no. 3, pp. 411–423, 1988.

[6] A. Asatiani and E. Penttinen, “Turning robotic process automation into

commercial success - Case OpusCapita,” J Inf Technol Teach Cases,

vol. 6, no. 2, pp. 67–74, 2016.

[7] F. G. Barbeite and E. M. Weiss, “Computer self-efficacy and anxiety

scales for an Internet sample: testing measurement equivalence of

existing measures and development of new scales,” Comput Human

Behav, vol. 20, no. 1, pp. 1–15, 2004.

[8] J. Becker, C. V. Uthmann, M. zur Muehlen, and M. Rosemann, “Identi-

fying the workflow potential of business processes,” in HICCS. IEEE,

1999, pp. 10–20.

[9] S. Choenni, R. Bakker, and W. Baets, “On the Evaluation of Workflow

Systems in Business Processes,” Electron J Inf Sys Eval, vol. 6, no. 2,

pp. 33–44, 2003.

[10] D. R. Compeau and C. A. Higgins, “Application of Social Cognitive

Theory to Training for Computer Skills,” Inf Syst Res, vol. 6, no. 2, pp.

118–143, 1995.

[11] F. D. Davis, “A technology acceptance model for empirically testing new

end-user information systems: Theory and results,” Ph.D. dissertation,

Massachussetts Institute of Technology, 1985.

[12] ——, “Perceived Usefulness, Perceived Ease of Use, and User Accep-

tance of Information Technology,” MIS Q, vol. 13, no. 3, pp. 319–339,

1989.

[13] W. B. Dodds, K. B. Monroe, and D. Grewal, “Effects of Price, Brand,

and Store Information on Buyers’ Product Evaluations,” J Mark Res,

vol. 28, no. 3, pp. 307–319, 1991.

[14] D. Fernandez and A. Aman, “Impacts of Robotic Process Automation on

Global Accounting Services,” Asian J Account Gov, vol. 9, pp. 123–132,

2018.

[15] M. Fishbein and I. Ajzen, “Belief, attitude, intention, and behavior: An

introduction to theory and research,” 1977.

[16] C. R. Franz and D. Robey, “Organizational context, user involvement,

and the usefulness of information systems,” Decis Sci, vol. 17, no. 3,

pp. 329–356, 1986.

[17] Gartner, “Predicts 2020: RPA Renaissance Driven by Morphing Offer-

ings and Zeal for Operational Excellence,” Tech. Rep., 2020.

[18] J. F. Hair, W. C. Black, B. J. Babin, and R. E. Anderson, “Multivariate

Data Analysis,” Essex Pearson Educ Ltd, 2014.

[19] J. F. Hair, C. M. Ringle, and M. Sarstedt, “PLS-SEM: Indeed a Silver

Bullet,” J Mark Theory Pract, vol. 19, no. 2, pp. 139–152, 2011.

[20] J. F. Hair, M. Sarstedt, C. M. Ringle, and J. A. Mena, “An assessment of

the use of partial least squares structural equation modeling in marketing

research,” J Acad of Mark Sci, vol. 40, no. 3, pp. 414–433, 2012.

[21] J. Hair, C. L. Hollingsworth, A. B. Randolph, and A. Y. L. Chong, “An

updated and expanded assessment of PLS-SEM in information systems

research,” Ind Manag Data Syst, vol. 117, no. 3, pp. 442–458, 2017.

[22] P. Hallikainen, R. Bekkhus, and S. L. Pan, “How OpusCapita Used

Internal RPA Capabilities to Offer Services to Clients,” MIS Q Exec,

vol. 17, no. 1, pp. 41–52, 2018.

[23] J. Hartwick and H. Barki, “Explaining the Role of User Participation

in Information System Use,” Manage Sci, vol. 40, no. 4, pp. 440–465,

1994.

[24] J. Henseler, G. Hubona, and P. A. Ray, “Using PLS path modeling in

new technology research: updated guidelines,” Ind Manag Data Syst,

vol. 116, no. 1, pp. 2–20, 2016.

TABLE VI

VARIABLE,TOKEN,QUESTION ITEM,AND REFERENCE.

Variable Token Question item Reference

General questions

GEN Which gender are you? [49], [50]

AGE How old are you? [49], [50]

USE How long have you been using RPA bots? [11]

Social influence

INF1 People who influence my behavior think that I should use RPA bots. [47]–[50]

INF2 People who are important to me recommend me to use RPA bots. [47]–[50]

INF3 The management has advised me to use RPA bots. [47], [49]

Job relevance

REL1 In my job, the usage of RPA bots is important. [47], [48]

REL2 In my job, the usage of RPA bots is relevant. [47], [48]

REL3 My workload could hardly be handled without RPA bots. [47] (adapted)

Result demonstrability

RES1 I have no difficulty telling others about the results of using RPA bots. [47], [48]

RES2 The results of using RPA bots are comprehensible to me. [47], [48]

RES3 I have no difficulty explaining why using RPA bots may or may not be beneficial. [47], [48]

User involvement

INV1 I (or the user group) was involved in the explanation and clarification of the automation needs and objectives. [16] (adapted)

INV2 I (or the user group) was heavily involved in testing RPA bots. [16] (adapted)

INV3 Prior to the implementation, I was informed about new possibilities the automation creates for me. self-developed

Trust

TRST1 I trust the RPA bots to behave in a privacy-protecting and tamper-proof manner. [35], [42] (adapted)

TRST2 I accept the results of RPA bots without subsequent checking. self-developed

TRST3 I trust the RPA bots to perform as designed and deliver the desired results without malfunctions. [35] (adapted)

Computer self-efficacy

CSE1 I can always manage to solve difficult computer problems if I try hard enough. [25] (adapted)

CSE2 I am confident that I can handle unexpected error messages from the computer efficiently. [25] (adapted)

CSE3 I feel confident troubleshooting computer problems. [7]

Innovation joy

JOY1 In general, I am not hesitant to try out new technologies. [1] (adapted)

JOY2 I feel confident and relaxed while trying out new technologies. [31]

JOY3 I am not afraid that my job will be done by computers in the near future. self-developed

Facilitating conditions

FAC1 I have the resources necessary to use RPA bots. [49], [50]

FAC2 I have the knowledge necessary to use RPA bots. [49], [50]

FAC3 A specific person (or group) is available for assistance when I have difficulties using RPA bots. [49], [50]

Hedonic motivation

HED1 Using RPA bots is fun. [50]

HED2 Using RPA bots is enjoyable. [50]

HED3 Using RPA bots is entertaining. [50]

Perceived usefulness

PU1 Using RPA bots in my job increases my productivity. [11], [49], [50]

PU2 Using RPA bots enables me to accomplish tasks more quickly. [11], [49], [50]

PU3 Overall, I find RPA bots useful in my job. [11], [49], [50]

Perceived ease of use

PEOU1 I find it easy to get RPA bots to do what I want them to do. [11], [47], [48]

PEOU2 Learning to work with RPA bots is easy for me. [11], [49], [50]

PEOU3 Overall, I find RPA bots easy to use. [11], [47]–[50]

Behavioral intention

BI1 I intend to use RPA bots frequently. [47], [48] (adapted)

BI2 I will always try to use RPA bots if my task are suitable. [50] (adapted)

BI3 I will use RPA bots in the near future. [49] (adapted)

TABLE VII

MEAN,STANDARD DEVIATION, VARIANCE INFLATION FACTOR, FACTOR LOADING, CRONBACH’SALPHA, COMPOSITE RELIABILITY,AND AVERAGE

VARIANCE EXTRACTED MEASURES FOR EACH QUESTION ITEM.

Variable Question item Mean Standard deviation Variance Inflation Factor Factor Loading Cronbach’s Alpha Composite Reliability Average Variance Extracted

Social influence

INF1 5.560 1.023 1.788 0.839

0.789 0.874 0.699INF2 5.300 1.136 1.574 0.868

INF3 5.560 1.431 1.660 0.801

Job relevance

REL1 5.460 1.099 1.742 0.911

0.622 0.792 0.581REL2 5.400 1.217 1.643 0.862

REL3 4.220 1.540 1.088 0.413

Result Demonstrability

RES1 5.940 1.008 2.000 0.881

0.812 0.888 0.727RES2 5.760 0.971 1.548 0.785

RES3 5.880 0.952 2.052 0.888

User involvement

INV1 5.680 1.434 1.290 0.734

0.519 0.756 0.509INV2 5.680 1.489 1.254 0.743

INV3 5.680 1.103 1.050 0.660

Trust

TRST1 5.900 1.063 1.141 0.695

0.619 0.797 0.569TRST2 4.480 1.578 1.308 0.714

TRST3 5.240 1.379 1.404 0.844

Computer self-efficacy

CSE1 4.920 1.339 1.830 0.769

0.801 0.873 0.697CSE2 4.900 0.964 1.542 0.897

CSE3 4.920 1.036 1.921 0.834

Innovation joy

JOY1 6.120 0.972 2.095 0.909

0.721 0.843 0.649JOY2 5.580 1.079 1.944 0.872

JOY3 5.300 1.628 1.186 0.600

Facilitating conditions

FAC1 5.940 0.988 1.204 0.784

0.523 0.744 0.511FAC2 5.940 0.925 1.275 0.861

FAC3 6.080 0.997 1.076 0.422

Hedonic motivation

HED1 5.460 1.135 1.997 0.891

0.651 0.807 0.605HED2 5.660 0.972 2.004 0.925

HED3 4.500 1.404 1.062 0.406

Perceived usefulness

PU1 5.900 0.943 2.178 0.887

0.830 0.898 0.746PU2 5.960 1.131 1.845 0.842

PU3 5.980 0.948 1.818 0.861

Perceived ease of use

PEOU1 5.480 1.170 2.041 0.831

0.846 0.906 0.764PEOU2 5.820 1.161 1.937 0.857

PEOU3 5.740 1.110 2.805 0.931

Behavioral intention

BI1 6.100 0.964 1.549 0.836

0.774 0.869 0.688BI2 6.000 1.149 1.574 0.824

BI3 6.060 0.947 1.657 0.829

TABLE VIII

FORNELL-LARCKER CRITERION.

BI CSE FAC HED JOY REL PEOU PU RES INF TRST INV

BI 0.83

CSE 0.50 0.84

FAC 0.62 0.49 0.72

HED 0.60 0.60 0.57 0.78

JOY 0.51 0.63 0.47 0.43 0.81

REL 0.60 0.35 0.46 0.50 0.33 0.76

PEOU 0.53 0.51 0.69 0.50 0.57 0.39 0.87

PU 0.81 0.43 0.58 0.67 0.61 0.58 0.54 0.86

RES 0.68 0.53 0.74 0.58 0.51 0.52 0.55 0.64 0.85

INF 0.44 0.12 0.41 0.39 0.22 0.51 0.32 0.58 0.38 0.84

TRST 0.47 0.43 0.50 0.45 0.35 0.36 0.44 0.49 0.49 0.36 0.75

INV 0.50 0.33 0.43 0.34 0.37 0.26 0.32 0.37 0.60 0.22 0.50 0.71

TABLE IX

CROSS LOADINGS.

INF REL RES INV TRST CSE JOY FAC HED PU PEOU BI

INF1 0.84 0.32 0.33 0.19 0.20 0.04 0.08 0.37 0.30 0.43 0.24 0.44

INF2 0.87 0.55 0.34 0.13 0.35 0.23 0.28 0.37 0.38 0.58 0.34 0.40

INF3 0.80 0.35 0.28 0.25 0.34 -0.02 0.16 0.28 0.28 0.39 0.20 0.25

REL1 0.50 0.91 0.52 0.33 0.36 0.30 0.29 0.42 0.46 0.55 0.38 0.62

REL2 0.43 0.86 0.42 0.11 0.32 0.29 0.36 0.47 0.42 0.49 0.36 0.45

REL3 0.11 0.41 0.12 0.13 0.01 0.26 -0.04 -0.01 0.20 0.18 0.03 0.24

RES1 0.43 0.52 0.88 0.47 0.39 0.48 0.52 0.64 0.43 0.58 0.50 0.65

RES2 0.36 0.41 0.79 0.48 0.45 0.30 0.23 0.46 0.46 0.45 0.44 0.43

RES3 0.20 0.39 0.89 0.59 0.42 0.54 0.52 0.76 0.60 0.59 0.47 0.63

INV1 0.14 0.23 0.43 0.73 0.36 0.22 0.27 0.27 0.17 0.23 0.13 0.38

INV2 -0.09 0.20 0.42 0.74 0.23 0.35 0.38 0.17 0.31 0.28 0.23 0.32

INV3 0.41 0.13 0.43 0.66 0.48 0.13 0.14 0.46 0.23 0.29 0.29 0.38

TRST1 0.24 0.24 0.40 0.42 0.70 0.35 0.42 0.26 0.27 0.37 0.26 0.40

TRST2 0.25 0.20 0.34 0.37 0.71 0.35 0.14 0.40 0.46 0.28 0.35 0.24

TRST3 0.32 0.34 0.36 0.35 0.84 0.29 0.21 0.47 0.32 0.43 0.40 0.41

CSE1 -0.09 0.12 0.27 0.27 0.31 0.77 0.47 0.28 0.50 0.33 0.23 0.40

CSE2 0.24 0.45 0.53 0.27 0.44 0.90 0.57 0.50 0.57 0.40 0.56 0.45

CSE3 0.01 0.18 0.43 0.29 0.28 0.83 0.53 0.39 0.41 0.33 0.36 0.41

JOY1 0.22 0.28 0.53 0.36 0.30 0.50 0.91 0.45 0.39 0.57 0.54 0.48

JOY2 0.20 0.33 0.39 0.20 0.19 0.66 0.87 0.43 0.40 0.50 0.50 0.35

JOY3 0.09 0.17 0.29 0.38 0.40 0.34 0.60 0.23 0.22 0.42 0.30 0.44

FAC1 0.33 0.25 0.51 0.25 0.46 0.33 0.39 0.78 0.37 0.49 0.53 0.46

FAC2 0.28 0.36 0.66 0.38 0.31 0.48 0.40 0.86 0.52 0.40 0.62 0.51

FAC3 0.57 0.62 0.44 0.38 0.40 0.19 0.16 0.42 0.35 0.43 0.21 0.40

HED1 0.27 0.32 0.49 0.22 0.24 0.50 0.27 0.40 0.89 0.48 0.40 0.49

HED2 0.42 0.56 0.61 0.33 0.49 0.59 0.57 0.65 0.93 0.75 0.58 0.61

HED3 0.15 0.20 0.08 0.33 0.37 0.19 -0.08 0.07 0.41 0.12 -0.03 0.19

PU1 0.57 0.49 0.60 0.35 0.36 0.28 0.50 0.45 0.51 0.89 0.40 0.68

PU2 0.51 0.50 0.41 0.28 0.55 0.44 0.58 0.50 0.60 0.84 0.51 0.64

PU3 0.42 0.51 0.63 0.33 0.37 0.39 0.52 0.54 0.62 0.86 0.49 0.76

PEOU1 0.31 0.32 0.32 0.24 0.32 0.27 0.43 0.45 0.35 0.41 0.83 0.35

PEOU2 0.23 0.35 0.50 0.30 0.37 0.51 0.50 0.62 0.46 0.43 0.86 0.46

PEOU3 0.31 0.35 0.59 0.29 0.45 0.51 0.55 0.70 0.49 0.56 0.93 0.54

BI1 0.31 0.53 0.45 0.36 0.40 0.44 0.48 0.42 0.52 0.72 0.43 0.84

BI2 0.37 0.42 0.60 0.37 0.47 0.42 0.37 0.67 0.53 0.64 0.55 0.82

BI3 0.42 0.56 0.65 0.53 0.30 0.38 0.43 0.47 0.45 0.64 0.32 0.83

[25] M. C. Howard, “Creation of a Computer Self-Ffficacy Measure: Anal-

ysis of Internal Consistency, Psychometric Properties, and Validity,”

Cyberpsychol Behav Soc Netw, vol. 17, no. 10, pp. 677–681, 2014.

[26] M. Lacity and L. Willcocks, “A New Approach to Automating Services,”

MIT Sloan Manag Rev, 2017.

[27] M. Lacity, L. Willcocks, and A. Craig, “Robotic Process Automation

at Xchanging,” Outsourcing Unit Work Res Pap Ser, vol. 15, no. 3, pp.

1–26, 2015.

[28] ——, “Robotic Process Automation: Mature Capabilities in the Energy

Sector,” Outsourcing Unit Work Pap Ser, pp. 1–19, 2015.

[29] ——, “Robotizing Global Financial Shared Services at Royal DSM,”

Outsourcing Unit Work Res Pap Ser, vol. 16, no. 2, pp. 1–26, 2016.

[30] ——, “Service Automation: Cognitive Virtual Agents at SEB Bank,”

London Sch Econ Polit Sci, pp. 1–29, 2017.

[31] D. Lester, B. Yang, and S. James, “A Short Computer Anxiety Scale,”

Percept Mot Skills, vol. 100, no. 3, pp. 964–968, 2005.

[32] J. Li, J. Wang, S. Wangh, and Y. Zhou, “Mobile Payment with Alipay:

An Application of Extended Technology Acceptance Model,” IEEE

Access, vol. 7, pp. 50 380–50 387, 2019.

[33] M. Limayem, S. G. Hirt, and C. M. K. Cheung, “How Habit Limits

the Predictive Power of Intention: The Case of Information Systems

Continuance,” MIS Q, vol. 31, no. 4, pp. 705–737, 2007.

[34] J.-B. Lohm¨

oller, Latent Variable Path Modeling with Partial Least

Squares. Physica, Heidlberg, 1989.

[35] C. Martins, T. Oliveira, and A. Popoviˇ

c, “Understanding the Internet

banking adoption: A unified theory of acceptance and use of technology

and perceived risk application,” Int J Inf Manage, vol. 34, no. 1, pp. 1–

13, 2014.

[36] V. Mezhuyev, M. Al-Emran, M. A. Ismail, L. Benedicenti, and D. A.

Chandran, “The Acceptance of Search-Based Software Engineering

Techniques: An Empirical Evaluation Using the Technology Acceptance

Model,” IEEE Access, vol. 7, pp. 101 073–101 085, 2019.

[37] G. C. Moore and I. Benbasat, “Development of an Instrument to Measure

the Perceptions of Adopting an Information Technology Innovation,” Inf

Syst Res, vol. 2, no. 3, pp. 192–222, 1991.

[38] B. Mutschler and M. Reichert, “A Survey on Evaluation Factors for

Business Process Management Technology,” pp. 1–21, 2006.

[39] B. Mutschler, M. Reichert, and J. Bumiller, “Unleashing the Effec-

tiveness of Process-Oriented Information Systems: Problem Analysis,

Critical Success Factors, and Implications,” IEEE Trans Syst Man

Cybern, vol. 38, no. 3, pp. 280–291, 2008.

[40] J. Nunnally and I. Bernstein, Psychometric Theory. McGraw-Hill

Education, New York City, 1967.

[41] A. Parkes, “Critical Success Factors in Workflow Implementation,” in

6th Pacific Asia Conf on Inf Sy, 2002, pp. 363–380.

[42] P. A. Pavlou, “Consumer Acceptance of Electronic Commerce: Inte-

grating Trust and Risk with the Technology Acceptance Model,” Int J

Electron Commer, vol. 7, no. 3, pp. 101–134, 2003.

[43] M. Sarstedt, C. M. Ringle, and J. F. Hair, “Partial Least Squares

Structural Equation Modeling,” Handb Mark Res, vol. 26, pp. 1–40,

2017.

[44] M. R. Simonson, M. Maurer, M. Montag-Torardi, and M. Whitaker,

“Development of a Standardized Test of Computer Literacy and a

Computer Anxiety Index,” J Educ Comput Res, vol. 3, no. 2, pp. 231–

247, 1987.

[45] M. Turner, B. Kitchenham, P. Brereton, S. Charters, and D. Budgen,

“Does the technology acceptance model predict actual use? A systematic

literature review,” Inf Softw Technol, vol. 52, no. 5, pp. 463–479, 2010.

[46] V. Venkatesh, “Determinants of Perceived Ease of Use: Integrating Con-

trol, Intrinsic Motivation, and Emotion into the Technology Acceptance

Model,” Inf Sys Res, vol. 11, no. 4, pp. 342–365, 2000.

[47] V. Venkatesh and H. Bala, “Technology Acceptance Model 3 and a

Research Agenda on Interventions,” Decis Sci, vol. 39, no. 2, pp. 273–

315, 2008.

[48] V. Venkatesh and F. D. Davis, “A Theoretical Extension of the Technol-

ogy Acceptance Model: Four Longitudinal Field Studies,” Manage Sci,

vol. 46, no. 2, pp. 186–204, 2000.

[49] V. Venkatesh, M. G. Morris, G. B. Davis, and F. D. Davis, “User

Acceptance of Information Technology: Toward a Unified View,” MIS

Q, vol. 27, no. 3, pp. 425–478, 2003.

[50] V. Venkatesh, J. Y. L. Thong, and X. Xu, “Consumer Acceptance

and Use of Information Technology: Extending the Unified Theory of

Acceptance and Use of Technology,” MIS Q, vol. 36, no. 1, pp. 157–178,

2012.

[51] J. Webster and J. J. Martocchio, “Microcomputer Playfulness: Develop-

ment of a Measure with Workplace Implications,” MIS Q, vol. 16, no. 2,

pp. 201–224, 1992.

[52] L. Willcocks and M. Lacity, “Robotic Process Automation: The Next

Transformation Lever for Shared Services,” Outsourcing Unit Work Res

Pap Ser, vol. 15, no. 7, pp. 1–35, 2015.

[53] ——, “Robotic Process Automation at Telef´

onica O2,” MIS Q Exec,

vol. 15, no. 1, pp. 21–35, 2016.

[54] J. M. Wooldridge, Introductory econometrics: A modern approach.

Nelson Education, 2016.

[55] H.-Y. Yoon, “User acceptance of mobile library applications in academic

libraries: an application of the technology acceptance model,” J Acad

Librariansh, vol. 42, no. 6, pp. 687–693, 2016.