Information Systems Frontiers

https://doi.org/10.1007/s10796-021-10219-4

Facilitators and Barriers of Artificial Intelligence Adoption in Business

– Insights from Opinions Using Big Data Analytics
Arpan Kumar Kar1   · Amit Kumar Kushwaha1

Accepted: 26 October 2021

© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2021

Abstract
Data-driven predictions have become an inseparable part of business decisions. Artificial Intelligence (AI) has started help-
ing the product and support teams perform more accurate experiments in various business settings. This study proposes a
framework for businesses based on inductive learnings related to success and barriers shared on social media platforms. Our
goal is to analyse the signals emerging from these conversational opinions from the early adoption of AI, with a focus towards
facilitators and barriers faced by teams. Factors like efficiency, innovation, business research, product novelty, manual inter-
vention, adaptability, emotion, support, personal growth, experiential learning, fear of failure and fear of upgradation have
been identified based on an exploratory study and then a confirmatory study. We present the learnings through a roadmap
for practitioners. This study contributes to the IS literature by delineating AI as a determinant of success and introduces a
lot of organizational factors into the model.

Keywords  Information systems adoption · Artificial intelligence · Machine learning · Chatbots · Marketing automation ·
Big data analytics

1 Introduction the adoption of AI in the business domain has started to

gain a great deal of attention from information systems (IS)
The significance of innovation and adoption has been widely researchers (Brynjolfsson & McAfee, 2017; Vimalkumar
accepted in organizational theories (Janssen et al., 2004; Van et al., 2021; Sharma et al., 2019; Dai & Singh, 2020; Bal-
de Ven, 1986; Woodman et al., 1993) for business effec- akrishnan et al., 2021; Kushwaha et al., 2021d; Kumar et al.,
tiveness and is considered an excellent aid for a business 2021).
organization to succeed (Yuan & Woodman, 2010; West & AI has traditionally been used to automate tasks through
Farr, 1989) in a competitive environment. Artificial Intel- robotics. Still, recent advances in computational power and
ligence (AI) (Meinhart, 1966) has taken the computer pro- enormous growth in data volume have allowed the business
grams, developed with proven mathematical theories, and to use AI in real-time decisions and projects. Pioneering
enabled the business houses in the digital domain to make research around AI usability was first proven in the 1950 s
real-time decisions. Today AI-driven projects and initiatives (Newell et al., 1959; Newell & Simon, 1956) in the manage-
have become part of product, operation, and customer teams’ ment domain. However, with a few initial kinds of research,
objectives and critical results (OKR). With these changes, the focus in exploring AI as a domain became more techni-
cal, and the management perspective was slowly liquidated
around the 1960 s (Cariani, 2010). Hence, the IS domain
Arpan Kumar Kar and Amit Kumar Kushwaha contributed equally research scholars started adopting a more contingency view
to this work. of research focused on the operational inefficiencies that can
* Arpan Kumar Kar be catered to using AI applications. Consequently, AI began
arpan_kar@yahoo.co.in attracting more computer science and operation scholars,
Amit Kumar Kushwaha while management and organization studies focused more on
kushwaha.amitkumar@gmail.com human aspects (Rahwan et al., 2019; Simon, 1987).
According to prior research work(von Krogh, 2018), the
1
Department of Management Studies, Indian Institute fundamental research topic is identifying how AI relates
of Technology Delhi, New Delhi 110016, India

13
Vol.:(0123456789)
 Information Systems Frontiers

well with overall organization stakeholders, primarily the to a state-dependent response issue, which might later influ-
employees and business stakeholders. Researchers (Metcalf ence future responses (de Jong et al., 2012). On the other
et al., 2019) have proven that without AI, it can amplify an hand, with the growing availability of text data in the form
employee’s capability to make informed decisions in real- of unstructured reviews, answers and opinions give the
time, which directly impacts customer experience and helps independence to various stakeholders to express themselves
the business houses grow. On the other hand, researchers independently without being confined to the structure of the
(Jarrahi, 2018) have also proved that employees having good survey on pre-selected items with a forced scale rating. They
knowledge, confidence, and belief in AI solutions will posi- speak their experience in the manner they want, like using
tively impact business (Bader & Kaiser, 2019). This leads “great” in conjunction with the noun “experience” or even
us to the first research gap to be addressed through the cur- “disappointment.”
rent study highlighted by previous researchers (Morikawa, The second unique point of the current study is the
2017; Brock & Von Wangenheim, 2019), which states that approach taken to analyze these unstructured data. The
the employees and managers should have good knowledge of research undertaken in the present paper is motivated by
AI for the successful operation of the organization’s business the literature gap in assessing AI’s adoption from an induc-
management. Therefore, this study highlights the uniqueness tive lens (Kar & Dwivedi, 2020; Kushwaha & Kar, 2021b;
of AI in terms of emotions, support, growth, and experiential Ghosh & Sanyal, 2021). With big data Kushwaha et al.,
learnings from stakeholders’ lenses. 2021a; Mohamed Ridhwan & Hargreaves, 2021; Kushwaha
The second research gap maps to the work that authors & Kar, 2020b) capturing a lot of veracity and volume, it
(Gunasekaran & Ngai, 2012) have highlighted: the need to becomes crucial that IS researchers focus more on capturing
develop AI-based models (Sharma et al., 2021; Kushwaha the unbiased signals first and concentrate on answering “why
et al., 2021a) and innovations driving more day-to-day tasks this behavior” with theoretical reasonings. This helps studies
businesses. As part of digital innovations by synthesizing establish knowledge among the relationships of factors more
and converting existing data to knowledge. Hence this contextually (Grover et al., 2017).
research presents various AI solutions from innovations, The sentiment and the subjectivity scores act as a proxy
efficiency, novelty, and less manual interventions (Kushwaha to first measure emotion and second the rationality of con-
& Kar, 2021a; Arjun et al., 2021). The last research gap tinued behavioral intention of experience of AI adoption
(Haenlein & Kaplan, 2019) addressed through the current to measure the overall effect of adoption factors. Overall,
study is the co-existence of multiple stakeholders with an emotion and rationality act as a collective measure for AI’s
aligned interest and motivation in an AI ecosystem. Hence overall experience of adoption. We use the social media ana-
the recent research explicitly analyses the AI adoption facets lytics (SMA) framework from the big-data analytics tool
from the lens of various stakeholders. kit applied to the conversations and experiences shared on
The above three research gaps identified highly motivates social media platforms (SMP) like Twitter by stakeholders in
use to perform the current study with the current research a business organization (business owners and employees as
is unique in two aspects: First, it draws all its inferences to AI application owners) by mining the original tweets posted
address the above three research gaps identified, using text- by them. We further use the opinion mining approaches
based analysis of user-generated content (UGC) in the form like topic modeling and sentiment mining. The below-listed
of experiences, views, and opinions shared by various stake- research questions guide our research through this paper.
holders which have received considerable attention on social The first two focus on the business owners as stakeholders,
media platforms (SMP). This approach has been widely and the last two focus on employees as stakeholders. The
accepted, used, and received traction with publication in article follows the editorial note’s research direction towards
top journals (Berger et al., 2010; Zhao et al., 2013; Ludwig inductive theory building in big data-based research in infor-
et al., 2013; Archak et al., 2011; Nair et al., 2021; Kushwaha mation systems (Kar & Dwivedi, 2020).
& Kar, 2020a). The approach addresses the shortcomings
of the conventional data collected on the fixed-point rating RQ1: How do facets of AI usage get shared by stakehold-
scale approach. The conventions approach has issues like ers while adopting AI through their opinions?
yea-saying, nay-saying, and scale use tendencies that chal- RQ2: What stakeholders emerge while using AI solutions
lenge inferences (Büschken et al., 2013). Some respondents who participate in these discussions?
might have the expertise to provide meaningful inputs, while RQ3: To what extent are the shared experiences represent
others do not, and the rest provide their independent evalu- the co-existence and interaction of various stakeholders
ations. This tends to create a halo in their responses (Gal & in the AI ecosystem?
Rucker, 2011).
Respondents also tend to substitute answers to the ques- The rest of the paper is structured into six sections. In
tions different from the survey, which exposes the responses the second section, we start with a synthesis of the available

13
Information Systems Frontiers

literature. The third section describes the research followed the outcome (Meske et  al., 2020; Dwivedi et  al., 2021;
for this research in detail. The fourth section discusses the Kushwaha & Kar, 2020a).
outcomes of the data signal processing in detail. We further In an organizational setup, “AI managers” are the ones
move the signal grounding phase into existing theories to who own the responsibility of controlling and supervis-
form constructs, and these constructs are later statistically ing the algorithm, followed by AI developers responsible
tested through the econometrics model. The fifth section for improvements in the AI systems. Independent AI users
discusses the practice and managerial implications of the receive the AI solutions and apply logical and domain rea-
findings. In the sixth section, we conclude the learnings from soning (for example, doctors and business heads) to either
the current research. In the seventh section, we discuss the go with the solution or overrule with manual decisions.
limitations of the recent study and propose future research This establishes three primary stakeholder groups: AI
directions for IS scholars. managers(employees), AI developers (employees), and AI
users, who exchange and share the AI system’s responsibil-
ity. This approach makes them agents in the AI system with
2 Literature Review a high level of interaction in a dynamic business situation
(Schuetz & Venkatesh, 2020). Having stated that, there is
The literature reviewed in this manuscript were identified very little research examining AI adoption from an agentic
from the Scopus database, using the following keywords: setup as equal stakeholders, as highlighted by researchers
(TITLE (“Adoption”) OR TITLE (“Artificial Intelligence”) (Baird & Maruping, 2021). This motivates us to take the
AND TITLE (“Experiences” OR “Hindrances” OR “Resist- current study of AI adoption and experience from multiple
ance” OR “Limitation” OR “Challenge” OR “Barriers” stakeholders’ lenses.
OR “Shortcoming” OR “agenda.” The literature review
is divided into four sub-sections, starting with stakehold- 2.1.2 Stakeholders’ Experience
ers, first, getting defined, followed by their experience and
knowledge with AI, which maps to the first research ques- While we move away from the traditional TAM-based mod-
tion. The second subsection presents various facets of AI els to understand the stakeholders’ (defined in Section 2.1)
applications, consumption in business, and why revisiting experiences, at the same time, prior research on AI risks and
the adoption experience literature is essential, mapping to failures highlights that it is vital to build sufficient expertise
the second research question. The third section offers the and opinions on AI systems (Kushwaha et al., 2021b; Neogi
co-existence of stakeholders in an AI ecosystem, which et al., 2021) However, whenever a new IS is implemented
aligns with the last research question. The final section in an organization, various factors’ sentiments and experi-
briefly discusses the limitations of AI adoption in tactical ences are driven by multiple factors. These factors vary from
decision-making. one stakeholder to another depending on how they interact
with the IS solutions (AI in the current context). These fac-
2.1 Stakeholders tors also drive the degree and strength of the knowledge
built of these IS systems. Researchers (Kim & Kankanhalli,
2.1.1 Defining Stakeholders 2009) have further divided these factors under the bucket of
psychological characteristics and behavioral factors listed
We start this section by first reviewing the historical work in as control(intervention), efficacy, perceived value, fear, and
defining the stakeholders in AI systems. The conventional uncertainty.
IS lens has dominated the work of scholars understanding Employees learning from their experience using new
the IS adoption from an individual’s perception of how technology, researching, and investigating the failures are
technology is adopted (Kim & Kankanhalli, 2009; Ahuja keys to an individual employee’s performance, leading to
& Thatcher, 2005; Wang et al., 2013). These have primarily an organization’s success (Simon, 1991; Argote & Miron-
used the technology acceptance model (TAM) that posits an Spektor, 2011). However, whether and to what extent the
individual’s attitude and belief while accepting an IS sys- fear of upgrading by learning new technology, fear of fail-
tem driven by survey data collected from experts. According ure in experimentation, and experience lead to being a bar-
to researchers (Baird & Maruping, 2021; Kushwaha et al., rier in employees adopting new technologies is still open to
2020a; Dai & Singh 2020), next-generation technologies like research and debate (KC et al., 2013; Deichmann & Ende,
big data, AI does not work in a human sub-ordinate form. 2013; Rajendran & Sundarraj, 2021; Kushwaha et al., 2021c;
Instead, they work in a setup in which transfer of responsi- Balakrishnan et al., 2021). Some scholars’ debate is the fear
bilities occurs from AI systems to agents like employees to of upgrading themselves that the employees are not adapting
develop the model and business practitioners to implement to new technologies. In contrast, others emphasize that the

13
 Information Systems Frontiers

fear of failure slows down an employee’s experimenting and employee to perform poorly (Andersson & Pearson, 1999).
learning new technologies. This peer pressure might impact an employee’s overall emo-
However, past research has established valuable learning tion, psychologically driving an employee to adopt AI.
opportunities from the new techniques’ experiential phase On the other hand, the workplace is considered a fertile
and failures (McGrath, 1999; Ellis et al., 2014). A differ- ground for research for personal development. With com-
ent set of scholars have highlighted the stress of learning panies adopting AI, the overall research mindset is advo-
new techniques, and being a painful experience of adopting cated within employees, which might affect the employee’s
new practices and technologies threatens an employee’ posi- behavior and emotion (Drexler, 1977). The organizations’
tive image (Edmondson, 2004) in a team or organization, internal climate while adopting AI or any new technology
driving taking a defensive stance when it comes to adopt- to advocate learnings and experimentations is one of the
ing (Kushwaha et al., 2020a, b, c) new techniques or tech- variables that might influence an employee’s positive emo-
nologies (Taylor, 1991). Such reactions and feelings restrict tion towards adopting AI. The organizational climate created
employees from processing further information, knowledge, by the business owners has been established by previous
and critical reflection in case of a hurdle while adopting research (Mcilroy et al., 2007). The above literature syn-
new techniques and technologies. This, in turn, prevents thesis points out a lack of existent research on the role of
employees from learning and experimenting with new tech- an employee’s emotional intelligence within the controlled
nologies (Aguinis et al., 2013). This section of the literature environment of an organizational structure as one of the pre-
review establishes four barrier factors from an employee’s dictor variables when it comes to overall behavioral senti-
experience standpoint by adopting new technologies (in our ment towards adopting AI.
research, it will be AI): (1) fear of upgradation, (2) fear of
experimentation, (3) fear of failure and (4) fear of losing an 2.2 Facets of AI
image.
There are different sets of scholars who establish experi- 2.2.1 Applications of AI
ential learning when it comes to adopting new technologies.
These experiential learnings will lead to overall growth in Researchers (Baird & Maruping, 2021) have highlighted the
an employee’s knowledge and success, leading to the overall agentic nature of AI in a human and business ecosystem.
team and the firm’s success. Because these theories (Sitkin, This means that AI works in a setup in which transfer of
1992) (Ellis et al., 2014; Kolb, 2015) are focused on indi- responsibilities occurs from time to time between AI systems
vidual learnings, they disregard the social structure and the and employees to develop the model and business practi-
fear of maintaining a positive image in an employee social tioners to implement the outcome. Management researchers
network in an organization (Reynolds & Vince, 2004; Argote have defined this as an automation-augmentation approach
& Miron-Spektor, 2011; KC et al., 2013). However, experi- (Raisch & Krakowski, 2020). While owing to the unique
ential learning by adopting AI within a business organiza- nature of AI in the business setup (von Krogh, 2018), pure
tion’s social context has not been researched so far. To curb automation(complete AI) vs. augmentation(mix of human
this non-availability of existing literature around employees and AI) becomes a critical corner case to study the facets of
adopting technology to go through experiential learning in AI. For instance: advertisement placement logic, delivery
the social context of an organization, we base our assump- personnel routing, warehouse selection. In this perspective,
tion on the organization learning theories that social context the metrics used to assess AI investments’ return are efficient
is essential for explaining an employee’s behavior and its processing (Cave & ÓhÉigeartaigh, 2019). If the business
outcomes in the organization (Johns, 2001, 2006, 2017). owners choose an automation-augmentation approach, then
We feel a gap in IS’s literature, which the proposed cur- relatively repetitive tasks can be automated with AI solutions
rent research can contribute to. This section of the literature with supervision and monitoring of an employee (Brynjolfs-
review establishes one crucial factor from an employee’s son & McAfee, 2017; Daugherty & Wilson, 2018). This
experience standpoint: adopting new technologies (in our brings out the time saved at the employees’ end, which they
study, AI) as (1) experiential learning in a social context. can utilize in other work as one of the OKR to measure the
Business organizations continuously look for purchasing, team’s success using AI. However, automation-augmenta-
installing, and using state-of-the-art technologies that money tion requires the constant collaboration of humans (employ-
can purchase to stay competitive. AI is the most recent ees) with the AI solution (Dwivedi et al., 2021; Sharma &
technology on similar lines. However, research (Cambre & Sharma, 2019);. The employee is expected to do the course
Cook, 1985) shows that introducing new technologies can correction in the event of AI not producing intuitive results.
always provoke cognitive and emotional reactions. These More enormous business organizations are now adopting
emotions can highly manipulate the way an employee is AI in human resources (HR) management to acquire talent
adopting AI. Peer pressure of adopting AI might cause an (Stephan et al., 2017). They cannot afford to spend valuable

13
Information Systems Frontiers

time working on live projects to screen the candidates in the perform product research and innovation through AI tech-
preliminary rounds. The use case mentioned in the JP Mor- nology stack: (1) time, (2) cost, (3) popularity, and (4) nov-
gan Chase article took an entire year of the first automation- elty. However, this use-case and experiences of a business
augmentation phase before automating the candidate screen- owner are unavailable for organizations still to adopt AI.
ing phase through AI. This section of the literature review There is a considerable gap in the literature to fill through
establishes four driving factors to address a tactical problem a thoroughly tested framework. This research is intended to
through the utilization of AI: (1) efficiency, (2) effectiveness, fill this gap with the work proposed.
(3) fairness, and (4) consistency. However, this use-case and
experiences a business owner had before entirely automating 2.3 Co‑existence of Stakeholders and AI
are unavailable for organizations that still adopt AI. There
is a considerable gap in the literature to fill through a thor- Over time, the close collaboration of human and AI solu-
oughly tested framework on what business owners look for tions enables employees to rule mining and optimizes a
when investing in an AI technology stack. This research is mathematical rule that maximizes the objective KPI. Sub-
intended to fill this gap with the work proposed. sequently, if these AI-driven models are sufficiently robust
and are stress-tested for field experiments, they can indepen-
2.2.2 Consumption of AI dently automate the same task, letting them focus on more
valuable tasks. Hence a phase of automation-augmentation
Product innovation and research are another business domain closely followed by a course corrected phase of AI adoption,
for large-scale and established organizations and small and and implementation can eliminate time-consuming tasks
medium scale organizations, which can be crucial for adop- more efficiently and effectively (Lindebaum et al., 2019).
tion and AI application. These product innovations could be AI solutions or systems are not adaptive enough to rede-
in cloud innovation (Benlian et al., 2018) by transforming fine the KPIs and objectives, which means that an AI man-
existing systems using AI or even more dynamic advertise- ager must state the success criteria (Braga & Logan, 2017).
ments recommended to customers providing experiential Defining and setting KPIs and goals closely resonates with
marketing (Schmitt, 1999). Another real-life example of owning and taking responsibility for associated business
using AI in experiential marketing for product innovation opportunities and tasks. While organizations can extend
is implemented by Symrise, a major global business house accountability to AI solutions, leadership requires intention
in the fragrance domain. The firm initially started with an exclusive to humans only (Floridi, 2008). This would mean
automation-augmentation approach to generate innovative that a business owner or a manager can take responsibility
ideas. The firm’s business owners like mater perfumers only if there is a constant involvement with AI solutions to
started using AI-driven solutions to establish and identify do the course corrections. In the business use-cases of AI
the relations between ingredients of a perfume and customer discussed so far, with the Symrise setting, the objectives
demographics. Every time, the AI solution had to browse to remain part of fragrance owners. The same applies to the
1.7 million combinations of fragrance to establish the rela- HR representatives and managers in the talent acquisition
tionship, which is real time could have been possible with process of JP Morgan and Chase.
human intervention (Sheridan, 2004).
Eventually, Symrise adapted to an automated AI solu- 2.4 Limitations for the Adoption of AI in Tactical
tion to predict a fragrance combination’s likeability and Business Decision Making
started using the AI recommendations to accept or reject
the combinations. They reached the entire automated state In terms of complex managerial tasks that a business owner
only after a whole year’s interaction and course correction must perform, AI solutions can always give various answers
by master perfumers. Symrise’s system is matured to the and probabilities associated with each of these options.
extent that, on a real-time basis and depending on customer However, AI lacks to decide in a particular context. A busi-
demographics and choices made in the past on fragrances, ness owner or the manager must then use intuition to rec-
the AI solution (Kushwaha et al., 2021c; Balakrishnan et al., oncile the possibilities suggested by AI to take the final call
2021; Mohamed Ridhwan & Hargreaves, 2021; Neogi et al., (Brynjolfsson & McAfee, 2017). This reconciliation is done
2021) can perform a real-time prediction recommend the by mapping the suggestions by AI to real-life field reality.
scent by searching for all possible combinations. This has For instance: AI has automated the candidate selection pro-
helped Symrise gain more popularity and gain novelty cess; however, an HR partner still has to intervene at the end
through product research and innovation. They have also of the process for the final selection (Riley, 2018). The AI
significantly reduced the time and search costs simultane- test can not entirely capture subjective predictors like the
ously (Bergstein, 2019). This section of the literature review cultural fit or interpersonal characteristics. Another aspect
establishes four driving factors for a business owner to of the AI solutions that a business owner might not like is

13
 Information Systems Frontiers

domain-specific learning. This has been highlighted as a fair tweets posted by 56,302 users out of original sample which
limitation for the adoption of AI in decision-making (Seufert had tweets posted by actual users.
et al., 2020). This means that the AI solution must again go We now deployed the second control mechanism to
through the phase, with product and business owners spend- restrict the sample creation to business and application
ing hours again before the solution can be used in an entirely owners (employees) by filtering user accounts that posted
automated manner. Synthesizing the prior paragraph of the the tweets. The filtering mechanism had a list of designa-
literature review and the current one establishes: (1) setting tions like “business head,” “line head,” “vertical head,”
objectives manually while owning the outcome of AI, and “sales head,” “advertising head,” “customer support heads,”
(2) limiting AI to adapt to change in the domain as the poten- and “principal engineer,” “technical architects,” “principal
tial barriers owner might see before adopting AI. data scientists,” product managers.” This helped us to cap-
ture the unbiased UGC for business and application own-
ers (employees) from experiences shared in conversations.
After cleaning the data for unnecessary text (like emoticons,
3 Research Methodology
URLs only, and non-English tweets). The resultant and the
final sample for the analysis was reduced to 1,56,345 tweets
The proposed study big-data study captures the unbiased
posted by 46,302 users.
conversational experience signals captured in UGC on both
the stakeholders (business and application). This approach
3.2 Content Analysis
solves any inherent bias captured while following an expert
opinion, interview, or survey-based methods. Twitter
The content analysis section consists of four separate and
(Grover et al., 2019) is one such microblogging SMP that
unique sections. We first start with each sample tweet’s sen-
allows business owners, employees, and application owners
timent and subjectivity score calculation. We deployed the
of all scale organizations to share their experience of adopt-
sci-kit learn library on Python for these scores’ calculation,
ing AI openly. We have validated the constructs using multi-
which helps execute the input tweets. As part of the second
variate regression analysis on overall polarity and emotional
step of content analysis, we now move to the next phase
scores of comments shared by business owners and applica-
of tweets summarization using Long short-term memory
tion owners (employees) of AI solutions for generalization
(LSTM) neural networks (NN) based topic modeling and
and robustness.
compared to results to conventional Latent Dirichlet Analy-
sis (LDA) topic modeling (Llewellyn et al., 2015) on topics
3.1 Data Collection coherency. The advantage of an LSTM NN architecture is
that it can learn and memorize the context and the frequency
The proposed study collected data (tweets) from SMP Twit- in natural language. Using LSTM NN topic modeling helps
ter for fourteen months, from 1st July 2019 till 31st August us learn about the topics and themes of emerging mathemati-
2020. Initially, a sample of 10,000 tweets was downloaded cally by controlling the number of words. Basis the themes
to understand the most trending and top hashtags (#) used that emerged, we identified the topics that best represent
by the business organizations to share views and experiences various aspects identified through the review of prior work.
using AI solutions. This exercise gave us ten top trending As part of the following steps, we try to search for the same
hashtags. Depending on the volume and frequency of the traces in available theories.
hashtags used, we shortlisted the top seven hashtags, finally Once the factors and the topics as an outcome of the topic
used them to build the analysis sample. The final list of modeling are established, we move on to the third step of
hashtags used were: [#Artificialintelligence, #Virtualreality, building a network diagram of words. A network diagram
#AI, #AIproducts, #AISolutions, #Chatbots, #industry4.0]. helps us to understand the co-occurrences of the words
This exercise gave us a sample of 3,54,782 tweets posted together in a visual representation. A network diagram bor-
by 60,142 users by executing the codes on the python tool rows the understanding of the network sciences (Rathore
using the streaming API of Twitter. As part of the control et al., 2017; Barabási, 2013; Börner et al., 2007) on word
mechanism, we deployed a thoughtful process to make that entities connected by edges. These edges help us understand
the tweets collected in the sample are neither marketing the degree of relationship, and the number of edges helps
tweets nor posted by a bot to increase the volume. By creat- us learn the community of co-occurrences of these entities
ing a control key of the pattern of tweets posted at the hour defining a group. This community formation within the net-
of the day, day of the week, many followers (greater than 1 work diagram helps us validate the correlation among the
Mn followers), a huge number of followings (greater than 1 themes and surrounding keywords that describe the con-
Mn following), accounts from which tweets are posted, any structs. We repeat this exercise overall, followed by a more
accounts tagged (@’s), we created a sample of size 2,82,543 focused positive and negative sentiment level.

13
Information Systems Frontiers

Once we have the community of the co-occurrences of variable captures the contextual (AI) usage experiential or
the topic words within each sentiment of positive and nega- satisfaction score (Asuncion & Lam, 1995; Venkatesh &
tives, we now take these keywords and try to find the cluster Speier, 1999; Venkatesh & Davis, 2000; Venkatesh & Bala,
of words spoken by the other members in the same respec- 2008). Empirically, we test the relationships built on sig-
tive group of sentiments: positive and negative, respectively nal grounding in existing theories driving the AI adoption
(Gargiulo et al., 2020). This helps us widen the understand- with these multivariate models. We conclude our analysis
ing of the comments and views shared by all the users who methodology through inferential analysis of four multivari-
fall in the positive sentiment or negative sentiments. An ate regression models, lasso regression: sentiment score and
explicit clustering exercise achieves this based on semantic emotion score, ridge regression: sentiment score, and emo-
similarity (Kushwaha et al., 2020b) of the words used by the tion score, to test the directional relationship of these factors.
rest of the members in the same group. A more detailed set The flow of these blocks of the methodology is represented
of steps are given below: in Fig. 1 below.

(a) We first translate the tweet words within each group

into a vector representation 4 Findings
(b) A k-means clustering exercise is performed on the vec-
tor representation within each sentiment group To better represent the learnings from the research findings,
(c) We made sure that the keywords do not end up forming this section is divided into three subsections. We first report
edges with the words within explicit similarity: instead, the exploratory results, topic modeling, and semantic simi-
the clusters should be formed with implicit connections larity clustering findings, and the last multivariate regression
basis the words identified through LSTM topic mod- analysis section concludes.
eling.
(d) We now run a principal component analysis (PCA) for 4.1 Signal‑capturing Through Visual Inferences
plotting the top 60 words in each cluster closest to the
respective centroids to a 2-dimension space. This helps We start with visual representations of the positive and nega-
to find the critical topics in each cluster tive conversations around AI adoption represented in Fig. 2.
(e) We plot the top 60 words in each cluster closest to Closely analyzing the positive and negative conversations,
the respective centroids to a 2-dimension space using we can visually extract some of the leading experiential key-
t-SNE is the respective sentiment group words. For instance, in the positive conversations, some of
the keywords used are: “best,” “solution,” “smart,” “boost,”
3.3 Inferential Analysis for Model Validation “customer,” “real-time.” These keywords visually correlate
to the theoretical assumptions (Davis, 1989; Al-Gahtani
Once we establish the co-occurrence and form clusters with & King, 1999) we made during the literature review sec-
semantic similarity within the positive and negative senti- tion. However, at this stage, we do not prove through topic
ment group, we move to the research methodology’s infer- modeling or regression analysis. Applying the same lens in
ential part. Traditional multivariate regression models might the negative conversations, some of the keywords used are:
have low errors in analysis predictions; however, they might “employee emotion,” “mistake, “broken,” “human,” “sup-
have scattered predictions due to over-finetuning within the port,” “lack.” These keywords are also visually correlated
analysis sample when generalized over a larger population. to the theoretical groundings built for AI adoption barriers
On the other hand, if the model is left un-tuned on the analy- in an organization. The subsequent phases of analysis will
sis sample, the inferences might not be significant. This phe- establish these findings more robustly.
nomenon is defined as a bias-variance tradeoff—the lasso
and ridge regression model help to cater to the bias-variance 4.2 Signal‑capturing Through Topic Modeling,
tradeoff through generalization. Lasso regression generalizes Network Analysis, and Clustering
using L1-norm regularization (Barrodale & Roberts, 1978),
and Ridge regression generalizes L2-norm regularization With words as the unit of analysis in the current research
(Hoerl & Kennard, 1970). This makes sure that the results and having established some directional visual correlation
and parameter estimates are correct if the prediction vectors among the organic themes emerging from the conversation
(dependent variable) are not orthogonal. and theoretically built hypothesis, we create a topic model to
This section runs a multivariate analysis with the key- develop these keywords more mathematically. For this pur-
words associated with each construct as independent vari- pose, we run LSTM topic modeling explained in detail in the
ables and the tweets’ sentiment and subjectivity score men- methodology section earlier. To equal weightage to drivers
tioned using AI as a dependent variable. The dependent and barriers to adoption, we run two separate LSTM topic

13
 Information Systems Frontiers

modeling in the groups of tweets emerging as contextually and process. It would be interesting to see how the topics
positive and contextually negative simultaneously. We then evolve when we look at their semantically similar clusters
plot the outcomes of these topic modeling in the respective formed in the cider sample across all the users.
groups illustrated in Fig. 3. We compared a similarity score With the signals of drivers and barriers from the lens
across topics and methods to vest the confidence and coher- of business and application owners (employees) established
ency among the topics and methodology and picked the visually, mathematically, and subcommunities formed, we
10-topic iteration. Also, it is evident from the diagram that now extend these sub-communities to the analysis sample to
LSTM based topic modeling presents better coherent topics understand how these topics relate to the users in the same
than LDA. However, the current study’s purpose is not to group talk about. For this purpose, we project the words
prove that LSTM topic modeling is better than LDA (Fig. 4). through ELMO embeddings to 512-dimensional vector
With the mathematical topic modeling, the topics within space and then run a principal component analysis (PCA)
the positive sentiment group of AI adoption that have to further reduce to a two-dimensional space to plot visu-
emerged have precise pivots of “enablement,” “automation,” ally. We now run a clustering algorithm to find the nearest
“efficiency,” “reach,” “advancement,” and “real-time.” This words from the respective sentiment groups and visualize
output further builds our confidence in the visually inter- using hierarchical clustering. The plots are represented in
preted keywords (drivers) from the previous sections, now Fig. 8 for positive and negative sentiment groups towards AI
being established mathematically. Using the same lens when adoption, respectively. This helps us to understand better the
we try to analyze the keywords in the negative sentiment sub-groups formed within each sentiment group.
group towards the adoption of AI, we notice that some of
the main keywords that emerge are: “precarious,” “slow,” 4.3 Theoretical Signal Grounding
“help,” “cost,” emotion” and “studies.”
These keywords that emerged through the topic modeling With the signal capturing and processing steps followed on
indicate signals which need to be further grounded in the UGC using SMA from big-data analytics, there are clear
existing theories and indications from the literature review indications of keywords like efficiency, business research,
leading to constructs meaning experiences of adopting AI. novelty with manual intervention from business owners’
We can confidently establish these keywords against the con- standpoint. At the same time, conversational exchange
struct statements mathematically now. However, as intuitive, among application owners (employees) indicates that expe-
these topics appear in the vicinity within each sentiment riential learning, personal growth under managerial and
group; it is also essential to see how these keywords are technical support, and potential fear of failure is another per-
intertwined as part of a more extensive conversation. To spective. We use these keywords and now proceed towards
cater to this, we incorporate a network diagram borrowed theory building, which will lead us to hypothesis statements.
from network sciences (Börner et al., 2007) but applied in The current section develops signal grounding into theories
the text domain with words (topics that emerged from topic from the management science domain of adoption, barriers,
modeling) as a unit of analysis. We perform the network and experience and chooses the adoption journey constructs.
establishment at two levels. We try to build a network dia- The dependent variable of the current research is overall
gram at an overall level, assuming the topics that emerged adoption experience (Janssen et al., 2004; Van de Ven, 1986)
from the respective positive and negative groups are essen- and (Woodman et al., 1993), defined as individual sentiment
tially part of the general sample illustrated in Fig. 5. and emotion towards adoption and usage of new technology
At the second level, we replicate the network diagram (AI in the current research). The present study considers the
within each sentiment group of AI adoption for a more sentiment and emotion expressed in AI usage a proxy for
granular understanding of the topics connected. These dia- the adoption experience. Next, we identify factors and driv-
grams are represented in Figs. 6 and 7, respectively. Closely ers through text summarization processes mapped to fifteen
analyzing the network diagram simultaneously, we learned theoretical constructs identified from the literature review
that keywords related to company branding, digitization, and synthesis represented in Table 1.
positioning are emerging as one group pivoting to AI. The Adoption experience: Initial perception and value stream
second group that emerged contains keywords like accelerat- mapping of the outcomes are the initial factors surrounding
ing, adapt, application, and accessibility. On similar lines, new technology’s adoption experience. However, based on
four more groups strongly appear at an overall level. our proposed research, AI adoption is successful only when
When we take the network analysis at the sentiment the business owners and employees who build and deliver
group level, we see closely associated groups emerging. For the AI solutions have a positive experience throughout the
instance, we see sub-groups emerging as smart experience, usage (Janssen et al., 2004). Multiple instances from IS lit-
scale, and dynamic in the positive sentiment group. In the erature (Grover et al., 2019; Brynjolfsson & McAfee 2017)
negative sentiment group, we see sub-groups like disruption indicate that the adoption experience is positive functionality

13
Information Systems Frontiers

Table 1  Factors affecting AI adoption experience

SNo. Factors Explanation Literature Citations

1 Efficiency Solving a tactical business problem efficiently (Cave & ÓhÉigeartaigh, 2019) 350
2 Innovation Embracing product and service offerings innovation to stay competi- (Floridi, 2008) 316
tive
3 Research Business research to identify new opportunities (Benlian et al., 2018) 95
4 Novelty To create a novelty in the product offering (Schmitt, 1999) 7888
5 Manual intervention Manual involvement of business owners (Seufert et al., 2020) 35
6 Adaptability Flexibility to adapt to any domain (within or outside the organization) (Riley, 2018) 1579
7 Emotion The emotion of employees towards AI as a technology (Andersson & Pearson, 1999) 3638
8 Support Providing technical and managerial support to employees (Johns, 2017) 479
9 Personal growth An employee’s motivation for personal growth by learning AI (Mcilroy et al., 2007) 323
10 Experiential learning An employee’s learning through experience, experimentation, and (Thumin & Thumin, 2011) 92
research with new technology adopted
11 Fear of failure An employee’s fear of failure in learning AI (Aguinis et al., 2013) 796
12 Fear of upgradation An employee’s fear of upskilling or cross-skilling on AI (Taylor, 1991) 18,672

Fig. 1  The research methodol-

ogy framework for the assess-
ment of AI adoption

and outcome performance with ease of usage is the crucial quantifiable parameter. However, these experiences are never
sets of drivers. The current research establishes the rela- shared for other organizations to learn and gain from the
tionship between overall adoption experience and usefulness experiences.
and ease of use (from the business owners and employees). More and more business owners start from chief technol-
Typically, these experiences are built over a period while ogy officers (CTO) until they build systems on the ground
adopting new technology (AI) in a business organization. and use SMP. The conversational exchange of experiences
At the end of a certain or trial period both business own- is built-in UGC. If mined through the social media analytics
ers and employees are asked to fill in their experiences in a framework (SMA) (Rathore et al., 2017), it may capture the

13
 Information Systems Frontiers

Fig. 2  Word clouds representing positive and negative conversations around AI adoption

Fig. 3  Word clouds representing topic modeling outputs in positive and negative conversations groups with AI adoption

Fig. 4  Topic coherency across

methods and number of topics

13
Information Systems Frontiers

Fig. 5  Network diagram based on topic association at an overall level

overall experience through contextual sentiments and emo- Management literature indicates that higher the efficiency
tions. In this study, we use these contextual sentiments and of an AI solution lowers the cost of manual and in-efficient
emotions as a proxy to define the overall adoption experi- work bringing down the original investment over a period
ence built over a period and is sharable across all the organi- leading to increased return on investments, which is a desir-
zations. This adoption experience is the dependent variable able state any business owner would like to be in. These
of our model to test the hypothesis’s robustness around fac- use-cases could be dynamic advertisement placement logic,
tors driving or hindering AI adoption in an organization. delivery personnel routing, warehouse selection. The metrics
used to assess AI investments’ return in this perspective are
Efficiency  This potential driver captures the efficiency efficient processing (Daugherty & Wilson, 2018). Hence, the
that adopting AI can bring by automating some manual hypothesis around efficiency can established as:
in-efficient and human error-prone business processes,
which could also be time-consuming in the same context H1: There is a positive relationship, and efficiency signifi-
(Cave & ÓhÉigeartaigh, 2019; Raisch & Krakowski, 2020). cantly drives overall adoption experiences of AI

13
 Information Systems Frontiers

Fig. 6  Network diagram based on topic association within the positive group

Innovation  Product innovation and research through AI Research  Business needs constant research to identify
solutions is another critical driver that emerged through the new opportunities to build a customer base (Kellogg et al.,
synthesis of management literature (Benlian et al., 2018; 2019; Lindebaum et al., 2019). This business research is
Schmitt, 1999). This applies to large-scale established organ- done through primary interviews, focused group feedback
izations and small and medium-scale organizations building sessions, and on-field conversations in a traditional environ-
up, a critical use case of AI adoption and application. This ment. However, in the contemporary business environment,
factor posits that AI solutions should drive more and more research on identifying a new business opportunity and even
product innovation use-cases (Stephan et al., 2017; Benlian research on the existing customer base are done real-time
et al., 2018), which would help business owners make real- using AI products like chatbots (Luo et al., 2019). This real-
time decisions to establish the value stream mapping of AI time research on the existing customer base and discover-
solutions in an organization. This value stream mapping will ing an untapped business opportunity helps business owners
point to AI adoption’s perceived benefits, eventually driving learn from failures quickly, referred to as “falling forward”
more AI techniques and solutions. Hence, the hypothesis in the management context (McGrath, 1999). Hence, the
around innovation can be established as: flexibility of doing these researches using AI establishes this
as one of the critical drivers in calculating the business’s
H2: There is a positive relationship, and product innova- ROI in AI solutions. Hence, the hypothesis around research
tions significantly drive overall adoption experiences of can be established as:
AI

13
Information Systems Frontiers

Fig. 7  Network diagram based on topic association within the negative group

H3: There is a positive relationship, and research sig- H4: Novelty would be positively associated with adoption
nificantly drives overall adoption experiences of AI experiences of AI

Novelty  With product innovations done and the business Manual‑Intervention  To build a robust and generalizable
opportunity identified through AI solutions, business organi- theoretical framework defining AI’s adoption, it is also
zations now look to create a novelty in the product offer- essential to consider some limitations a business owner
ing. In the literature review section, we have quoted several might see or feel. Synthesis of the literature review leads
examples where more prominent organizations have created us to believe that AI solutions or systems require manual
a novelty in offering AI (Sheridan, 2004; Luo et al., 2019). involvement to redefine the KPIs and objectives with the
These novel solutions could be less time-consuming search change in domain chain time. This means that a business
and reduce costs (Bergstein, 2019). Furthermore, teams owner has to evaluate the AI solution’s goals (Braga &
working on AI development and solution building can enor- Logan, 2017; Floridi, 2008) with closer human interven-
mously enable the product team with various novel solu- tion. Suppose a human solution replaces the AI solution. In
tions, leading a business owner to invest more in building that case, the humans working on building the solution can
and adopting AI capability (Meinhart, 1966). This leads to revisit the KPIs according to the domain they are working
the hypothesis that the product novelty that an AI solution in, which would mean minor augmentation with business
can bring could be one of the drivers leading to the adop- leaders. This leads to hypothesize that the constant need for
tion of AI. Hence, the hypothesis around novelty can be manual intervention (Kellogg et al., 2019; Luo et al., 2019)
established as: to redefine the KPIs of an AI system might act as a negative

13
 Information Systems Frontiers

Fig. 8  Clustering association graphs within positive and negative groups, respectively

factor that could hinder the adoption of AI by a business manual phase, with product and business owners spending
owner. Hence, the hypothesis around manual intervention hours again before the solution can be used in an entirely
can be established as: automated manner. Hence, the hypothesis around adopt-to-
domain can be established as:
H5: There is a negative relationship, and manual interven-
tion significantly drives negative overall adoption experi- H6: There is a negative relationship, and the adopt-to-
ences of AI domain limitation of AI significantly drives negative
overall adoption experiences of AI
Adopt to the Domain  With manual intervention to set up
the KPIs and objectives, AI solutions are also not evolved Emotion  Business and management literature (Woodman
to adapt to the domain in which the AI must be applied. This et al., 1993; Yuan & Woodman, 2010) suggests that busi-
domain could be various functions like HR (Riley, 2018), ness owners always try to use the state of the art solutions
advertisement placement, or even a vertical domain in which (for instance: AI) that money can purchase, which keeps
the AI solution has to be implemented, like finance and man- them competitive in the market. However, the organization’s
ufacturing. Synthesizing the literature leads us to believe theories (Cambre & Cook, 1985; Deichmann & Ende, 2013)
that failure to adopt is a factor that a business owner might suggest that the companies’ introduction can provoke vari-
not like. Most of the AI solution is built through domain- ous emotions. The structured synthesis of the management
specific learning. IN MOST SCENARIOS, an AI model is and organizational theories encourages us to believe that
trained in a specific business domain or vertical and might emotion can be crucial in driving AI adoption in an organi-
not work well if the domain is changed (Seufert et al., 2020). zation. However, we are unsure about employees’ feelings
This means that the AI solution must again go through the about AI’s overall adoption experience from their len at this

13
Information Systems Frontiers

research stage. Theories suggest that peer pressure of adop- motivation) and sharing are helpful and are highly likely to
tion might cause an employee to perform poorly (Anders- acquire new knowledge and apply them in problem-solving
son & Pearson, 1999) which could further slowdown the using new technology (AI) in an organization. Therefore, the
adoption of AI and can have negative sentiment and emo- ongoing discussion leads us to combine PKC and PKA as
tion. At the same time there are theories that suggest that one driver and hypothesize that learning driven by personal
with a good organizational environment might positively growth might be one of the significant drivers that could
affect employees’ emotion (Drexler, 1977). In either case, lead organizations to quickly adopt AI through employees.
our assumption is that emotion will have significant impact Hence, the hypothesis around personal growth can estab-
in driving the overall adoption experience. Towards the end lished as:
of the current research, we will learn whether emotion of
employees towards AI positively drives or negatively drives H9:There is a positive relationship, and personal growth
the adoption of AI. Hence, the hypothesis around emotion significantly drives overall adoption experiences of AI
can established as: in an organization from employees’ standpoint

H7: There is a strong relationship and emotion of employ- Experiential Learning  Scholars have established experiential
ees towards AI significantly drives overall adoption expe- learning through experimentations as research performed
riences of AI by employees on new technology as critical steps in the
organizations’ quest to adopt any technology (Sitkin, 1992;
Ellis et al., 2014; Kolb, 2015). However, these organizational
Support  Researches suggest that organizations and team theories for adoption disregard the social structure and the
owners should provide a fertile ground for the team mem- fear of maintaining a positive image in an employee social
bers to adopt solutions (in the current research AI) (Mcilroy network in an organization (Reynolds & Vince, 2004; Argote
et al., 2007; Drexler, 1977). The organizations’ internal cli- & Miron-Spektor, 2011; KC et al., 2013). However, experi-
mate while adopting AI or any new technology to advocate ential learning by adopting AI within a business organiza-
learnings and experimentations is one of the variables that tion’s social context has not been tested in IS research so far.
might expedite the employee’s adoption of AI. This internal Hence, the literature synthesis hypothesizes that experiential
climate is built on the pillars of managerial support (Davis, learning during AI adoption will speed up the adoption rate.
1989; Al-Gahtani & King, 1999) and technical and func- Thus, the hypothesis around experiential learning can be
tional training (Igbaria et al., 1996) given to the employ- established as:
ees. These support factors have been significantly proven
in the adoption of information technology (IT). This leads H10: There is a positive relationship, and experiential
us to believe that cultivated and fertile environmental sup- learning significantly drives overall adoption experi-
port through appropriate training and managerial support ences of AI in an organization from the employees’
will enable employees to adopt AI faster than organizations standpoint
where this support is missing. Hence, the hypothesis around
support can be established as: Fear of Failure  Having established the learnings gained by
employees of the organization adopting AI, through experi-
H8: There is a positive relationship, and support signifi- ence, experimentation, and research on the new technolo-
cantly drives overall adoption experiences of AI in an gies, there is a group of scholars who have proposed fear of
organization from employees’ standpoint failure while learning new techniques (McGrath, 1999; Ellis
et al., 2014) hindering an employee’s motive of adopting
Personal Growth  There is much research that perceived new technology. Another set of scholars have highlighted the
knowledge articulation (PKA) and perceived knowledge stress of learning new techniques and being a painful experi-
codification (PKC) (Furlan et al., 2019) as one of the driv- ence of adopting new practices and technologies threaten-
ers at the individual level learning within organizational ing an employee’ positive image (Edmondson, 2004) in a
learning mechanism (Hansen et al., 1999; Helfat & Peteraf, team or organization driving taking a defensive stance while
2015). As the organization learns to adopt new technology adopting new techniques or technologies (Taylor, 1991).
(AI) and learn about it, the employees are driven by PKC Synthesis of literature establishes the internal fear as one
and PKA to develop their cognitive capabilities (Dittrich factor that might hinder the adoption of AI within a business
et al., 2016). The motivation here is personal development organization. Hence, the hypothesis around fear of failure
and growth with the new technology. Overall, the literature can be established as:
(Day & Lord, 1992) suggests that employees with larger
PKC and PKA believe that personal learning (driven the

13
 Information Systems Frontiers

H11: There is a negative relationship, and fear of failure AI adoption, was executed with lasso and ridge regression.
in learning AI significantly drives negative overall adop- Consumer behavior research (Trudel, 2019) indicates that an
tion experiences of AI from an employee’s perspective individual’s evaluation can be rational or emotional. Analyti-
cal, emotional behavior is expressed with some beliefs, feel-
Fear of Upgradation  When the organization introduces new ings, and views. Simultaneously, the emotional evaluation
technology, literature (Al-Gahtani & King, 1999) suggests can be expressed as non-tangible emotions captured in the
that employees fear going through the cycle of upgrading degree of emotions. In NLP, the first is captured through a
themselves through learning upskilling or cross-skilling text’s subjectivity score, and the second is captured using the
themselves. This fear prevents employees from learning and sentiment score. Hence to learn and interpret the complete
experimenting with new technologies (Aguinis et al., 2013). experience of AI adoption, it becomes crucial to consider
The complex the technology, the larger the fear is, enhanc- both.
ing the barriers and slowing down AI adoption. Hence, the The model setup contains sentiment and subjectivity
hypothesis around fear of up-gradation can be established as: scores representing the overall AI adoption experience and
the dependent variable. The independent variables are the 12
H12: There is a negative relationship, and fear of upgra- hypothesis statements expressed in Table 1, with each of the
dation significantly drives negative overall adoption expe- keywords highlighted as regressors. Only those constructs
riences of AI from an employee’s perspective were accepted at the model-building exercise, whose p-value
was more significant than or equal to 0.05. The hypothesis
The theoretical framework with all the factors hypoth- must also come out significant across all the four models to
esized towards adopting AI in an organization is represented be placed in the final framework for adopting AI in an organ-
in Fig. 9 below. ization. This rigorous condition helps to make the results
more generatable and robust. The model summary across
4.4 Hypothesis Validation Through a Multivariate four models is represented in Table 2 below.
Model With the synthesis and analysis of the multivariate regres-
sion analysis, we concluded that factors surrounding “effi-
Multivariate regression analysis across sentiment and sub- ciency achieved,” “product innovation,” “unique product
jectivity scores, representing emotion and rationality towards value,” “employee emotion,” “technical and managerial

Fig. 9  The theoretical framework for the assessment of AI adoption

13
Information Systems Frontiers

Table 2  Results of multivariate Model Sentiment scores as the dependent vari- Subjectivity scores as the dependent
regression analysis able variable
Lasso Ridge Lasso Ridge
Coefficient P>|t| Coefficient P>|t| Coefficient P>|t| Coefficient P>|t|

Constant 0.138 0 0.138 0 0.305 0 0.305 0

Efficiency 0.215 0.04 0.243 0.176 0.276 0.16 0.315 0.161
Innovation 0.136 0.003 0.259 0.112 0.230 0.049 0.401 0.045
Research 0.120 0.25 0.236 0.256 1.096 0 2.177 0
Novelty 0.141 0.000 0.558 0 0.075 0.031 0.301 0.031
Manual intervention -0.01 0.08 -0.007 0.917 -0.070 0.42 -0.074 0.424
Adaptability -0.166 0.04 -0.101 0.412 -1.103 0 -0.693 0
Emotion 0.303 0.62 5.09E-14 0 0.148 0.457 5.62E-14 0
Support 0.042 0.001 0.038 0.001 0.016 0.299 0.014 0.308
Personal growth 3.551 0 8.46E-18 0.097 1.035 0.066 6.57E-18 0.293
Experiential learning 0.028 0.009 0.029 0.003 0.140 0.002 0.144 0.002
Fear of failure -0.134 0 -0.153 0 -0.247 0 -0.279 0
Fear of upgradation -1.216 0 -229.607 0.433 -3.483 0.87 -4.847 0.336

support provided to employees for adoption,” and “experi- by business owners and employees in the sample created.
ential learning” that employees gain are the factors which The viable reasons and factors were found in the manage-
can significantly be accepted as the drivers of adoption of ment and organizational literature (Daugherty & Wilson,
AI. At the same time, the hypothesis around “manual inter- 2018; Raisch & Krakowski, 2020) with the assumption of
vention from time to time” and “employees’ fear of failure” automation-augmentation. This led us to uncover the aug-
are the factors that can be significantly accepted as barri- mentation aspect of AI, human involvement in development,
ers of adoption from business and AI application owners’ and solution building. The human interaction aspects that we
(employees) perspectives. uncovered were related to: “employee emotion,” “mistake,
“broken,” “human,” “support,” “lack” from Fig. 3. We found
initial supporting arguments behind the emerged human
5 Discussion aspects rooted in organizational and behavioral science lit-
erature (Argote & Miron-Spektor, 2011; Day & Lord, 1992;
In the proposed research, we started with the theoreti- Deichmann & Ende, 2013).
cal building blocks from management theories. We found Through the second phase of our research, the second
prior research surrounding innovation and adoption widely finding helped us learn about the significant themes through
accepted in organizational theories (Janssen et al., 2004; Van topic modeling. With words as the unit of analysis and hav-
de Ven, 1986; Woodman et al., 1993) for business effective- ing established some directional visual correlation through
ness. With the synthesis of the literature, we further identi- the first phase of the learnings, we further extended this
fied that the overall behavior of adopting any new technol- by the mathematical topic modeling. The topics within the
ogy is an excellent aid for business organizations to succeed positive sentiment group of AI adoption that emerged had
in an ever-competitive world (Yuan & Woodman, 2010; precise pivots of “enablement,” “automation,” “efficiency,”
West & Farr, 1989). This motivated us to pursue and come “reach,” “advancement,” and “real-time.” This output further
up with a framework that best defines the overall adoption built our confidence in the literature’s visually interpreted
experience of AI in the current digital world as a function of keywords (drivers) (Woodman et al., 1993; Janssen et al.,
the drivers and barriers of a business owner who sponsors 2004). Using the same analysis, we learned the keywords in
the investment in AI capability building and the application the negative sentiment group towards the adoption of AI; we
owner (employee) of AI who builds the AI solution that is noticed that some of the main keywords that emerged were:
used on the field. “precarious,” “slow,” “help,” “cost,” emotion” and “stud-
The first finding of the first phase of our research (Fig. 2) ies.” This had the theoretical groundings in organizational
developed some prominent themes that emerged from the and behavioral science literature (Argote & Miron-Spektor,
visual analysis around AI like “experience,” “latest,” “auto- 2011; Day & Lord, 1992; Deichmann & Ende, 2013). We
mation,” “accelerate,” “success,” “search,” “study.” These extended this analysis to identify these themes’ commu-
keywords have emerged basis the frequency of their usage nity and then looked at the positive and negative sentiment

13
 Information Systems Frontiers

groups. Closely analyzing the network diagram, we found “support,” “personal growth,” “experiential learning,” “fear
two significant groups emerging: company branding, digiti- of failure,” “fear of upgradation” from not only business and
zation, and positioning, and the second group as accelerate, management literature (Bartunek & Ragins, 2015; Becker &
adapt, application, and accessibility (Figs. 4, 5, and 6). Jaakkola, 2020) but also combining the human augmentation
In the last section of the research, we have established part through organization and behavior literature (Cave &
the framework with four multivariate regressions models ÓhÉigeartaigh, 2019).
using lasso and ridge regressions on the contextual (adop- Adoption experience overall depends on two crucial fac-
tion experience of AI) sentiment and subjectivity scores. tors: the overall value addition AI can with the investments
Generalizing using L1-norm regularization (Barrodale & done to build the capability and how it is effectively used
Roberts, 1978) and L2-norm regularization (Hoerl & Ken- and built by the employees. All the theoretical factors that
nard, 1970) interprets the model more robust. Hence the we have identified rolls-up to one of these two groups of
initial proposed 12 hypothesis statements were tested; eight factors. At the same time, prior studies have either value
were finally approved at the statistical significance of 0.05 addition or employee experience. Combining both the fac-
across all four models. tors and testing in the context of AI has never been tried
to the best of our knowledge. Our research highlights that
5.1 Literature Implications apart from the efficiency that an AI solution can bring into
an organization, the product design novelty and innovation
The proposed research contributes to the IS literature AI can get to the overall product offering drives significant
in two folds. First, we make a theoretical contribution to organizations to invest in it and build the capability. While
the organizations that are still unsure of the advantages of the decision to adopt AI brings slight unrest among employ-
adopting AI. Second, we make a vital framework contribu- ees, it is also the fact that with technical and managerial sup-
tion from the lens of business and application owners of port, AI leads to experiential learning for employees, which
AI, which does not exist when writing this manuscript to motivates them to start using AI. Since each of these hypoth-
the best of our literature synthesis done. In this research, esis statements has gone through multiple validation phases
we extended the theoretical understanding of factors like and tested through four different regression models, we are
“efficiency,” “innovation,” “business research,” “product confident in finalizing the framework driving AI adoption
novelty,” “manual intervention, “adaptability,” “emotion,” and illustrated in Fig. 10.

Fig. 10  Framework driving the adoption of AI in an organization

13
Information Systems Frontiers

The proposed framework is a significant contribution to 6 Conclusions

the IS literature for future researchers. It can act as a tested
business framework for originations looking for the expe- The research undertaken in the present paper was entirely
rience of adopting AI to plan the investment at their end. motivated by the fact that there is a literature gap in assess-
The framework results from prior management and organi- ing AI’s adoption from different lenses, which was not
zational behavior literature, with unbiased conversations on attempted earlier and to act as a reference for suture IS
SMP Twitter using a combined SMA framework (Grover researchers and even business owners the management
et al., 2019) tested thoroughly by multiple models. This standpoint. This gap exists even more from the viewpoint
keyword extraction approach, association through network of inductive theory building from multi stakeholders’
sciences, and testing on polarity and emotion scores also adoption literature. The unbiassed sample collected was
form a structured hypothesis testing framework for future methodologically controlled to build technology and busi-
IS researchers. The proposed research also indicates that ness savvy participants and capture the right conversations
it is wise to test the hypothesis through multiple models catching their experience. The structured analysis under
before integrating a theoretical construct as part of the final the big-data analytics framework (Grover et al., 2019),
framework. This makes the model much more robust to any drawing insights from SMA (Lakhiwal & Kar, 2016), led
change in the domain and makes the results generalizable. us to remove some initial signals from the conversational
exchange of experiences around the stakeholders AI being
business owners and application owners (employees). We
5.2 Practice Implications used summarization techniques to establish the keyword
associating automation, efficiency, business research,
Our research has highlighted that factors like efficiency, experiential learning, and a few barriers like manual inter-
product and service innovations, and novelty have made vention and the fear of employees’ failure. We interpreted
more influential business owners develop and invest in AI that these keywords are strongly connected through net-
capability. Hence, this acts as the biggest management ref- work sciences, which established that they are not isolated.
erence for the business owners in small and medium-scale We established these keywords are generalizable even
organizations to build a potential use case for AI’s applica- when we extend to a broader group of conversations. Post
bility. Furthermore, they can also learn from the barriers the theory building for hypotheses and testing through
that more prominent organizations have faced; for instance, multivariate model analysis, the findings of the research
employees’ fear of failure or potential lack of managerial establish that efficiency, innovation, and novelty (unique-
and technical support can slow down the overall adoption. ness) in the product and service offerings drive a business
This can act as learning for the business owners in small owner to invest in AI while also at the same time being a
and medium scale organizations to ensure that the overall skeptic on the level of manual involvement required. From
adoption of AI successfully provides these barriers are well an application owners’ standpoint, emotion, managerial
catered to and answered through management strategies. and technical support, and experiential learning drive an
Technical and business managers will further confirm that employee to adopt AI more; however, the skepticism of
they adopt AI by indicating the fertile culture of innovation failure might hinder the adoption rate.
and research within organizations. However, they also try to
build on top of the baseline solutions to innovate the product
and service offerings.
This research methodology and framework also act as one 7 Limitations and Future Research Scope
of the first of its kind. Even business managers who have
already invested in building AI capabilities can assess inter- The proposed research has a few limitations; first, we have
mediate success or barriers faced internally. Furthermore, only considered two AI adoption lenses through a business
this framework can be used internally within an organiza- owner and second through application owners (employ-
tion by taking anonymous contextual and descriptive tex- ees). However, both the owners work towards providing a
tual feedbacks rather than only categorical feedback. For state-of-the-art experience to customers using AI. So, the
example, suppose the business managers run this framework basic assumption here is that customers feel the change
on textual descriptive feedback collected with business and or enhancement in their experience after the companies
team heads. In that case, they can assess the best value addi- started catering to them through AI solutions. Again, this
tions that AI has bought in the organization’s business field can lead to two assumptions: the customers are happy with
and the teams’ hurdles. This can help the business organi- the AI-driven solutions, more customized solutions, or do
zations do any course corrections and churn out the AI’s not feel comfortable interacting with the AI products and
maximum ROI.

13
 Information Systems Frontiers

solutions business organizations have launched by keeping Balakrishnan, J., Dwivedi, Y. K., Hughes, L., & Boy, F. (2021). Ena-
their customers in the middle. A prominent example is cus- blers and inhibitors of AI-powered voice assistants: a dual-fac-
tor approach by integrating the status quo bias and technology
tomer support through chatbots. In the proposed research, acceptance model. Information Systems Frontiers. https://d​ oi.o​ rg/​
we have not made any assumptions or considered any con- 10.​1007/​s10796-​021-​10203-y
structs around customer experiences adopting AI-driven Barabási, A. L. (2013). Network science. Philosophical Transactions
products and companies’ solutions. We feel that this is a of the Royal Society A: Mathematical, Physical and Engineering
Sciences, 371, 20120375
limitation of the currently proposed research. However, Barrodale, I., & Roberts, F. D. (1978). Solution of the constrained, ℓ1
future IS researchers can take up the same limitation as linear approximation problem. ACM Transactions on Mathemati-
part of new research using the currently proposed research cal Software, 6(9), 231–235
as a baseline and borrow the overall framework proposed Bartunek, J. M., & Ragins, B. R. (2015). Extending a provocative tradi-
tion: book reviews and beyond at AMR. Academy of Manage-
in the current manuscript. ment Review, 40(3), 474–479. https://​doi.​org/​10.​5465/​amr.​2015.​
0029
Becker, L., & Jaakkola, E. (2020). Customer experience: Fundamental
Declarations  premises and implications for research. Journal of the Academy
of Marketing Science, 48(4), 630–648. https://​doi.​org/​10.​1007/​
s11747-​019-​00718-x
Conflicts of Interest  Authors have no conflict of interests to declare. Benlian, A., Kettinger, W. J., Sunyaev, A., Winkler, T. J., & EDITORS,
G. (2018). Special section: the transformative value of cloud
computing: a decoupling, platformization, and recombination
theoretical framework. Journal of Management Information Sys-
tems, 35(3), 719–739. https://​doi.​org/​10.​1080/​07421​222.​2018.​
References 14816​34
Berger, J., Sorensen, A. T., & Rasmussen, S. J. (2010). Positive effects
Aguinis, H., Gottfredson, R. K., & Culpepper, S. A. (2013). Recom- of negative publicity: when negative reviews increase sales.
mendations for estimating cross-level interaction effects using Marketing Science, 29(5), 815–827. https://​doi.​org/​10.​1287/​
multilevel modeling. Academy of Management Proceedings, mksc.​1090.​0557
2013(1), 10839. https://​doi.​org/​10.​5465/​ambpp.​2013.​10839​ Bergstein, B. (2019). Can AI pass the smell test? MIT Technology
abstr​act Review, 122(2): 82–86
Ahuja, M. K., & Thatcher, J. B. (2005). Moving beyond Intentions Börner, K., Sanyal, S., & Vespignani, A. (2007). Network science.
and toward the Theory of trying: Effects of work environment Annual Review of Information Science and Technology, 41(1),
and gender on post-adoption information technology use. MIS 537–607
Quarterly, 29(3), 427–459. https://​doi.​org/​10.​2307/​25148​691 Braga, A., & Logan, R. K. (2017). The emperor of strong AI has no
Al-Gahtani, S. S., & King, M. (1999). Attitudes, satisfaction and usage: clothes: limits to artificial intelligence. Information, 8(4), 156.
Factors contributing to each in the acceptance of information https://​doi.​org/​10.​3390/​info8​040156
technology. Behaviour & Information Technology, 18(4), 277– Brock, J. K. U., & Von Wangenheim, F. (2019). Demystifying AI: What
297. https://​doi.​org/​10.​1080/​01449​29991​19020 digital transformation leaders can teach you about realistic artifi-
Andersson, L. M., & Pearson, C. M. (1999). Tit for Tat? The spiraling cial intelligence. California Management Review, 61(4), 110–134
effect of incivility in the workplace. Academy of Management Brynjolfsson, E., & McAfee, A. (2017). The business of artificial intel-
Review, 24(3), 452–471. https://d​ oi.o​ rg/1​ 0.5​ 465/a​ mr.1​ 999.2​ 2021​ ligence. Harvard Business Review, July Issue
31 Büschken, J., Otter, T., & Allenby, G. M. (2013). The dimensionality
Archak, N., Ghose, A., & Ipeirotis, P. G. (2011). Deriving the pricing of customer satisfaction survey responses and implications for
power of product features by mining consumer reviews. Manage- driver analysis. Marketing Science, 32(4), 533–553. https://​doi.​
ment Science, 57(8), 1485–1509. https://​doi.​org/​10.​1287/​mnsc.​ org/​10.​1287/​mksc.​2013.​0779
1110.​1370 Cambre, M. A., & Cook, D. L. (1985). Computer anxiety: defini-
Argote, L., & Miron-Spektor, E. (2011). Organizational learning: from tion, measurement, and correlates. Journal of Educational
experience to knowledge. Organization Science, 22(5), 1123– Computing Research, 1(1), 37–54. https://​doi.​org/​10.​2190/​
1137. https://​doi.​org/​10.​1287/​orsc.​1100.​0621 FK5L-​092H-​T6YB-​PYBA
Arjun, R., Kuanr, A., & Kr, S. (2021). Developing banking intelligence Cariani, P. (2010). On the importance of being emergent. Constructivist
in emerging markets: Systematic review and agenda. Interna- Foundations, 5, 86–91
tional Journal of Information Management Data Insights, 1(2), Cave, S., & ÓhÉigeartaigh, S. S. (2019). Bridging near- and long-
100026. https://​doi.​org/​10.​1016/j.​jjimei.​2021.​100026 term concerns about AI | Nature Machine Intelligence. Nature
Asuncion, A. G., & Lam, W. F. (1995). Affect and impression forma- Machine Intelligence, 1, 5–6
tion: influence of mood on person memory. Journal of Experi- Dai, T., & Singh, S. (2020). Conspicuous by its absence: diagnos-
mental Social Psychology, 31(5), 437–464. https://​doi.​org/​10.​ tic expert testing under uncertainty. Marketing Science, 39(3),
1006/​jesp.​1995.​1019 540–563. https://​doi.​org/​10.​1287/​mksc.​2019.​1201
Bader, V., & Kaiser, S. (2019). Algorithmic decision-making? The user Daugherty, P., & Wilson, H. J. (2018). Human + machine: Reimagining
interface and its role for human involvement in decisions sup- work in the age of AI. Harvard Business Review
ported by artificial intelligence—Verena Bader, Stephan Kaiser, Davis, F. (1989). Perceived usefulness, perceived ease of use, and user
2019. Organization Science, 26(5), 655–672 acceptance of information technology. MIS Quarterly, 13(3),
Baird, A., & Maruping, L. M. (2021). The next generation of research 319–340
on is use: a theoretical framework of delegation to and from Day, D. V., & Lord, R. G. (1992). Expertise and problem categori-
agentic is artifacts. MIS Quarterly, 45(1), 315–341. https://​doi.​ zation: The role of expert processing in organizational sense-
org/​10.​25300/​MISQ/​2021/​15882 making. Journal of Management Studies, 29(1), 35–47

13
Information Systems Frontiers

de Jong, M. G., Lehmann, D. R., & Netzer, O. (2012). State-depend- Haenlein, M., & Kaplan, A. (2019). A brief history of artificial intel-
ence effects in surveys. Marketing Science, 31(5), 838–854. ligence: On the past, present, and future of artificial intelligence.
https://​doi.​org/​10.​1287/​mksc.​1120.​0722 California Management Review, 61(4), 5–14
Deichmann, D., & van den Ende, J. (2013). Rising from failure and Hansen, M. T., Nohria, N., & Tierney, T. (1999). What’s your strat-
learning from success: the role of past experience in radical ini- egy for managing knowledge? Harvard Business Review, 77(2),
tiative taking. Organization Science, 25(3), 670–690. https://d​ oi.​ 106–116
org/​10.​1287/​orsc.​2013.​0870 Helfat, C. E., & Peteraf, M. A. (2015). Managerial cognitive capabili-
Dittrich, K., Guérard, S., & Seidl, D. (2016). Talking about routines: ties and the microfoundations of dynamic capabilities. Strategic
The role of reflective talk in routine change. Organization Sci- Management Journal, 36(6), 831–850
ence, 27(3), 678–697 Hoerl, A. E., & Kennard, R. W. (1970). Ridge regression: biased
Drexler, J. A. (1977). Organizational climate: Its homogeneity within estimation for nonorthogonal problems. Technometrics, 12(1),
organizations. Journal of Applied Psychology, 62(1), 38–42. 55–67. https://​doi.​org/​10.​1080/​00401​706.​1970.​10488​634
https://​doi.​org/​10.​1037/​0021-​9010.​62.1.​38 Igbaria, M., Parasuraman, S., & Baroudi, J. J. (1996). A motivational
Dwivedi, Y. K., Hughes, L., Ismagilova, E., Aarts, G., Coombs, C., model of microcomputer usage. Journal of Management Infor-
Crick, T., & Williams, M. D. (2021). Artificial Intelligence (AI): mation Systems, 13(1), 127–143. https://​doi.​org/​10.​1080/​07421​
Multidisciplinary perspectives on emerging challenges, opportu- 222.​1996.​11518​115
nities, and agenda for research, practice and policy. International Janssen, O., van de Vliert, E., & West, M. (2004). The bright and dark
Journal of Information Management, 57, 101994. https://d​ oi.o​ rg/​ sides of individual and group innovation: A Special Issue intro-
10.​1016/j.​ijinf​omgt.​2019.​08.​002 duction. Journal of Organizational Behavior, 25(2), 129–145.
Edmondson, A. C. (2004). Learning from mistakes is easier said than https://​doi.​org/​10.​1002/​job.​242
done: group and organizational influences on the detection and Jarrahi, M. H. (2018). Artificial intelligence and the future of work:
correction of human error. The Journal of Applied Behavioral Human-AI symbiosis in organizational decision making. Busi-
Science, 40(1), 66–90. https://​doi.​org/​10.​1177/​00218​86304​ ness Horizons, 61(4), 577–586
263849 Johns, G. (2001). In Praise of Context.  Journal of Organizational
Ellis, S., Carette, B., Anseel, F., & Lievens, F. (2014). Systematic Behavior
reflection: implications for learning from failures and successes. Johns, G. (2006). The essential impact of context on organizational
Current Directions in Psychological Science, 23(1), 67–72. behavior. Academy of Management Review, 31(2), 386–408.
https://​doi.​org/​10.​1177/​09637​21413​504106 https://​doi.​org/​10.​5465/​amr.​2006.​20208​687
Floridi, L. (2008). Information ethics: A reappraisal. Ethics and Infor- Johns, G. (2017). Reflections on the 2016 decade award: Incorporat-
mation Technology, 10, 189–204 ing context in organizational research. Academy of Management
Furlan, A., Galeazzo, A., & Paggiaro, A. (2019). Organizational and Review, 42(4), 577–595. https://​doi.​org/​10.​5465/​amr.​2017.​0044
perceived learning in the workplace: a multilevel perspective Kar, A. K., & Dwivedi, Y. K. (2020). Theory building with big data-
on employees’ problem solving. Organization Science, 30(2), driven research – Moving away from the “What” towards the
280–297. https://​doi.​org/​10.​1287/​orsc.​2018.​1274 “Why. International Journal of Information Management, 54,
Gal, D., & Rucker, D. D. (2011). Answering the unasked question: 102205. https://​doi.​org/​10.​1016/j.​ijinf​omgt.​2020.​102205
response substitution in consumer surveys—David Gal, Derek KC, D., Staats, B. R., & Gino, F. (2013). Learning from my success
D. Rucker 48(1), 185–195 and from others’ failure: evidence from minimally invasive car-
Gargiulo, F., Cafiero, F., Guille-Escuret, P., Seror, V., & Ward, J. K. diac surgery. Management Science. https://d​ oi.o​ rg/1​ 0.1​ 287/m​ nsc.​
(2020). Asymmetric participation of defenders and critics of vac- 2013.​1720
cines to debates on French-speaking Twitter. Scientific Reports, Kellogg, K. C., Valentine, M. A., & Christin, A. (2019). Algorithms
10(1), 6599. https://​doi.​org/​10.​1038/​s41598-​020-​62880-5 at work: the new contested terrain of control. Academy of Man-
Ghosh, I., & Sanyal, M. K. (2021). Introspecting predictability of agement Annals, 14(1), 366–410. https://d​ oi.o​ rg/1​ 0.5​ 465/a​ nnals.​
market fear in Indian context during COVID-19 pandemic: An 2018.​0174
integrated approach of applied predictive modelling and explain- Kim, H., & Kankanhalli, A. (2009). Investigating user resistance to
able AI. International Journal of Information Management Data information systems implementation: a status quo bias perspec-
Insights, 1(2), 100039. https://​doi.​org/​10.​1016/j.​jjimei.​2021.​ tive. MIS Quarterly, 33(3), 567–582. https://​doi.​org/​10.​2307/​
100039 20650​309
Grover, P., Kar, A. K., Dwivedi, Y. K., & Janssen, M. (2019). Polariza- Kolb, D. A. (2015). Experiential learning: experience as the source of
tion and acculturation in US Election 2016 outcomes – Can twit- learning and development. Pearson Education
ter analytics predict changes in voting preferences. Technological Kumar, S., Kar, A. K., & Ilavarasan, P. V. (2021). Applications of text
Forecasting and Social Change, 145, 438–460. https://​doi.​org/​ mining in services management: A systematic literature review.
10.​1016/j.​techf​ore.​2018.​09.​009 International Journal of Information Management Data Insights,
Grover, P., Kar, A. K., & Ilavarasan, P. V. (2017). Understanding nature 1(1), 100008. https://​doi.​org/​10.​1016/j.​jjimei.​2021.​100008
of social media usage by mobile wallets service providers –An Kushwaha, A. K., & Kar, A. K. (2020a). Language model-driven chat-
exploration through SPIN framework. Procedia Computer Sci- bot for business to address marketing and selection of products.
ence, 122, 292–299. https://​doi.​org/​10.​1016/j.​procs.​2017.​11.​372 In S. K. Sharma, Y. K. Dwivedi, B. Metri, & N. P. Rana (Eds.),
Grover, P., Kar, A. K., Janssen, M., & Ilavarasan, P. V. (2019). Per- Re-imagining Diffusion and Adoption of Information Tech-
ceived usefulness, ease of use and user acceptance of blockchain nology and Systems: A Continuing Conversation (pp. 16–28).
technology for digital transactions – insights from user-generated Springer International Publishing. https://​d oi.​o rg/​1 0.​1 007/​
content on Twitter. Enterprise Information Systems, 13(6), 771– 978-3-​030-​64849-7_3
800. https://​doi.​org/​10.​1080/​17517​575.​2019.​15994​46 Kushwaha, A. K., & Kar, A. K. (2020b). Micro-foundations of arti-
Gunasekaran, A., & Ngai, E. W. T. (2012). The future of operations ficial intelligence adoption in business: making the shift. In
management: An outlook and analysis. International Journal S. K. Sharma, Y. K. Dwivedi, B. Metri, & N. P. Rana (Eds.),
of Production Economics, 135(2), 687–701. https://​doi.​org/​10.​ Re-imagining Diffusion and Adoption of Information Technol-
1016/j.​ijpe.​2011.​11.​002 ogy and Systems: A Continuing Conversation (pp. 249–260).

13
 Information Systems Frontiers

Springer International Publishing. https://​doi.​org/​10.​1007/​978- Springer International Publishing. https://​doi.​org/​10.​1007/​978-

3-​030-​64849-7_​22 3-​319-​24592-8_​36
Kushwaha, A. K., & Kar, A. K. (2021a). Information Labelling Ludwig, S., de Ruyter, K., Friedman, M., Brüggen, E. C., Wetzels, M.,
of Medical Forum Posts by Non-Clinical Text Information & Pfann, G. (2013). More than words: the influence of affective
Retrieval. 12 content and linguistic style matches in online reviews on conver-
Kushwaha, A. K., & Kar, A. K. (2021b). MarkBot – A language model- sion rates. Journal of Marketing, 77(1), 87–103. https://​doi.​org/​
driven chatbot for interactive marketing in post-modern world | 10.​1509/​jm.​11.​0560
SpringerLink. Information Systems Frontiers, 1–18. https://​doi.​ Luo, X., Tong, S., Fang, Z., & Qu, Z. (2019). Frontiers: machines vs.
org/​10.​1007/​s10796-​021-​10184-y humans: the impact of artificial intelligence chatbot disclosure on
Kushwaha, A. K., Kar, A. K., & Vigneswara Ilavarasan, P. (2020a). customer purchases. Marketing Science, 38(6), 937–947. https://​
Predicting information diffusion on Twitter a deep learning doi.​org/​10.​1287/​mksc.​2019.​1192
neural network model using custom weighted word features. McGrath, R. G. (1999). Falling forward: real options reasoning and
Responsible Design, Implementation and Use of Information and entrepreneurial failure. Academy of Management Review, 24(1),
Communication Technology, 456–468. https://​doi.​org/​10.​1007/​ 13–30. https://​doi.​org/​10.​5465/​amr.​1999.​15804​38
978-3-​030-​44999-5_​38 Mcilroy, D., Sadler, C., & Boojawon, N. (2007). Computer phobia
Kushwaha, A. K., Kar, A. K., & Vigneswara Ilavarasan, P. (2020b). and computer self-efficacy: Their association with undergradu-
Predicting information diffusion on Twitter a deep learning ates’ use of university computer facilities. Computers in Human
neural network model using custom weighted word features. Behavior, 23(3), 1285–1299. https://​doi.​org/​10.​1016/j.​chb.​2004.​
Responsible Design, Implementation and Use of Information and 12.​004
Communication Technology, 456–468. https://​doi.​org/​10.​1007/​ Meinhart, W. A. (1966). Artificial intelligence, computer simulation of
978-3-​030-​44999-5_​38 human cognitive and social processes, and management thought.
Kushwaha, A. K., Mandal, S., Pharswan, R., Kar, A. K., & Ilavarasan, Academy of Management Journal, 9(4), 294–307. https://d​ oi.o​ rg/​
P. V. (2020c). Studying online political behaviours as rituals: a 10.​5465/​254948
study of social media behaviour regarding the CAA. In Sharma, Meske, C., Bunde, E., Schneider, J., & Gersch, M. (2020). Explain-
S. K., Dwivedi, Y. K., Metri, B., & Rana, N. P. (Eds.), Re-imag- able artificial intelligence: objectives, stakeholders, and future
ining Diffusion and Adoption of Information Technology and research opportunities. Information Systems Management, 0(0),
Systems: A Continuing Conversation (pp. 315–326). Springer 1–11. https://​doi.​org/​10.​1080/​10580​530.​2020.​18494​65
International Publishing. https://​doi.​org/​10.​1007/​978-3-​030-​ Metcalf, L., Askay, D. A., Rosenberg, L. B., Askay, D. A., & Rosen-
64861-9_​28 berg, L. B. (2019). Keeping humans in the loop: pooling knowl-
Kushwaha, A. K., Kar, A. K., & Dwivedi, Y. K. (2021a). Applications edge through artificial swarm intelligence to improve business
of big data in emerging management disciplines: A literature decision making—Metcalf, L., Askay, D. A., Rosenberg, L. B..
review using text mining. International Journal of Information California Management Review, 61(4), 84–109
Management Data Insights, 1(2), 100017. https://​doi.​org/​10.​ Mohamed Ridhwan, K., & Hargreaves, C. A. (2021). Leveraging Twit-
1016/j.​jjimei.​2021.​100017 ter data to understand public sentiment for the COVID-19 out-
Kushwaha, A. K., Kar, A. K., & Ilavarasan, P. V. (2021b). Predict- break in Singapore. International Journal of Information Man-
ing retweet class using deep learning. Trends in Deep Learning agement Data Insights, 1(2), 100021. https://​doi.​org/​10.​1016/j.​
Methodologies, 89–112. https://​doi.​org/​10.​1016/​B978-0-​12-​ jjimei.​2021.​100021
822226-​3.​00004-0 Morikawa, M. (2017). Firms’ expectations about the impact of ai and
Kushwaha, A. K., Kumar, P., & Kar, A. K. (2021c). What impacts robotics: Evidence from a survey. Economic Enquiry, 55(2),
customer experience for B2B enterprises on using AI-enabled 1054–1063
chatbots? Insights from Big data analytics. Industrial Market- Nair, R. S., Agrawal, R., Domnic, S., & Kumar, A. (2021). Image min-
ing Management, 98, 207–221. https://​doi.​org/​10.​1016/j.​indma​ ing applications for underwater environment management—A
rman.​2021.​08.​011 review and research agenda. International Journal of Informa-
Kushwaha, A. K., Pharswan, R., & Kar, A. K. (2021d). Always Trust tion Management Data Insights, 1(2), 100023. https://d​ oi.o​ rg/1​ 0.​
the Advice of AI in Difficulties? Perceptions Around AI in Deci- 1016/j.​jjimei.​2021.​100023
sion Making. In Dennehy, D., Griva, A., Pouloudi, N., Dwivedi, Neogi, A. S., Garg, K. A., Mishra, R. K., & Dwivedi, Y. K. (2021).
Y. K., Pappas, I., & Mäntymäki, M. (Eds.), Responsible AI and Sentiment analysis and classification of Indian farmers’ protest
Analytics for an Ethical and Inclusive Digitized Society (pp. using twitter data. International Journal of Information Man-
132–143). Springer International Publishing. https://​doi.​org/​10.​ agement Data Insights, 1(2), 100019. https://​doi.​org/​10.​1016/j.​
1007/​978-3-​030-​85447-8_​12 jjimei.​2021.​100019
Lakhiwal, A., & Kar, A. K. (2016). Insights from Twitter Analytics: Newell, A., Shaw, J. C., & Simon, H. A. (1959). Report on a general
Modeling Social Media Personality Dimensions and Impact of problem solving program. International Conference on Informa-
Breakthrough Events. In Dwivedi, Y. K., Mäntymäki, M., Ravis- tion Processing, 256–264
hankar, M. N., Janssen, M., Clement, M., Slade, E. L., Rana, N. Newell, A., & Simon, H. (1956). The logic theory machine—A com-
P., Al-Sharhan, S., & Simintiras, A. C. (Eds.), Social Media: The plex information processing system. IRE Transactions on Infor-
Good, the Bad, and the Ugly (pp. 533–544). Springer Interna- mation Theory, 2, 61–79
tional Publishing. https://d​ oi.o​ rg/1​ 0.1​ 007/9​ 78-3-3​ 19-4​ 5234-0_4​ 7 Rahwan, I., Cebrian, M., Obradovich, N., Bongard, J., Bonnefon, J.-F.,
Lindebaum, D., Vesa, M., & den Hond, F. (2019). Insights from “The Breazeal, C., Crandall, J. W., Christakis, N. A., Couzin, I. D.,
Machine Stops” to better understand rational assumptions in Jackson, M. O., Jennings, N. R., Kamar, E., Kloumann, I. M.,
algorithmic decision making and its implications for organiza- Larochelle, H., Lazer, D., McElreath, R., Mislove, A., Parkes,
tions. Academy of Management Review, 45(1), 247–263. https://​ D. C., Pentland, A. ‘Sandy,’ … Wellman, M. (2019). Machine
doi.​org/​10.​5465/​amr.​2018.​0181 behaviour. Nature, 568(7753), 477–486. https://d​ oi.o​ rg/1​ 0.1​ 038/​
Llewellyn, C., Grover, C., Alex, B., Oberlander, J., & Tobin, R. (2015). s41586-​019-​1138-y
Extracting a topic specific dataset from a Twitter archive. In Raisch, S., & Krakowski, S. (2020). Artificial intelligence and manage-
S. Kapidakis, C. Mazurek, & M. Werla (Eds.), Research and ment: the automation-augmentation paradox. Academy of Man-
Advanced Technology for Digital Libraries (pp.  364–367). agement Review. https://​doi.​org/​10.​5465/​2018.​0072

13
Information Systems Frontiers

Rajendran, D. P. D., & Sundarraj, R. P. (2021). Using topic models Venkatesh, V., & Bala, H. (2008). Technology acceptance model 3 and
with browsing history in hybrid collaborative filtering recom- a research agenda on interventions. Decision Sciences, 39(2),
mender system: Experiments with user ratings. International 273–315. https://​doi.​org/​10.​1111/j.​1540-​5915.​2008.​00192.x
Journal of Information Management Data Insights, 1(2), Venkatesh, V., & Davis, F. D. (2000). A theoretical extension of the
100027. https://​doi.​org/​10.​1016/j.​jjimei.​2021.​100027 technology acceptance model: four longitudinal field studies.
Rathore, A. K., Kar, A. K., & Ilavarasan, P. V. (2017). Social media Management Science, 46(2), 186–204. https://​doi.​org/​10.​1287/​
analytics: literature review and directions for future research. mnsc.​46.2.​186.​11926
Decision Analysis, 14(4), 229–249. https://​doi.​org/​10.​1287/​ Venkatesh, V., & Speier, C. (1999). Computer technology training in
deca.​2017.​0355 the workplace: a longitudinal investigation of the effect of mood.
Reynolds, M., & Vince, R. (2004). Critical management education Organizational Behavior and Human Decision Processes, 79(1),
and action-based learning: synergies and contradictions. Acad- 1–28. https://​doi.​org/​10.​1006/​obhd.​1999.​2837
emy of Management Learning & Education, 3(4), 442–456. Vimalkumar, M., Sharma, S. K., Singh, J. B., & Dwivedi, Y. K. (2021).
https://​doi.​org/​10.​5465/​amle.​2004.​15112​552 ‘Okay google, what about my privacy?’: User’s privacy percep-
Riley, T. (2018). Get ready, this year your next job interview may tions and acceptance of voice based digital assistants. Computers
be with an A.I. robot. CNBC. https://​www.​cnbc.​com/​2018/​03/​ in Human Behavior, 120, 106763. https://​doi.​org/​10.​1016/j.​chb.​
13/​ai-​job-​recru​iting-​tools-​offer​ed-​by-​hirev​ue-​mya-​other-​start-​ 2021.​106763
ups.​html von Krogh, G. (2018). Artificial intelligence in organizations: new
Schmitt, B. (1999). Experiential marketing. Journal of Marketing opportunities for phenomenon-based theorizing. Academy of
Management, 15(1–3), 53–67. https://​doi.​org/​10.​1362/​02672​ Management Discoveries, 4(4), 404–409. https://​doi.​org/​10.​
57997​84870​496 5465/​amd.​2018.​0084
Schuetz, S., & Venkatesh, V. (2020). The rise of human machines: Wang, Y., Meister, D. B., & Gray, P. H. (2013). Social influence and
how cognitive computing systems challenge assumptions of knowledge management systems use: evidence from panel data.
user-system interaction. Journal of the Association for Infor- MIS Quarterly, 37(1), 299–313
mation Systems, 21(2), 460–482 West, M. A., & Farr, J. L. (1989). Innovation at work: Psychological
Seufert, S., Guggemos, J., & Sailer, M. (2020). Technology-related perspectives. Social Behaviour, 4(1), 15–30
knowledge, skills, and attitudes of pre- and in-service teach- Woodman, R. W., Sawyer, J. E., & Griffin, R. W. (1993). Toward a
ers: The current situation and emerging trends. Computers in theory of organizational creativity. The Academy of Management
Human Behavior, 106552. https://​doi.​org/​10.​1016/j.​chb.​2020.​ Review, 18(2), 293–321. https://​doi.​org/​10.​2307/​258761 JSTOR
106552 Yuan, F., & Woodman, R. W. (2010). Innovative behavior in the work-
Sharma, S. K., Sharma, H., & Dwivedi, Y. K. (2019). A hybrid SEM- place: the role of performance and image outcome expectations.
neural network model for predicting determinants of mobile The Academy of Management Journal, 53(2), 323–342 (JSTOR)
payment services. Information Systems Management, 36(3), Zhao, Y., Yang, S., Narayan, V., & Zhao, Y. (2013). Modeling con-
243–261. https://​doi.​org/​10.​1080/​10580​530.​2019.​16205​04 sumer learning from online product reviews. Marketing Science,
Sharma, S. K., & Sharma, M. (2019). Examining the role of trust 32(1), 153–169. https://​doi.​org/​10.​1287/​mksc.​1120.​0755
and quality dimensions in the actual usage of mobile banking
services: An empirical investigation. International Journal of Publisher’s Note Springer Nature remains neutral with regard to
Information Management, 44, 65–75. https://​doi.​org/​10.​1016/j.​ jurisdictional claims in published maps and institutional affiliations.
ijinf​omgt.​2018.​09.​013
Sharma, S., Rana, V., & Kumar, V. (2021). Deep learning based seman-
tic personalized recommendation system. International Journal
of Information Management Data Insights, 1(2), 100028. https://​ Arpan Kumar Kar  Associate Pro-
doi.​org/​10.​1016/j.​jjimei.​2021.​100028 fessor in Information Systems in
Sheridan, C. (2004). A taste of the future. Nature Biotechnology, Indian Institute of Technology
22(10), 1203–1205. https://​doi.​org/​10.​1038/​nbt10​04-​1203 Delhi, India. His research inter-
Simon, H. A. (1987). Two heads are better than one: the collaboration ests are in the domain of data
between AI and OR. INFORMS Journal on Applied Analytics, science, machine learning, digi-
17(4), 8–15. https://​doi.​org/​10.​1287/​inte.​17.4.8 tal transformation, internet eco-
Simon, H. A. (1991). Bounded rationality and organizational learn- systems, social media, and ICT-
ing. Organization Science, 2(1), 125–134. https://​doi.​org/​10.​ based public policy. He has
1287/​orsc.2.​1.​125 authored over 150 peer reviewed
Sitkin, S. B. (1992). Learning through failure: the strategy of small articles and edited 7 research
losses. Research in Organizational Behavior, 14, 231–266 books. As of 2021, he has an H
Stephan, M., Brown, D., & Erickson, R. (2017). Talent acquisition index of 35 and I-10 index
through predictive hiring | Deloitte Insights. https://​www2.​deloi​ exceeding 80 in Google Scholar
tte.​com/​us/​en/​insig​hts/​focus/​human-​capit​al-​trends/​2017/​predi​ with over 4000 citations. He is
ctive-​hiring-​talent-​acqui​sition.​html the Editor in Chief of Interna-
Taylor, S. E. (1991). Asymmetrical effects of positive and negative tional Journal of Information Management Data Insights, published by
events: The mobilization-minimization hypothesis. Psychologi- Elsevier. He further supports established journals like Journal of Public
cal Bulletin, 110(1), 67-85 Affairs, International Journal of Electronic Government Research,
Thumin, F. J., & Thumin, L. J. (2011). The measurement and inter- Information Systems Frontiers and Global Journal of Flexible Systems
pretation of organizational climate. The Journal of Psychology, Management as associate or coordinating editor. He is also on the edi-
145(2), 93–109. https://​doi.​org/​10.​1080/​00223​980.​2010.​538754 torial board of Journal of Business Research, The Bottom Line and
Trudel, R. (2019). Sustainable consumer behavior. Consumer Psychol- International Journal of Information Management. He has been confer-
ogy Review, 2(1), 85–96. https://​doi.​org/​10.​1002/​arcp.​1045 ence chair / track chair / associate editor for IFIP and AIS conferences.
Van de Ven, A. H. (1986). Central problems in the management of Prior to joining IIT Delhi, he has worked for IIM Rohtak, IBM
innovation. Management Science, 32(5), 590–607 (JSTOR)  Research and Cognizant. He has received numerous awards and

13
 Information Systems Frontiers

recognitions from organizations like IFIP, TCS, PMI, AIMS, IIT Delhi, has published more than 07 peer-reviewed papers in high-quality jour-
BK Birla (BimTech), NIT Goa, Elsevier and IIM Rohtak. nals and conference proceedings. He has received the best conference
paper award at the ISIC conference held in Delhi 2021. He has also
Amit Kumar Kushwaha  Amit works in the part-time capacity as a guest faculty in various educational
Kumar Kushwaha is a part-time institutes and is currently associated as well. In the past, he has con-
research scholar in Information ducted courses as “Into to machine learning”, “Mathematics for
Systems at the Indian Institute of machine learning”, ”Optimization techniques for machine learning”,
Technology Delhi, India, and a ”Soft computing”, ”Big-Data Analytics”, and ”Natural language pro-
full-time AI/ML practitioner cessing” in various institutes like the ”Indian Institutes of Information
with over 14 years in the indus- Technology Lucknow”, ”National Institute of Technology, Rourkela”,
try, currently working in the ”Delhi Technical University” to name a few on the list. He is also
capacity of data science and associated as a reviewer at the International Journal of Information
engineering team head at the Management Data Insights, published by Elsevier.
OTT platform wing of Dish Net-
work, USA. A gold-medallist
engineering graduate with an
Hons. degree, he completed his
Masters (MBA) from SIBM,
Pune. His interests lie in research
cutting across business applications of AI/ML innovations and produc-
tionalization of these solutions for the consumption of end users. He

13