Journal of Computer Information Systems

ISSN: (Print) (Online) Journal homepage: www.tandfonline.com/journals/ucis20

The Role of AI in Financial Services: A Bibliometric

Analysis

Oluwafemi Ayotunde Oke & Nadire Cavus

To cite this article: Oluwafemi Ayotunde Oke & Nadire Cavus (19 Jan 2024): The Role of AI in
Financial Services: A Bibliometric Analysis, Journal of Computer Information Systems, DOI:
10.1080/08874417.2024.2304545

To link to this article: https://doi.org/10.1080/08874417.2024.2304545

Published online: 19 Jan 2024.

Submit your article to this journal

Article views: 319

View related articles

View Crossmark data

Citing articles: 1 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=ucis20
JOURNAL OF COMPUTER INFORMATION SYSTEMS
https://doi.org/10.1080/08874417.2024.2304545

The Role of AI in Financial Services: A Bibliometric Analysis

Oluwafemi Ayotunde Oke and Nadire Cavus
Department of Computer Information Systems, Near East University, Nicosia, Turkey

ABSTRACT KEYWORDS
Financial Services (FS) and its integration of Artificial Intelligence (AI) has led to an improvement in Artificial intelligence;
various aspects of the field, such as fraud detection, investment performance, and risk manage­ financial service; fraud
ment. With a focus on the investment, insurance, and banking industries, this study implements detection; risk management;
a Bibliometric Analysis (BA) by extensively analyzing existing works of literature in the field which Investment performance
provides an overview of the roles of AI in FS. In achieving a comprehensive overview of the
objectives of the study, articles published in the Clarivate Web of Science (WoS) database from
1989 to 2023 were collected and analyzed using CiteSpace and VOS Viewer software applications.
The results from this BA show a growth in the interest rate scholars’ have toward the role of AI in FS.
One of the recommendations of the study is the integration of AI and human expertise in financial
decision making.

Introduction respect to customer data alongside a potential for biased

algorithms and how the adoption of AI in FS raises
Overtime, the adoption of Artificial Intelligence (AI)
concerns about potential risks and challenges.6
has seen a rise by various fields and industries including
Additionally, the integration of AI systems in FS has
finance. The investment, banking, and insurance aspects
shown to require specialized skills and expertise, as well
of the financial sector has experienced performance and
as significant investments in the technology and
productivity spikes as a result of the adoption and inte­
gration of AI.1 Hence, leading to an increase in research infrastructure.4 Hence, making it crucial to understand
interests by scholars.2 Use of AI in financial services the state of the art (SOTA) on the role of AI in FS and
(FS) has become increasingly important due to the ensuring that it is beneficial with minimized risks with
increasing complexities of financial products and ser­ some of the relevant theories supporting issues of
vices, as well as the velocity and of volume financial money laundering, smart contracts and banking activ­
data.3 As a result of efficiency, speed, and accuracy, AI ities with regards to AI in FS.2
has the potential to transform the finance industry in
terms of cost reduction and improved customers (i) Money Laundering: Few of the relevant theories
experience.4 For instance, personalized recommenda­ underpinning issues of money laundering
tions of products can be given to customers based on includes:
their preferences using AI-powered chatbots, and with ● Regulatory Arbitrage: This concept refers to
the use of ML algorithms, identification of patterns for a practice in which regulatory differences in
detecting fraudulent activities in financial data has been various jurisdictions are taken advantage of.
made possible.5 The bibliometric analysis (BA) in this Thus, individuals who are involved in money
study makes a significant contribution to literature laundering exploit financial regulations gaps
based on the roles AI play in FS by providing an over­ or loopholes across different countries for the
view of existing research in this field. The outcome of purpose of legitimizing illegal funds. Hence,
this contribution is the identification of key themes and the need arises for global cooperation in com­
trends in the literature, identification of influential bating money laundering.7
authors and publications, and examination with ● Rational Choice Theory: From the works of,4
a focus on how AI has improved the various finance it theorizes that individuals who engage in
fields. In this study, the analysis covers the current state crime such as money laundering, do so after
of research in the areas such as ethical issues with weighing their chances and potential threats

CONTACT Oluwafemi Ayotunde Oke [email protected]; [email protected] Department of Computer Information Systems, Near
East University, Cyprus via Mersin 10, Nicosia 99138, Turkey
© 2024 International Association for Computer Information Systems
2 O. A. OKE AND N. CAVUS

against the gains. This helps to understand the AI-driven algorithms in banking often rely
decision-making process of those individuals. on. These strategies seek to predict market
● Crime Opportunity Theory: This theory movements and exploit market inefficiencies.3
encompasses how people engage in criminal ● Modern Portfolio Theory (MPT): By analyz­
activities, such as money laundering as soon ing risk and return, the integration of AI in
as they feel less threatened of being caught banking often leverages MPT to optimize
with a high potential for financial profit. It investment portfolios. Asset allocation and
highlights criminal behavior and roles which risk management which all align with the
are pertinent in understanding how activities principles of MPT can be achieved by the
of money laundering are mitigated and assistance of AI systems.1
conducted.8
(ii) Smart Contracts: Some of the relevant theories The above theories provide frameworks for an under­
include: standing of the complex issues surrounding money
● Contract Law and Theory: Contract law and laundering, smart contracts, and the integration of AI
related theories makes the legal and contrac­ in banking activities within the financial sector and how
tual aspects supporting smart contracts they can be addressed, thus providing information on
a possibility. Smart contracts predefined con­ decision-making processes, regulatory efforts, and tech­
ditions, relying on principles of contract the­ nological advancements, aimed at tackling challenges
ory such as offer, acceptance, and and optimizing operations in these domains. The struc­
consideration are aimed to be automatically tural organization of this research paper is as follows.
executed.6 The methodological framework used in conducting the
● Game Theory: When it comes to smart con­ BA, followed by the result section of the analysis, with
tracts, game theory examines how the presentation of findings on the key themes, authors,
involved participants maximize their out­ publications, and trends in the literature. Finally, the
comes via their actions and strategic deci­ result of this study will be valuable for future research
sions. Thus, giving an understanding on the and developments by researchers, policymakers, and
behaviors of their interaction within smart practitioners who are interested in understanding the
contracts. Most especially in terms of trust current state of research on uses of AI in FS, as well as its
and potential vulnerabilities.4 potential impact on the industry. Overall, this biblio­
● Efficient Market Hypothesis (EMH): EMH metric analysis paper will provide a guideline for future
recommends all available information are research and policy makers through its contribution to
reflected on asset prices and when applied to ongoing research on AIFS roles.
smart contracts, this theory emphasizes the
notion that the terms and conditions within
Methodology
these contracts can foster trust and reliability
by accurately reflecting the participants’ Search strategy
intentions and agreements.9 In this study, a BA search strategy based on precise
(iii) Banking Activities and AI: Relevant theories indicators was developed using keywords search and
relating to banking activities include: terminologies that can be found in the literature. The
● Behavioral Finance: Behavioral finance plays
BA in this study provides an insight to the current state
a crucial role as its theories are used to of AI-based FS with the final search conducted on 7th
examine how psychological factors influence April, 2023 in Core Clarivate Collection of Web of
financial decisions and AI applications in Science databases for the retrieval of existing high-cited
banking take cues from these theories. and quality AIFS records. Only the Social Science
Thus, a proper understanding of human Citation (SSCI) and Science Citation Index Expanded
behaviors and biases will help AI systems (SCIE) indexes of WoS were selected after which different
in offering personalized risk management analysis such as spatial and temporal, word co-
and financial advice.5 occurrence, co-authorship, and co-country analysis were
● Algorithmic Trading Strategies: Various trad­ conducted. The search keywords related to AI in FS were
ing strategies based on financial theories like coined from preliminary literature and based on the
technical analysis, fundamental analysis, and aspect of FS intended. Thus, a search query was formed
quantitative finance are benchmarks in which using the combination of search keywords and Boolean
 JOURNAL OF COMPUTER INFORMATION SYSTEMS 3

operators (AND & OR): ((“artificial intelligence” OR proceed. Records excluded based on all open access
“machine learning” OR “deep learning”) AND (“financial screening was 637 leaving the new record to 1067.
service” OR “banking” OR “insurance” OR “investment”) Afterwards, screening using publication, record, and
AND (“application” OR “risk management” OR “fraud access type was carried out with 531 meeting the cri­
detection” OR “investment performance”)). With search teria, leading to the removal of 536 records. To further
filters such as indexing and timespan, there was no time­ streamline the records, the database indexing strategy
span limit for the initial search in this study as the time­ was implemented. The records were indexed to only
span was set as ALL from 1989–2023 (34 years), the initial SCIE and SSCI publications, leading to the removal of
database indexes was set to ALL indexing types, and other 115 records. Final records included in the study for the
filters were set to default. Hence, a total of 1704 published BA was 416 records. However, in contrast to
records were registered and retrieved from the search and a systematic literature review, in a BA, only records
accessed for further processing. screening of the abstract is required except in scenarios
where full text when is necessary.10 Hence, to ensure the
included records are a good fit for the study, their titles
Screening strategy
and abstracts were screened. Every record in terms of
In this phase, the 1704 records gotten from the search titles and abstract were correlating to the objectives of
strategy for AI in FS were included for screening. The the study. Since there was no change, the final records in
screen strategy entailed different criteria for inclusion the study retained the same value of 416 records as
and exclusion. Although, despite setting the initial shown in Figure 1.
search year to 1989, the retrieved publication records
in AIFS started in 1992. The inclusion criteria for
retrieved records had to meet and satisfy the following Bibliometric analysis
requirement: i. written in English language, ii. a Journal
record, iii. an open access only record, iv. a review and Bibliometric according to Oxford Dictionary of English
research record, v. in the computer and business field vi. is the statistical analysis of books, articles, and publica­
Within 5-year range. The growth and improvement of tions in general.11 Also, according to,12 BA is the quan­
AI has led to the significant progress in other field as titative study of publications methodological approach
well as there have been lots of publications in the to analytically measure and discover trends and devel­
research areas. In this study, only peer review journals opments in a research area to get insights on objectible,
were included after which the authors carried out a BA obtainable, quantifiable, and reproducible data. The BA
screening. Categorization of papers based on search in this study computes different matrices such as growth
strategy screening process were based on certain inclu­ rates of publications, research characteristics such as
sion and exclusion criteria. Before, during and after the keywords and topics, patterns of publication such as
categorization screening, there were discussions and journals and countries, and research hotspot tendencies
a consensus to be for each record to meet in order to such as map timelines and citation.

Search results from Web of

Science (n=1704)

Records excluded based n

screening (n=1067)

Records after screening inclusion

(n=531)
Records removed based on
indexing (n=115)

Final records included for

Bibliometric analysis (n=416)

Figure 1. Flowchart detailing the record retrieval, screening and selection criteria.
4 O. A. OKE AND N. CAVUS

Growth rate of publications keywords. Trending topics such as fraud detection and
machine learning under FS and AI technology, respec­
To get publications growth rate percentage, a mathematical
tively, were identified and defined using percentage,
formula was implemented to get the growth over time. The
citation rate, and frequency of keywords used by the
formula was calculated by raising “the number of current
authors. The keyword centralities were computed using
publication rate as at 7th April 2023” divided “the number
CiteSpace software based on statistical equations com­
of publications since 1st January 1989” raise to the power of
bination that measures the selected words text content
“1/the differences in current and first year (1/34)” multi­
occurrences based on degree of closeness, clustering
plied by 100, as shown below.
coefficient, frequency scores eccentricity, PageRank
Growth rate ¼ ððCurrent year publication number (Google LLC) etc.13
�first year number of publicationsÞ
1=ðcurrent year first yearÞ
1Þ � 100:
Research hotspot tendencies
Trends in terms of publications number for each year
The CiteSpace software was used to develop a timeline
were also reported.
map and citation bursts. The publications between 1989
and 2023 were used in the generation of cluster view,
Publication patterns and the keywords in the records were used in the label­
ing of each cluster. The duration set for the time slice
For the publication record patterns, different top cita­
was 1 year. The threshold interpolation of co-
tion trends were analyzed. They include the top 10
authorship, co-occurrence, and co-citations, citation,
research domains, journals, and citations. The publica­
and bibliographic coupling were set to a minimum of
tions statistical mode and percentages in each country
5 documents respectively.
and journal were analyzed using year of publication.
The authors affiliations were used in the identification
of the listed countries. Using the citation rates, the Results
research impact rankings of each country and journal
were extracted using Mendeley Software. Growth rate of publications
The plotted graph of yearly citation rating and trends of
the publications on AI in FS is shown is Figure 2. Using
Characteristics of research activities
the growth rate formula in section 3.1, the growth rate of
Analysis of research characteristics activities was carried research publications in AI-based financial service
out based on the publication records topics and the research from 1989 to 2023 is 120%. From a total of

Figure 2. Yearly bibliographic records distribution.

 JOURNAL OF COMPUTER INFORMATION SYSTEMS 5

416 publications, 3 publications were published between a visualization of other strong collaborative countries
1992 till 2008, making up 0.7211% of publications in the like Cyprus and Iran which are also from Europe and
first 16 years. However, 42 citations were recorded in Asia, respectively.
those years. From 2009 till 2018 (9 years) began the Further to the analysis of the publications and as
gradual growth of interest of researchers in the area of shown in Table 2, out of 205 Journals present in the
AIFS with 37 publications, making up 8.8942% of study, IEEE Access holds the record for the most pub­
research in AIFS, and an average of 4.1 records not lished journal in the area of AIFS research with a record
exceeding 20 publication mark were carried out by of 36 publications and 8.6% of the entire journal pub­
researchers. But from 2019 till 2023 (4 years), the years lications present in the study. Followed by Sustainability
recorded 26, 70, 119, 145, 16, publications, respectively, Journal with 23 records and 5.529% of the publications.
passing the minimum 20 publication marker and sum­ The minimum number of documents from the to 10
ming up to 376 publications and 90.3846% of the total journals is 6 records per journal. Different domain cate­
publication since inception in 1992. These past 4 years gories exist in WoS database. However, for the focus on
have shown how the interest rate of researchers have this study, the specific domains of publications in AIFS
drastically increased with each year getting more pub­ is analyzed with Computer Science domain ranking as
lications than the former. first with a frequency of 134 making up 32.212% of the
entire 416 publications. This is followed by Engineering,
Telecommunications, Science technology other topics,
Publication patterns Mathematics, Environmental sciences ecology, Business
With 76 countries present in the study, the top 10 economics, Chemistry, Physics, and Materials science.
countries bibliographic distribution of publication is From the domains, it can be deduced that the area of
shown in Table 1. From the table, the represented Computer Science is a major domain based on Artificial
countries all have a minimum of 17 document pub­ intelligence. Also, from the FS standpoint, Business
lications, with Peoples Republic of China ranking as economics is ties with that domain with a frequency of
number 1 and responsible for impacting 27.163% of 40 and 9.615% of the total publications as shown in
the research in AIFS. The top 10 countries according Table 3.
to ranking are Peoples Republic of China, USA,
England, Germany, South Korea, Taiwan, Australia,
India, Saudi Arabia, andSpain. With 5 of the top 10 Characteristics of research activities
countries originating from Asia continent, the conti­
nent dominates the list, followed by Europe continent The essential words given by the researchers in their
with 3 countries. North America and Oceania both study are keywords which would help in the easy iden­
have 1 country respectively. In terms of citation rank­ tification and access to their records. The top 10 key­
ing, USA tops the chart being the most cited country words with the most occurrences in the study are
with 1810 citations. No country amongst the top 10 machine learning, artificial intelligence, deep learning
countries have less than 100 citations. Furthermore, (DL), prediction, classification, model, performance,
a cluster analysis of countries that collaborated neural networks, system, and algorithm. Table 4 and
together on research in the area of AIFS is shown in Figure 4 show information about all the keywords co-
Figure 3 using CiteSpace software. It can be seen that occurrences and their cluster arrangement respectively
there exist strong collaborations between European in the study. The top 5 AI keywords include algorithm,
and Asian Countries with USA being the only North classification, machine learning, prediction, deep learn­
American Country in the mix. Also, there is ing. The top 5 FS keywords are investment, big data,
fraud detection, management, and risk. These keywords
were retrieved by adjusting the threshold parameters of
Table 1. Top 10 countries bibliographic distribution of records.
the keywords using VOS Viewer software. Out of the
Rank Countries Publications % of 416 Citations
2342 keywords being the total number of keywords in
1. PEOPLES R CHINA 113 27.163 576
2. USA 75 18.029 1810 this study, 15 keywords met the threshold with the
3. ENGLAND 37 8.894 437 settings parameter of making the minimum number of
4. GERMANY 29 6.971 435
5. SOUTH KOREA 25 6.010 122 occurrences of a keyword to be 16. Figure 4, however,
6. TAIWAN 24 5.769 140 shows the visual representation of Table 4 and the
7. AUSTRALIA 23 5.529 435
8. INDIA 21 5.048 297 relationship that exists in their co-occurrence. The gen­
9. SAUDI ARABIA 21 5.048 117 eration of the clusters were based on keywords occur­
10. SPAIN 20 4.808 245
rence in a research domain.
6 O. A. OKE AND N. CAVUS

Figure 3. Cluster analysis of countries that collaborated (CiteSpace).

Table 2. Top 10 journal bibliographic distribution of records. Research hotspot tendencies

Record % of
Publication Titles Count 416 Citation Major milestones in the study in terms of development
IEEE ACCESS 36 8.654 320 of AI in FS were presented the by analyzing the keyword
SUSTAINABILITY 23 5.529 146 list with citations with strong bursts between 1995 to
APPLIED SCIENCES BASEL 19 4.567 106
COMPUTATIONAL INTELLIGENCE AND 14 3.365 7 2023 as shown in Table 5. Using CiteSpace software with
NEUROSCIENCE parameter settings of g-index=25, top N (selecting the
MATHEMATICS 14 3.365 86
SENSORS 12 2.885 55 top 15 keywords) = 15 and years per slice = 1 year, five
PLOS ONE 9 2.163 114 prominent keywords (algorithm, machine learning,
MOBILE INFORMATION SYSTEMS 8 1.923 4
ELECTRONICS 6 1.442 18 deep learning, model, and big data) were generated.
INTERNATIONAL JOURNAL OF 6 1.442 27 Algorithm burst was first in 2010. However, its most
ENVIRONMENTAL RESEARCH AND PUBLIC
HEALTH
recent burst is from 2019 till 2021. Machine learning
burst was initially in 2014 and deep learning in 2018.
Both have their recent bursts from 2020 till 2023.
Furthermore, Model with an initial burst in 2011 and
big data with previous burst in 2021, both have their
Table 3. Top 10 research domains bibliographic distribution of
recent burst since 2021 till 2023. However, the keyword
records.
with the most burst strength is machine learning with
Rank Research Areas Frequency % of 416
1. COMPUTER SCIENCE 134 32.212
a strength of 10.76. which also doubles alongside deep
2. ENGINEERING 125 30.048 learning as the most recent keywords as at the time of
3. TELECOMMUNICATIONS 54 12.981 this research study in April 2023. Both have the recent
4. SCIENCE TECHNOLOGY OTHER TOPICS 48 11.538
5. MATHEMATICS 47 11.298 longest citation burst run. In addition, a visualization of
6. ENVIRONMENTAL SCIENCES ECOLOGY 44 10.577 the keywords timeline was generated to show the key­
7. BUSINESS ECONOMICS 40 9.615
8. CHEMISTRY 36 8.654 words cluster trends since 1995 till April, 2023 along
9. PHYSICS 32 7.692 a horizontal axis as shown in Figure 5. With the year
10. MATERIALS SCIENCE 22 5.288
shown on top of the visualization, the clusters we
 JOURNAL OF COMPUTER INFORMATION SYSTEMS 7

Table 4. Co-occurrences keywords.

All keywords

Cluster 1 Cluster 2 Cluster 3 Cluster 4

Algorithm Artificial intelligence Classification Machine learning
Investment Big data Deep learning Models
Model Framework Fraud detection Prediction
Neural-networks Impact System Regression
Performance Management
Selection Risk

Figure 4. Cluster arrangement of all keywords.

Table 5. Strongest citation bursts of top 5 keywords.

Keywords Year Strength Begin End 1995–2023
algorithm 2010 4.35 2019 2021
machine learning 2014 10.76 2020 2023
deep learning 2018 6.58 2020 2023
model 2011 5.94 2021 2023
big data 2021 4.03 2021 2023

analyzed and listed vertically on the right in keywords are in information theory, ensemble learning,
a descending order based on timeline occurrence ana­ data mining, anomaly detection, accident situation, and
lysis from left to right. Co-citation of keywords along robust optimization, with communication being the
with its corresponding year are represented with colored focus area on ensemble learning.
curves. Also, the specific nodes represent particular level
of interest based on citation burst ratings. Hence, the
larger the nodes, the bigger the interest of researchers. It Discussion
can further be seen from Figure 5 that in 1995, informa­
tion theory focused on behavior in AIFS research. The implementation of a BA in this study for financial
However, since then, there was a research gap until service AI-based research has led to the exploration of
2010 with algorithms and classification being the focus publications growth rate, publication patterns,
area in information theory and imputation respectively. research hotspot tendencies, and research activities
However, the current areas n term of timeline of characteristics.
8 O. A. OKE AND N. CAVUS

Figure 5. Timeline visualization of keywords.

Publications growth rate However, it is not regarded among the years of bare
minimum since the research was conducted in the first
Top years (10) in terms of publication and growth
quarter of 2023. Hence, there’s still room for growth.
percentage as shown in Figure 2 are 2023 (19, 4.567%),
Given from the growth of publication in previous years,
2022 (143, 34.375%), 2021 (118, 28.365), 2020 (70,
the growth publication projection of 2023, however, is
16.827%), 2019 (26, 6.250%), 2018 (8, 1.923%), 2017
(10, 2.404%), 2016 (5, 1.202%), 2014 (4, 0.962%), and estimated to surpass the former 140 thresholds of 2022
2013 (3, 0.721%). Since the first publication in 1992, FS and get to a growth of 160 publications. Furthermore,
AI-based research had a slow increase for over 20 years. multiple reasons are responsible for the recent rapid
However, in 2013, there was an awakening which led to growth such as AI technological breakthroughs in the
the slight significant rise in number of publications. time period and contributing to its adoption in FS AI-
This slight growth was at its bare minimum for 5 years based research.5 Also, Fintech firms growth are on the
(in 2018) with less than 20 publications in those years. rise to challenge existing banks.14 Automation of FS
Although, this hiatus came to an end in 2019 with the processes is possible by AI,1 increase efficiency,15 cost
growth rate of publication exceeding the bare minimum reduction,4 and customer service improvement.2 Tasks
of 20 publications in the field of AIFS. The year 2019 such as predictive modeling by AI are already imple­
was the beginning of a new dawn for research in AI&FS. mented in baking sector16 and fraud detection.7,17 In
The grown since then skyrocketed and have seen expo­ banking, conversational systems such as chatbots pow­
nential increases with the latter years greater than the ered by image recognition,18 and NLP now help con­
former. Hence, within the bare minimum years from sumers with transactional inquiries and provide
1992 till 2018, they make up 9.616% of the entire pub­ human-like responses to frequently asked questions
lications till date. However, within the skyrocketed four involving bank products, financial and non-financial
years from 2019–2023, there has been 90.384% publica­ transactions.19 Investing in machine learning based
tion growth rate, which is nine times more than the solutions in FS can provide a competitive edge to orga­
growth rate from 1992 to 2018. Also, as at the time of nizations over rivals.4,20 Furthermore, given the growth
conducting this research, the year 2023 from Figure 2 is of financial service AI-based research, the prospective
also in the bare minimum as at the time of this research. publications will see exponential growth in the future.21
 JOURNAL OF COMPUTER INFORMATION SYSTEMS 9

Patterns of publication the role of AI in FS, there are areas of FS that the
integration of AI has led to increase in performances
AI&FS has spiked the interests of many across the globe
such as banking,2 insurance,8 and investment
with the top and high-income world countries being the
industries.20 As there are pros in its integration, there
main forces in FS AI-based research. Being one of the
are also cons. According to,6 there are risks associated
most powerful countries in the world, The Peoples
with the use of AIFS in banking, such as customer low
Republic of China contributed to a huge number of
or zero understanding of complexities in the AI-based
research with 113 publications, making up 27.163% of
FS solutions which are difficult to understand. Iit was
the entire research records in the field. Every country in 22
also pointed out the risks of lack of adequate informa­
the top 10 country list s from a developed country.
tion and regulation in automated consulting of insur­
Hence, in respect with the rapid advancement of finan­
ance FS. In investment, problems of digitization
cial service AI-based research, the low income countries
involving the use of new channels of communication
have very little contributions to the field. But with
like smartphones and the issues of their screens being
collaboration of developing countries with developed
too small. This can lead to the customers missing out on
countries, the developed countries would definitely
investments opportunities and pre-contractual informa­
influence putting the developing countries on the map
tion such as updates on new sales was stated by.6 In
and on spotlight. Furthermore, from the curations
addition, other issues in FS includes inflation,4 violation
standpoint, The United States of America impacted
of principles of privacy which can lead to customer
the research area by being a major course of reference
records manipulation,2 improper revenue
and contributing 1810 citations to the field. In terms of
recognition, overstatement of assets,1 understatement
14
journals and sources, the top percentage of the journals
of liabilities,22 misappropriation of assets and improper
in terms of publications and citations are from the
disclosure in fraud detection,17 misinterpretation of
sciences, with IEEE Access journal having the most
data in prediction,15 lack of accountability in
records for 36 publications and 320 citations. Making
management,23 and misunderstanding of performance
up 8.654% of the entire publications. Given the publica­
and financial information which leads to bad market
tion growth of the records, with respect to the journal
timing in risks.19
implies that majority of researchers prefer to have their
In the banking sector, customer analysis is one of the
research submitted to this journal given it is a high
biggest problems for analyzing the loan defaults or for
quality journal. This might be as a result of high accep­
detecting any fraud transaction according to.7 However,
tance rate, fast processing or low article processing
these problems can be eliminated and customers con­
charges. From citation counts, AI-based research is
fidence can be increased as Blockchain-based payments
widely accepted by top journals in business and com­
fixes that Hence, bringing transformation to the bank­
puter science with ongoing research in the field leading
ing sector in alternative ways asides traditional bank
to ground-breaking improvement and discoveries that
money transfers.4 In the case of customer experiences
has led to the development of AI.9 From the study, the
in the banking sector,2 states that churn rates can be
research domain of computer science had the highest
controlled by seamlessly integrating mandatory intelli­
frequency of publications with 134 publications and
gent services and deregulating the market, leading to the
32.212% of publications. Business economics domain
amplification for the need of technology in the banking
was also present with 40 records and 9.615% of the
sector prospectively. In terms of banking in India,
publications. The results of these domains can be attrib­
according to research conducted in 2021 by,19 a record
uted to the keywords findings from the included
of 36% large financial institutions have started investing
records.
in these technologies and other financial institutions
have proposed to invest in the future. Also, AI being
the future of banking provides resources to rapidly
Characteristics of research activities
detect and combat fraudulent transactions with also
In terms of AI publications characteristics, AI is dis­ resources to analyze data more efficiently. Thus,
rupting and redefining the way FS operations are being improving compliance. Also, cybersecurity implemen­
carried out across diverse levels. From the cluster cate­ tation in the insurance sector of FS can lead to exclusion
gories identified keywords, the top areas of FS in AI as a result of lack of specific digital skills in the analysis
research are investment, big data, fraud detection, pre­ of big data.6 However,3 pointed out that, in manage­
diction, management and risk. Solutions to many tradi­ ment, AI can also analyze data and identify patterns and
tional banking problems have led to the consideration of provide insights into project performance and trends,
AI-based solutions by the banking industry.6 Regarding allowing project managers to identify potential
10 O. A. OKE AND N. CAVUS

problems and prevent disruptions in their projects. In forests. The study further suggested the technique be
cyber security fraud detection, there will be several adopted and implemented in the future. However, in the
topics for integration of cyber security and AI technol­ case of big data techniques in financial banking crimes,
ogies can be building for Novel and special AI algo­ increasing integration of ML in future big data applica­
rithms in the future.7 In the investment industries,7 tions will be required to draw useful insights. The emer­
stated in a study recommending that the LightGBM gence of big data would lead to the following identity
algorithm built on the GBDT framework aims at theft, financial statement & insurance fraud, anti-money
improving efficiency computationally, especially with Laundering and anti-bribery and Corruption.24 In the
regards to prediction problems in big data. The author case of anti-money laundering, application of reinforce­
further stated that use of predictive models based on ML ment learning in model training is recommended given
techniques in analysis can be used to create insights to how agile financial operation can be.21 Also, in the case
predict future risks probabilities and also in the detec­ of card fraud detection, the implementation of hybrid
tion of unnatural pattern in data. In terms of forecasts models such as CatBoost, hyperparater tunings and the
and predictions, studies of9 shows that investors gener­ use of stronger and better hardware may bring better
ally have different tolerance thresholds. Hence, the need results according to.17
for a future dividend predictor.7 suggested in the study Thread classification and investment prediction are
that the use of knowledge integration model ranked the common AI technologies applications in FS.5 Use of AI
best predictor in the prediction of future dividends for in investment performance which would be helpful to
customers regardless of their levels of tolerance. A study investors according to8,9 is that accurate forecasting of
by22 also states that accurate prediction and forecast of investments can be made possible by combining AI and
cash flow can be made possible. And a study by7 states digital analytics as one of the important aspect of invest­
that further improvements in the technology will inevi­ ment for investors is timing in the past but with predic­
tably make smart investments spread to various finan­ tion of the future forecasts, that timing problem has
cial institutions. been taken care of. Furthermore, according to the
study of,9 in terms of volatility and liquidity in invest­
ments, certain models like fat tails and volatility cluster­
Research hotspot tendencies
ing can now understand and forecast financial markets
Over the last three decades, there has been variations in characteristics and evaluate their system properties.
terms of research topics with some keywords having Singh states that to draw useful insights from big data
consistencies of appearing over time while other key­ in the future, the integration of ML is a prerequisite to
words show up once in a while in the field. A keyword achieve the insights.
burst of multiple financial—related AI-based keyword According to,18 complex and essential AIFS issues
can be seen from 2019 to 2022 as shown in Figure 4. can be addressed and corrected if their ethical consid­
Investments in AI in the field of finance is leading to erations get reported. That way guiding future research­
rapid improvements of the FS system.5 The potential ers in the direction to steer on pressing issues. This will
impact of AI in FS research has led to the witnessing of require a consistent financial investment by institutions
an extensive acceptance in the classification, detection, who should commit to the notion of ever greening AI
prediction, regression, performance, management, and infrastructure for the future.25 The impact of govern­
selectin of FS anomalies.4 In this study, the most com­ mental regulations and managerial decisions on banks
monly used AI technique in FS is ML, followed by with risk-taking can be researched in future studies.2,20 also
deep learning. In terms of applications, applications of states that the lack of patenting the AI technologies
AIFS technology research area includes the use of AI to makes it a difficulty factor in terms of comprehensions
improve risk management in terms of operational and understanding of the technology. The author
Research and AI methods could be used to improve further suggested the implementation of non-patent AI-
efficiency forecasting. Also, in terms of machine learn­ related publications as a consideration in journals which
ing model implementation, the fitness of the models can provide a comprehensive understanding on how the
used should be put into consideration.9 Furthermore, technology works from a user testing perspective.
Internet of Things (IoT) is crucial for future banking as According to,21 continuous research focused on
a result of AI and the Blockchain clusters thereby redu­ discovering a fast and reliable approach in detecting
cing professionalism problems and allowing regular money laundering in transactions can be achieved by
investors have access to good FS.22 implementing Graph mining model and Social net­
With the use of AI in fraud detection,22 study sug­ work analysis. Thus, the evaluation of the model will
gests that a ML techniques for detecting fraud is random be used in a large data also, the application of
 JOURNAL OF COMPUTER INFORMATION SYSTEMS 11

another Ensemble learning which is another AI tech­ Conclusions

nique is an alternative. Furthermore, in banking,
So far, in this research study, the entirety of the AI-
a suggestion from the study of17 is that the imple­
based FS service has been provided. Thus, providing
mentation of a second-generation gradient boosting
an extensive analysis of AI-based related research.
machines (GBM) known as XGBoost and LightGBM,
With different parameters in place such as multiple
performs better than the first-generation GBM
search and screening with 20 final search terms, and
ensemble method known as RFs. In addition, this
a retrieval year span of 34 years, from 1989–2023,
study recommends that in banking research, future
this study with a BA framework has been carried out
studies can also explore the role of unsupervised
to identify the inclusive AI-based FS. This study also
learning AI methods as the method is capable of
represented research trends of AI-based FS in terms
learning and drawing inferences on its own without
of: (1) rate of rapid growth over the past 5 years and
input from human. k-nearest neighbors’ algorithm
a trend of continuous growth since 2019; (2) high-
(KNN) model technique is recommended to achieve
income and 1st world countries being the main force
this.
in terms of research and development in AI-based
In future, use of self-executing contracts known as
FS; (3) great customer experiences were the motive
“smart contracts’’ would play an important role in
behind a vast majority of AI models research inte­
managing numerous events in the finance sector as the
gration into banking sector; (4) the AI techniques
smart contracts are built Blockchain framework. This
frequently used for detection, selection, prediction
leads to the automation of compliance human
and classification are DL and ML; (5) the ubiquitous
operations.4 For future research, the suggestion of the
use of Blockchain in the FS are currently research
hybridization of ML models and capturing of bank
hotspots of AI-based FS. AIFS is rapidly accelerating,
transaction variables. This would ensure high efficiency
with various demonstrations of potential applications
and accuracy while training and during testing. Hence,
across various business domains. An AI system in
leading to accuracy of predictions and an improved
the in the banking, insurance and investment indus­
reliable insights model result.16
try can examine millions and billions of datasets, and
find patterns, trends that humans miss, and even
predict future patterns. In addition, there are limits
Limitations to the current examples of such technology. For
For the progress of the research body, limitations in instance, the analytical themes drawn from addres­
this study needs to be discussed. Firstly, for the sing the frameworks on the complex issues sur­
analyses in this study, some relevant publications rounding the integration of AI in banking activities,
have been retrieved from a large single database, money laundering, and smart contracts within the
WoS. It is recommended that future research studies financial sector. Hence, tackling challenges and opti­
make use of other databases, such as Scopus, mizing operations in these domains by provision of
Dimensions, Lens and PubMed for retrievals of pub­ information on regulatory efforts, decision-making
lications. Secondly, “banking,” “insurance,” and processes, and technological advancements.
“investment” were the generic search keywords Blockchain being successfully deployed to improve
used. Hence, constrained to only a particular aspect risk management, detect fraud and improve invest­
of FS field. For instance, AI-based studies in ment performance. This study projects the use of
Regulated advice, Accounting, Life assurance pen­ multi-channel robots, the use of smart home mobile
sions etc. Third, the keywords ranking amongst devices secured using a customizable risk assessment
others, were gotten based on set metrics and para­ layer, and autonomous banking services as the future
meters on the implemented software. Hence, other of customer banking. However, research has to be
parameter tuning could have led to other analyzed done to advance the field for an understanding of
results. Fourth, gray literatures such as books and how future bank and their respective branches will
records published in languages other than English look based on the integration of new technologies to
were excluded from the research. Hence, it is possi­ bank applications; and how client/users/customers
ble that relevant studies conducted in these litera­ will engage physically and virtually with the SOTA
tures might have been missed. This study bank channels and FS as a whole. Finally, it is
recommends the broadening of the research scope recommended that future AI research work with
in terms of search and retrieval of relevant literatures existing FS before bridging the gaps with new AI
that can enrich FS as a whole in future studies. and SOTA technologies.
12 O. A. OKE AND N. CAVUS

Disclosure statement approach. Front Big Data. 2022:5. doi:10.3389/fdata.

2022.961039.
No potential conflict of interest was reported by the author(s). 9. Milana C, Ashta A. Artificial intelligence techniques in
finance and financial markets: a survey of the literature.
Strateg Change. 2021;30(3):189–209. doi:10.1002/jsc.
ORCID 2403.
10. Polanin JR, Pigott TD, Espelage DL, Grotpeter JK.
Oluwafemi Ayotunde Oke http://orcid.org/0000-0001- Best practice guidelines for abstract screening
9040-0531 large-evidence systematic reviews and
Nadire Cavus http://orcid.org/0000-0001-7470-7752 meta-analyses. Res Synth Methods. 2019;10
(3):330–42. doi:10.1002/JRSM.1354 .
11. Mayr P, Scharnhorst A. Scientometrics and informa­
Abbreviations tion retrieval: weak-links revitalized. Scientometrics.
2015 [accessed 2023 Apr 19];102(3):2193–99. doi:10.
AI artificial intelligence 1007/s11192-014-1484-3.
FS financial service 12. Donthu N, Kumar S, Mukherjee D, Pandey N,
AIFS artificial intelligence in financial service Lim WM. How to conduct a bibliometric analysis: an
SCI Science Citation Index overview and guidelines. J Bus Res. 2021;133:285–96.
SSCI Social Science Citation doi:10.1016/J.JBUSRES.2021.04.070.
WoS Web of Science 13. Vega-Oliveros DA, Gomes PS, Milios EE, Berton L. A
multi-centrality index for graph-based keyword
extraction. IInf Process Manag. 2019;56(6):102063.
References doi:10.1016/J.IPM.2019.102063.
1. Al-Ababneh HA, Borisova V, Zakharzhevska A, 14. Ashta A, Biot-Paquerot G. FinTech evolution: strategic
Tkachenko P, Andrusiak N. Performance of artifi­ value management issues in a fast changing industry.
cial intelligence technologies in banking Strateg Change. 2018;27(4):301–11. doi:10.1002/JSC.
institutions. WSEAS Trans Bus Econ. 2023;20 2203.
(December 2022):307–17. doi:10.37394/23207.2023. 15. Met I, Erkoc A, Performance SS. Efficiency, and target
20.29 . setting for bank branches: time series with automated
2. Vergallo R, Mainetti L. The role of technology in machine learning. IEEE Access. 2022;11(January).
improving the customer experience in the banking sec­ doi:10.1109/ACCESS.2022.3233529
tor: a systematic mapping study. IEEE Acces. 2022;10 16. Moreira MÂL, De Souza Rocha Junior C, Silva DFDL,
(November):118024–42. doi:10.1109/ACCESS.2022. De Castro Junior MAP, De Araújo Costa IP,
3218010. Gomes CFS, Dos Santos M. Exploratory analysis and
3. Dattatraya PL. A study of applications of artificial intel­ implementation of machine learning techniques for
ligence in banking and finance sector international predictive assessment of fraud in banking systems.
journal for innovative research in multidisciplinary Procedia Comput Sci. 2022;214(C):117–24. doi:10.
field a study of applications of artificial intelligence in 1016/j.procs.2022.11.156.
banking and finance sector. Article In Int J Innovative. 17. Hashemi SK, Mirtaheri SL, Greco S. Fraud detection in
2022;(May):35–37. www.ijirmf.com . banking data by machine learning techniques. IEEE
4. Javaid M, Haleem A, Singh RP, Suman R, Khan S. Acces. 2022;11(January):3034–43. doi:10.1109/
A review of blockchain technology applications for ACCESS.2022.3232287.
financial services. BenchCouncil Trans Benchmarks 18. Schwalbe N, Wahl B. Artificial intelligence and the
Stand Eval. 2022;2(3):100073. doi:10.1016/j.tbench. future of global health. Lancet. 2020;395
2022.100073. (10236):1579–86. doi:10.1016/S0140-6736(20)30226-9.
5. Fares OH, Butt I, Lee SHM. Utilization of artificial 19. Lakhangaonkar SM, Kamath KK. Artificial intelli­
intelligence in the banking sector: a systematic litera­ gence applications in banking sector. Wesley J Res.
ture review. J Financ Serv Mark. 2022;28:835–52. 2021;14(1):185-190. doi:10.34218/ijm.11.6.2020.049 .
doi:10.1057/s41264-022-00176-7. 20. Santos RS, Qin L. Risk capital and emerging technol­
6. Nizioł K. The challenges of consumer protection law ogies: innovation and investment patterns based on
connected with the development of artificial intelli­ artificial intelligence patent data analysis. J Risk
gence on the example of financial services (chosen Financ Manag. 2019;12(4):189. doi:10.3390/
legal aspects). Procedia Comput Sci. jrfm12040189 .
2021;192:4103–11. doi:10.1016/j.procs.2021.09.185. 21. Kute DV, Pradhan B, Shukla N, Alamri A. Deep learn­
7. Narsimha B, Raghavendran CV, Rajyalakshmi P, Kasi ing and explainable artificial intelligence techniques
Reddy G, Bhargavi M, Naresh P. Cyber defense in the applied for detecting money laundering-A critical
age of artificial intelligence and machine learning for review. IEEE Acces. 2021;9:82300–17. doi:10.1109/
financial fraud detection application. Int J Electr ACCESS.2021.3086230.
Electron Res. 2022;10(2):87–92. doi:10.37391/IJEER. 22. Li L, Wang J, Li X. Efficiency analysis of machine
100206. learning intelligent investment based on K-Means
8. Lokanan M. The determinants of investment fraud: algorithm. IEEE Acces. 2020;8:147463–70. doi:10.
a machine learning and artificial intelligence 1109/ACCESS.2020.3011366.
 JOURNAL OF COMPUTER INFORMATION SYSTEMS 13

23. Aguado-Correa F, de la Vega-Jiménez JJ, López- 24. Singh J, Singh G, Gahlawat M, Prabha C. Big data as
Jiménez JM, Padilla-Garrido N, Rabadán-Martín I. a service and application for Indian banking sector.
Evaluation of non-financial information and its contri­ Procedia Comput Sci. 2022;215:878–87. doi:10.1016/j.
bution to advancing the sustainable development goals procs.2022.12.090.
within the Spanish banking sector. Eur Res Manage Bus 25. Kassam A, Kassam N. Artificial intelligence in health­
Econ. 2023;29(1):100211. doi:10.1016/j.iedeen.2022. care: a Canadian context. Healthc Manage Forum.
100211. 2020;33(1):5–9. doi:10.1177/0840470419874356.