2.
1 Introduction of AI in Finance
Artificial intelligence (AI) has been a topic of discussion since the 1950s, but recent
developments in big data, data analytics, and supercomputing have brought it into the spotlight.
In finance, AI is widely used to gain insights, measure performance, make predictions, perform
real-time calculations, enhance customer service, and retrieve information intelligently. It
consists of technologies that help financial organizations better understand markets and
customer behaviors, learn from digital interactions, and engage with clients in ways that
resemble human-like intelligence on a large scale. AI's key contributions in finance include
personalizing products and services, identifying new opportunities, managing risks and
detecting fraud, promoting transparency and regulatory compliance, and streamlining operations
to cut costs.
AI will be a major driver of growth in financial services, supporting digital advancements that
enhance sales, streamline processes, and maximize data use. Moving forward, organizations
will need to scale personalized, relationship-centered customer interactions. AI will enable
financial institutions to offer tailored responses, recommend products and services more safely
and responsibly, and build trust by expanding around-the-clock concierge services.
Additionally, financial institutions must create comprehensive, permission-based digital
customer profiles, but their data is often siloed. By breaking down these barriers, applying AI,
and integrating human interaction, they can craft unique, scalable experiences that meet the
specific needs of each customer.
2.2 Functions of AI
AI systems operate through a combination of some components like - data collection, data
processing, model training, model evaluation, deployment, feedback and improvement.AI is
driving significant efficiency improvements across various functions, reducing customer support
response times by 33% and halving some code-generation times. AI serves multiple functions
which are :
1.Automation : Industries have increasingly used technology to boost productivity and
lower production costs by automating repetitive tasks. By reducing human involvement,
machines and computers can now handle routine activities and adjust to changing
conditions. This trend of automation has become common in both manual labor and
office-based roles.
2. Machine learning : Machine learning is a transformative concept where machines
analyze large datasets to learn from the information and enhance their algorithms over
time. A key aspect of machine learning is neural networks, which are inspired by the
structure of the human brain. These networks consist of interconnected nodes, known as
neurons or perceptrons, which work together to process and learn from data. ( Shaan
Ray, published in codebrust, Aug 17, 2018)
Neural networks can recognize patterns and improve their ability to categorize new data
through experience. For instance, a neural network designed to identify dog breeds will
analyze labeled images of different dogs, gradually learning to associate specific
features with each breed. Through this process, the machine becomes increasingly
accurate in its identification. ( Shaan Ray, published in codebrust, Aug 17,2018)
3.Problem solving and decision making
A core goal of AI is to create systems capable of analyzing vast datasets, recognizing
patterns, and making decisions based on data insights. This skill in efficient problem-
solving and decision-making is highly valuable across sectors like healthcare, finance,
transportation, and manufacturing.( By Sneha Kothari, simpli learn, sep 17,2024)
4.Customer Support
Generative AI offers more than just automation and efficiency in customer support; it can
also proactively reduce the need for support by predicting and resolving potential issues.
Its applications in customer service include analytics to anticipate problems, chatbots for
self-service, algorithms for matching customers with suitable agents, and tools to help
agents work more efficiently.
Generative AI can speed up agent response times by up to 35%, assist consultants by
organizing knowledge sources, and improve service quality by up to 40%. For instance,
a tech and manufacturing firm created two AI-based prototypes for field services: a
maintenance assistant to enhance technician productivity and systems to process vast,
varied sensor data, aiding emergency responders in decision-making.
5.Fraud Detection:
AI systems identify fraudulent transactions by examining patterns in financial data.(Samir
bazatu, software engineer at spacex,May 2024).Banks have long used machine learning
techniques within their credit card portfolios, as transaction data offers a valuable
foundation for training unsupervised learning models. These models are particularly
effective in detecting credit card fraud, given the extensive data available for
development, training, and validation.
Workflow engines within credit card payment systems continuously monitor transactions
to evaluate fraud risk. The detailed transaction histories allow banks to identify distinct
characteristics that differentiate fraudulent from legitimate transactions.
2.3 Benefits of AI in financial sector
AI in finance has improved process efficiency by minimizing manual tasks and supporting
various areas, such as market research, data analysis, and more.Here are few benefits of AI for
finance :
● Cost reduction: In finance, labor costs—particularly in compensation and benefits—
represent a major expense for many institutions. AI in accounting can significantly boost
worker efficiency and productivity. For instance, Decision Management Systems (DMS)
enable quicker, more informed decision-making. DMS can also streamline customer
onboarding by using preset responses to common questions, reducing the need for
front-line employees. Additionally, by standardizing data entry, DMS minimizes human
errors, which can protect a company's reputation and avoid costly government fines.
● Automating the investment process
AI is transforming investment decision-making in retail banking by streamlining
processes and uncovering new opportunities. Financial organizations now use AI for
tasks like identifying investment prospects, enhancing algorithmic trading strategies, and
delivering personalized advice through robo-advisors. With machine learning and
predictive analytics, AI tools can reveal patterns and insights that human analysts might
miss, leading to better-informed investment choices.
However, integrating AI into investment management comes with challenges, particularly
around model transparency. Financial institutions must ensure their AI systems are
understandable to meet regulatory standards and provide clear justifications for
recommendations.
● Evolutionary improvement :
For over a decade, investment firms have used machine learning and neural networks to
drive high-speed, automated trading, especially in highly liquid markets. The recent
development of large language models (LLMs) is advancing this capability further,
allowing investors to process vast amounts of unstructured, text-heavy data to boost
their analytical power. This advancement not only enhances broad forecasting but also
aids quantitative investors in quickly analyzing complex documents like bond
agreements and earnings reports, leading to better price discovery across various
assets. (Tobias Adrian, September 6, 2024)
● Improving financial forecasting
AI-based forecasting models are extensively used in the finance sector to improve
prediction accuracy by incorporating diverse traditional and non-traditional data
sources.AI-based models focus on predicting company bankruptcies, stock price
changes, and credit risk.
AI models also play a role in asset return predictions, with Li and Mei (2020) using deep
learning with multiple hidden layers for asset forecasting, and Ruan et al. (2020)
applying machine learning to gauge stock market returns based on investor sentiment.
Petrelli and Cesarini (2021) combined AI methods to predict high-frequency asset prices.
In the insurance industry, AI has been adopted for claim prediction. For example, models
can predict reimbursement outcomes for claims related to accidents, property damage,
or medical emergencies. Both feed-forward and recurrent neural networks are used,
enabling insurers to analyze relationships between test data and standardized data for
annual claim predictions (Rawat et al., 2021).
● Automating key business process in customer service and insurance
Financial institutions have increasingly adopted Robotic Process Automation (RPA) to
streamline core processes and enhance customer service, especially in retail banking
and wealth management. These RPA tools often function as robo-advisors, providing
automated financial services such as tax guidance, account opening, insurance
recommendations, and investment advice (Wittmann & Lutfiju, 2021; Kruse et al., 2019).
By leveraging an AI-first approach, banks can expand into new client segments, reduce
acquisition costs, and increase usage of existing services, all while meeting customer
demand for convenient, round-the-clock service (McKinsey, 2020).
AI-powered RPA allows banks to offer personalized services, enhancing user
experience and reducing decision fatigue for customers. This strategic emphasis on
RPA helps banks stay competitive, improve profitability, and operate more efficiently by
integrating real-time analytics and messaging, which requires modernized IT systems for
processing client data (Zeinalizadeh et al., 2015; Villar & Khan, 2021).
2.4 AI impact on financial performance and financial services
AI's integration in the financial sector has begun to improve efficiency, drive new services, and
enhance revenue for institutions. However, quantifying AI's impact on bank performance is
challenging due to difficulties in measuring AI-specific contributions. This raises questions about
AI's broader effects on businesses, consumers, and the economy.
When it comes to disclosing AI practices, particularly in banking, disclosure remains mostly
voluntary. With no established standards, companies independently decide what AI-related
information to share. Financial services with high data processing needs often lead in disclosing
AI use, especially in natural language processing. However, there is no unified framework
guiding these disclosures, and current practices fail to fully address AI's unique impact.
The OECD emphasizes AI's impact in financial markets, particularly in increasing
competitiveness through cost efficiency, productivity, and enhanced service quality. By
streamlining processes, automating tasks, and strengthening risk management, AI can lead to
higher profitability, better decision-making, and improved regulatory compliance. The technology
allows for new, customized financial services, improving customer experiences and creating
opportunities for personalized solutions like robo-advisors.
AI adoption also reduces the need for low-skilled labor, boosting the productivity of remaining
employees. Many banks report benefits like cost reduction, revenue growth, and operational
improvements, especially in lending, fraud detection, and compliance. While AI offers
substantial economic advantages for both stakeholders and shareholders, challenges remain in
measuring its direct impact on bank performance, leaving questions about its broader effect on
businesses, consumers, and the economy.
2.5 AI voluntary disclosure and financial performance
AI advancements have notably influenced sectors like banking, where companies are
increasingly adopting AI-driven solutions. In banking, this includes applications that process vast
volumes of text and speech data, particularly for customer service and operational efficiency
(Sætra 2021). According to the AI McKinsey Global Surveys, financial services tend to disclose
more about their AI-driven natural language capabilities due to these high data demands.
However, AI-related disclosures remain voluntary, and decisions about what, how much, and
which aspects to disclose largely depend on each company. No universally accepted guidelines
currently exist for AI disclosure, as the field is still emerging and lacks established international
reporting standards. Current practices do not yet fully address the unique impacts that AI
technologies can have.
In today’s AI-driven world, the opacity of AI systems poses a significant risk. S. Lu (2021)
argues that current corporate disclosure frameworks in securities law have limited effect in
reducing this opacity and calls for a more effective framework to ensure transparency in AI
products and services. Enhanced disclosure is essential to reveal potential algorithmic risks,
protect stakeholders, and foster stable, sustainable markets.
Regulators are increasingly considering AI disclosure requirements. The OECD, for instance,
has emphasized the need for transparency and responsible disclosure in AI systems to enable
people to understand and question AI-driven outcomes (OECD, 2019). Consumers, especially
in financial services, should be informed if AI is involved in delivering a product, so they can
make informed choices. In 2021, the OECD further emphasized that disclosures can help
financial service providers assess whether clients understand AI’s impact on services. Likewise,
the International Organization of Securities Commissions (IOSCO) has advocated for clearer
information on AI systems’ abilities and limits within disclosures.
The EU AI Act (2023) introduces transparency requirements for AI systems used within the
European Union, marking an important step in AI regulation that could significantly influence AI
development and usage across the region. The Act outlines requirements for transparency,
documentation, auditing, and user information provisions (Article 13). Although still in the early
stages, this legislation has the potential to shape the future of AI in both Europe and the United
States. Early preparation, including establishing effective disclosure procedures, is key to
adapting to these new standards.
2.6 AI implications in the banking industry
AI, along with advancements like machine learning, big data analytics, cloud computing, and
social media, is transforming digital business and is widely integrated into modern life.
Technology not only alters physical and operational processes but also improves efficiency,
competence, and enables forward-thinking business solutions (Tekic and Koroteev, 2019). AI’s
impact is multifaceted: it can predict and interpret the environment by processing audio, text,
and computational data. Through natural language processing (NLP), it enhances human-
machine interaction and operates independently without human intervention (Purdy and
Daugherty, 2016; Rao and Verweij, 2017; Tákacs et al., 2018; Ottosson and Westling, 2020).
Additionally, AI stands out from traditional machines due to its self-learning ability, allowing it to
improve continually based on past experiences (Öztemel and Gursev, 2020). These innovations
are reshaping competitive business strategies and contributing to new methods of value
creation.
The integration of AI in banking has significantly improved decision-making, enabling
compliance with regulatory standards while enhancing accuracy and efficiency. AI technologies
help minimize false contracts, predict resource needs, and adhere to regulations (Han et al.,
2020; Couchoro et al., 2021; Garcia-Bedoya et al., 2020; Kute et al., 2021). Various AI methods,
including data mining, fuzzy logic, machine learning, sequence alignment, and genetic
programming, aid banks in reducing fraud (Raj & Portia, 2011). Autonomous data management
has boosted the speed and reliability of banking processes (Soni, 2019). Predictive analytics
tools—such as SSL, encryption, multi-level authorization, device fingerprinting, and endpoint
protection—help prevent fraud before it occurs (Kikan et al., 2019). Deep learning and artificial
neural networks allow personalized banking services by helping banks assess customer
preferences and responses to marketing (Kim et al., 2015; Zakaryazad & Duman, 2016). AI has
streamlined processes, reduced costs, minimized risks, and enhanced customer experience
with chatbots and robo-advisors.
2.7 Case studies and industrial example
AI implementations in financial institutions, offering insights into their applications, advantages,
and challenges faced.
Examination of successful AI implementation in financial institutions
Case 1 : JPMorgan Case
JPMorgan Chase, a leading global financial institution, has integrated AI into various aspects of
its operations to improve efficiency and customer experience. The bank utilizes AI algorithms for
fraud detection, risk management, customer service, and trading. By harnessing machine
learning and natural language processing, JPMorgan Chase has enhanced the accuracy of
fraud detection, minimized false positives, and improved customer interactions through
personalized recommendations and chatbots.
Case 2 : Vanguard
Vanguard, a prominent investment management firm, has incorporated AI-driven robo-advisors
into its wealth management offerings, delivering automated portfolio management and
investment guidance. These robo-advisors use machine learning algorithms to assess investor
preferences, risk profiles, and market trends, enabling the creation of tailored investment
strategies and asset allocations at a lower cost than traditional advisory services. This
innovation has democratized access to wealth management, appealing to a wider array of
investors and contributing to the firm's growth.
Case 3 : Blackrock
BlackRock, the largest asset manager globally, has implemented AI-driven solutions to enhance
its investment strategies, risk management, and portfolio construction. The firm employs
machine learning algorithms to analyze market data, discover investment opportunities, and
optimize asset allocations within its portfolios. These AI-enhanced strategies have provided
clients with superior risk-adjusted returns compared to traditional investment methods,
contributing to the firm’s growth.
Case 4 : Ping and insurance
Ping An Insurance, a top insurer in China, has integrated AI throughout its operations to boost
efficiency, manage risks, and enhance customer service. Using AI for underwriting, claims
processing, CRM, and fraud detection, Ping An has optimized insurance processes, shortened
claims handling times, and improved customer satisfaction. Additionally, the company uses AI-
driven predictive analytics to evaluate and reduce risks, strengthening its competitive edge and
profitability.
2.8 Development of hypothesis
Prior research on voluntary disclosure explores different theoretical viewpoints. For instance,
agency theory emphasizes that managers communicate specialized information to increase firm
value and reduce capital costs (Jensen & Meckling, 1976). By providing voluntary disclosure,
managers can effectively communicate on topics such as investment opportunities and
financing strategies, which can enhance the firm’s overall value. Scholars within agency theory
suggest broadening the categories of voluntary disclosure to better monitor managers' actions
and enhance firm performance.
Research shows that AI positively influences nonfinancial metrics in firms, such as enhancing
the efficiency of accounting systems and internal workflows (Hashem & Alqatamin, 2021).
Wamba-Taguimdje et al. (2020) highlight that AI supports organizations in improving
performance across various areas—financial, marketing, and administrative—at both
organizational and process levels. Additionally, Fethi and Pasiouras (2010) reviewed 179
studies on AI’s financial impacts in banking and found that AI notably boosts banks' financial
performance.
Therefore, the following first hypothesis is suggested :
H1 : AI related terms disclosure will be positively related to the bank's financial
resources
The banking industry is particularly affected by stricter regulations compared to other sectors,
which may encourage banks to voluntarily disclose more information about their AI systems.
Such transparency demonstrates compliance with ethical standards and helps banks gain a
competitive advantage in markets that prioritize responsibility (Krakowski et al., 2022).
Generally, banks highlight favorable AI developments in these disclosures, aiming to positively
influence economic outcomes and appeal to investors. This approach enables investors to
better understand the banks' investments in AI technologies (Bonsón et al., 2021).
From theoretical perspectives and literature, companies that see financial benefits from AI—
such as higher revenue or reduced costs—are often motivated to share positive updates about
their AI use and development. Companies that disclose more about their AI activities are likely
to have integrated AI into their business strategies, experiencing measurable financial benefits.
As a result, it’s expected that frequent mentions of AI in annual reports correlate with improved
financial performance, specifically in terms of increased revenue and reduced costs.
The following second hypothesis is suggested :
H2 : AI related terms disclosure will be negetively related to the bank's total expenses
2.9 Challenges of AI in financial sector
The above section highlights key issues arising from the ongoing and widespread adoption of AI
in finance. However, additional challenges will persist. If not effectively managed, these
challenges could hinder the broader implementation of AI-driven systems within the financial
industry.
1. Availability and quality of Training data
Training AI models requires extensive data to enhance their accuracy and reliability.
Large datasets allow AI to learn from diverse examples, enabling it to recognize complex
patterns and relationships that smaller datasets might miss. This abundance of data also
helps reduce the risk of overfitting, where models perform well on training data but
struggle with new data.
In the financial sector, however, much of the available data remains underutilized for AI
training. Many financial services have yet to fully digitize their processes, limiting the
amount of data readily available for AI applications (BoE and FCA, 2022; Cao, 2022;
Kruse et al., 2019; Milana and Ashta, 2021).
Data quality is equally important, as incomplete, inaccurate, or biased data can
undermine model performance and lead to unfair outcomes. High-quality, consistent
data is essential, particularly for deep learning models and when data is unstructured or
gathered from multiple sources (Greenspan et al., 2016; Lee, 2017).
2. Lack of appropriate skill
Using AI presents challenges for financial organizations, as many employees lack the
specialized technical skills needed to effectively operate AI systems (Kruse et al., 2019).
These systems often require knowledge in programming, data analysis, and machine
learning, which can make it difficult for employees to use them correctly and interpret
their results. Additionally, the fast-paced advancements in AI make it challenging for
employees to stay current on the latest techniques, suggesting a need for ongoing
training and development programs.
Furthermore, the adoption of AI can lead to shifts in job roles and responsibilities, as
some tasks may be automated while others may require new skills or approaches
(Culbertson, 2018; LinkedIn, 2017). Organizations, therefore, should prepare for these
changes by providing necessary support and training to help employees transition into
their evolving roles.
3. AI model development challenges
Despite recent advancements in AI, financial organizations still face difficulties in
developing accurate and high-performing AI models. Beyond general skill and expertise
issues, some challenges are inherent to the AI techniques themselves. For instance,
natural language processing (NLP) is often used in sentiment analysis to predict stock
prices or generate trading signals from financial text data (Osterrieder, 2023). However,
NLP faces unique challenges in interpreting language accurately. Unlike humans, NLP
models struggle to understand context when words have multiple meanings and find it
difficult to recognize different words expressing the same idea (Khurana et al., 2023).
Additionally, homonyms can complicate tasks such as question-answering and speech-
to-text applications, as they lack written context.
4. Selecting the optimal ML model
No single AI algorithm is universally effective for all problems; using an inappropriate
algorithm can lead to poor performance, inaccurate predictions, or failure to solve the
problem. Traditional methods may even outperform AI in some cases, as discussed in
section 2. Selecting the right algorithm requires understanding the problem and data
characteristics, as well as the strengths and weaknesses of various models.
Organizations often choose different algorithms based on needs for accuracy,
interpretability, and data type (Lee and Shin, 2020).
Similarly, using robotic process automation (RPA) can be challenging, particularly for
financial institutions that often struggle to identify the best use cases for RPA. Targeting
overly complex processes with RPA is a common issue that results in high costs that
could be better allocated elsewhere (Lamberton et al., 2017). This issue partly arises
because many organizations lack a clear understanding of bots' capabilities and
operations (Cooper et al., 2019). Another concern is the need to protect business
processes and manage information flow across jurisdictions (Cooper et al., 2019).
5. Requirement of better agility and faster adaptability
Agility and adaptability are essential for managing AI-related risks in the financial
industry, such as data bias, security and privacy issues, and model opacity (Thowfeek et
al., 2020). These risks can impact both businesses and customers, making it vital for
companies to respond quickly and flexibly. Additionally, companies with advanced AI
capabilities can gain a competitive edge, underscoring the need for adaptability in
adopting AI to stay relevant. As AI reshapes business operations and decision-making,
companies must adjust existing processes and structures to maximize AI's benefits.
