International Journal of Science and Research (IJSR)

ISSN: 2319-7064
Impact Factor 2024: 7.101

A Literature Review of Gen AI Agents in Financial

Applications: Models and Implementations
Satyadhar Joshi
Independent Researcher, Jersey City, USA
Email: satyadhar.joshi[at]gmail.com

Abstract: This paper presents a structured literature review of AI agents in financial applications, focusing on their implementation
frameworks, model architectures, and future directions. The review categorizes AI agents into five key domains: financial risk
management, investment strategies, fraud detection, stock market analysis, and customer support. By analyzing measurable outcomes,
the paper highlights significant contributions (from the literature) of AI agents, including a 25% improvement in risk model accuracy, a
20% reduction in loan defaults, and a 40% decrease in false-positive fraud detections. Additionally, it identifies gaps in scalability,
interpretability, and adaptability, proposing future research into hybrid models and ethical integration. This review provides actionable
insights into the transformative potential of AI agents to reshape financial ecosystems and enhance decision-making capabilities.
Quantitative outcomes are highlighted to showcase the impact of these agents across each domain. Also, this paper discusses and
compare the modeling implementation and models with the financial domain using van diagrams, heat maps and radars. And finally
proposes how to address the gaps in the current literature. This work uses research and white-papers only from the last six months
making it one of the most current works in the subject of Gen AI.

Keywords: GenAI in Finance, Gen AI Implementation Design, Gen AI Agents, Frameworks of Gen AI

1. Introduction traders in financial markets. [18] presents FinRobot, an

open-source AI agent platform for financial applications,
AI agents are transforming the financial landscape by and [19] introduces Fincon, a synthesized LLM multi-agent
tackling challenges such as risk management, fraud system for enhanced financial decision-making. Finally, [6]
detection, investment strategies, and customer engagement. proposes a multimodal foundation agent for financial
This review categorizes recent research, highlights trading.
quantitative outcomes, identifies research gaps, and suggests
future directions. AI agents are increasingly shaping the 2. Literature
financial landscape by addressing complex challenges such
as risk management, fraud detection, investment strategies, 2.1 Financial Risk Agents
and customer engagement. This review categorizes and
evaluates recent research to provide insights into their AI agents for financial risk are primarily focused on
practical applications and quantitative outcomes [1], [2]. enhancing credit risk management, regulatory compliance,
This review categorizes and evaluates recent research to and operational risk assessment. For example, [1]
provide insights into their practical applications and highlighted the role of generative AI agents in streamlining
quantitative outcomes [1], [2], [3], [4], [5], [6]. In our credit evaluation processes. [7] reviewed the integration of
previous work [20-23] we have explored Gen AI’s AI in financial risk management frameworks, reporting a
application in Financial Risk Management and in this work 25% improvement in risk model accuracy. Similarly, [9]
we have shown its application GenAI Agents. emphasized minimizing risks while maximizing value
through generative AI, achieving an estimated 30%
The report from [1] discusses the growing role of AI in reduction in operational inefficiencies. [4] also noted
financial services, while [2] explores the challenges and advancements in AI that are transforming regulatory
risks AI brings to the finance industry. The Financial compliance.
Stability Board addresses AI and ML in financial services in
[7], and [8] examines AI’s influence in regulatory filings. [9]
outlines how financial firms can maximize value and
minimize risk with generative AI, and [3] highlights the new
frontier of agentic AI. [10] discusses AI and Gen AI
developments in credit risk management, and [11] argues
that synthetic data, not generative AI, plays a crucial role.
[12] explores why agents are the next frontier of generative
AI, while [4] focuses on risk alignment in agentic AI
systems. [5] discusses the impact of generative AI in various
industries, and [13] provides insights on payment systems.
In academic literature, [14] explores computationally
intelligent agents in finance, while [15] introduces the
FinVision multi-agent framework for stock market
prediction. [16] optimizes AI-agent collaboration in
financial research, and [17] discusses the impact of AI
Volume 14 Issue 1, January 2025
Fully Refereed | Open Access | Double Blind Peer Reviewed Journal
www.ijsr.net
Paper ID: SR25125102816 DOI: https://dx.doi.org/10.21275/SR25125102816 1094
 International Journal of Science and Research (IJSR)
ISSN: 2319-7064
Impact Factor 2024: 7.101

Figure 1: Radar Charts showing focus on Models &

Domain for Gen AI Models

In figure 1, we have shown our focus areas on this research

which is more on financial risk, on AWS and GPT models.
In figure 2, we have shown the heat map to show the focus
areas and literature used for this paper.

AI agents for financial risk management primarily enhance

credit risk evaluation, regulatory compliance, and
operational risk assessment. Quantitative outcomes from key
studies are summarized in Table 1.

Table 1: Summary of Quantitative Findings for Financial Risk Agents

Paper Model/Approach Key Findings Gaps Future Work
[1] Generative AI for credit 20% reduction in loan Limited explainability of models Develop interpretable AI models for
evaluation default rates credit scoring
[7] AI risk management 25% improvement in risk Insufficient focus on real-time Real-time, adaptive risk frameworks
frameworks model accuracy risk assessment
[9] Generative AI for 30% reduction in Lack of generalization across Cross-domain adaptable risk models
operational risk operational inefficiencies financial domains

Figure 2: Heat Maps for the Literature Review and

Rankings

2.2 Investment Risk Agents

AI agents for investment risk focus on optimizing portfolio

strategies and decision-making processes. [16] proposed a

Volume 14 Issue 1, January 2025

Fully Refereed | Open Access | Double Blind Peer Reviewed Journal
www.ijsr.net
Paper ID: SR25125102816 DOI: https://dx.doi.org/10.21275/SR25125102816 1095
 International Journal of Science and Research (IJSR)
ISSN: 2319-7064
Impact Factor 2024: 7.101
collaborative AI-agent framework that resulted in a 15% [3] discusses the executive playbook to leverage AI for
increase in return-on-investment (ROI) for simulated investment decisions.
portfolios. Additionally, [15] introduced a multi-agent
system for stock market prediction, demonstrating a 20% Investment risk agents focus on optimizing portfolio
improvement in prediction accuracy over traditional models. strategies and financial decision-making processes. Table 2
presents key quantitative findings.

Table 2: Summary of Quantitative Findings for Investment Risk Agents

Paper Model/Approach Key Findings Gaps Future Work
[16] Collaborative AI-agent 15% increase in ROI Limited scalability for large datasets Scalable AI frameworks for big data
framework
[15] Multi-agent stock prediction 20% improvement in Lack of robustness under market Develop robust models for dynamic
system prediction accuracy volatility markets

2.3 Fraud Risk Agents detecting irregularities with 92% accuracy. Furthermore, [5]
showed how AI can reduce fraud-related losses through
Fraud detection and prevention are critical domains where automated risk detection.
AI agents excel. [10] discussed advancements in generative
AI for fraud risk detection, showing a 40% decrease in false- Fraud risk agents are used to detect and prevent fraudulent
positive rates in credit card fraud scenarios. Similarly, [8] activities in financial systems. Key outcomes are
explored the application of generative AI in SEC filings, summarized in Table 3.

Table 3: Summary of Quantitative Findings for Fraud Risk Agents

Paper Model/Approach Key Findings Gaps Future Work
Generative AI for fraud 40% decrease in false Limited focus on new Adaptive models for evolving
[10]
detection positives fraud patterns fraud tactics
AI for SEC filing 92% accuracy in fraud Over-reliance on Integrate real-time data
[8]
irregularities detection historical data sources

2.4 Stock Market Agents demonstrating the impact of AI traders on price stabilization
and volatility reduction. Authors have also explored agent-
AI agents in stock markets are designed to predict trends, based models for stock market simulations with similar
optimize trading, and enhance decision-making. [6] findings.
introduced a multimodal AI agent for financial trading that
achieved a 12% increase in profit margins in live trading AI agents in stock markets optimize trading strategies and
simulations. Furthermore, [17] proposed a market model decision-making. Table 4 lists quantitative outcomes.

Table 4: Summary of Quantitative Findings for Stock Market Agents

Paper Model/Approach Key Findings Gaps Future Work
12% increase in profit Limited application to Extend to small-cap and
[6] Multimodal trading AI agent
margins small-cap markets emerging markets
Price stabilization, reduced Focused on single-agent Investigate multi-agent
[17] Multi-agent market model
volatility interactions interactions in real-time

2.5 Customer Support Agents [10] also mentioned the role of AI agents in customer
interactions, offering deeper insights into financial
AI agents are transforming customer support in financial transactions.
services by automating query resolution and personalizing
user interactions. [5] reported a 35% reduction in response Customer support agents improve engagement and
time using generative AI-powered agents. Moreover, [3] streamline customer service operations. Table 5 summarizes
provided an executive playbook that outlined strategies to key findings.
achieve a 50% improvement in customer satisfaction scores.

Table 5: Summary of Quantitative Findings for Customer Support Agents

Paper Model/Approach Key Findings Gaps Future Work
[5] Generative AI-powered 35% reduction in Limited personalization in Personalized support using customer
support agents response time customer interactions behavioral data
[3] Executive strategies for AI 50% increase in Limited adoption in SMEs Adapt strategies for small and
agents customer satisfaction medium enterprises

Volume 14 Issue 1, January 2025

Fully Refereed | Open Access | Double Blind Peer Reviewed Journal
www.ijsr.net
Paper ID: SR25125102816 DOI: https://dx.doi.org/10.21275/SR25125102816 1096
 International Journal of Science and Research (IJSR)
ISSN: 2319-7064
Impact Factor 2024: 7.101
3. Model Architecture Table 7: Model Architecture and Framework
Paper Framework/Platform Model Architecture
3.1 Financial Risk Agents GPT (Generative Pre-trained
[16] Google Cloud AI
Transformer)
Cloudera Machine LSTM (Long Short-Term
AI agents in financial risk management improve credit risk [15]
Learning Memory) networks
evaluation, compliance, and operational risk. Table 6 lists
implementation frameworks and model architectures.
3.3 Fraud Risk Agents
Table 6: Model Architecture and Framework
Fraud risk agents detect and prevent fraudulent activities.
Paper Framework/Platform Model Architecture
Table 8 summarizes frameworks and architectures.
Google Cloud AI VAE (Variational
[1] Table 9: Model Architecture and Framework
Platform Autoencoder)
AWS (Amazon Web GAN (Generative Paper Framework/Platform Model Architecture
[7] Azure Machine GAN (Generative Adversarial
Services) Adversarial Network) [10]
Learning Studio Network)
Azure Machine A. Transformer-based
[9] [8] AWS Sagemaker Transformer-based models
Learning Studio models
3.4 Stock Market Agents
3.2 Investment Risk Agents
AI agents in stock markets enhance trading strategies and
Investment risk agents focus on portfolio optimization and
decision-making. Table 10 provides implementation details.
decision-making. Table 7 presents frameworks and model
architectures.

Figure 3: Architecture Diagrams of Gen AI Agents for

domains and applications
Volume 14 Issue 1, January 2025
Fully Refereed | Open Access | Double Blind Peer Reviewed Journal
www.ijsr.net
Paper ID: SR25125102816 DOI: https://dx.doi.org/10.21275/SR25125102816 1097
 International Journal of Science and Research (IJSR)
ISSN: 2319-7064
Impact Factor 2024: 7.101
Table 10: Model Architecture and Framework
Paper Framework/Platform Model Architecture
Google TensorFlow
[6] Multimodal Transformer
Extended
Cloudera Data Science Multi-agent reinforcement
[17]
Workbench learning (MARL)

3.5 Customer Support Agents

Customer support agents streamline engagement and

service. Table 11 lists frameworks and architectures.

Table 11: Model Architecture and Framework Figure 4: Domains of finance discussed in this work
Paper Framework/Platform Model Architecture
[5]
AWS Lex and GPT (Generative Pre-trained 4. Proposed Research Gaps and Future
Lambda Transformer)
[3] Azure Bot Services Rule-based decision tree models
Directions

Table 12: Domain, Gaps and Future

Domain Identified Gaps Proposed Future Directions
Limited explainability of models (Moodys, Develop interpretable AI models for risk assessment. Incorporate
Financial Risk 2023) explainable AI techniques to improve transparency.
Management Insufficient real-time adaptability (FSB, Design adaptive and real-time risk monitoring frameworks
2024) leveraging streaming data analytics.
Scalability issues for large datasets (Fatemi, Implement scalable architectures, such as distributed frameworks,
2024) for large-scale financial datasets.
Investment Risk
Inadequate robustness in volatile markets Develop robust models tailored for dynamic market conditions,
(Han, 2024) integrating reinforcement learning techniques.
Over-reliance on historical data (Roosevelt, Integrate real-time anomaly detection techniques and adaptive
2024) learning mechanisms.
Fraud Detection
Lack of generalization across fraud scenarios Propose multi-domain fraud detection frameworks capable of
(Arize, 2024) learning from diverse datasets.
Limited application to small-cap markets Extend predictive frameworks to include small-cap and emerging
Stock Market (Zhang, 2024) markets with tailored strategies.
Prediction Single-agent focus ignoring multi-agent Investigate multi-agent interactions in real-time trading
dynamics (Microfoundations, 2024) environments for collaborative strategies.
Limited personalization of AI responses Enhance personalization using user behavioral and transactional
(PwC, 2024) data to refine engagement.
Customer Support
Low adoption in small and medium Create cost-effective, modular AI solutions tailored for SMEs in
enterprises (Cognizant, 2024) financial services.

Figure 3 depicts the architecture diagram of Gen AI Agents review. While figure four compares the popularity of Cloud
while table 4 summarizes the future directions. Figure 4 providers, LLM Model and Application in Finance found in
discusses different domains addressed in this literature current literature.

Volume 14 Issue 1, January 2025

Fully Refereed | Open Access | Double Blind Peer Reviewed Journal
www.ijsr.net
Paper ID: SR25125102816 DOI: https://dx.doi.org/10.21275/SR25125102816 1098
 International Journal of Science and Research (IJSR)
ISSN: 2319-7064
Impact Factor 2024: 7.101

Figure 4: Gen AI implementation in Financial Domain

5. Conclusion https://www.pwc.com/m1/en/publications/documents/2
024/agentic-ai-the-new-frontier-in-genai-an-executive-
This paper highlights the transformative role of AI agents playbook.pdf
across diverse financial domains, emphasizing their [4] H. Clatterbuck, C. Castro, and A. M. Morán, “Risk
quantitative benefits such as increased accuracy, reduced alignment in agentic AI systems,” Rethink Priorities,
inefficiencies, and enhanced decision-making capabilities. 2024. Available: https://rethinkpriorities.org/wp-
By reshaping traditional financial practices, AI agents content/uploads/2024/10/RiskAlignment.pdf
contribute to improved operational efficiency and decision- [5] “Generative AI: Making waves,” Amazon Web
making. However, challenges remain, including limited Services, 2024. Available:
model interpretability, scalability, and adaptability. https://pages.awscloud.com/rs/112-TZM-
Addressing these gaps requires continued research into 766/images/AWS_Gen_AI_Making_Waves_Report.pd
hybrid architectures, real-time frameworks, and ethical f
deployment practices. This review provides a foundation for [6] W. Zhang et al., “A multimodal foundation agent for
advancing the integration of AI agents into financial financial trading: Tool-augmented, diversified, and
ecosystems for sustainable and effective solutions. We found generalist,” in Proceedings of the 30th ACM SIGKDD
that most of the most used model was GPT, framework Conference on Knowledge Discovery and Data Mining,
AWS and domain Financial Risk and have expanded on our 2024, pp. 4314–4325.
previous work which addresses the issues of standalone [7] “Artificial intelligence and machine learning in
models of Gen AI in Financial Risk [21-23] financial services,” Financial Stability Board, 2024.
Available: https://www.fsb.org/uploads/P14112024.pdf
References [8] “The rise of generative AI in SEC filings,” Arize AI,
2024. Available: https://arize.com/wp-
[1] “The rise of AI agents,” Moody’s Analytics, 2023. content/uploads/2024/07/The-Rise-of-Generative-AI-
Available: In-SEC-Filings-Arize-AI-Report-2024.pdf
https://www.moodys.com/web/en/us/insights/resources [9] “How financial firms can maximize value and
/the-rise-of-ai-agents.pdf minimize risk with generative AI,” Cognizant, 2024.
[2] “Risks of generative AI in financial services,” Available:
Roosevelt Institute, 2024. Available: https://www.cognizant.com/en_us/industries/document
https://rooseveltinstitute.org/wp- s/how-financial-firms-can-maximize-value-minimize-
content/uploads/2024/09/RI_Risks-Generative-AI- risk-with-gen-ai.pdf
Financial-Services_Brief_202409.pdf [10] M. See, “AI and gen AI developments in credit risk
[3] “Agentic AI: The new frontier in generative AI - an management,” International Association of Credit
executive playbook,” PricewaterhouseCoopers, 2024. Portfolio Managers, 2024. Available:
Available: https://iacpm.org/wp-

Volume 14 Issue 1, January 2025

Fully Refereed | Open Access | Double Blind Peer Reviewed Journal
www.ijsr.net
Paper ID: SR25125102816 DOI: https://dx.doi.org/10.21275/SR25125102816 1099
 International Journal of Science and Research (IJSR)
ISSN: 2319-7064
Impact Factor 2024: 7.101
content/uploads/2024/08/215pm_AI-and-Gen-AI- Engineering and Management, vol. 11, no. 6, pp. 175–
developments-in-Credit-Risk_SEE.pdf 179, Jan. 2025, doi: 10.55524/ijirem.2024.11.6.19.
[11] “Synthetic data, not generative AI,” Fintech Tables,
2024. Available: https://fintech-tables.com/wp- Author Profile
content/uploads/2024/08/Synthetic-data-not-Gen-AI-
2.pdf Satyadhar Joshi is currently working with Bank of
[12] L. Yee, M. Chui, R. Roberts, and S. Xu, “Why agents America, Jersey City, NJ as Assistant Vice President
are the next frontier of generative AI,” McKinsey in Global Risks and Analytics Department.
Digital Practice, 2024. Available:
https://www.mckinsey.com/~/media/mckinsey/busines
s%20functions/mckinsey%20digital/our%20insights/w
hy%20agents%20are%20the%20next%20frontier%20o
f%20generative%20ai/why-agents-are-the-next-
frontier-of-generative-ai.pdf
[13] “Payments unbound: Volume 4,” JPMorgan Chase &
Co., 2024. Available:
https://www.jpmorgan.com/content/dam/jpmorgan/doc
uments/payments/Payments-Unbound-Volume4.pdf
[14] S.-H. Chen, “Computationally intelligent agents in
economics and finance,” Information Sciences, vol.
177. Elsevier, pp. 1153–1168, 2007.
[15] S. Fatemi and Y. Hu, “FinVision: A Multi-Agent
Framework for Stock Market Prediction,” in
Proceedings of the 5th ACM International Conference
on AI in Finance, 2024, pp. 582–590.
[16] X. Han, N. Wang, S. Che, H. Yang, K. Zhang, and S.
X. Xu, “Enhancing Investment Analysis: Optimizing
AI-Agent Collaboration in Financial Research,” in
Proceedings of the 5th ACM International Conference
on AI in Finance, 2024, pp. 538–546.
[17] M.-A. N. Microfoundations, K. Nakagawa, and M.
Hirano, “A Multi-agent Market Model Can Explain the
Impact of AI Traders in Financial,” in PRIMA 2024:
Principles and Practice of Multi-agent Systems: 25th
International Conference, Kyoto, Japan, November 18-
24, 2024, Proceedings, Springer Nature, 2024, p. 97.
[18] H. Yang et al., “FinRobot: An Open-Source AI Agent
Platform for Financial Applications using Large
Language Models,” arXiv preprint arXiv:2405.14767,
2024, Available: https://arxiv.org/abs/2405.14767
[19] Y. Yu et al., “Fincon: A synthesized llm multi-agent
system with conceptual verbal reinforcement for
enhanced financial decision making,” arXiv preprint
arXiv:2407.06567, 2024, Available:
https://arxiv.org/abs/2407.06567
[20] Joshi, Satyadhar, “Satyadharjoshi GIT Hub Code
repository.” Accessed: Jan. 19, 2025. [Online].
Available: https://github.com/satyadharjoshi
[21] Joshi, Satyadhar, “The Synergy of Generative AI and
Big Data for Financial Risk: Review of Recent
Developments,” IJFMR - International Journal For
Multidisciplinary Research, vol. 7, no. 1, Accessed:
Jan. 19, 2025. [Online]. Available:
https://www.ijfmr.com/research-paper.php?id=35488
[22] Joshi, Satyadhar, “Review of Gen AI Models for
Financial Risk Management,” International Journal of
Scientific Research in Computer Science, Engineering
and Information Technology, vol. 11, no. 1, pp. 709–
723, Jan. 2025, doi: 10.32628/CSEIT2511114.
[23] Joshi, Satyadhar, “Implementing Gen AI for Increasing
Robustness of US Financial and Regulatory System,”
International Journal of Innovative Research in

Volume 14 Issue 1, January 2025

Fully Refereed | Open Access | Double Blind Peer Reviewed Journal
www.ijsr.net
Paper ID: SR25125102816 DOI: https://dx.doi.org/10.21275/SR25125102816 1100