Received 27 October 2024, accepted 23 November 2024, date of publication 9 December 2024, date of current version 2 January 2025.

Digital Object Identifier 10.1109/ACCESS.2024.3513279

AI-Enhanced Robotic Process Automation: A

Review of Intelligent Automation Innovations
SADIA AFRIN 1, SHOBNOM ROKSANA 1, AND RIAD AKRAM2 , (Member, IEEE)
1 Department of Computer Science, The University of Texas at San Antonio, San Antonio, TX 78249, USA
2 Intel Corporation, Hillsboro, OR 97124, USA

Corresponding author: Sadia Afrin ([email protected])

ABSTRACT The rapid technological growth in recent decades due to the integration of robust technologies
and automation have led to the rise of digital services and the emergence of Industry 4.0. This paper explores
the concept and potential of AI-powered intelligent automation based on the synergistic use of Robotic
Process Automation (RPA) and Artificial Intelligence (AI) to enhance organizational and business processes
across various sectors. RPA automates routine, rules-based tasks, thereby allowing human workers to engage
in more innovative activities. When integrated with AI, RPA systems gain the capacity to analyze data,
identify patterns, classify information and forecast which leads to significant improvement in accuracy and
productivity. This literature review investigates the current state of RPA and AI integration while highlighting
its applications in different sectors such as manufacturing, agriculture, healthcare, finance, and retail. Along
with discussing the drawbacks and restrictions, such as technological issues and moral dilemmas, this paper
also discusses the advantages of this integration, which include decreased costs, increased output, and
simplified operations. By leveraging AI techniques such as classification, text mining of neural network,
RPA technologies optimize business operations and advance Industry 4.0. This study also illustrates the
challenges and limitations of this integration such as technical difficulties and ethical considerations. The
aim of this review is to provide a comprehensive understanding of the synergistic potential of RPA and AI
while offering insights into their contribution in shaping the future of intelligent automation.

INDEX TERMS Artificial intelligence (AI), business process, intelligent process automation (IPA), robotic
process automation (RPA).

I. INTRODUCTION Robotic Process Automation (RPA) is a new technol-

The rapid advancement of technology over the past years ogy that allows one to easily automate such routine and
has, in many ways, introduced a new wave of digital rules-based tasks performed by humans. Such tasks can
transformations that are now changing what, how, and with include repetitive actions—for instance, data entry, moving
whom organizations work and compete worldwide. At its files, and a number of other back-office tasks. The use of the
core is an integration largely driven by advanced automation RPA bots enables one to do mundane, repetitive tasks much
technologies forming this new era—a reshaping of business faster than a human being can do them, with precision, thus
and competition—called Industry 4.0. At the core of this reducing errors and greatly increasing productivity. They lack
revolution, there is Robotic Process Automation and Arti- the cognitive ability compared to AI; they simply follow a set
ficial Intelligence—two complementary technologies that, of pre-defined rules and are used in operations that require
when combined, unleash extraordinary potential to improve zero to very little human judgment. According to a Grand
business processes, boost productivity, and trim operational View Research, the RPA market is poised to have a compound
costs. growth rate (CAGR) of more than 40% to $25 billion in
2027 [1]. With its aptitude in cost-saving, error reduction, and
The associate editor coordinating the review of this manuscript and business process improvement, it has resulted in increased
approving it for publication was Okyay Kaynak . levels of deployment by different business concerns.

2024 The Authors. This work is licensed under a Creative Commons Attribution 4.0 License.
VOLUME 13, 2025 For more information, see https://creativecommons.org/licenses/by/4.0/ 173
 S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

However, it is when RPA gets combined with Artificial details of patients, automation of administrative functions,
Intelligence that its true potential comes out because AI and other diagnostic-related processes have become easier.
gives the process of automation cognitive abilities. The key The problems related to this industry, including data privacy
activities in which AI is associated are those that require problems and ethical issues of AI in healthcare, are discussed
human-like intelligence, including data analytics, pattern in the following sections of this paper.
recognition, and decision-making. By adding AI to RPA, Recruitment Industry: The recruitment industry uses AI
companies can upscale their automating operations into for resume screening, ranking, and interview scheduling
complex ones that are out of reach for ordinary RPA. This activities. We will also discuss issues related to the bias of
brings in the combination of Intelligent Process Automation AI models and transparency in automated decisions.
(IPA), which offers firms a strong tool for automating simple Manufacturing: How AI-based RPA is utilizing automa-
and complex activities. It does not just automate repetitive tion to optimize production processes, supply chain man-
processes; rather, thanks to AI’s learning and adaptive agement, and quality control in the manufacturing space.
properties, it brings about enhanced improvements in terms Discussion will cover the challenges in integrating AI with
of accuracy, productivity, and innovation across a wide range current manufacturing systems and complexities in managing
of sectors [2]. industrial operations.
The popularity of both RPA and AI increases each Back Office Operations: The paper has provided a business
year, and the rate at which enterprises deploy them is case of a real AI-powered RPA project for back-office
increasing in industries that seek digitization. In earlier operations, explaining how the RPA can be put in place
days, automation used to be a huge activity made possible without disturbing the existing systems. The section then
by only a few very skilled professionals who built and discusses technical challenges of AI-RPA integration in data-
deployed large automation systems. Today, the integration intensive environments.
of AI with RPA has enabled the development of lean Furthermore, we will discuss challenges and limita-
end-to-end business processes such that companies can tions in the implementation of AI-powered RPA, such as
innovate and bring about revolution in workplace practices. technical complexity, integration problems, data privacy
Most industries have started infusing artificial intelligence concerns, and ethical considerations [3]. Overcoming such
into advanced technologies to drive the power of digital problems needs effective planning and well-prepared gov-
automation in improving both operational efficiencies and ernance with a commitment to continuous learning and
customer experiences. improvement.
Throughout this paper, we will explore the evolution of Finally, the paper will explore open challenges for research
RPA, how it has moved from mere automation to becoming regarding the context of AI-powered RPA. Among other
an important feature in contemporary business processes. subjects, it will address change management, ethical and
We will then delve into how AI is integrated with RPA legal considerations, model selection for machine learning,
so that we understand how the synergy between these two and process explainability within the AI-driven process.
technologies enhances the capabilities of automation systems We will also examine the potential for Intelligent Process
and permits more complex and intelligent operations. The Automation in sectors that are still in the early stages of
paper will also cover widely used tools that provide the digital transformation, such as agriculture.
platform for integration for both RPA and AI, among others, This paper aims to offer a comprehensive review of
including UiPath, Kofax, Automation Anywhere, and Blue AI-powered RPA systems across multiple industries as
Prism. mentioned above. The workflow of our study is presented
We will provide an in-depth analysis of the application of in Figure 1. Unlike many recent literature reviews that focus
AI-driven RPA across the following sectors: narrowly on a single domain or technology (either RPA
Finance and Banking: AI-powered RPA automates loan or AI), as summarized in Table 1, this paper integrates
processing, increases fraud detection with automated credit findings from diverse areas, providing a holistic perspective.
checking, and offers improved customer service. Further, Additionally, it examines the latest advancements and
challenges of data security, compliance, and integration will features of AI in leading RPA tools such as UiPath,
also be discussed. Kofax, Automation Anywhere and Blue Prism, detailing
Auditing: The efficiency and accuracy of the process how AI enhances their capabilities. By investigating AI-
in the audit industry are improved by AI-based RPA, and enhanced RPA systems across various industries, this study
it reduces the likelihood of human error, allowing one to consolidates existing knowledge, highlights detailed use
continue monitoring financial activities. We discuss specific cases, identifies challenges and gaps in the existing systems,
use cases and the associated challenges here, which include and proposes future directions for integrating AI into RPA.
large volumes of data and maintaining transparency in AI- Furthermore, this paper also proposes an AI powered RPA
driven decisions. based framework that can be beneficial to the agricultural
Medical and Health Services: Health, too, has been one sector, an area that, to the best of our knowledge, has not been
of the beneficiaries. Use of AI-enabled RPA in managing explored in existing studies.

174 VOLUME 13, 2025

S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

TABLE 1. A summary of the review papers focused on RPA and AI based systems.

FIGURE 1. Workflow of AI-enhanced RPA study.

II. ORGANIZATION OF THE PAPER Framework (III) describes the systematic methodology used
The organization of this paper follows a logical structure in this study to investigate recent AI-powered RPA systems.
to comprehensively address the goal of our study. The The Evolution of Robotic Process Automation(RPA) (IV)
next section,The Research Approach and Methodological provides the history of RPA and how it has evolved

VOLUME 13, 2025 175

 S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

FIGURE 2. Research procedure.

over the past few years. The Integration of AI with RPA

(V) section explores how AI enhances RPA systems for FIGURE 3. U.S RPA market size analysis by year [17].
greater adaptability and efficiency. The Tools That Adapt AI
Techniques in RPA Automation (VI) section provides insights
into platforms such as UiPath, Kofax, Automation Anywhere
and Blue Prism, showcasing their AI-driven features. The 32.8%. The growth is largely fueled by RPA service models,
following sections focus on domain-specific applications: AI- with the RPA-as-a-service segment capturing over 61% of
Driven RPA in Finance and Banking (VII), Audit (VIII), the market share in 2022 and experiencing the fastest growth
Medical and Health Services (IX), Recruiting Industry (X) rate [16]. Figure 3 illustrates a market analysis or RPA.
and Manufacturing (XI). Each domain includes subsections The growth reflects the widespread recognition of RPA
on applications, use cases and challenges. The study also and its potential in different industries where automation can
highlights Another Real AI-Powered RPA Project(XII) for provide with transformational benefit [7].
Back Office, detailing its associated challenges. The Open RPA has a rich history that started from the early days
Research Challenges section discusses (XIII) critical issues of computing and automation in 1990s. It emerged with the
such as change management, communication strategies, use of screen scraping and workflow automation software.
ethical and legal considerations, selecting machine learning These technologies allowed the automation of repetitive
models, explainability, and future applications. Finally, the tasks by capturing data from graphical user interface (GUI).
Conclusion (XIV) synthesizes the findings and proposes The concept got more recognized when organization started
directions for future research. adopting automation in their business process automation
altering their IT infrastructure [17]. The term ‘‘Robotic
III. RESEARCH APPROACH AND METHODOLOGICAL Process Automation’’ was addressed as a separate automation
FRAMEWORK category in the automation industry in 2010. The early
This paper adopts a literature review approach inspired by adoption of RPA started with the use of automating simple,
the Systematic Literature Review (SLR) methodology, which rule-based tasks.
is structured into three distinct phases: planning, conducting, Around 2012, companies like Blue Prism, Automation
and reporting as shown in Figure 2. In the planning phase, Anywhere and UiPath appeared to be as leading vendors
we identified the necessity for this study (as outlined above) that provides platform to design business automation pro-
and developed a comprehensive search protocol that includes cesses [17]. The RPA market started to gain recognition
inclusion/exclusion criteria. For identifying relevant articles, for its potential of cost saving and operational efficiency
we applied these criteria using Google Scholar and reputable during the year between 2013 and 2015 [18]. Gradually the
databases such as IEEE Xplore and SpringerLink. The adoption flowed across different industries such as financial,
search process was guided by two specific keywords: ‘‘RPA’’ healthcare, manufacturing and so on. The flexibility of
and ‘‘AI.’’ The initial selection involved screening titles to deploying RPA played a significant role into the growth of
evaluate their relevance against predefined inclusion criteria, RPA.
followed by a detailed review of abstracts. Articles passing During 2017, businesses started to publish report on high
these stages were further checked for full-text accessibility. returns on investment (ROI) from their automation processes.
In the conducting phase, we systematically curated a final Gradually, RPA’s evolution started to be advanced by the
set of articles for review. These articles were analyzed with introduction of AI technologies which began enhancing
a specific focus on methodologies related to RPA and AI. the potential of RPA and leaded towards IPA [19]. That’s
Finally, in the reporting phase, we synthesized the findings, when UiPAth, Automation Anywhere, Blue Prism and other
providing our interpretations, critical analyses and insights vendors started to receive funds to accelerate the development
into the reviewed literature [15]. of RPA by integrating AI technologies. This advancement
and integration of AI and RPA allowed bots to handle
IV. EVOLUTION OF ROBOTIC PROCESS AUTOMATION unstructured data and make complex predictive decisions.
(RPA) This development made significant shift in the automation
As mentioned before, the RPA market is projected to reach industry during 2019 [20]. The COVID-19 accelerated the
$25 billion by 2027, with a CAGR of over 40%. And, USA digital transformation across the organization with the need
alone will exceed $12B by 2028, with a CAGR growth of of maintaining social distancing [21]. [22] investigates the use

176 VOLUME 13, 2025

S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

of emerging RPA and AI in mitigating the negative impacts of

the COVID-19 pandemic on the global economy and public
health. It explores how these technologies can be applied
across various industries to effectively manage the challenges
posed by the pandemic.
For the last two to three years, RPA has become a
mainstream technology as large enterprises and government
agencies are implementing it to improve their operational
efficiency and reduce costs. The focus shifted towards other
technologies such as, cloud computing. As the need of
RPA with cognitive power is increasing the vendors are
offering more advanced features with flexibility of selecting
AI technologies and deploying RPA models into cloud. This
is how the evolution or RPA from performing simple tasks to FIGURE 4. An overview of IA [24].
do complex and sophisticated tasks has reshaped industries
and their business processes globally. In our next section we
will talk about this integration in details.
capable of executing processes that require routine tasks with
V. INTEGRATION OF ARTIFICIAL INTELLIGENCE (AI) minimal human input [26] as well as cognitive capabilities
WITH RPA such as infusing NLP technology [27], [28], [29]. The IPA
In today’s world, AI algorithms and Machine learning(ML) technology need low level of human intervention in compared
techniques has been implemented in almost all sectors where to RPA process. However, they need fine tuning to select
digital transformation is introduced. ML is used to train optimal ML models. IPA has been utilized in both internal
machine how to deal with large dataset efficiently. By using (back office) and external (front office) services in industries
AI techniques, it is possible to explore data, get deep insight such as banking [30]. In recent years, several studies have
from the data, extract information, analyze and predict future been conducted on the applicability, challenges and potential
trends. Having these scopes of AI, the implementation of of IPA. These studies include case examples [31], such as
RPA has been powerful than ever [23]. There are use cases automatic discovery and data transformation, auditing [32],
where an outcome is not 100% guaranteed to be correct. business process management, and productivity optimiza-
For example, value of a property, analyzing the risk of a tion [33]. There have been studies on different tools that have
loan default, and predicting inventory. If we want to value a made integrating AI in RPA easier. In our next section, we will
property, we need to consider some key factors such as the age focus on the tools.
of the house, the location and so on. Each of this needs to be
analyzed against each other to forecast the value. Earlier, only VI. TOOLS THAT ADAPT AI TECHNIQUES IN RPA
human brain could do such complex calculation. Now, AI can AUTOMATION
help robots to accurately predict the results using its advanced A. UiPath
techniques.There are cases where the outcome depends on UiPath is a software company which provides platform to
multiple variables. For example, finding matching resume create software bots for business process automation. It has an
for job placements. This a hectic job for recruiters. With the architecture based on web Orchestrator that is built on .NET.
help of AI, robots can automatically match candidate’s skill It facilitates the development and execution of RPA function-
sets for relevant job posting. Apart from that, robot alone alities through a block-based interface and various plugins
cannot deal with unstructured data (images, video or text). for business process automation. The platform operates with
But when advanced AI is combined with the robot, the robot three specific modules; UiPath Studio, UiPath Robot and
can learn the pattern of unstructured data too. Also, RPA UiPath Orchestrator. UiPath Studio is used to design, model
always provides a single kind of outcome. On the other hand, and execute workflows, while ensuring the management and
AI can work on both structured and unstructured data to transfer of packages between robots. The workflows are
produce different kind of outcomes and predictions. created by RPA developers. UiPath Orchestrator is used to
To enhance the functionality of both of these technologies, upload bots in cloud, deploy them and manage the resources.
there is a need to combine these two technologies. [24] Various configurations such as, saving login credentials of
mentions the integration of RPA, AI and soft computing websites or platforms, scheduling the runs, storing multiple
as Intelligent Automation (IA). The authors furthermore, bots are managed in Orchestrator. An RPA admin usually
categorize IPA as a part of IA. In our paper, we will mostly takes care of such things. There are two kinds of robots;
focus on IPA. An overview of IA technology is illustrated in attended and unattended. The attended robots need human
figure 4. intervention to complete tasks whereas unattended bots can
IPA [25] combines RPA and AI with other advanced run independently. UiPath offers five types of recorders:
technologies into one preconfigured software instance that is basic recording for individual activities, desktop recording for

VOLUME 13, 2025 177

 S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

updates to new versions and provides the ability to revert

to previous versions if necessary. AI Center enhances
understanding and engagement not only for data scientists
and RPA developers but also is a beneficial tool for business
users and leaders. By observing models and automation in
action within a single platform, leaders and managers gets
more visibility which helps them to understand what works
for their business and what does not.

4) CONSTANTLY MAKE THE MODELS SMARTER

The power of AI Center increases over time as models get
improved with the validation and learning of experienced
developers. Developers identify low-confidence predictions
FIGURE 5. Four stages of integrating ML models in AI Center.
made by model and update the model by retraining and
changing parameters. This eventually increases the potentials
capturing multiple actions across different applications, web of automation.
recording for recording activities in web browsers, and image
and Citrix recording for virtual environments, which support B. Kofax
image, text, and keyboard automation [34]. Kofax [36] is a company that automates business processes
According to UiPath documentation [35], UiPath has for companies and organizations. They have Kofax intelligent
now connected AI in the tool by introducing AI Center. Automation platform that includes capabilities such as AI,
AI Center allows to coordinate all the stages of AI life cycle, RPA, Cognitive Capture, Process Orchestration, Advanced
including deploying, utilizing, managing, and enhancing Analytics and so on. It allows data extraction from web,
machine learning models as shown in the following image 5. email, local files and ERP systems. It now incorporates AI
techniques and algorithms for document content, context
1) EASILY DEPLOY MODELS recognition, and information classification in emails, web
It allows for seamless integration of AI into the automation portals and documents. The tool uses machine learning
processes. It has made possible to oversee the models, approaches for supervised learning in OCR document recog-
monitor their development, interact with them, and validate nition and classification and natural language processing
predictions to improve accuracy over time. AI Center for both supervised and unsupervised learning to analyze
simplifies the deployment and scaling of AI in production content. Supervised learning is one of the versions of machine
environments. Users can upload their own models or choose learning. The algorithm in supervised learning learns from
from a variety of pre-built models offered by UiPath, UiPath’s the data about the labels during training phase by making
partners or the open-source community.It has made deploying repetitive prediction and correction. On the other hand, the
easy with just a few clicks. The user-friendly interface algorithm in the unsupervised learning generates inferences
allows data scientists and RPA developers to focus more without knowing the labels [37]. Natural language processing
on addressing business challenges and less on development is considered as a type of both supervised and unsupervised
operations and engineering tasks. machine learning which creates a way to analyze texts by
computerized means [38].
2) DRAG AND DROP MODELS INTO AN RPA WORKFLOW
It allows injecting ML models with its simple drag and drop C. AUTOMATION ANYWHERE
interface. It is easy to choose ML models from dropdown Automation Anywhere [34] is a leading provider of RPA
and inject them into the workflows developed in UiPath tools. It offers a client-server-based architecture. It has
Studio. As the ML models get trained on data, they recognize three main components; Bot Creator, Control Room, and
unstructured data and their patterns to make predictions. Bot Runner. The Bot Creator facilitates the easy design
This increases the scope of automation capacity by allowing and automation of bots. The Control Room handles the
companies to automate complex business models that has execution and scheduling of bots, manages credentials,
cognitive requirements. Also, as UiPath allows users to addresses security issues, and oversees client permissions
choose ML models, the data science team do not need to and assessments. The Bot Runner is responsible for running
spend much time on implementing models. This will increase bots and recording analytics which are then sent back to the
productivity and free up resources. Control Room. Automation Anywhere supports three types of
bots which are Task Bots, Meta Bots, and IQ Bots. Task Bots
3) MONITOR AND MANAGE THE MODELS are used for automating rule-based, repetitive tasks, while
The platform enables users to monitor machine learning Meta Bots serve as reusable building blocks for other bots.
model consumption at the tenant level. It allows for model IQ Bots, on the other hand, have cognitive and intelligent

178 VOLUME 13, 2025

S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

capabilities for processing unstructured data. The platform environments require AI to learn and adapt. The type of AI
offers three types of recorders—Screen Recorder, Smart agents are as listed below [40]:
Recorder, and Web Recorder—to automate tasks which Simple Reflex Agents: They rely on predefined rules and
mimics user actions. Additional features of Automation current data or signals. They are effective for straightforward
Anywhere include BOT INSIGHTS, an analytics engine that cognitive tasks that needs quick responses without in-depth
visualizes user data to provide business insights. Then there reasoning. For instance, they can reset customer passwords
is BOT FARM, which allows companies to purchase RPA when certain keywords appear in a conversation or adjust the
tools based on usage rather than capacity or licenses. Lastly it temperature in smart homes based on thermostat readings.
offers BOT STORE, a marketplace where various plug-and- Model-Based Reflex Agents: Unlike simple reflex agents,
play bots are available. Model-based reflex agents utilize an internal model con-
Automation Anywhere has added a functionality called structed from data inputs and perceptions to make informed
AI Agent Studio in their system to build AI Agents [39]. decisions. This capability makes them suitable for more
AI agents are software programs designed to interact with complex tasks that involve predicting outcomes such as
their operating environment that perform cognitive tasks recommending driving routes based on maps and real-time
and make decisions autonomously or semi-autonomously location updates.
by analyzing data using machine learning algorithms. The Goal-Based Agents: Goal-based agents do not just follow
features of AI agents are as follows [40]: rules in straight-forward manner. Rather they, evaluate
Purpose-Driven: AI agents are designed with specific methods using its reasoning capability for achieving a task.
objectives rather than following a set sequence of steps. They For instance, these agents can assist in scheduling tasks that
utilize all available capabilities and information, including need to meet specific deadlines or constraints.
organizational knowledge and environmental perceptions to Learning Agents: As their name suggests, learning agents
conduct their assigned tasks. constantly adapt and enhance their performance through
Autonomy/Partial Autonomy: AI agents can operate based experiences and collected. This approach makes them ideal
on their on intellectual without the need of any human for tasks like recommendation engines, which better align
intervention. They are smart enough to take decisions and act with user preferences as they learn from feedback and
accordingly to target their goals. interactions.
Memory: AI agents keep track of their plans, experience
and communication to ensure continuity of their work, inform D. BluePrism
future actions and update in their performance. BluePrism is another leading company that provides solution
Planning: AI agent create sequence of their tasks and work for automation using RPA [34]. It uses Java and .NET
according to the objective. Their planning capabilities have frameworks and offers drag-and-drop interface for designing
improved with the integration of Large Language Models RPA solutions. It has four main components; Process
(LLM). diagrams, process studio, object studio, and application mod-
Perception: AI agents continuously perceive and process eller. Process diagrams represent business workflows which
new information in real-time. Using sensors and data inputs, are created utilizing core programming concepts. Process
they can assess their surroundings and adjust their actions studio offers a platform to create these diagrams using
when needed. various drag-and-drop activities. Object studio is used to
Reasoning: AI agents have the ability to make decisions create Visual Basic objects that enable communication with
and solve problems. By processing new information and other applications. Application Modeller creates application
connecting their perceptions of the environment or situation models within Object Studio which exposes the UI elements
to their foundational knowledge and predefined objectives, of a target application to the Blue Prism program.
AI agents is able to determine the optimal course of action. According to the documentation of BluePrism [41] they
Action: AI agents are capable of taking action. They have introduced Intelligent Automation Platform (IAP)
can interact with and influence their environment to where it combines the power of AI and ML in RPA. The
meet their assigned goals and objectives by being con- features of IAP are described as follows:
nected to actuators and execution systems, such as ERP Design Studio: Design Studio allows to build automated
applications. processes with a simple drag-and-drop interface so we don’t
Reference [40] also describes different types of AI agents. need to write any code. It also allows to create reusable
Each AI agent type, whether simple reflex or hierarchical, ‘objects’ that can trigger events within a process. This makes
possesses distinct strengths and characteristics that make it easy to scale automation across the entire business.
them appropriate for different enterprise automation situa- Digital Workforce: A digital workforce consists of
tions. AI agents are categorized based on traits like reactivity autonomous software robots that use AI technologies to
or proactivity, the characteristics of their environment, mimic human intelligence and learning.
and whether they function individually or within multi- Control Room: The Control Room trains digital workers
agent systems. For instance, static environments may need and assign them specific tasks while providing real-time
agents that can perform responsive tasks, whereas dynamic insights into how these processes are performing.

VOLUME 13, 2025 179

 S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

Blue Prism’s IPA solutions deliver value in five different

ways [41]. One is creating journeys. It helps assessing,
advancing and improving processes. Second one is, accel-
erating work. This adds benefits by automating tasks using
insights for complex works such as processing documents.
The third one is, transforming experience which helps to
transform business by empowering workforce with new
experience. Last one is, offering an enterprise platform and
robotic operating model that makes the solution scalable, FIGURE 6. The combined architecture of RPA and AI [47].

secure and offers flexible deployment. So, the overall

technology that are used to create intelligent automation in
BluePrism are: RPA, NLP, AI, Business process management defined tasks. Their system architecture can be divided into
(BPM), ML, process discovery and process mining. A three phases like the following figure 6.
comparison between the highly used RPA tools (UiPath, Data Collection: Here, RPA is used to collect data from
Automation Anywhere and Blue Prism) is also presented in different sources such as financial systems, emails and local
Table 2. servers. It captures and records voice commands received by
Apart from the above mentioned tools, there are few more microphone. RPA streamlines this data collection process by
tools where AI and RPA are integrated together to automate conducting repetitive data gathering tasks.
processes. Such as, AssistEdge [42], Automagica [43] and Data Processing: After that AI is used to recognize
WinAutomation [44]. In the next few sections, we will talk and transform user’s voice into text, which is basically
about the use cases and applications of IPA solution across instructions from users. By using feature extraction and other
various industries. advanced techniques, the robot management and dispatching
center turns the texts into meaningful instructions and
VII. AI DRIVEN RPA IN FINANCE AND BANKING forwards them to robot actuator for task execution.
The development of financial process intelligence has Report Generation: Lastly, the robot actuator conducts
leveraged advances in technologies such as cloud computing business actions by following the instructions. In terms of
and AI to enhance process automation. RPA is utilized result, it provides financial query, tobacco marketing query,
to mimic user interfaces through cloud computing which large screen control and etc, and returns the final output to the
facilitates extensive automation of tasks like data entry, end user through a user interface.
form filling, and email handling in financial operations. The architecture incorporates additional technologies like
IPA integrates AI algorithms to handle unstructured data, cloud computing and blockchain to improve interaction
make complex decisions, and optimize processes through between RPA and AI. The main goal of this study is to smartly
learning and feedback. This combination enables financial automate routine tasks and improve decision-making through
organizations to automate back-office operations, reduce collaboration between RPA and AI.
labor costs, and increase productivity [46]. Banking industry
can also be benefited with AI based RPA systems. It helps In another study [48], they discuss about some use cases
banks in various ways starting from fraud detection to loan where RPA and AI can be applied in the field of Banking.
processing. And finally, this can increase bank’s revenue,
lower operating costs and compliance and increase customer 1) USE CASE 1: LOAN PROCESSING
satisfaction. The loan process generally involves the customer completing
numerous forms, which the bank then verifies. The bank
A. APPLICATIONS AND USE CASES: FINANCE AND checks customer’s credit history before approving the loan
BANKING and disbursing the funds. Intelligent automation powered by
Kai et al. [47] describe how AI and RPA are combined to AI and RPA can play a major role here to automate the end-to-
enhance the financial office’s intelligence and automation end processes and transform business operations. The figure 7
capabilities. This architecture utilizes RPA for conducting bellow depicts a recommended way for an automated load
repetitive tasks accurately while AI provides cognitive processing focusing on the applicability of AI and RPA.
abilities for intelligent decision-making. Here, they create
a robot that includes dual architecture where RPA acts like a: LOAN APPLICATION AND DOCUMENT SUBMISSION
‘‘hands’’ and AI acts like ‘‘brain’’ of the robot. In the loan application process, while some banks have
This design is based on two main components: digitalized systems for submitting applications, others still
human-computer interaction and software workflow. The rely on paper forms. For these banks, RPA using its
human-computer interaction focuses on how financial staff OCR (Optical Character Recommendation) technology can
deal with software to ensure smooth communication and automate data extraction and entry to bank’s loan system.
task completion. On the other hand, the software workflow Since the data entry is a manual and error prone process, here
component makes interaction between AI and RPA to execute RPA can help improving accuracy and efficiency. also, as loan

180 VOLUME 13, 2025

S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

TABLE 2. Comparison of RPA Tools: UiPath, Kofax, Automation anywhere and Blue Prism.

FIGURE 8. AI component for document verification [48].

features within a specified proximity. Naïve Bayes classifier

is a probabilistic classification that uses Bayes theorem
of probability to identify the class of a new or unknown
data. Neural Network focuses on underlying relationships of
FIGURE 7. AI and RPA integration in loan processing [48].
characters and words to extract data which is considered as
a strong ability in [51]. Lastly, SVM has been found to be
a better performer than any other classification algorithm in
the context of OCR as it does not work based on assumptions
documents usually follow a standard format, using RPA is as Naïve Bayes classifier does. The figure 8 illustrates the
often sufficient without applying advanced AI. However, this system component where AI is applied.
approach should be assessed for feasibility. After data extraction, the information from documents
is verified against trusted sources via system integration
b: VALIDATION AND SCREENING that may run in real-time or batch mode. Customer profile
The next step is to validate the documents submitted in and transaction screening involve specialized systems that
banking system. To validate the data, system needs to compile data from multiple sources. This requires bank
extract correct data and then verify them against correct agents to input data manually for analysis which is a labor-
source. For example, the driving license should be verified intensive process. This is where RPA can again help and
against trusted motor vehicle database. For this, there is automate data entry, especially when third-party systems
a need to apply artificial intelligence to do document may not always support API calling or other methods.
processing and data extractions. This requires techniques Additionally due to dealing with sensitive data, these systems
like segmentation, text processing, text retrieval, paragraph are typically hosted on-premises within the client’s network.
retrieval and key value pair extractions. This will provide Therefore, RPA offers a non-invasive and cost-effective way
them text version of documents such as PDF or images. to automate the repetitive data entry process. This helps from
After that, implementation of AI will do text analysis to preventing manual error and speeds up the process as well.
extract the correct meaning of the data specially from
paragraph like texts. This approach is referred to as Intelligent c: UNDERWRITING AND RISK ASSESSMENT
Automation (IA) [49]. Research indicates that text analysis The subprocess of evaluating a customer’s creditworthiness
from documents like PDFs requires both OCR and NLP. is vital in the loan process which is shown in the following
Most OCR methods use an algorithm that is trained on a figure 9.
known dataset to accurately label or classify the character It involves analyzing data from various sources to assess
set in the dataset. According to [50], some of the popular lending risks, detect potential fraud, and evaluate the cus-
algorithms used in OCR techniques are K Nearest Neighbor tomer’s financial stability. RPA bots automates data collec-
(KNN), Naïve Bayes Classifier, Neural Network and Support tion from multiple sources or even crawl portals to fetch data,
Vector Machine (SVM). KNN classifies objects with similar while AI uses algorithms to predict risk assessment scores.

VOLUME 13, 2025 181

 S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

FIGURE 10. RPA and AI solutions for banking fraud operations [48].

FIGURE 9. Data extraction and validation solution [48].

raised by the use cases that were covered are listed below,
along with some more information from pertinent research
This combination of RPA and AI enhances the efficiency papers.
and accuracy of credit evaluations while assisting bankers
in making informed decisions. Common AI algorithms used 1) DATA QUALITY AND INTEGRATION ISSUES
include K-Nearest Neighbors, Logistic Regression, Decision Both studies highlighted the critical role of data collection,
Trees, and Neural Networks, with service providers offering processing and validation in the financial and banking areas.
simplified solutions for risk assessment modeling. However financial institutions often rely on legacy systems
that may not be compatible with modern AI and RPA tech-
d: LOAN APPROVAL AND DISBURSEMENT nologies that leading to data inconsistencies and integration
Next step is, loan approval and payment disbursement. difficulties. And so ensuring high data quality and seamless
This subprocesses can be automated using workflows, API integration across multiple systems is a critical challenge
integration, or RPA, depending on the requirements. While when implementing AI and RPA in banking. Inaccurate or
RPA offers quick, temporary solutions, API integration is incomplete data entries can lead to incorrect AI predictions,
considered as a long-term, approach. The choice between which can be particularly problematic in processes like loan
RPA and API should be based on factors such as speed, approval or risk assessment [47]. Integrating various systems
cost, and benefit. Applying AI and RPA in the loan process and ensuring data flows seamlessly between AI and RPA
can transform operations which significantly improves KPIs components is also challenging due to legacy systems and
through straight-through processing and better customer fragmented IT environments.
experiences. These technologies have a direct and positive
impact on the efficiency and effectiveness of loan services. 2) REGULATORY COMPLIANCE AND RISK MANAGEMENT
When using AI and RPA to automate tasks like fraud
2) USE CASE 2: FRAUD DETECTION detection and loan processing one must take into account the
The fraud detection operations process involves data gather- dynamic and complicated regulatory environment. A major
ing and analysis to find patterns, anomalies and outliers that problem in the banking business is that it adheres to tough
may suggest fraud. It includes alert responses, transaction regulations such as GDPR, AML (Anti-Money Laundering)
investigations and potentially taking actions such as reversing and KYC (Know Your Customer) etc. For regulatory
transactions or initiating legal proceedings. RPA and AI can compliance it is imperative that AI predictions are auditable
significantly enhance efficiency and KPIs in fraud detection and explicable particularly when it comes to loan processing
and investigation. RPA bots streamline data preparation and fraud detection [48]. Furthermore, updating AI models
by navigating systems and organizing data, while AI and and RPA workflows on a regular basis is necessary to
machine learning models identify patterns and relationships maintain compliance due to the changing nature of regulatory
to support investigations. Both supervised and unsupervised environments. When it comes to the application of RPA
machine learning algorithms are used to detect hidden in banking, Patri [52] explores the difficulties associated
patterns and relationships which improves the effectiveness with regulatory compliance. Through the implementation
of fraud detection and decision-making. Figure 10 shows four of strong governance frameworks and transparent and
phases of fraud detection operation where RPA and AI can be explainable AI decisions, the paper offers insights into how
implemented together in three phases. banks may effectively manage these obstacles.

B. CHALLENGES: FINANCE AND BANKING 3) SECURITY AND PRIVACY CONCERNS

Although AI-powered robotic process automation (RPA) has Security and privacy are major concerns since the banking
great potential to revolutionize a number of banking and industry uses AI and RPA to process massive amounts of
financial services, putting it into practice presents a number sensitive personal and financial data. Financial institutions
of limitations. These limitations are a result of operational, need to make sure that data protection laws are followed while
ethical, legal, and technological aspects. The main issues safeguarding this information from cyberbullies. AI models

182 VOLUME 13, 2025

S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

used in fraud detection and other critical banking operations invest in sophisticated data cleaning tools to address problems
can become targets for adversarial attacks, where malicious with data integration and quality. This will help close
actors attempt to deceive the AI system. Moreover the the gap between legacy systems and contemporary AI-
integration of AI and RPA increases the attack surface RPA technology. To address security and privacy concerns,
and potentially exposing banks to higher risks of data financial institutions must implement advanced encryption,
breaches [48]. The study by Iyamu and Mlambo [53] provides regular security audits, and adversarial defense mechanisms
an analysis of security and privacy challenges in the context to safeguard sensitive data. Finally, ethical concerns around
of RPA implementation in the banking sector. The authors bias in AI models require ongoing monitoring, regular
emphasize the importance of addressing these concerns re-training on unbiased datasets, and active human oversight
through robust security measures and privacy frameworks to to ensure fairness and equity in processes of the banking
protect sensitive data. The study by Choubey and Sharma [54] industry. By adopting these measures, organizations can
examines the impact of RPA on sustainable banking in India, mitigate the challenges and fully leverage the potential of
highlighting that despite potential risks related to security and AI-powered RPA in financial services.
privacy, the adoption of RPA in banking is expected to grow
significantly. The research emphasizes the future integration VIII. AI DRIVEN RPA IN AUDIT
of robots with human workers, predicting that RPA will Auditors usually deal with heavy workloads as they have
enhance efficiency and support sustainable economic models to meet strict deadlines in most cases. A survey of over
in the banking sector. 700 auditors examined the relationship between audit work-
loads, perceived audit quality and job satisfaction. It shows
4) SCALABILITY AND SYSTEM INTEROPERABILITY that auditors need to work an average of five hours per
The challenge is in ensuring that AI and RPA projects week beyond their schedule which decreases the quality
by banks and financial institutions can handle growing of their work [57]. Strict deadlines and the shortage in
workloads and seamlessly interface with a diverse range number of resources are considered to be the primary factors
of other tools and systems. When bottlenecks arise from for the excessive workloads. Furthermore, the survey found
interoperability problems, automated operations become less that auditors’ job satisfaction decreases significantly when
efficient overall Kai et al., [47]. In the context of smart workloads surpass the level that negatively impacts audit
RPA application in financial operations, Gotthardt et al. quality. These findings align with the PCAOB’s concerns
[55] describes the difficulties with scalability and system that heavy workloads threaten audit quality and suggest that
compatibility. To guarantee that AI and RPA systems can the main drivers of excessive workloads could be the root
scale efficiently and interface with other systems without cause of workload-related audit deficiencies. If the audit is
any problems, the study emphasizes the necessity of strong conducted by external agencies, there is a possibility that
frameworks. there will be corruption and loopholes [58]. To address
these challenges, public accounting firms are exploring
the potentials of RPA [59], [60], [61], [62] and AI [63],
5) ETHICAL AND BIAS CONCERNS
[64], [65] to enhance the capability of their audit software.
AI models have the potential to unintentionally add biases
The term automation is not new to managers of auditing
and provide unfair results, especially when applied in
firms as it always requires multiple repetitive tasks such
procedures like fraud detection and loan processing. For
as reconciliations, internal testing and detail testing. Earlier
instance, if the training data has ingrained biases, AI-
audit firms were dependent on scripts written in Python
driven loan approval systems may prejudice against particular
or R language. For that they needed to hire developers or
demographic groups. Since AI models frequently learn from
programmers which was a major investment of money. That’s
potentially biased historical data, it is imperative yet difficult
why audit firms are not eager to implement automation
to ensure that these algorithms are transparent, equitable, and
through RPA [58] that requires low code or no code.
devoid of prejudice [48]. Addressing these ethical concerns
However, there have not many researches conducted on AI
is essential to maintain trust and avoid reputational damage.
powered RPA audit processes. We will talk about some of the
The ethical and bias-related issues surrounding the use
applications and use cases in our next section.
of AI and RPA in the banking industry are addressed in
the paper by Jaiwani and Gopalkrishnan [56]. The authors
mention how rigorous examination of AI models is necessary A. APPLICATIONS AND USE CASES: AUDIT
in high-stakes domains like loan approvals and fraud Reference [66] provides a framework for implementing IPA
detection to guarantee that they do not reinforce preexisting in audit. They divide their framework in three sections; Ana-
biases. lyzing the audit workflow, automating individual audit tasks
While the challenges discussed above are significant, and forming the IPA structure. Analyzing the audit workflow
they can be addressed through targeted strategies and involves determining the primary, secondary and lower-level
innovations. Banks may create middleware solutions and workflows. They mention that, a detailed understanding

VOLUME 13, 2025 183

 S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

B. CHALLENGES: AUDIT
The integration of AI and RPA into audit procedures raises
a number of important challenges that differ from those
faced by the banking and financial industries. Although
there are many advantages to IPA in auditing, there are
also many drawbacks, including the intricacy of audit
activities, the requirement for accuracy, and the integration
of many technologies. Here, we focus on the particular
difficulties caused by various implementations, as supported
FIGURE 11. Framework of IPA implementation in audit [66]. by corresponding studies.
Processing unstructured data is an important challenge
for automating audit tasks. Artificial intelligence (AI)
of the workflows is necessary to evaluate if the current approaches such as natural language processing (NLP) and
workflows need any modifications. cognitive tools are necessary to handle unstructured data,
Once the analysis is done, next step is to automate whereas RPA solutions can handle structured data with ease.
individual audit tasks. The tasks can be categorized into These technologies must accurately interpret and analyze
structured, semi-structured and unstructured task. Tools from unstructured information such as client communications or
RPA are utilized for automating structured tasks such as data complex financial statements which is often nuanced and
extraction and file organization. Semi structured tasks can context-dependent. Ensuring that AI systems can reliably
be conducted by AI technologies such as natural language handle these tasks without human oversight is a significant
processing and computer vision. Unstructured data are challenge. The study by Zhang [66] emphasizes the difficulty
supported by cognitive tools as they need expert’s attention. of automating semi-structured and unstructured audit tasks.
This is to ensure final output is accurate. The paper discusses the need for sophisticated AI tools to
The final step is to integrate both the automated and support these tasks and highlights the risks of errors if these
augmented tasks in audit workflow to form a loop called IPA. tools are not properly implemented.
The figure 11 below illustrates the IPA loop. Moreover, the integration of these automated tasks back
After automating or augmenting individual audit tasks, into the audit workflow poses another layer of complexity.
these tasks are reintegrated into the layered workflow until Each automated or augmented task must seamlessly interact
the primary workflow loop is completed. For instance, during with others within the layered workflow. Any disruptions
the audit planning phase, previously automated tasks might or failures in one part of the loop can have cascading
include file organization via RPA, file processing through effects, leading to delays or errors in the audit process. For
NLP, and client risk assessment supported by a cognitive example, if NLP tools misinterpret critical data during the file
assistant. The next step involves organizing these tasks back processing phase, this can lead to incorrect information being
into the workflow. Files organized by RPA are sent to passed to cognitive assistants, potentially resulting in flawed
NLP for essential information extraction, and the results are risk assessments. The integration challenge is exacerbated by
then passed to the cognitive assistant, which helps auditor the need for these technologies to work harmoniously in a
understand client’s risk level. Once the auditor assesses the dynamic and often unpredictable audit environment.
client’s risk with cognitive tool assistance, the information Additionally, there is a challenge in maintaining the
is forwarded to the internal control testing workflow. This continuity and accuracy of the IPA loop over time. As work-
process of completing lower-level tasks and passing results to flows evolve and new audit requirements emerge, the IPA
higher levels continues until the audit engagement concludes framework must be adaptable enough to accommodate these
which completes the primary workflow. If necessary, work- changes without significant reconfiguration. This requires
flows can be re-triggered, updating all subsequent workflows ongoing updates and fine-tuning of the automated tools,
and forming a continuous loop. which can be resource-intensive. Moreover, ensuring that the
According to another study [63], EY, one of the Big entire workflow remains synchronized is a complex task,
4 audit firms, develops analytics tools for the purpose of as each update or modification to the workflow can affect
analyzing huge amount of data in a shorter period of time. multiple layers of the IPA structure.
EY also employs RPA technologies to automate most time- An additional difficulty lies in the possibility of auditors
consuming, repetitive, traditional and rule-based tasks and experiencing cognitive overload when utilizing these sophis-
leave the complex tasks for AI. ticated technology. Although the goal of cognitive tools is
Reference [67] mentioned that the HR group of Coca-Cola to help auditors make better decisions, they also present
is running HR Automation to run audits in SAP which is auditors with additional interfaces and data streams to handle.
enabled by Blue Prism’s intelligent automation. So far, more The possibility exists that auditors could get overpowered
than 50 processes are automated by this beverage giant across by these technologies’ complexity, especially if they lack the
multiple SAP systems which helps them deliver services necessary training to utilize them properly. This increases the
24 hours a day with no additional staff required. possibility of errors or a dependence on technology without
184 VOLUME 13, 2025
S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

enough critical monitoring, especially in the setting of an

audit where judgment and accuracy are crucial.
Finally, it is impossible to overlook the difficulty of
guaranteeing the IPA process’s auditability and transparency.
The necessity to make sure automated processes are visible
and auditable independently is developing as audits depend
more and more on them. This means that all choices made
FIGURE 12. Telemedicine workflow with RPA and AI.
by AI tools, as well as the procedures they adhere to,
ought to be completely transparent and intelligible to human
A. APPLICATIONS AND USE CASES: MEDICAL AND
auditors. It is challenging to maintain this degree of openness,
HEALTH SERVICES
particularly as AI models grow increasingly intricate and
difficult to understand. There may be questions regarding the According to [69], the design of Randomized Clinical Trials
dependability of the IPA-integrated audit process if there are (RCT) involves collecting huge amount of data to utilize them
unclear audit trails, as it may be difficult to confirm the audit’s accurately. As data needs to be entered manually, RPA can
accuracy. help managing the data in the main file by following a fixed
Gotthardt et al. [55] highlight several critical challenges data structure. After that, AI enabled system can do data
in implementing Smart Robotic Process Automation (RPA) verification to check if there is any missing data. With trained
in auditing, including data quality and integration issues, bots, the missing data can be filled as well.
cybersecurity risks, and the need for transparency and Reference [70] mentions several use cases on AI enabled
governance. The paper also emphasizes the complexity RPA system which are as follows:
of implementation and the resistance to change within
organizations, pointing out that successful adoption requires 1) CALCULATING THE PRECISE USAGE OF MEDICINES
careful planning, ongoing maintenance, and a focus on This study shows that RPA can be used to collect and manage
training to bridge skill gaps. data related to drugs. They give an example of a system
To address these challenges holistically, organizations that can be installed to monitor the use of medicines or
should adopt a proactive approach that combines technolog- drugs. The system can be used to predict the number of
ical innovation with robust governance and training frame- medicines required based on the saved record of previously
works. Collaborative efforts between AI developers, auditors, admitted patients and their common health issues. Then
and management can ensure that automation tools are both using RPA, the system can automatically generate emails to
effective and aligned with audit requirements. Additionally, authorities so that they can place order of medicines before
fostering a culture of adaptability and continuous learning they get out of stock. Furthermore, the AI-enabled hospitals
within audit teams will help mitigate resistance to change can communicate between each other to estimate the number
and ensure long-term success. Together with continued of drugs and predict the rise in the price of the same drugs.
investments in modular automation systems and explainable Through communication, hospitals can also forecast medical
AI, these strategies can help reduce the gap between emergencies, alert doctors, get supply of medical facilities,
the present constraints and the revolutionary potential of prepare infrastructure, plan enough resource in terms of
AI-powered RPA in auditing. shifts.

IX. AI DRIVEN RPA IN MEDICAL AND HEALTH SERVICES 2) TELEMEDICINE

Technologies applied to medical administrative services are There are apps that can turn flashlights into an electrocardio-
changing over the years and it has now impact on how dis- gram (ECG) device which helps doctors to diagnose various
eases are treated. Doctors are also depending on automation heart-related conditions. Reference [70] shows, how apps like
process to diagnose problems and even treat patients. Besides, these can be used to collect data through RPA and create
automation can help in managing people, supplies and taking a database system with that data. While, RPA can be used
care of financial documents in medicals, hospitals and health to manage the data, AI system can recommend medications
care clinics. RPA has gained its popularity in healthcare based on patient’s health condition [71]. In the final stage,
for automating repetitive tasks, optimizing processes and doctors can review, modify and approve the medications
minimizing human errors. Leveraging RPA has also increased for patients. Once approved and payment is completed, the
productivity and helped in resources management which prescriptions can be sent to nearby pharmacies where patient
eventually results in better customer experience and job is located. The entire telemedicine workflow, including data
satisfaction [68], [69]. On the other hand, AI and ML collection, management, AI-based recommendations, and the
have proven its diagnostic capabilities by analyzing large doctor’s final review, is visually represented in Figure 12.
datasets, identifying patterns and generating insights. While
RPA and AI technologies offer unique benefits when used 3) CELL AND TISSUE ENGINEERING
individually, their integration shows a paradigm shifts in Cell and tissue engineering are at the cutting edge of medical
healthcare diagnostics. research. Cell growth is categorized into autologous therapy,

VOLUME 13, 2025 185

 S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

which is personalized using the patient’s own cells and does respond to other systems and handling repetitive tasks such
not require mass production, and allogeneic therapy, which as data entry, file transfers, document filing, and information
aims to scale production for widespread use. However, due to access. This automation helped healthcare organizations
phenotypic drift, cell engineering is generally feasible only minimize tedious tasks for their employees, leading to
on a small scale. To address these challenges, engineers are increased productivity and social distance during COVID-19.
turning to intelligent automated systems, such as AI-enabled Systems like these also helped saving cost during crisis.
RPA [72], [73], [74].
8) AUTOMATION IN HEALTH INSURANCE DURING
4) DATA CURATION COVID-19
While data availability is not an issue today, data quality Hospitals used blue Prism to implement RPA bot in their
remains a concern. Medical images, such as CT and MRI machine. These bots in health insurance can skip heavy
scans, can be challenging to interpret, which may lead to coding and work on desktop software such as Excel to create
errors in AI classification tasks. Using properly curated data a standardized process. By taking care of these tedious jobs,
with benchmark evaluations can enhance model accuracy it allows machines to do higher order tasks such as using
and robustness. RPA can facilitate data collection and natural language processing to improve documentation.
curation, including steps like obtaining ethical approval,
accessing databases, querying, re-identifying, transferring 9) IP MANAGEMENT IN HEALTHCARE DURING COVID-19
data, performing quality control and preprocessing tasks like During COVID-19 in healthcare, RPA with AI was employed
resizing and labeling. According to a framework proposed in Intellectual Property (IP) Management to protect innova-
by [75], RPA can also be helpful in eliminate duplicates while tions, automate workflows, and streamline business opera-
data preprocessing. tions. This integration is beneficial for IP Audit and Data
Mining as it safeguards procedures, and identifies potential
5) RADIOLOGY
IP assets while minimizing costs and effectively managing
intellectual property.
An average radiologist on an 8-hour shift usually interprets an
image every 4 seconds which creates significant pressure on
B. CHALLENGES: MEDICAL AND HEALTH SERVICES
the individual, especially with multiple patients waiting [76].
The integration of AI-powered Robotic Process Automation
This workload can have negative impact on decision-making
in the medical and health services sector presents several
which may lead to provide wrong treatment to patients [77].
significant challenges, particularly concerning the implemen-
In such situations, RPA and AI can be highly beneficial.
tation and operationalization of these advanced systems.
While RPA can be used to do repetitive tasks such as to
Data collecting and processing by AI powered RPA system
sort and collect image data, an AI system would be used
is one of the major challenges when it comes to ensuring its
as a supervised learning model, trained with both input and
accuracy and completeness. In the context of Randomized
output data. Once trained, the model can classify anomalies in
Clinical Trials (RCTs), for instance while RPA can automate
images automatically which will eventually reduce workload
data entry and management the quality of the data entered
for radiologists.
is critical. AI systems that perform data verification rely
heavily on the initial data’s accuracy. Errors in data collection,
6) SELECTING ALGORITHM BASED ON TASKS whether due to incomplete or incorrect information, can lead
Medical imaging includes a wide range of specialties, includ- to flawed analyses and, consequently, incorrect clinical trial
ing ophthalmology, pathology, cardiology, and radiation outcomes. Moreover, filling missing data with trained bots,
oncology, each with unique data distributions. Algorithms although innovative, introduces risks of compounding errors
that work for one specialty may not be suitable for another. if the imputation models are not sufficiently accurate or
It also depends on the type of tasks. RPA can assist in representative.
identifying the appropriate AI tools and algorithms to apply Cell and tissue engineering, especially for large-scale
based on the specific task and data. Once these technicalities allogeneic therapies faces significant hurdles due to the
are established, the data can be transferred into the training complexity and variability of biological data. The phenotypic
pipeline. drift and the need for precise control over the environment for
Another paper [21] focuses on how RPA with the help of cell growth pose unique challenges for AI and RPA. Scaling
AI emerged as a transformative tool in the healthcare industry up these processes while maintaining quality and consistency
during the COVID-19 pandemic. Few of the cases are listed is difficult and current AI-powered RPA systems may not yet
below: be fully equipped to handle these complexities at a larger
scale without extensive human oversight.
7) SOCIAL DISTANCING DURING COVID-19 Telemedicine applications powered by AI and RPA
By utilizing software robots or AI, RPA significantly reduced introduce challenges related to patient data privacy, informed
the need for interpersonal interactions. It automated business consent and the reliability of AI recommendations. The
processes by upgrading applications to interpret, interact, and prescription process can be automated by these systems

186 VOLUME 13, 2025

S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

which might increase efficiency, however there are concerns as resistance could impede the effective implementation of AI
regarding the transparency and explainability of decisions in rural healthcare settings.
made by artificial intelligence. For example, patients and Ensuring data accuracy in RCTs and telemedicine systems
doctors may have limited insight into how an AI system necessitates robust validation frameworks, with AI models
reaches its conclusions. It can also raise ethical issues about trained on high-quality, diverse datasets. For cell and tissue
the system’s accountability, especially if an AI’s recom- engineering, hybrid models combining AI automation with
mendation leads to adverse patient outcomes. Additionally, expert human oversight can help manage biological variabil-
regulatory frameworks may not be fully adapted to govern the ity and ensure quality control at scale. To mitigate privacy and
use of such technologies which complicates the widespread ethical concerns in telemedicine, integrating explainable AI
adoption of AI-powered RPA in telemedicine. (XAI) and transparent decision-making processes into RPA
While IPA systems can significantly reduce the work- systems will enhance trust and accountability. In radiology
load of radiologists by automating image classification and medical imaging, domain-specific AI algorithms that
and data management tasks, there are challenges related undergo rigorous validation and are continuously updated
to the system’s reliability and the potential for errors. with diverse imaging data can help improve reliability.
The high-stress environment in radiology, combined with It is also necessary to invest in user-friendly interfaces
the critical nature of accurate image interpretation means and thorough training programs for patients and healthcare
that even minor errors in AI-driven image analysis can personnel in order to overcome opposition to AI in rural
have severe consequences. Ensuring that AI systems are healthcare settings. Encouraging broad adoption will require
sufficiently trained and validated to handle the diverse range establishing trust by open and honest communication about
of imaging data encountered in practice remains a substantial the advantages and drawbacks of AI-powered solutions.
challenge.
The diversity of tasks in medical imaging, such as varying X. AI DRIVEN RPA IN RECRUITING INDUSTRY
specialties and data types, necessitates highly customized AI Another excellent application for deploying an Intelligent
algorithms. RPA can assist in selecting these algorithms but Process Automation or AI enabled RPA system in large
the challenge lies in ensuring that the chosen algorithms are enterprises is in recruitment industry [49]. IPA can streamline
appropriate for the specific context. The heterogeneity of the recruitment process by helping recruiters collect resumes
medical data means that a one-size-fits-all approach is not from various online portals, analyze specific skill sets, assess
feasible and substantial effort is required to match the right candidate value, and filter out spam and unwanted appli-
AI tools to the right tasks. This requires ongoing monitoring cations. This technology allows recruiting team to identify
and adjustment as well as deep domain knowledge to guide suitable candidates more efficiently and cost-effectively. IPA
the selection process. not only discards non suitable applications but also aids in
Kamala Venigandla’s paper [75] identifies several key sorting relevant resumes and ensure access to applications
challenges in integrating RPA with AI and ML for improving that closely match job requirements. It supports HR personnel
diagnostic accuracy in healthcare sector. Interoperability throughout the entire recruitment process, from screening and
issues, concerned about data security and privacy and the evaluation to final integration and management. Recruitment
necessity of rigorous adherence to regulatory compliance, industries get benefited not only in the hiring process but
such as the Health Insurance Portability and Accountability also in financial functions. According to Robert Half, the
Act (HIPAA) in the US are some of these barriers. Healthcare world’s first and largest specialized talent solutions firm,
companies also need to spend heavily in infrastructure and many companies, including theirs, are now adopting IPA to
guarantee personnel preparedness by providing sufficient streamline and enhance complex tasks, such as tax and com-
training and assistance. The paper emphasizes the importance pliance reporting and financial statement reconciliation [79].
of ethical data usage and careful planning during the
implementation of these integrated systems. A. APPLICATIONS AND USE CASES: RECRUITING
Saurabh A. Pahune’ [78] highlights several critical chal- In this section we will talk about two use cases from two
lenges in deploying AI-powered healthcare systems in rural different studies.
areas. Among these the integration of AI into existing
clinical workflows poses significant difficulties, particularly 1) USE CASE 1
if the technology demands substantial changes to established Reference [80] proposes a system including RPA and AI
processes. Physicians and patients who may be beware that will help HR recruiters to rank resumes based on job
of AI technology or concerned about how it may affect descriptions. At first, the job requirements are sent to the bot
clinical decision-making have also demonstrated substantial called PrimeBot via emails. The robot then downloads the
resistance. In order to enable the effective adoption of AI email to a specific folder. The description can be C++/SQL
systems, the paper [78] highlights how crucial it is to make which will act as a keyword for the robot. The robot then scans
sure that these systems are easy to use and smoothly integrate various resumes for these keywords and if a keyword is found
into current processes. Moreover gaining the trust and in a resume, the resume is ranked based on the percentage of
acceptance of both healthcare providers and patients is crucial keyword matches. The robot parses resumes in any format

VOLUME 13, 2025 187

 S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

(PDF, DOC, TXT) and ranks them accordingly. The output is

generated in a CSV file. After that, the resumes are moved
from one folder to a final folder based on the percentage
criteria. The final folder including the resumes in CSV format
is then emailed to the HR recruiter.
For ranking, they propose to use the SVM (Support Vector
Machine) classifier. They describe the logic of how SVM
works. Given two items, item1 and item2, the SVM classifier
determines their order. If item1 is meant to be ranked before
item2, the input to the classifier is (item1, item2) and the
output is 1. This signifies the correct order. Conversely, if the
input is (item2, item1), the output is -1. This indicates that the
order is incorrect.
The ranking process for a set of three items, (item1, item2,
item3), using the SVM classifier is illustrated by assigning
scores based on pairwise comparisons:

(item1, item2) = 1
(item1, item3) = 1
Score of item1 = 1 + 1 = 2 FIGURE 13. AI and RPA-powered resume screening system [80].

(item2, item1) = −1 words to pull information such as their name, phone number,
(item2, item3) = 1 address and technical skills. An account will be created
Score of item2 = −1 + 1 = 0 automatically in the portal with the information. At this point,
the system should be able to determine whether the candidate
is suitable for the position or not. If the candidate is not a
(item3, item1) = −1 good for the given position but possibly suited for a different
(item3, item2) = −1 position listed in their portal, they will be notified by email
Score of item3 = −1 − 1 = −2 and encouraged to apply for the position. If the candidate
is suitable for that position, an interview will be arranged
Based on these scores, the ranking is: item1, item2, and virtually by the system. There will be options to choose time
item3. through the system that meets with HR’s availability. The
This work not only demonstrates that SVM-Rank is par- email will also suggest a set of times for the candidate to
ticularly effective for ordering items represented as feature choose from.
vectors but also provides a solution that utilizes both the Although, this paper doesn’t mention the technical imple-
power of RPA and AI. The whole process is illustrated in the mentation in details, according to our findings, we have come
following image 13. to this analysis that this solution can use RPA to enter job
description in the system, create profile for candidates and
2) USE CASE 2 send out emails automatically to the candidates. On the other
Reference [81] provides a solution including artificial hand, AI models such as LLMs can help extracting candidates
intelligence and robotic process automation to release some relevant information and qualifications even if the CV doesn’t
of the duties from recruiters. The whole solution is illustrated follow a single pattern. LLMs can also verify if the candidate
in two in 14 and 15. is suitable by analyzing the relevance between qualification
Phase 1: The process starts with HR people deciding on the of the candidate and the job requirements [82]. Decision
necessary job positions and required skill for the company. tree can help categorize candidates (suitable, not suitable,
After they create the description, the system is able to make suitable for other positions) based on their specific skills and
a list of technical skill and soft skill needed for the role from experience.
the description provided. An HR can modify the list as well. Phase 2: Once the candidates are selected from the first
The proficiency of the skill is determined by rating on a scale phase, they will be asked to participate in the interview online.
of 1 to 5 stars. The system also allows job offers to be marked Their facial expressions can be assessed by using advanced
as inactive at any point which will result in removing the job software during the interview. Recruiters and interviewers
offers from the system will take tests in multiple stages to evaluate the candidates.
Once job offers are posted on the company’s internal portal, At the end of the recruitment process candidate will discuss
it is expected to see several CVs and cover letters. The system their salary expectations and benefits. System will then
will then traverse through candidate’s CV search for key check if the expectations are reasonable based on the job

188 VOLUME 13, 2025

S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

FIGURE 15. Flowchart of use Case 2 - Phase 2 [81].

to train the model certain demographic groups might be

overrepresented or underrepresented. To ensure that the
automated system does not unfairly disadvantage certain
candidates, bias mitigation is essential. Studies have shown
that AI-based recruitment tools can perpetuate biases if not
carefully managed [84].
The process involved in collecting, storing, and processing
candidate data raises concerns about data privacy and
security. Ensuring compliance with data protection standards
like GDPR is important for RPA and AI systems since
FIGURE 14. Flowchart of Use Case 2 - Phase 1 [81]. resumes often contain sensitive personal information. The
public release of SVM classifiers could unintentionally
expose private details of candidates if the support vectors used
in the classification are not properly anonymized.
descriptions and current market analysis. All results including Another challenge lies in the trust and acceptance of
camera reports collected from software will be saved in the AI-driven recommendations by human recruiters. Despite
database and can be evaluated again later if any positions open the efficiency gains, there is often skepticism among HR
up. Here, RPA can again help to insert data into database and professionals about relying solely on algorithmic decisions
scheduling video interview with description on how to use for candidate selection. This skepticism can lead to either
the video platform [83]. Whereas, AI can help to do market over-reliance on the AI system (automation bias) or a
analysis. preference for human judgment, even when the algorithm
is more accurate (algorithm aversion). Research has shown
B. CHALLENGES: RECRUITING that recruiters may distrust AI systems for making subjective
The implementation of AI-powered Robotic Process decisions, which can impact the successful adoption of these
Automation for resume screening in recruitment processes, technologies.
as described in the use case involving PrimeBot, introduces Beyond these technical and privacy concerns, Diwakar
several significant challenges as well. One of the primary Chaudhary et al. [85] emphasize the importance of addressing
challenges is the risk of bias in the AI models used for ranking broader organizational and cultural challenges in adopting
resumes. The SVM classifier, while effective in ordering AI-powered RPA in HR. This includes tackling skepticism
resumes based on keyword matching, may inadvertently and concerns within the organization, ensuring clear and
reinforce existing biases present in the training data. Based consistent communication throughout the AI implementation
on the selection of keywords or the historical data used process, and fostering collaboration between humans and

VOLUME 13, 2025 189

 S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

machines. Additionally, they stress the need for equipping that were previously conducted by human. Automation in
employees with essential technical skills through comprehen- manufacturing refers to reducing the requirement for human
sive training programs which encompass computer science, involvement in data entering process. IPA employs a variety
statistics and AI literacy. Creating a culture that values of machines and robots to streamline production processes
ongoing education and mental flexibility is essential for with the aim of increasing production capacity and reducing
adapting to how AI technology is changing and how it will costs. By replacing human labor with industry-specific
affect human resources practices. Leadership plays a pivotal robots manufacturing companies use RPA-based systems for
role in maintaining a work environment that encourages crucial tasks such as product assembly, quality monitoring
innovation, openness and the integration of AI tools while and packing. According to [87], the combination of RPA,
preserving the human touch in recruitment processes. AI and machine learning in automation is referred to as
Seamlessly integrating AI and RPA into existing HR Hyperautomation.
workflows presents another challenge. The HR department
must make substantial modifications to its operations in order A. APPLICATIONS AND USE CASES IN MANUFACTURING
to make the transition from traditional resume screening to In [67], the author has mentioned several uses cases of
an automated AI-driven process. These improvements might intelligent automation in manufacturing. Some of them are
refer to retraining employees, modifying hiring procedures, listed below:
and ensuring the AI system aligns with the business’s current
standards and values. The challenge is particularly acute in 1) PRODUCT TRACKING
ensuring that the AI tools complement rather than disrupt the Product tracking is an approach that is used to measure,
human aspects of recruitment, such as maintaining a personal analyze and advance visibility across the manufacturing
touch in candidate interactions. process. Using intelligent automation, manufacturers can
Bajzikova and Smerdova’s paper [81] highlights critical create an almost real-time visualization of progress of orders
ethical and legal challenges in AI-powered RPA for recruit- and ongoing needs of components and materials. This helps
ment, particularly regarding data privacy and compliance manufacturers optimize their process, utilize resource and
with the General Data Protection Regulation (GDPR). keep margin low.
According to the paper, businesses need to make sure that
any automated methods for profiling and making decisions
2) BILL OF MATERIALS
comply to GDPR regulations, which include gaining appli-
cants’ explicit and firm consent before processing their Maintaining an accurate bill of materials (BOM) is crucial
personal data. It emphasizes the significance of openness in for manufacturers because it gives them information about
automated hiring procedures and the ban on the use of private the parts, sub-components, intermediate assemblies, raw
information (including sexual orientation, political beliefs, materials and their corresponding amounts that are required
race or ethnicity) in any way that can result in discriminatory to make a high-quality final product. Manufacturers can
actions. Furthermore, candidates must be informed about the effectively gather product data and replicate the human
purpose, methods and consequences of data processing and procedures required to build an accurate BOM quickly and
they retain the right to audit the data collected about them at with little errors by utilizing intelligent automation.
any stage without barriers or fees.
Reference [86] discusses several challenges associated 3) CUSTOMER SERVICE AND SUPPORT
with the adoption of AI-powered RPA in recruitment. Enhancing purchase order processing, delivering responsive
One major challenge identified is uncertainty regarding the customer service and improving quality assurance are key to
correct choice of technology and the cost of implementation. increase customer satisfaction. Sysco is the world’s largest
HR managers often face difficulties in deciding which tech- distributor of food and related products. They demonstrate
nologies to adopt and ensuring that the investment will yield how intelligent automation can enhance customer service.
significant returns. Additionally, resistance from employees To do so, they integrated intelligent automation into its
and management toward these technologies is a barrier. distribution network to provide better service to its 600,000
Concerns arise regarding the impact on HR professionals’ client. When the pandemic strained global economies,
roles, with some fearing that the increasing reliance on AI Sysco expanded its automation to include over 60 digital
and automation could diminish their professional value over workers. This expansion enabled the processing of 6.2 million
time. This could result in technology that initially enhances transactions which saved the company over 250,000 work
competence but eventually leads to competence destruction hours.
as AI and RPA take over more tasks.
4) PERFORMANCE REPORTING
XI. AI DRIVEN RPA IN MANUFACTURING Digital workers are capable of generating reports that show-
The manufacturing sector is utilizing AI-powered RPA case regulatory compliance and adherence to service level
systems at a growing rate in order to increase productivity agreements (SLAs). ABB, a Swiss industrial manufacturing
and efficiency by automating repetitive and manual processes leader, utilized intelligent automation to compile and produce

190 VOLUME 13, 2025

S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

more than 200 reports every morning. This gives employees

in over 25 countries the insights necessary to enhance their
performance.

5) QUOTE MANAGEMENT
Intelligent automation allows manufacturers to automatically
generate on-demand quotes that reduces delays and enhances
customer experience. AGCO utilized intelligent automation
to manage its essential quotation process which is vital for
their revenue generation. Previously, staff spent significant
time completing each quote. Therefore, AGCO aimed to
optimize this process to allow employees to focus on
negotiating better contracts. Now, when a quotation request
is received, a digital worker reads the email, retrieves the
necessary data, updates the systems and sends the quote. This
automation has saved thousands of hours and helps staff to
concentrate on more valuable tasks.
FIGURE 16. Instructional sequence for AI and RPA in manufacturing [88].

6) SUPPLY CHAIN DEMAND PLANNING

Forecasting the demand for goods and materials across creates an order by inserting all the information into an SAP
different regions in the supply chain requires strong digital transaction.
integration. Supply chain automation eliminates silos and A major challenge occurs if there is an exception or
provides insights to ensure that goods and materials are business rule. The robot has to choose from a pool of available
available where needed. Norsk Stål, a leading supplier fabric rolls while optimizing the selection to minimize waste
of steel and metals presented the potential of RPA and in fabric rolls. To address this, an AI algorithm was integrated
intelligent automation in supply chain demand planning. into the RPA system to enhance decision-making which
Across their facilities, a team optimizes steel product eventually allows for faster selection of fabric rolls. After
manufacturing to meet customer deadlines and reduce waste. exploring different options of integrating AI algorithms into
Each day, production planners sends an estimate of the plants’ RPA, they found that inserting code directly into robot’s
maximum workload to the automated system. The system programming was the best option. This way, robot remains
then retrieves approved orders for steel products from their aware of its input and output. After deciding on the most
ERP system, calculates the optimal production schedule, optimal selection of fabric rolls while following the business
and prepares a manufacturing plan for the plant before the rules and requirements, the robot creates a purchase order in
morning shift begins. SAP. At the end, it creates a PDF file and saves it in a specific
Reference [88] presents a case study in the manufactur- folder that follows a specific name convention and sends to
ing industry that involves the automation of sending out organizational manager via email.
production orders to a fabric finishing plant by developing
artificial intelligence on RPA. This allows optimization in c: IPA PROCESS MINING
the allocation and cutting of fabric rolls which is a primary In this step they have described the sequence in which
raw material in the process. The steps of the framework is IPA carries out its tasks. They also mention exceptional
presented in figure 16. cases where error may occur for example, when RPA robot
receives an email it validates its information with the existing
materials inserted in SAP. If information doesn’t match,
a: IPA OBJECTIVE
it sends out an email including the reason behind the error
The purpose of this step is to create purchase orders for
to a desired mailbox. The information that are sent to
fabric stamping. Here, RPA robot follows a templated process
RPA Orchestrator are: material, delivery cycle, value, brand,
to create the purchase order in SAP and generate an email
quality, the quantity of fabric required, ID of the fabric
response by attaching the order in a PDF format.
supplier, minimum quantity to take, and priority of the order,
among others.
b: DESCRIPTION OF THE IPA PROCESS
The process starts with receiving a customer email which d: AI INTEGRATION CAPABILITY
includes a file that has fabric purchasing requirements such as The AI component optimizes batch selection and ensures
quality, color and type. This information is initially validated that certain business conditions are met, such as selecting
before being sent to the Orchestrator to activate the robot. rolls from the same batch, avoiding the splitting of rolls
If there are pending orders, the robot logs into SAP and from external suppliers, and preventing rolls from being

VOLUME 13, 2025 191

 S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

left with less than 50 meters. The system manages order

tolerances by allowing a maximum of 10% additional fabric
for requirements up to 600 meters, 6% for 601 to 1000 meters,
and 2% for over 1000 meters. The AI component, was written
in Python programming. It was integrated with RPA through
a built-in activity that converts Python code to .NET code.
It helps enabling to monitor error within the robot platform.
By combining AI and RPA with specific business rules, the
automation gains decision-making capabilities. This allows
the software to make optimal decisions that aligns with
business expectations.

e: RESULTS AND BENEFITS

The implementation of IPA resulted in the following impacts:
• Two employees’ time was freed up. FIGURE 17. A global view of RAIL project [92].
• The time spent processing each order was reduced from
12–15 minutes to 5.1 minutes.
• The integration of AI software decreased fabric waste by fabric roll allocation, seamless integration between AI, RPA,
30% which leads to economic benefits. and ERP systems is critical for maintaining operational
• The IPA’s order prioritization helped meet urgent needs efficiency. Discrepancies or integration failures can lead
and improved the efficiency of the processing pipeline. to order mismanagement or delays, impacting the overall
productivity [91].
B. CHALLENGES: MANUFACTURING AI-powered RPA systems in manufacturing are designed
The integration of AI-powered RPA systems in manufac- to reduce human labor; however, achieving an optimal
turing presents several specific challenges, particularly in balance between automation and human oversight presents
areas such as product tracking, supply chain demand planning challenges. Ensuring that humans remain involved in key
and production automation. One of the key challenges is decision-making processes, particularly when dealing with
ensuring the accuracy and consistency of data throughout exceptions or complex business rules, is essential for the
the manufacturing processes. Automated systems heavily success of these systems.
rely on data integrity for tasks such as bill of materials
(BOM) and product tracking. Any inconsistency or error in XII. ANOTHER REAL AI POWERED RPA PROJECT FOR
data could result in production delays, increased costs or BACK OFFICE
faulty products. For example, as seen in the implementation Several organizations have embraced RPA to advance the
of AI in fabric roll allocation, errors in data validation technology they use for back office. As discussed so far, the
or processing could disrupt the entire workflow, causing current trends mostly lie in infusing a loop called ‘‘human-
significant inefficiencies [89]. in-the-loop’’ which supports human interaction with robot.
AI integration in manufacturing, such as in Norsk Stål’s This is important as most RPA bots need human interaction
supply chain demand planning, requires real-time decision- to some extent.
making that often involves complex business rules. Ensuring
that AI systems can make accurate and optimal decisions in
1) USE CASE 1
such dynamic environments remains a significant challenge.
This becomes especially difficult when balancing multiple Reference [92] demonstrates a real project called ‘‘RAIL’’
factors, such as resource allocation, delivery schedules and which is developed by Servinform Inc. and the University of
production constraints [32]. Seville. The aim of this project is to identify and automate
As AI systems become more deeply embedded in business processes from back office without disrupting the
manufacturing, ensuring scalability and flexibility to existing system. It is made of a series of modules which is
accommodate growing production needs and varying depicted in figure 17.
business conditions becomes a challenge. The integration
of AI in fabric roll optimization, for instance, must a: NON-INTRUSIVE MONITORING
account for changes in production volume, materials, and First of all, RAIL creates a document to capture a structured
external market factors without significantly disrupting dataset in real time from the interaction between user or
operations [90]. computer without hampering anyone’s activity from back
Another challenge lies in effectively integrating AI- office, breaching security and jeopardizing confidentiality
powered RPA with existing enterprise resource planning of the data. It uses a non-intrusive monitoring software to
(ERP) systems, such as SAP. In complex processes like capture the data.

192 VOLUME 13, 2025

S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

b: AUTOMATIC PROCESS SURVEY MODULE non-intrusive monitoring and guarantee compliance with
Then it includes an automatic process survey module to GDPR and other regulatory standards. AI models must
create workflow variants from logs and images collected include methods for continuous learning that adjust to
from previous step using image-hash, image-match and OCR changes in workflow and take unexpected inputs into account
algorithms. in order to handle complicated processes and exceptions.
When data scientists and domain specialists work together,
c: QUALITATIVE EVALUATION MODULE domain-specific information may be incorporated into AI
The automatic survey of the process goes through a number of algorithms, increasing the predictive models’ resilience.
evaluation and fine tuning by changing different parameters AI and RPA must be seamlessly integrated with current
of algorithms. This helps to improve the process. The infrastructure, which calls for modular designs and phased
integration of PoM software helps users to implement process implementation tactics that minimize interruptions and
generation algorithms and determining evaluation matrices permit gradual adaptation. AI systems may maintain align-
as output. The outcome is an accurate demonstration of the ment with organizational procedures and provide reliable
executed process which highlights the edge or exceptional performance without sacrificing security or efficiency with
cases that are typically more difficult to find. It is important regular audits and feedback loops.
to identify such cases as it may lead to failure in robotization.
XIII. OPEN RESEARCH CHALLENGES
d: MODULE FOR THE IDENTIFICATION AND DEPLOYMENT
Based on our in depth review and survey we have identified
OF PREDICTIVE ALGORITHMS
the following areas where there are limited work or lack in
research is noticeable:
This module includes the learning components and the
predictive components. A neural network based model is
utilized to learn from user inputs such as mouse click, A. CHANGE MANAGEMENT AND COMMUNICATION
keystrokes, text and so on to convert the tasks into a corpus. STRATEGIES
Once the model is trained, it can be used to predict robot’s To adapt IPA technology in any organization, it is needed
behavior so that RAIL can be asked for the next course of to make changes in change management through proper
actions. communication. Wherever, there is a need for digital
transition, the structure of the team needs to be changed. If we
combine technologies like RPA and AI then there should be
A. CHALLENGES: BACK OFFICE
an open communication channel between RPA team and AI
The RAIL project and similar AI-powered RPA systems in
engineers. While RPA team will take care of managing bots,
back office processes face several key challenges as well.
their subscriptions and other configurations, AI engineers
Non-intrusive monitoring techniques as employed in the
will focus on developing suitable ML models to advance
RAIL project, raise concerns about data privacy and security.
the process. The whole process needs strategic movement,
Although designed to avoid disrupting user activities, ensur-
discussion and communication.
ing that the data collected through such monitoring remains
confidential and compliant with regulations like GDPR is
a persistent challenge [93]. The ability of AI systems to B. ETHICAL AND LEGAL CONSIDERATION
handle complex business processes and identify exceptions, The data-driven nature of IPA raises important ethical and
such as unexpected input during automated tasks remains a legal questions. This is particularly related to data privacy and
significant challenge. AI algorithms must be robust enough algorithmic bias. It is important to do research into developing
to manage variations in data and workflows and errors in ethical guidelines and legal frameworks that address these
predicting robotic behavior that can cause system failures issues. Ensuring that collecting data through RPA process is
or inefficiencies. Developing effective predictive models to not impacting data security, AI algorithms are transparent,
optimize robotic actions, as seen in the RAIL project, requires fair and compliant with regulations is crucial to gaining
extensive training and data collection. Ensuring that neural people’s trust and ensuring the ethical deployment of IPA
networks and other AI models can accurately predict future technologies.
tasks without significant error is a key challenge, especially
in dynamic back office environments where manual input can C. SELECTING ML MODELS
still be necessary [94]. Integrating AI and RPA systems with Data curation is an important part of IPA process. Also,
existing back office infrastructure while ensuring seamless accessing to data through ML models is a significant step.
automation across various processes can be a daunting task. However, there has not been much research on the selection
Managing exceptions and ensuring that AI-driven systems do process of ML models. Selecting biased or not suitable
not interfere with established workflows requires meticulous models will have negative impact on the output of IPA
planning and ongoing adjustments [32]. process. There is a pressing need for research on listed models
Organizations should use end-to-end encryption and in RPA tools. An open documentation on selecting optimal
anonymization solutions to address data privacy issues in models for particular use cases should be published.

VOLUME 13, 2025 193

 S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

D. EXPLAINABILITY data processing is essential. Research work [103] explores

One critical challenge in the IPA process is the need for the potential and efficiency of container technology for AI
Explainable AI (XAI). Reference [95] defines XAI as a application. They experimented with container for object
suit of techniques and algorithms designed to advance the detection application and shows that container is suitable for
trustworthiness and transparency of AI systems. The goal of handling complex data.
these technique is to explaining the decisions or predictions
of the models using several methods. As complex models XIV. CONCLUSION
such as deep learning models are now integrated with RPA, Robotic process automation and artificial intelligence have
their black box nature can undermine the trust and prevent come together to bring about a new phase of intelligent
informed decision making. Research on XAI can focus on process automation, marking a significant advancement in
developing techniques that can balance model’s accuracy the field of automation. In this paper we analyze how
with interpretability. By taking care of these challenges the integration of these technologies transforms business
organizations will gain more trust on adopting IPA solutions processes in several industries by incorporating cognitive
and believe in its success. capabilities into automation, rather than just automating
repetitive activities. This change constitutes the core of the
ongoing digital revolution under the name Industry 4.0, which
E. FUTURE APPLICATIONS is based on efficiency, precision, and adaptability.
Adopting AI has evolved and increased recently in agricul- RPA has broad application from enhancing productivity to
tural sector too [96]. Earlier researches were based on crop reducing operational costs in the fields of finance, healthcare,
yield with regression and clustering based analysis [97]. Now, manufacturing, and many other industries. But it actually
there are advanced ongoing researches on water management, sees its potent realization when subsumed into the application
seed or soil analysis with Deep Learning techniques [98]. of AI, which brings along such abilities as data analysis,
For example, [99] and [100] describes how drones are used pattern recognition, and decision-making. Such synergy
to collect image data from crop field which will eventually allows the automation of more intricate processes, thus letting
be used by vendor specific applications to analyze. However, businesses innovate and respond to market demands more
predicting crop yield is one of the fields where we have effectively.
found limited research conducted purely based on IPA. For In this paper, we have discussed the journey that RPA has
example, [101] has used RPA to collect data from different taken from its evolution and merging with AI to give insights
websites (weather, soil) to feed the data into predictive model into how it changes functions, from finance and banking to
to do crop yield analysis. As technology advances, integrating health, manufacture, and recruitment. Case studies and use
smart drones and RPA can significantly enhance agricultural cases have shown how AI-powered RPA is used to provide
practices. Smart drones can collect real-time data on soil operational efficiencies, a better customer experience, and
moisture, temperature, crop health, and pest presence, which optimized decision-making. One such example is finance,
can be stored in a centralized database using RPA bots. These where AI-driven RPA has modernized fraud detection and
bots can automate data entry, ensuring timely and accurate the processing of loans. Another one would be in healthcare,
updates. AI models analyze this data to assess soil conditions, where it has revolutionized patient care and administrative
predict crop yields and recommend appropriate fertilizers to workflows.
farmers. Additionally, creating an AgroBot can be beneficial The benefits are clear, but there have been challenges
to answer basic questions related to planting, harvesting, and following the implementation of AI-powered RPA. Some of
crop diseases which will provide farmers as well as local these include data privacy, ethical considerations, bias in AI
people expert advice. models, and complexity in integration with existing systems.
With those data a model can analyze the soil condition, This is only possible through multi-dimensional solutions,
predict crop yield and recommend farmers for appropriate such as solid governance frameworks, relentless learning, and
fertilizers. RPA bots can also inform farmers the exact time to putting prime importance on transparency and accountability
use the fertilizers or pesticides by sending them alert. An agro in AI-driven processes.
bot can also be created for answering basic questions related The future, or rather further transformation in industries,
to planting harvest or crop diseases. would be powered by AI RPA. However, it is extremely
Another future application can be, integrating RPA important to tread the waters carefully, whether within the
in applications using container technologies. This offers ethical and legal environments, the kind of machine learning
promising avenues for enhancing automation capabilities at model used, and one needs to be explainable in order to
the edge [102]. Containers, with their lightweight nature trust the AI systems. With digital transformation speeding
and flexibility, can facilitate deploying RPA processes on up, sectors like agriculture, at the very outset of massive
resource-constrained ARM-based edge devices, allowing technology adoption, offer fresh opportunities for growth and
automated tasks to be efficiently executed closer to data innovation.
sources. This approach could significantly reduce latency In conclusion, the convergence of RPA and AI will deter-
and improve response times in environments where real-time mine the future of intelligent automation by enabling firms

194 VOLUME 13, 2025

S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

with the tools they need to drive efficiency, cut costs, and [18] R. Syed, S. Suriadi, M. Adams, W. Bandara, S. J. J. Leemans, C. Ouyang,
foster innovation. With this understanding of their synergetic A. H. M. ter Hofstede, I. van de Weerd, M. T. Wynn, and H. A. Reijers,
‘‘Robotic process automation: Contemporary themes and challenges,’’
potentials, businesses can unlock new opportunities and, as a Comput. Ind., vol. 115, Feb. 2020, Art. no. 103162.
result, maintain a sustainable competitive advantage in the [19] J. Kokina and S. Blanchette, ‘‘Early evidence of digital labor in account-
fast-moving digital ecosystem. ing: Innovation with robotic process automation,’’ Int. J. Accounting Inf.
Syst., vol. 35, Dec. 2019, Art. no. 100431.
[20] S. A. Khan, M. M. H. Chowdhury, and U. Nandy, ‘‘AI robotics
REFERENCES technology: A review,’’ J. Eng. Res. Rep., vol. 25, no. 10, pp. 187–194,
[1] F. Kanakov and I. Prokhorov, ‘‘Analysis and applicability of artificial Nov. 2023.
intelligence technologies in the field of RPA software robots for automat- [21] A. Nath and U. Saha, ‘‘Intellectual property management in healthcare
ing business processes,’’ Proc. Comput. Sci., vol. 213, pp. 296–300, using robotic process automation during COVID-19,’’ in Proc. Conflu-
Jan. 2022. ence Artif. Intell. Robotic Process Autom. Cham, Switzerland: Springer,
[2] R. Götzen, J. v. Stamm, R. Conrad, and V. Stich, ‘‘Understanding Jan. 2023, pp. 177–197.
the organizational impact of robotic process automation: A socio- [22] D. Rao and P. Pathak, ‘‘Evolving robotic process automation (RPA) &
technical perspective,’’ in Proc. Working Conf. Virtual Enterprises. Cham, artificial intelligence (AI) in response to COVID-19 and its future,’’
Switzerland: Springer, Jan. 2022, pp. 106–114, doi: 10.1007/978-3-031- AIP Conf. Proc., vol. 2519, no. 1, 2022, Art. no. 030025, doi:
14844-6_9. 10.1063/5.0109615.
[3] R. Palaniappan, ‘‘An overview on robot process automation: Advance- [23] J. Ribeiro, R. Lima, T. Eckhardt, and S. Paiva, ‘‘Robotic process
ments, design standards, its application, and limitations,’’ Informatica, automation and artificial intelligence in Industry 4.0—A literature
vol. 48, no. 1, Feb. 2024, doi: 10.31449/inf.v48i1.5058. review,’’ Proc. Comput. Sci., vol. 181, pp. 51–58, Jan. 2021.
[4] L.-V. Herm, C. Janiesch, H. A. Reijers, and F. Seubert, ‘‘From symbolic [24] K. K. H. Ng, C.-H. Chen, C. K. M. Lee, J. Jiao, and Z.-X. Yang, ‘‘A
RPA to intelligent RPA: Challenges for developing and operating systematic literature review on intelligent automation: Aligning concepts
intelligent software robots,’’ in Proc. 19th Int. Conf. Bus. Process Manag., from theory, practice, and future perspectives,’’ Adv. Eng. Informat.,
Rome, Italy. Cham, Switzerland: Springer, Jan. 2021, pp. 289–305. vol. 47, Jan. 2021, Art. no. 101246.
[5] W. A. Ansari, P. Diya, S. Patil, and S. Patil, ‘‘A review on robotic process [25] UiPath. (2024). Introducing Uipath AI Fabric. Accessed: Jul. 30, 2024.
automation-the future of business organizations,’’ in Proc. 2nd Int. Conf. [Online]. Available: https://www.uipath.com/blog/product-and-
Adv. Sci. & Technol. (ICAST), 2019, doi: 10.2139/ssrn.3372171. updates/introducing-uipath-ai-fabric
[6] N. T. Da, H.-S. Le, T.-D.-N. Nguyen, H.-T. Lam, T.-A. Tran, and [26] IEEE Guide for Terms and Concepts in Intelligent Process Automation,
Q.-T. Tran, ‘‘A survey of AI-based robotic process automation for Standard IEEE 2755-2017, IEEE Standards Association, Sep. 2017.
businesses and organizations,’’ Sci. Technol. Develop. J., vol. 26, no. 3, [27] P. Lewicki, J. Tochowicz, and J. van Genuchten, ‘‘Are robots taking
pp. 2959–2966, Jan. 2023. our jobs? A RoboPlatform at a bank,’’ IEEE Softw., vol. 36, no. 3,
[7] T. Chakraborti, V. Isahagian, R. Khalaf, Y. Khazaeni, V. Muthusamy, pp. 101–104, May 2019.
Y. Rizk, and M. Unuvar, ‘‘From robotic process automation to [28] R. M. Cronin, D. Fabbri, J. C. Denny, S. T. Rosenbloom, and
intelligent process automation: Emerging trends,’’ in Proc. Int. Conf. Bus. G. P. Jackson, ‘‘A comparison of rule-based and machine learning
Process Manag., Seville, Spain. Cham, Switzerland: Springer, Jan. 2020, approaches for classifying patient portal messages,’’ Int. J. Med.
pp. 215–228. Informat., vol. 105, pp. 110–120, Sep. 2017.
[8] L. Patrício, L. Varela, and Z. Silveira, ‘‘Integration of artificial [29] H. Tu, Z. Lin, and K. Lee, ‘‘Automation with intelligence in drug
intelligence and robotic process automation: Literature review and research,’’ Clin. Therapeutics, vol. 41, no. 11, pp. 2436–2444, Nov. 2019.
proposal for a sustainable model,’’ Appl. Sci., vol. 14, no. 21, p. 9648, [30] M. Lacity and L. P. Willcocks, Robotic Process Automation and Risk
Oct. 2024. Mitigation: The Definitive Guide. Kerala, India: SB Publishing, 2017.
[9] R. Bhatnagar and R. Jain, ‘‘Robotic process automation in healthcare—A [31] V. Leno, M. Dumas, M. La Rosa, F. M. Maggi, and A. Polyvyanyy,
review,’’ Int. Robot. Autom. J., vol. 5, no. 1, pp. 12–14, Jan. 2019. ‘‘Automated discovery of data transformations for robotic process
[10] C. Lamberton, D. Brigo, and D. Hoy, ‘‘Impact of robotics, RPA and AI automation,’’ 2020, arXiv:2001.01007.
on the insurance industry: Challenges and opportunities,’’ J. Financial [32] J. Siderska, L. Aunimo, T. Süße, J. V. Stamm, D. Kedziora, and
Perspect., vol. 4, no. 1, pp. 8–20, Jan. 2017. S. N. B. M. Aini, ‘‘Towards intelligent automation (IA): Literature
[11] S. Kumar, S. Khanna, N. Ghosh, and S. O. D. Kumar, ‘‘Importance review on the evolution of robotic process automation (RPA), its
of artificial intelligence (AI) and robotic process automation (RPA) challenges, and future trends,’’ Eng. Manage. Prod. Services, vol. 15,
in the banking industry: A study from an Indian perspective,’’ in no. 4, pp. 90–103, Dec. 2023.
Proc. Confluence Artif. Intell. Robotic Process Autom., Jan. 2023, [33] D. Fluss, ‘‘Smarter bots mean greater innovation, productivity, and value:
pp. 231–266. Robotic process automation is allowing companies to re-imagine and re-
[12] S. Moreira, H. S. Mamede, and A. Santos, ‘‘Process automation using invest in all aspects of their businesses,’’ CRM Mag., vol. 22, no. 10,
RPA—A literature review,’’ Proc. Comput. Sci., vol. 219, pp. 244–254, pp. 38–39, 2018.
Jan. 2023. [34] S. Khan, ‘‘Comparative analysis of RPA tools-uipath, automation
[13] C. H. V. D. Moraes, J. Scolimoski, G. Lambert-Torres, M. Santini, anywhere and blueprism,’’ Int. J. Comput. Sci. Mobile Appl., vol. 8, no. 11,
A. L. A. Dias, F. A. Guerra, A. Pedretti, and M. P. Ramos, ‘‘Robotic pp. 1–6, Nov. 2020.
process automation and machine learning: A systematic review,’’ [35] UiPath. (2024). Rpa and AI Integration With AI Center. Accessed:
Brazilian Arch. Biol. Technol., vol. 65, Jan. 2022, Art. no. e22220096. Jul. 30, 2024. [Online]. Available: https://www.uipath.com/product/rpa-
[14] J. K. Ray, R. Sultana, R. Bera, S. Sil, and Q. M. Alfred, ‘‘A comprehensive ai-integration-with-ai-center
review on artificial intelligence (AI) and robotic process automation [36] T. Automation. (2019). Kofax Enhances Industry-leading Intelligent
(RPA) for the development of smart cities,’’ in Proc. Confluence Artif. Automation Platform. Accessed: Jul. 30, 2024. [Online]. Available:
Intell. Robotic Process Autom., Jan. 2023, pp. 289–311. https://www.tungstenautomation.de/about/press-releases/2019/kofax-
[15] Miswar, Suhardi, and N. B. Kurniawan, ‘‘A systematic literature review enhances-industry-leading-intelligent-automation-platform
on survey data collection system,’’ in Proc. Int. Conf. Inf. Technol. Syst. [37] K. Meena and S. Suriya, ‘‘A survey on supervised and unsupervised
Innov. (ICITSI), Oct. 2018, pp. 177–181. learning techniques,’’ in Proc. Int. Conf. Artif. Intell., Smart Grid Smart
[16] G. V. Research. Robotic Process Automation (RPA) Market City Applications (AISGSC). Cham, Switzerland: Springer, Jan. 2020,
Analysis. Accessed: Aug. 10, 2024. [Online]. Available: pp. 627–644.
https://www.grandviewresearch.com/industry-analysis/robotic-process- [38] A. Gangal, A. Shrivastava, N. M. Hussien, P. Singh, M. Diwakar, K. Joshi,
automation-rpa-market S. Bisht, and N. C. Joshi, ‘‘Natural language processing: A review,’’ Int.
[17] ElectroNeek. (2024). History of RPA: From Early Beginnings to J. Res. Eng. Appl. Sci., vol. 2849, Jan. 2023, Art. no. 020010.
Intelligent Automation. Accessed: Aug. 10, 2024. [Online]. Available: [39] A. Anywhere. (2024). Home. Accessed: Jul. 30, 2024. [Online].
https://electroneek.com/rpa/history-of-rpa/ Available: https://www.automationanywhere.com/

VOLUME 13, 2025 195

 S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

[40] Automation Anywhere. (2024). AI Agents. Accessed: Aug. 3, 2024. [63] A. R. Hasan, ‘‘Artificial intelligence (AI) in accounting & auditing: A
[Online]. Available: https://www.automationanywhere.com/rpa/ai-agents literature review,’’ Open J. Bus. Manage., vol. 10, no. 1, pp. 440–465,
[41] Blue Prism. (2024). Intelligent Automation Guide. Accessed: 2022.
Aug. 3, 2024. [Online]. Available: https://www.blueprism.com/guides/ [64] A. Zemankova, ‘‘Artificial intelligence in audit and accounting: Develop-
intelligent-automation/ ment, current trends, opportunities and threats–literature review,’’ in Proc.
[42] EdgeVerve. (2024). Robotic Process Automation (RPA). Accessed: Int. Conf. Control, Artif. Intell., Robot. Optim. (ICCAIRO), Dec. 2019,
Jul. 30, 2024. [Online]. Available: https://www.edgeverve.com/ pp. 148–154.
assistedge/robotic-process-automation-rpa/ [65] S. M. Al-Sayyed, S. F. Al-Aroud, and L. M. Zayed, ‘‘The effect of
[43] GitHub. (2024). Automagica. Accessed: Jul. 30, 2024. [Online]. Avail- artificial intelligence technologies on audit evidence,’’ Accounting, vol. 7,
able: https://github.com/automagica/automagica no. 2, pp. 281–288, 2021.
[44] WinAutomation. (2020). Endeavor 2020. Accessed: Jul. 30, 2024. [66] C. Zhang, ‘‘Intelligent process automation in audit,’’ J. Emerg. Technol.
[Online]. Available: http://rpa.winautomation.com/Endeavor-2020.html Accounting, vol. 16, no. 2, pp. 69–88, Sep. 2019.
[67] Blue Prism. (2024). 20 Effective RPA Use Cases in Manufacturing.
[45] J. Krzywy, K. Dorofiejczuk, F. Nowak, and M. Jasiulewicz-Kaczmarek,
Accessed: Aug. 3, 2024. [Online]. Available: https://www.blueprism.
‘‘Study of the use of robotic process automation in supporting
com/resources/blog/20-effective-rpa-use-cases-in-manufacturing/
customer order process,’’ IFAC-PapersOnLine, vol. 58, no. 19, [68] M. Ratia, J. Myllärniemi, and N. Helander, ‘‘Robotic process automation–
pp. 1018–1023, 2024. creating value by digitalizing work in the private healthcare?’’ in Proc.
[46] X. Zhan, Z. Ling, Z. Xu, L. Guo, and S. Zhuang, ‘‘Driving efficiency and 22nd Int. Academic Mindtrek Conf., Oct. 2018, pp. 222–227.
risk management in finance through AI and RPA,’’ Unique Endeavor Bus. [69] S. V. Belkum, N. Brun, S. Cleve, P. McGovern, M. Lumpkin, P. E.
& Social Sci., vol. 3, pp. 189–197, Oct. 2024. Schaeffer, and T. Netzer, ‘‘Artificial intelligence in clinical development
[47] L. Kai, L. Wenxin, Z. Ran, K. Suihua, Y. Dazhu, Z. Min, and T. Wenhong, and regulatory affairs�preparing for the future,’’ Regulatory
‘‘Research on ‘AI+ RPA’ interactive technology of intelligent financial Rapporteur, vol. 15, no. 10, pp. 17–21, 2018.
management platform,’’ in Proc. 19th Int. Comput. Conf. Wavelet Act. [70] N. H. Barla, S. M. Almeida, and M. S. Almeida, ‘‘RPA revolution in
Media Technol. Inf. Process. (ICCWAMTIP), Dec. 2022, pp. 1–5. the healthcare industry during COVID-19,’’ in Proc. Confluence Artif.
[48] D. Dasgupta, ‘‘Impact of AI and RPA in banking,’’ in Proc. Confluence Intell. Robotic Process Autom. Cham, Switzerland: Springer, Jan. 2023,
Artif. Intell. Robotic Process Autom. Cham, Switzerland: Springer, pp. 199–229.
Jan. 2023, pp. 41–72. [71] S. Mahmoodzadeh, M. Moazenzadeh, H. Rashidinejad, and
[49] G. Shidaganti, K. N. Karthik, Anvith, and N. A. Kantikar, ‘‘Integration of M. Sheikhvatan, ‘‘Diagnostic performance of electrocardiography in the
RPA and AI in Industry 4.0,’’ in Proc. Confluence Artif. Intell. Robotic assessment of significant coronary artery disease and its anatomical size
Process Autom. Cham, Switzerland: Springer, Jan. 2023, pp. 267–288. in comparison with coronary angiography,’’ J. Res. Med. Sci., Off. J.
[50] D. Baviskar, S. Ahirrao, V. Potdar, and K. Kotecha, ‘‘Efficient automated Isfahan Univ. Med. Sci., vol. 16, no. 6, p. 750, Jan. 2011.
processing of the unstructured documents using artificial intelligence: A [72] E. Costariol, M. Rotondi, A. Amini, C. J. Hewitt, A. W. Nienow,
systematic literature review and future directions,’’ IEEE Access, vol. 9, T. R. J. Heathman, M. Micheletti, and Q. A. Rafiq, ‘‘Establishing the
pp. 72894–72936, 2021. scalable manufacture of primary human T-cells in an automated stirred-
[51] Y. Ye, S. Zhu, J. Wang, Q. Du, Y. Yang, D. Tu, L. Wang, and J. Luo, ‘‘A tank bioreactor,’’ Biotechnol. Bioeng., vol. 116, no. 10, pp. 2488–2502,
unified scheme of text localization and structured data extraction for joint Oct. 2019.
OCR and data mining,’’ in Proc. IEEE Int. Conf. Big Data (Big Data), [73] D. de Sousa Pinto, C. Bandeiras, M. de Almeida Fuzeta,
Dec. 2018, pp. 2373–2382. C. A. V. Rodrigues, S. Jung, Y. Hashimura, R. Tseng, W. Milligan,
[52] P. Patri, ‘‘Robotic process automation: Challenges and solutions for the B. Lee, F. C. Ferreira, C. L. da Silva, and J. M. S. Cabral, ‘‘Scalable
banking sector,’’ Int. J. Manage., vol. 11, no. 12, p. 2020, Dec. 2020. manufacturing of human mesenchymal stromal cells in the vertical-
[53] T. Iyamu and N. Mlambo, ‘‘Actor-network theory perspective of robotic wheel bioreactor system: An experimental and economic approach,’’
process automation implementation in the banking sector,’’ Int. J. Inf. Biotechnol. J., vol. 15, no. 8, Aug. 2020, Art. no. 1800716.
Technol. Syst. Approach, vol. 15, no. 1, pp. 1–17, Jul. 2022. [74] S. Hamad, D. Derichsweiler, S. Papadopoulos, F. Nguemo, T. Šarić,
A. Sachinidis, K. Brockmeier, J. Hescheler, B. J. Boukens, and
[54] A. Choubey and M. Sharma, ‘‘Implementation of robotics and its impact
K. Pfannkuche, ‘‘Generation of human induced pluripotent stem cell-
on sustainable banking: A futuristic study,’’ J. Phys., Conf. Ser., vol. 1911,
derived cardiomyocytes in 2D monolayer and scalable 3D suspension
no. 1, May 2021, Art. no. 012013.
bioreactor cultures with reduced batch-to-batch variations,’’ Theranos-
[55] M. Gotthardt, D. Koivulaakso, O. Paksoy, C. Saramo, M. Martikainen,
tics, vol. 9, no. 24, pp. 7222–7238, 2019.
and O. Lehner, ‘‘Current state and challenges in the implementation of
[75] K. Venigandla, ‘‘Integrating RPA with AI and ml for enhanced diagnostic
smart robotic process automation in accounting and auditing,’’ ACRN J.
accuracy in healthcare,’’ Power Syst. Technol., vol. 46, no. 4, pp. 1–10,
Finance Risk Perspect., vol. 9, no. 1, pp. 90–102, 2020.
2022.
[56] M. Jaiwani and S. Gopalkrishnan, ‘‘Adoption of RPA and AI to enhance [76] R. J. McDonald, K. M. Schwartz, L. J. Eckel, F. E. Diehn, C. H. Hunt,
the productivity of employees and overall efficiency of Indian private B. J. Bartholmai, B. J. Erickson, and D. F. Kallmes, ‘‘The effects of
banks: An inquiry,’’ in Proc. Int. Seminar Appl. Technol. Inf. Commun. changes in utilization and technological advancements of cross-sectional
(iSemantic), Sep. 2022, pp. 191–197. imaging on radiologist workload,’’ Academic Radiol., vol. 22, no. 9,
[57] J. S. Persellin, J. J. Schmidt, S. D. Vandervelde, and M. S. Wilkins, ‘‘Audi- pp. 1191–1198, Sep. 2015.
tor perceptions of audit workloads, audit quality, and job satisfaction,’’ [77] R. Fitzgerald, ‘‘Error in radiology,’’ Clin. Radiol., vol. 56, no. 12,
Accounting Horizons, vol. 33, no. 4, pp. 95–117, Dec. 2019. pp. 938–946, Dec. 2001. [Online]. Available: https://www.clinical
[58] A. Chakraborty, S. Bhattacharyya, D. De, P. Sarigiannidis, and radiologyonline.net/article/S0009-9260
J. S. Banerjee, ‘‘Confluence of artificial intelligence and robotic process [78] S. A. Pahune, ‘‘A brief overview of how ai enables healthcare sector rural
automation: Concluding remarks,’’ in Proc. Confluence Artif. Intell. development,’’ 2024, doi: 10.13140/RG.2.2.16675.63525.
Robotic Process Autom. Cham, Switzerland: Springer, Jan. 2023, [79] R. Half. (2024). Why Future-focused Talent Management Hinges
pp. 389–399. on Aligning People, Process, Technology, Data and AI. Accessed:
[59] F. Huang and M. A. Vasarhelyi, ‘‘Applying robotic process automation Aug. 2, 2024. [Online]. Available: https://www.roberthalf.
(RPA) in auditing: A framework,’’ Int. J. Accounting Inf. Syst., vol. 35, com/us/en/insights/management-tips/why-future-focused-talent-
Dec. 2019, Art. no. 100433. management-hinges-on-aligning-people-process-technology-data-and-ai
[60] K. C. Moffitt, A. M. Rozario, and M. A. Vasarhelyi, ‘‘Robotic process [80] H. Mhaske, S. Kulkarni, V. V. Menon, P. Nikam, and B. Niras,
automation for auditing,’’ J. Emerg. Technol. Accounting, vol. 15, no. 1, ‘‘Development of PrimeBot as an assistant to HR in recruitment process
pp. 1–10, Jul. 2018. using RPA,’’ Int. J. Eng. Res. Technol., vol. 8, no. 5, pp. 97–99,
[61] M. Eulerich, J. Pawlowski, N. J. Waddoups, and D. A. Wood, ‘‘A May 2019.
framework for using robotic process automation for audit tasks,’’ [81] L. Bajzikova and T. Smerdova, ‘‘Improving the recruitment process
Contemp. Accounting Res., vol. 39, no. 1, pp. 691–720, 2022. in multinational organizations using robotic process automation and
[62] R. Lacurezeanu, A. Tiron-Tudor, and V. P. Bresfelean, ‘‘Robotic process artificial intelligence,’’ in Data-Centric Busines and Applications:
automation in audit and accounting,’’ Audit Financiar, vol. 18, no. 160, Advancements in Information and Knowledge Management. Cham,
pp. 752–770, Oct. 2020. Switzerland: Springer, 2024, pp. 29–60.

196 VOLUME 13, 2025

S. Afrin et al.: AI-Enhanced RPA: A Review of Intelligent Automation Innovations

[82] Z. Zheng, Z. Qiu, X. Hu, L. Wu, H. Zhu, and H. Xiong, ‘‘Generative job [101] K. K. Devi and J. P. Kumar, ‘‘An efficient data collection tool for crop
recommendations with large language model,’’ 2023, arXiv:2307.02157. recommendations model using robotic process automation,’’ in Proc.
[83] M. Michailidis, ‘‘The challenges of AI and blockchain on HR recruiting 14th Int. Conf. Comput. Commun. Netw. Technol. (ICCCNT), Jul. 2023,
practices,’’ Cyprus Rev., vol. 30, no. 2, pp. 169–180, Jan. 2018. pp. 1–5.
[84] S. Li, K. Li, and H. Lu, ‘‘National origin discrimination in deep-learning- [102] S. Kaiser, Md. S. Haq, A. S. Tosun, and T. Korkmaz, ‘‘Container
powered automated resume screening,’’ 2023, arXiv:2307.08624. technologies for ARM architecture: A comprehensive survey of the state-
[85] M. D. Chaudhary, R. Baliyan, D. S. K. Verma, K. S. Madhukar, and of-the-art,’’ IEEE Access, vol. 10, pp. 84853–84881, 2022.
M. A. Kushwaha, ‘‘Challenges of adopting digitalization in human [103] S. Kaiser, A. S. Tosun, and T. Korkmaz, ‘‘Benchmarking container
resources management through artificial intelligence,’’ J. Res. Admin., technologies on ARM-based edge devices,’’ IEEE Access, vol. 11,
vol. 6, no. 1, pp. 15–42, 2024. pp. 107331–107347, 2023.
[86] J. Saukkonen, P. Kreus, N. Obermayer, Ó. R. Ruiz, and M. Haaranen, ‘‘AI,
RPA, ML and other emerging technologies: Anticipating adoption in the
HRM field,’’ in Proc. Eur. Conf. Impact Artif. Intell. Robot. (ECIAIR).
Oxford, U.K.: EM-Normandie Business School, Oct./Nov. 2019, pp. 287–
296. SADIA AFRIN received the B.Sc. degree in com-
[87] R. Kavitha, ‘‘Hyperautomation-beyond RPA: Leveraging automation to puter science and engineering from North South
transform the manufacturing industries,’’ in Proc. Int. Conf. Comput. University (NSU), Bangladesh, in 2018, and the
Commun. Informat. (ICCCI), Jan. 2023, pp. 1–5. M.Sc. degree from The University of Texas at San
[88] F. A. Lievano-Martínez, J. D. Fernández-Ledesma, D. Burgos, Antonio (UTSA), in 2022. During her graduate
J. W. Branch-Bedoya, and J. A. Jimenez-Builes, ‘‘Intelligent process studies, she was a Graduate Research Assistant
automation: An application in manufacturing industry,’’ Sustainability,
with the Computer Science Department, UTSA.
vol. 14, no. 14, p. 8804, Jul. 2022.
[89] A. D. Badmus, ‘‘Leveraging software automation to transform the After her graduation, she has been a Software
manufacturing industry,’’ J. Knowl. Learn. Sci. Technol., vol. 2, Developer with Robert Half Inc., since 2022. Her
no. 1, pp. 84–92, Apr. 2023. [Online]. Available: https://api. research interests include artificial intelligence,
semanticscholar.org/CorpusID machine learning, and robotic process automation.
[90] J. Kim, ‘‘A study on the success cases about AI RPA (robotic process
automation) in manufacturing industry,’’ Int. J. Wireless Mobile Commun.
Ind. Syst., vol. 6, no. 1, pp. 15–20, Dec. 2019. [Online]. Available:
https://api.semanticscholar.org/CorpusID
[91] A. M. Radke, V. Vinsmart, M. T. Dang, and A. Tan, ‘‘Using robotic SHOBNOM ROKSANA received the B.Sc.
process automation (RPA) to enhance item master data maintenance degree in computer science and engineering from
process,’’ Logforum, vol. 16, no. 1, pp. 129–140, Mar. 2020. American International University–Bangladesh,
[92] R. Cabello, M. J. Escalona, and J. G. Enríquez, ‘‘Beyond the hype: in 2017, and the M.Sc. degree from The University
RPA horizon for robot-human interaction,’’ in Proc. Int. Conf. Bus. of Texas at San Antonio, in 2024. After graduation,
Process Manag., Seville, Spain. Cham, Switzerland: Springer, Jan. 2020, she was a Senior Web Developer with METech
pp. 185–199. Ltd., for nearly four years in Bangladesh. She was
[93] Y. Al-Slais and M. Ali, ‘‘Robotic process automation and intelligent a Graduate Research Assistant with the Computer
automation security challenges: A review,’’ in Proc. Int. Conf. Cyber Science Department, The University of Texas
Manage. Eng. (CyMaEn), Jan. 2023, pp. 71–77.
at San Antonio. Her research interests include
[94] R. F. Negoita and T. Borangiu, ‘‘AI-driven RPA for back-office
management of work capacity,’’ EMERG Energy. Environ. Efficiency.
the intersection of AI and computational design to develop innovative
Resour. Globalization, vol. 9, no. 3, pp. 7–22, 2023. solutions in various domains, including parallel discrete event simulation,
[95] A. Das and P. Rad, ‘‘Opportunities and challenges in explainable artificial data analysis, robotic process automation, and predictive modeling.
intelligence (XAI): A survey,’’ 2020, arXiv:2006.11371.
[96] O. B. Akintuyi, ‘‘AI in agriculture: A comparative review of developments
in the USA and Africa,’’ Res. J. Sci. Eng., vol. 10, no. 2, pp. 60–70, 2024.
[97] S. Afrin, A. T. Khan, M. Mahia, R. Ahsan, M. R. Mishal, W. Ahmed, and
R. M. Rahman, ‘‘Analysis of soil properties and climatic data to predict RIAD AKRAM (Member, IEEE) received the
crop yields and cluster different agricultural regions of Bangladesh,’’ in B.Sc. degree in computer science and engineering
Proc. IEEE/ACIS 17th Int. Conf. Comput. Inf. Sci. (ICIS), Jun. 2018, from Bangladesh University of Engineering and
pp. 80–85. Technology (BUET), Bangladesh, in 2009, and
[98] I. Attri, L. K. Awasthi, T. P. Sharma, and P. Rathee, ‘‘A review of deep the Ph.D. degree in computer science from The
learning techniques used in agriculture,’’ Ecological Informat., vol. 77, University of Texas at San Antonio, San Antonio,
Nov. 2023, Art. no. 102217.
TX, USA, in 2017. He is currently a Software
[99] M. Kulbacki, J. Segen, W. Kniec, R. Klempous, K. Kluwak, J.
Nikodem, J. Kulbacka, and A. Serester, ‘‘Survey of drones for agriculture Engineer with Intel Corporation, Hillsboro, OR,
automation from planting to harvest,’’ in Proc. IEEE 22nd Int. Conf. Intell. USA. His research interests include approximate
Eng. Syst. (INES), Jun. 2018, pp. 353–358. computing and its application in parallel programs.
[100] A. Ahmad, J. Ordoñez, P. Cartujo, and V. Martos, ‘‘Remotely piloted He is also working on developing approximate performance model for next
aircraft (RPA) in agriculture: A pursuit of sustainability,’’ Agronomy, generation products.
vol. 11, no. 1, p. 7, Dec. 2020.

VOLUME 13, 2025 197