Impact of Artificial Intelligence on Future of Accounting Profession

This paper aims to review and analyse the research efforts and debates on the application of complex artificial intelligence in accounting profession. Is artificially intelligent system capable of performing more complex and important task such as materiality assessments, detect fraud, going-concern decisions, analytical review procedures?

A thesis
On
The Impact of Artificial Intelligence on Future of
Accounting Profession
Name of the Student
Name of the University

CONFIDENTIALITY AGREEMENT OF GRADUATING PROJECT
Family name:
....................................................................................................................................................
First name:
....................................................................................................................................................
Programme attended at RENNES School of Business:
....................................................................................................................................................
TITLE OF THE GRADUATING PROJECT:
......................................................................................................................................................
..... ..............................................................................................................................................
.............. .....................................................................................................................................
.......................
Date of submission:
......................................................................................................................................................
......
Appendices: YES NO
Confidentiality: YES duration: ................. NO
Comments: ..................................................................................................................................
.......................... .........................................................................................................................
................................... ................................................................................................................
............................................ .......................................................................................................
.....................................................

OATH OF PERSONNAL WORK
I undersigned ..................................................................................... declares that the
following graduating project is my own work. No part of this research has been submitted in
the past for publication or for degree purposes.
I am fully responsible for the truthfulness of this declaration.
Date: .......................................................
Signature:

Acknowledgement
I would first like to thank my thesis advisor [title] [Name Surname] of the [School / Faculty
name] at [University name]. The door to Prof. [Last name] office was always open whenever
I ran into a trouble spot or had a question about my research or writing. He/she consistently
allowed this paper to be my own work, but steered me in the right the direction whenever he
thought I needed it.
I would also like to acknowledge [title] [Name Surname] of the [School / Faculty name] at
[University name] as the second reader of this thesis, and I am gratefully indebted to his/her
for his/her very valuable comments on this thesis.
Finally, I must express my very profound gratitude to my parents and to my [partner, spouse,
girl/boyfriend] for providing me with unfailing support and continuous encouragement
throughout my years of study and through the process of researching and writing this thesis.
This accomplishment would not have been possible without them. Thank you.
Author
[Name Surname]

Executive Summary
Computer systems and software of the present age are capable of creating and exhibiting
intelligence. Artificial intelligence in accounting profession has automated several accounting
tasks and helped accountants and auditors to perform better and play a more effectual role.
Numerous accounting researches have been done to improve the application of artificial
intelligence in accounting and auditing tasks. Artificial Intelligence has mostly been applied
successfully only to the more structured, programmable and repetitive tasks in which
gathering human expertise is not extremely difficult (Moudud-Ul-Huq, 2014). Frey and
Osborne’s (2013) predicted that the accounting profession will face extinction. However,
advanced study and researches claim that technologies cannot replace human intelligence.
According to (Greenman, 2017), Artificial Intelligence in the accounting world will not
replace accountants, it will simply change the focus. This paper aims to review and analyse
the research efforts and debates on the application of complex artificial intelligence in this
profession. Is artificially intelligent system capable of performing more complex and
important task such as materiality assessments, detect fraud, going-concern decisions,
analytical review procedures?

Table of Contents
Chapter 1: Introduction........................................................................................................9
1.1 Background of the Study............................................................................................9
1.2 Problem Statement...................................................................................................10
1.3 Research Questions..................................................................................................12
1.4 Aim and Objectives..................................................................................................12
1.5 Rationale of the Study..............................................................................................12
1.6 Key Terms................................................................................................................13
Chapter 2: Literature Review.............................................................................................14
2.1 Introduction..............................................................................................................14
2.2 Introduction to Accounting:.....................................................................................14
2.3 What is Artificial Intelligence?................................................................................15
2.4 Automation and Artificial Intelligence in Accounting:...........................................16
2.5 Impact of AI on the accounting industry..................................................................20
2.5.1 Accounting Tasks Which Machines Can Do....................................................21
2.5.2 AI Doesn’t Mean Job Losses............................................................................22
2.6 New AI and Digital Technologies in Accounting profession..................................23
2.6.1 AI as a Tool for Business Enablement and Productivity Enhancement............23
2.6.2 Robotics Process Automation...........................................................................23
2.6.3 Advanced Analytics..........................................................................................24
2.7 Why Accountants Must Embrace Machine Learning..............................................25
2.7.1 The Spectrum of Artificial Intelligence............................................................25
2.7.2 Accounting Already Experienced Machine Learning.......................................26
2.7.3 The Impact of Biased Data on Inductive Reasoning.........................................26
2.7.4 Machine Learning Implications for Auditors....................................................27
2.7.5 Machine Learning Implications for Management Accountants........................27
2.7.6 Openings in Enterprise Use of Machine Learning............................................28

2.7.7 Bookkeepers Must Embrace Machine Learning...............................................28
2.8 Conclusion................................................................................................................29
Chapter 3: Research Methodology.....................................................................................30
3.1 Introduction..............................................................................................................30
3.2 Research Design.......................................................................................................30
3.3 Data Collection for analysis.....................................................................................31
3.3.1 Dataset related to predicting Fraud in Financial Payment Services..................31
3.4 Data Analysis Model................................................................................................31
3.5 Data analysis tool.....................................................................................................33
3.6 Limitations...............................................................................................................33
Chapter 4: Data Analysis...................................................................................................34
4.1 Introduction..............................................................................................................34
4.2 How to predict and detect fraud with Machine Learning........................................34
4.2.1 Which types of transactions are fraudulent?.....................................................34
4.2.2 What determines whether the feature isFlaggedFraud gets set or not?............34
4.2.3 Looking at Account Types................................................................................35
4.2.4 Looking at Transaction Types...........................................................................36
4.2.5 Looking balances before and after the transaction............................................36
4.2.6 Artificial Neural Networks................................................................................36
4.2.7 Model 2: Random Forest...................................................................................37
4.3 Discussion................................................................................................................38
Chapter 5: Conclusion and Recommendations..................................................................42
5.1 Conclusion................................................................................................................42
5.2 Recommendations....................................................................................................42
5.2.1 AI will drive advancement in bookkeeping......................................................43
5.2.2 Bookkeeping assignments that machines can figure out how to do..................43
5.3 Future direction........................................................................................................43

5.3.1 The open door in AI..........................................................................................43
5.3.2 The cutoff points of AI......................................................................................44
5.3.3 Critical thinking, as once huge mob..................................................................44
Reference............................................................................................................................45
Appendix............................................................................................................................48

List of tables
Table 1: AI in accounting...................................................................................................19
Table 2: Automation in Accounting...................................................................................26

Table of figures
Figure 1: AI funding by accounting firms..........................................................................27
Figure 2: Transaction type to identify fraud......................................................................43
Figure 3: Fraudulent feature...............................................................................................44
Figure 4: Accounting type..................................................................................................45
Figure 5: Transaction Type................................................................................................45
Figure 6: ANN network.....................................................................................................47
Figure 7: Random Forest....................................................................................................48

DEB: Double-entry bookkeeping
AIS: Accounting Information System
AI: Artificial Intelligence
EY: Ernst & Young
PwC: Price Waterhouse and Coopers
KPMG: Klynveld Peat Marwick Goerdeler
IDP: Integrated Data Processing
AMP: AUDIT MASTERPLAN
ICOR: Integrated Capital and Operations Reporting
ES: Expert System
FSA: Financial Statement Analyzer
EDGAR: Electronic Data Gathering, Analysis and Retrieval
TDE: Transaction Data Enrichment
RPA: Robotic Process Automation
IA: Intelligent Automation

Chapter 1: Introduction
1.1 Background of the Study
Earlier, Robots, Artificial Intelligence, Machine learning were found only in the science
fiction stories and films. In today’s world, these technologies have taken the stage in
workstations across the world. Industries such as accounting, retail, customer service,
agriculture and manufacturing have the AI replacement of various jobs, which has left the
workers and staffs to find the other career option because of the AI replacement. The
revolution of AI is not expected to slow down any time rather increase over time. Moreover,
experts extrapolate that AI technology can replace nearly about 800 million jobs by the year
2030 (Hoffman 2019). Automation and AI technology in organizations can affect the blue
and pink-collar workers. As the AI technology is being more sophisticated and powerful, the
white-collar workers, professionals and accountants are worried about their career in future
and if the AI makes their jobs untenable.
AI is a technology of intelligent machine, which are able to finish the mundane and repetitive
tasks in no time with greater accuracy than the human. The emergence of the machine
learning allows the AI platforms for self-learning, analysing, processing and observing the
task for improving the accuracy and performance over the time. AI technology is able to
handle and manage accounting functions like, audits, payroll and tax preparation. Many
leading accounting software such as Sage, Intuit and Zero have incorporated with the AI
technology to handle and manage the basic tasks of accounting like risk assessment, audit
process, invoice categorization, bank reconciliations, invoice payment and expense
submission (Sutton, Holt and Arnold 2016).
It is noted that the accountants are performing several works which is usually time
consuming. Now the introduction of AI will help institution to reduce the time required to
perform such works and thus directly affects the pay structure of accountants. The other
concern about AI technology is that it may replace the requirement of accountants in the
professional world. There is no doubt that the AI technology is able to handle the standard
accounting tasks and work faster with more efficiency and the capabilities is expected to
increase only over the time. However, this does not mean that the role of the accountants will
come to an end. There will be always a requirement of the human element, which is human
intelligence beside the AI technology. According to the research firm, Gartner, the AI
technology is set to create more jobs for the workers than it will replace the accountants.

The accountants do not have to worry about their profession that it can be replaced by the AI
technology in the future. Companies and organizations will always need accountants to
analyse and interpret the AI data or information and the AI technology can transform the
tasks, an accountant performs. With machine learning and AI technology that handle many
mundane and repetitive tasks of the accountants, they will have more time to focus on the
other important aspects of their jobs like data analysis and consulting. Spending hours in
completing the menial tasks, the future accountants will be able to analyse and use the AI
data to provide the clients with better business solution. In the other ways, the AI technology
can help the accountants by improving the accuracy of data entry and lower the risks of
various task of the accountants. Moreover, the emerging AI technology is efficient in
detecting fraud and it can add extra layer of security and protection for the accountants and
the clients. AI technology provides the real time data that permits the accountants for
providing the real time solution. It is more impressive that the ability of the machine learning
can analyse huge data instantly. The technology evaluates the past failures and success for
predicting the future outcomes accurately.
1.2 Problem Statement
The financial sector is one of the first ever disciplines where the researchers thought AI was
capable of a causing dramatic change and evolve the accounting industry. Accountancy is not
different from the other finance disciplines and the world continues to read headlines that
explains the future of the accounting using the AI. After many years of research, the AI
technology is ready for the key period of transforming, if not damaging all the aspects of
economy, which creates many data from technology to energy, healthcare, finance,
manufacturing, mobility and communication (Hays 2019). The accounting department has
not seen much of innovation apart from the creation of the double entry bookkeeping, which
is the process to record the profit and loss as well and consider the great advances in the
business and commerce history. The revolution of AI technology is here. People have many
AI assistants such as Siri and Alexa in their homes and offices. People have accessible and
user-friendly analytics tools, which are AI powered such as IBM Watson analytics, which is
the platform of the same AI technology that created the culturally defined moments of the
sophisticated AI technology which defeated two Jeopardy human players.
Major firms such as PwC, EY and Deloitte have been investing and considering to invest
more in the AI technology to streamline the tax process and auditing. The demand for
accountants is set to grow faster than the average through in the year 2026. The clear area of

the accountancy to see the AI technology in the action is with these auditing firms such as
KPMG and Deloitte that are using the AI technology to streamline the audit process. The
vital lesson is the use of techniques the AI technology. However, the AI technology will not
replace the auditors. AI technology will help the auditors to review the information in short
time and provide essential and crucial information to make recommendations.
The tasks that consist of extracting and processing of the extreme number of data are the
prime area for the AI technology. For an instance, the firms are using the NLP technology in
coordination with the other AI technology and the tools of data extraction can parse the
information from the contracts in no time while the same task takes considerable human
hours. Sophisticated NLP algorithms can extract the unstructured data such as the emotional
sentiment of any chat or email that can pinpoint the complex issues such as fraud with the
competitors.
There are two major aspects where the AI technology struggles and the two aspects
encompass the most essential parts of the accountancy:
 Responding to the novel situations and
 Extracting the insight from the data and determining the following steps.
For IBM Watson and its historic win at the Jeopardy, it is very easy to see the two aspects
that come into the action. However, Jeopardy is considered unstructured than games such as
chess or checker. Once the AI technology understands the clues, syntax and requirements, it
answers in form of the question, as data extraction issues can be seen in the game. This is
why the platform of the IBM beat the human opponents in section such as naming particular
songs of Beatles, history of Olympic moments and literary criminals while struggling with
the clues that are related to the message like Harry Potter. This does not mean that technology
is to understand the syntax or languages. However, the AI technology continues to struggle
for holding the clues for the future accounting with the AI.
In the Accountancy program of DePaul’s online Master of Science, it emphasizes the vital
importance of understanding the theory and practice of accounting and the AI technology is
perfect example for making other people understand why moving forward is important in
many tasks, and the AI technology will outperform the humans each time.
Accountants will always require to be well-versed in the particular methodologies, which
goes for conducting the tasks. However, the accountants, which excel in the AI-powered

world can be those, who will be able to take the practices and apply them to the new and the
different situations. Moreover, they can use the accounting practices for informing the
business decisions. If accountancy is just the series of the tasks that is conducted in the
similar way to the same type of information each time, then the technology far less
sophisticated than the AI would have threatened the jobs of the accountants. The reality is
that the information is unstructured and humans continue for encounter new situations, which
creates demands for human insight.
The question to be asked isn’t whether the AI will replace the accountants or not. Instead, it is
vital to think how the accountants can use AI technology for being more effective. Below are
some of the top three aspects they will be useful for the accountants.
1. Self-service AI technology: Complex software targets the business users and has the
unique challenge in which it is the complicated functionality. It is easily usable and
accessible. Self-service is the key priority in various areas of the business technology from
cloud computing to the business analytics and the intelligence.
One of the AI platforms have followed the path is IBM’s Watson, for instance, now has a
self-service portals offer functionality ranging from the business analytics with the Watson
Analytics to come up with the Chef Watson with his new recipes.
This trend will go further with the self-service of machine learning. The algorithms need
many complex data science and math knowledge for creating and optimizing. However, the
leaders in the AI technology are working for making it easier to give training to the
algorithms of machine learning. For example, Google uses a tool, which lets the users to train
the image-recognition AI technology by dropping and dragging.
For the accountants, training the AI technology will have many far-reaching impacts.
However, the practical and near-term training will be in the auditing. The accountants will
take larger audit and higher-quality samples, for instance, as they will train the algorithm of
machine learning to identify the types of important data to them and decrease the time it takes
to compile the relevant data for the audit.
2. Turnkey AI modules: Although the machine learning is impressive, not everyone requires
own algorithm of machine learning to do their jobs efficiently and this may not be the best
measure for each issue. In smaller businesses, the accountants that run their own process and
practices, and the organizations which do not have the plethora of the historical data for

training the AI technology that will find it more beneficial to think out-of-the-box with the
software that is built into the AI technology.
We can find AI embedded technology in the popular accounting software such as
QuickBooks, Humans can see the purpose-built AI technology modules, which are designed
to help with the particular series or task of the accounting. The predictive functionality will
be the first true innovations to watch out in this area. That means that the AI modules is
capable of evaluating the accuracy of revenue, forecast the sales, flag while any organization
is to miss the payment and create its own cash flow prediction using the abundance of the
data, which used to be the time consuming.
1.3 Research Questions
Considering the above problem statement, the author has considered the below mentioned
research question for this study:
Is Artificial Intelligence Set to Replace Accountants in the Future?
1.4 Aim and Objectives
The aim of this study was to comprehend how Artificial Intelligence is changing the
landscape of accounting profession and minimize human errors in accounting aspects.
The objectives of the study were:
 To evaluate the way AI is poised to transform the finance industry;
 To estimate the opportunities, it presents for the accounting profession;
 To provide recommendations regarding how AI should be utilized for betterment of
accounting profession;
1.5 Rationale of the Study
A discussion about the AI technology will be incomplete except exploring role of the
automation. The AI technology will make it possible to automate the tasks which was not the
case before, however these tasks will be repetitive and data-centric heavily in nature. The
main advantages of the technologies such as NLP, is that it is able to bring together the
structured and unstructured data. For instance, imagine that you have the data type’s mixture
such as PDF contracts, image files of receipts and email attachments. As the NLP becomes
more advance, the AI technology will make possible to extract the relevant information easily
like the financial input and data into the database for a thorough analysis.

This will cut down the time of the accountants that is spent drastically on the tasks such as
data extraction and data entry leaving much time to create the strategic decisions and provide
the insights for the businesses.
One of the lessons that comes from the evolutions in the technology is the requirement for
staying agile. The revolution of AI technology cannot replace the accountants, however it
will need that the accountants position differently whether they are running the own practice
or in any large organizations. The emphasis in the terms of a high-demand skills will shift to
insightful recommendations and complex thinking, which the accountants can create in the
world where the AI technology can perform the myriad time-consuming tasks. This means
that the accountants will require to show wide breadth of the knowledge and the expertise of
deep accountancy domain as well, which executes the required traditional tasks. The main
aspect is that the AI technology changes the equipment of accountants with the advance AI
that will help in improving the speed and quality of the work thus, minimizing the time to
give value to the organization.

Chapter 2: Literature Review
2.1 Introduction
The purpose of this chapter is to provide a background of the study and literature searched to
support the research question. In this, the researcher has done a comprehensive literature
search related to application of AI in accounting and in the later section, the researcher
specified about machine learning which is broadly used as tool for reducing manual
intervention.
2.2 Introduction to Accounting:
The history of accounting dates back to the ancient times. Accounting is the language of
business. It is a structured process of identifying, recording, classifying, measuring, verifying,
and communicating the financial information. (Kokina and Davenport 2017) explains that
“accounting is concerned with collecting, analysing and communicating financial
information.” (Fisher, Garnsey and Hughes 2016) The art of double-entry bookkeeping
(DEB) was put in business practice around the thirteenth century. In 1494, an Italian
mathematician named Luca Pacioli first published the concept of DEB. Pacioli continues to
famously be recognized as “The father of Accounting and Bookkeeping”. The system has
ever since been used for centuries and is the foundation of accounting.
Accounting is a tool which effectively measures the financial pulse rate of a business. The
result of this systematic process exhibits performance of a business in monetary terms. The
accounting process helps answer questions that first arise in our minds when we come across
any business such as ‘is the business earning profit or operating in loss?’ (Hays 2019). This
continuous cycle of recording and reporting of financial information helps decision makers
make crucial decisions. A person who performs the job of bookkeeping and accounting is
called an accountant. It is used not only in businesses but also helps manage our finances in
our personal lives.
The document which communicates the results and findings of processing the financial
information are called financial statements. Financial statements show you where a company
gets its money from, where the money was spent and where it is now. There are four financial
statements are:
1. Statement of Financial Position
2. Cash Flow Statement

3. Statement of profit and loss
4. Statement of Changes in Equity
The three main branches of accounting are:
1. Financial accounting
2. Management accounting
3. Cost accounting
Apart from the branches mentioned above there are other popular branches of accounting
which help professionals in practice. With time as different types of accounting information
needed by decision makers started increasing and because of which different branches of
accounting came into existence.
1
)
Auditing i. Internal audit evaluates the fairness of internal control
structure of a company.
ii. External audit refers to inspection of financial statements
by an independent party.
2
)
Tax accounting Tax accounting assists in following regulations set by tax
authorities.
3
)
Accounting
Information System
(AIS)
AIS deals with development, execution and monitoring of
systems and procedures in the accounting process.
4
)
Fiduciary accounting Fiduciary accounting is evaluation and accounting of a third
party’s business.
5
)
Forensic accounting Forensic accounting deals with areas that involve legal matters
(fraud investigation, court and ligation cases, disputes, etc.). It is
one of the most trending form of accounting today.
Table 1: AI in accounting
2.3 What is Artificial Intelligence?
Artificial Intelligence (AI) has been a subject of interest for decades and is still one of the
most interesting and elusive concepts because this subject matter is so vast and has so much
more potential that still needs to be discovered. The definition of AI can vary depending on

the discipline it is applied to. Merriam-Webster defines artificial intelligence as “A branch of
computer science dealing with the simulation of intelligent behavior in computers. The
capability of a machine to imitate intelligent human behavior.” To put it in simple words AI
could be described as any task performed by a machine or program that if a human
performed, would require intelligence and effort to accomplish the same task. The outcome
of the task performed by AI would be achieved more efficient, accurate and reliable than that
of a human’s result.
(Kokina and Davenport 2017) AI in its developing stages and progressing years had many
influential ideas from various disciplines. These ideas came from people working in logic and
philosophy (for example, Alan Turning, Alonzo Church, and Carl Hempel), communication
theory ( for example, Claude Shannon’s theoretical work), engineering (such as Norbert
Wiener’s work on cybernetics), biology (for example, W. Ross Ashby and Warren
McCulloch and Walter Pitt’s work on neural networks in simple organisms), game theory (by
John Von Neumann and Oskar Morgenstern), experimental psychology (by Newell and
Simon), mathematics and statistics (such as Irving J. Good), and linguistics (for example
Noam Chomsky’s work). It was not more than half a century ago that powerful programming
languages and computational devices were developed to test and conduct experiments to
understand what intelligence is. Alan Turning’s seminal paper, ‘Mind’ published in 1950 in
the philosophy journal is a significant milestone in the history of AI. (Issa, Sun and
Vasarhelyi 2016) Even though the concept of AI of can be traced back to antiquity, the term
‘Artificial Intelligence’ was first coined by John McCarthy when the first conference on AI
was held. He was a computer scientist and is famously known as the father of AI.
To describe AI from a technical standpoint, it is a mathematical algorithm. It is a code that
helps a machine recognize patterns in data. These days we use devices equipped with sensors
that produce massive data and information that helps AI learn and understand. Artificial
intelligence is the backbone for innovative technologies that we use every day. In almost
every technology we use, we can find the application of AI. It creates a higher degree of
efficiency and productivity by automating repetitive task, by understanding human
sentiments and emotions. It trains and learns similarly to the way humans do. Just as humans
do not need continuous information input to process and utilize data for specific purposes, AI
was created in order to learn just like us and so it does not require continuous instructions like
humans don’t. AI is designed to automatically learn functions on its own.

There are three types of artificial intelligence:
1. Narrow Artificial Intelligence: It is an AI system that can perform a particular task
within a limited context. It is very good at routine and cognitive tasks such as identifying
patterns and correlations.
2. Artificial General Intelligence: It is an AI system that has an ability to apply
intelligence to any problem rather than just one specific problem. It has general cognitive
abilities that can solve unfamiliar task it comes across. Since it can reason and understand the
environment as a human, it is also known as strong AI. Computers can process data faster
than humans but cannot innovate or think abstractly as humans.
3. Artificial Super Intelligence: ASI is a situation where machines will be more
intelligent and will outperform humans. Machines would be in a position where it has higher
cognitive abilities and also capable of mimicking human intelligence and thoughts. It would
gain superiority over humans.
2.4 Automation and Artificial Intelligence in Accounting:
We’re living in a time of constantly evolving environment where change travels in an
accelerating rate. Innovation in a competitive market demands swift adaption and integration.
Modern accounting is more than pen and paper, and number crunching unlike in the past.
(Cao, Xu and Li 2019) Accounting in the 21st century sets the tone of exception. Accounting,
at present should be a mixture of traditional accounting and modern advisory services that are
designed to meet current and future needs. Whether it’s the introduction of DEB, invention of
calculator to simplify calculation, invention of computers, the development of accounting
software or the rapid advancement of AI, we understand the value of information and data.
Accounting at its cores involves massive amount of unfiltered information and data which is
further complied, systematically processed and presented to communicate useful information
and to provide accountability. These information and data, which is often without pattern and
noisy is more than beyond the understanding of an accountant or a team of accountants.
Although accounting has had subtle changes all along, the invention of computer and
intelligent accounting softwares changed it all. This innovation eliminated the concept of
accounting being just paper records of numbers. Cognitive learning and computer-based
automation with human supervision is the perfect combination for the accounting industry.
(Sun 2017) “AI in accounting looks to go beyond the likes of process automation and defined

calculation to stimulate human intelligence as applied in our field.” Data analysis is a very
important aspect in accounting as it involves vast quantities of information and data, both
structured and unstructured. Implementing pre-defined algorithm or logic over huge amount
of structured data could be one thing but it is not the same to iterate through unstructured
information and to learn and scavenge useful and meaningful information.
(Keenoy, 1958) Automation in accounting history has been around since the 1920’s with the
invention of traditional accounting product-the cash register. The concept of Integrated Data
Processing (IDP) was developed with a goal to reduce physical human effort in data
processing. IDP was developed around a machine language that could link together different
operations and machines in a continual automatic machine. (Pannu 2015) “In the 1950’s,
there was a lot of excitement for the potential of automation as it could help accountants and
bankers speed up their work by helping them analyse, calculate and process their work at an
efficiency not possible by humans.”
The financial sector is said to be one of the first industry to steer interest in applying AI.
However, majority of the AI researches done during the 50’s and 60’s did not focus on its
practical application. During that time, a lot of researches focused on the Bayesian statistics.
It is a method that is applied in machine learning and is also used for auditing and stock
market prediction. It was during the 1980’s that AI started flourishing prominently in the
finance and accounting world. (Torresen 2018) “During that time, over two thirds of Fortune
1000 companies had at least one AI project being developed.” A lot of mathematics that had
originally originated during the early 1900’s is now currently being used in AI solutions.
The foundation of applying advanced mathematics in the financial industry began in 1900
with the publication of ‘Théorie de la Spéculation’ (Theory of Speculation) by Louis
Bachelier, which marked the dawn of primitive AI. His research was one of the first papers to
explore the application of mathematics as a technique to evaluate stocks (Pannu 2015).
Robert Schlaifer’s research work in the field of statistics, based on Bayesian Decision Theory
made the application of Bayesian statistics popular. Bayesian statistics constructs informed
decisions that was based on probabilities. Schlaifer published a book, ‘Probability and
Statistics for Business Decision’ in 1959 which lead to exploring the potential capacity of
statistics in the business world. (Pannu 2015) argued how auditors could apply the theory for
auditing in their assessment and hence it was possible to produce a mathematical model that
could help auditors to precisely evaluate the value of assets. Similarly, (Al-Htaybat, von

Alberti-Alhtaybat and Alhatabat 2018) was another researcher who attempted to apply the
Bayesian Analysis in auditing.
During the 1980’s Expert Systems or also called knowledge-based systems grabbed a lot of
interest among many other techniques such as Fuzzy Systems and Artificial Neural
Networks. For example, “Dupont had built 100 expert systems which helped them save close
to USD 10 million a year.” The Protrader expert system, designed by Ting-Peng Lian from
the University of Illinois and K. C Chen of California State University, was one among the
first programs that was hypothesized to predict the market. (du Chenne 2019) “The major
functions of PROTRADER include the following: monitor the premium in the market;
determine the optimum investment strategy, execute transactions, when appropriate, and
modify the knowledge base through a learning mechanism.”
System Name Application
AUDITPLANNE
R
Steinbert developed Auditplanner, an EMYCIN knowledge
engineering language in 1986/87 (Steinbart, 1987).This expert system
assists the auditors in materiality judgement in different stages of audit
planning. (Woods and Linsley 2017) “Auditplanner is a prototype system
emulating the process by which the auditor arrives at judgements about
materiality and tolerances on the basis of quantitative and qualitative
information.” This language is a computational model which copies the
cognitive learning process of an expert reasoning on the same. It is used
by companies such as Deloitte, Haskins & Sells Foundation
RISK ADVISOR Risk Advisor is a FFAST developing tool developed by Coopers &
Lybrand. It is an Automated Reasoning Tool (ART) Shell. Risk advisor
analyzes the audit risks and can also be used in audit planning stage to
identify and document the potential risks. It can also be applied to
substantiate where the risk identified in the audit plan has been tackled
with or not. (Anderson and McGrew 2017) “It is based on Audit
Strategy Memorandum and was used by Coopers & Lybrand to identify
audit risk and how to approach it.”
INTERNAL-
CONTROL-
ANALYZER
Internal-Control-Analyzer is an EMYCIN Knowledge engineering
language which is used to assess internal control. It is used in helping the
auditors to evaluate the internal control in the income cycle of
accounting. To assess the accounting internal control, the data and

information is drawn from the source database of the client.
EXPERTEST Expertest is a programming language written in LISP language. This
programming language (Ayoubi et al. 2018) “produces audit schedule
which include every audit aspect, or partial audit aspect about 19
standard audit schedules contained in the system knowledge base.” The
system also produces a review record that assists audit managers evaluate
the generated audit-schedule.
AUDITOR Auditor is an AL/X Shell (Advice Language/X) invented by Michie.
It is built on general inference engine. “It assists the external auditors to
assess the adequacy of the forecast made by the client in order to cover
the bad debt risk.”
CHECK-GAAP CHECK-GAAP was invnted by Deloitte, Haskins, and Sells. It is
developed in C and assembler language. This system assures whether the
audit report prepared by the auditor abides by the UK Companies Act.
AUDIT
MASTERPLAN
(AMP)
AMP helps the internal auditors in the planning process of audit. It
was developed by Institute of Internal Auditors (IIA).
ICOR Material
Subsystem
ICOR is a computer-based system that amalgamates 11 subsystems.
“One of these modules, the Material Subsystem, is an expert system that
assists in the pricing of material transfer at EXXON. The system has
three knowledge bases, which are used to transfer pricing suitable for an
inventory or transferred equipment with or without transfer of
ownership.”
XPR XPR is a PROLOG language that verifies management control systems
and was developed by Institut Superieur des Affaires (ISA) en Jouy-en-
Josas. This language performs a financial and economic diagnose on the
management control components. It is also used to perform a technical
analysis on the management control items.
AGGREGATE AGGREGATE was developed in PROLOG computer language. Its
function is to help the accountants in drafting an accounting information
system (AIS) and also assist in preparing the financial statements.
COMPTA COMPTA is wriiten in SNARK language. It is a “theoretic
development of an ES to carry out the accounting entries if a sale

invoice.”
FSA FSA was invented by Arthur Andersen & Co and developed in LISP
language during the preliminary stage of EDGAR project. It
systematically inspects each entry in a business financial statement. “The
system enables to identify frauds in any kind of format, page foot or
wherever comments. It executes ratio analysis using as an information
source business annual reports (10K). It understands the balance sheet,
profit and loss statement and page foots.” The system contains two
accounting knowledge bases. The first base to interpret financial
information and financial statements and the second contains semantic
structures that is managed by page foot process.
CONSOLIDEX CONSOLIDEX is a prototype ES developed in CRYSTAL Shell. It
focuses on understanding the consolidated annual accounts standards of a
corporate group.
CASHVALUE CASHVALUE was developed in BASIC programming language by
Heuros Development Ltd. It was appropriate to analyze investment
projects. The system potential uses include project investments,
enhancing the existing operations, cost reduction schedules, purchasing,
new risked business and business assessments. It utilizes a cash flow
methodology, discounted to assess investments. Once the consultation is
complete, it generates an assessment report on the project. The generated
result can also be communicated graphically and in spreadsheets to
create a sensitivity analysis.
MANAGEMENT
ADIVSOR
MANAGEMENT ADVISOR is developed in LISP language and was
first known as FINANCIAL ADVISOR. The system guides financial
managers and accountants on capital allowances and budget planning for
big businesses. It is built on discounted cash flow method and also helps
in business functions such as purchase, cost control, merging, etc.
Personal Financial
Analysis
Personal Financial Analysis was developed by PwC and gives
employee financial guidance to its clients. The system, based on one
family and economic status information (i.e. revenues, taxes,
investments, marital status, specific financial goals as retirement
financing or children education, …), prepares a long report providing

suggestions for asset management, invest strategies, tax saving strategies,
education saving planning, life insurance needs, retirement funds, etc.”
Table 2: Automation in Accounting
2.5 Impact of AI on the accounting industry
Considering the expanding guideline and requests from customers, many bookkeeping firms
are presently moving to various sorts of workforce to assist them with their tedious errands.
This new supply of help can assist organizations with the most perplexing undertakings
without requesting a month to month compensation. That is on the ground that this new
workforce isn't human, yet isn’t fake. Man-made brainpower (AI) stretches out the abilities of
registering to a level that is unheard of. It lets frameworks to make forecasts and make
changes in the manner – similarly as humans would. It empowers computers to perform
machine-based realizing which prior was human’s job. In the bookkeeping calling, where
individuals manage repetition errands, AI is a substitution of human abilities to improve
things. As per an ongoing overview led by MIT-Boston Consulting Group, over 80%
individuals accept that AI prompts upper hand and 79% accept that innovation builds
profitability of an organization. Man-made consciousness is being utilized by many
bookkeeping firms where it breaks down an enormous volume of information at fast which
would not be simple for people.
Figure 1: AI funding by accounting firms
Organizations today are grasping and executing new advances to streamline their business
activities, and one of the tasks that is on the highest priority on their rundown is bookkeeping.
That is on the grounds that AI is giving positive outcomes, for example, expanded

profitability, improved precision, and diminished expense. With such a large number of
advantages, AI is utilized progressively for regulatory errands and bookkeeping bringing
about different auxiliary changes. With AI, preparing and taking care of information is totally
mechanized and is, in this manner, one of the key advantages of AI in the zone of
consistence. Along these lines, the information produced by any expense report would have
guaranteed precision levels and will be created rapidly. Furthermore, with the assistance of
AI, information can be perceived and ordered from various sources to the correct
bookkeeping head (Sutton, Holt and Arnold 2016). Numerous other devoted errands which
were done by bookkeepers like preparing payable and receivables records are effectively
taken care of by AI. This prompts improved cost of the board of organizations.
2.5.1 Accounting Tasks Which Machines Can Do
Alongside freeing up people from complex errands, AI would likewise enable organizations
to improve their activities. Bookkeepers who grasp the most recent developments in
innovation will pick up skills which would make them significant in business process
changes.
1. Month to month or quarterly close strategy
The sooner you get the numbers, the additional time your organization needs to make
procedures about what should be possible with the numbers. Simulated intelligence can give
you information from different sources, combine, and consolidate it. This would accelerate
the month to month process as well as be increasingly precision in light of association with
machines.
2. Procurement
The following and obtaining technique for some, organizations is loaded up with
administrative work and utilize diverse document designs which may not be good with one
another. Be that as it may, machines with APIs can be incorporated, and unstructured
information can be handled. This makes the obtainment procedure paperless and simpler.
With the assistance of AI, one can without much of a stretch track the adjustments in cost
among numerous providers.
3. Records payable/receivable
The current framework as of now has an AI-controlled receipt the executives procedure
which can make creditor liabilities/receivable handling increasingly streamlined with the

assistance of computerized work process. They can gain proficiency with the bookkeeping
code for the separate receipt.
4. Review
Digitalization tracks which records, to whom, by whom and when, builds the security of
information and documents. During a review, evaluators need not look file organizers for
documentation as they can undoubtedly approach the computerized records. This, thusly,
expands exactness and effectiveness of reviews and makes it conceivable to review 100
percent of an association's monetary exchanges, rather than just examples.
5. Cost the executives
Auditing and affirming every one of the costs to ensure that they are consistent with the
organization's approaches can be a tedious procedure. Man-made intelligence makes it a lot
simpler as machines can check receipts, survey costs, and caution individuals if there is any
rupture.
6. Simulated intelligence chatbots
With AI, machines can proficiently resolve normal inquiries from clients, for example, when
bills are expected, most recent record equalization, and status on records.
2.5.2 AI Doesn’t Mean Job Losses
As indicated by an investigation, it is normal that bookkeeping undertakings, for example,
charge, finance, reviews, and banking would be totally computerized utilizing AI. Things
being the what they are, what will bookkeepers do now and in what manner will AI change
the calling? Man-made consciousness is certainly opening up bookkeepers from doing
modest errands, and this enables them to enlarge their jobs. We can take a case of inspecting
of cost claims where AI can be utilized to recollect and actualize an organization's cost
strategy. This would imply that bookkeepers would never again need to glance through the
receipts and classify dates and VAT numbers. Like cloud bookkeeping disturbance in the
bookkeeping area, bookkeeping experts should re-examine their profile (Issa, Sun and
Vasarhelyi, 2016). This soothes the quantity of hours spent on authoritative undertakings and
enables them to convey more worth and better assistance to customers, just as spotlight on the
development of the organization and generally speaking achievement. Having better
innovation wouldn't mean anything in the event that one can't follow up on it. That is the
place where the job of a bookkeeper comes. Regardless of whether machines can play out

every figure or introductory review related assignments, somebody would be expected to
break down the procedure and reach a significant determination.
With the ascent of AI, bookkeepers are required to move their jobs into counselling and
encouraging to guarantee that they are exploiting mechanization while helping their business
develop. SMBs don't have similar assets when contrasted with greater firms to make their
very own AI items. In any case, specialists accept that in the coming years, AI would be
broadly accessible in any event, for littler firms at a standard admission. Man-made
consciousness is as of now affecting the advertising segment and would likewise acquire
radical changes in the bookkeeping division with an immense probability of the decrease in
expenses and expanded profitability. This is notwithstanding the exactness and accuracy
which AI can get any humble and monotonous manual assignments which have been
mechanized. In perspective on the momentous advancement made by AI, this new world
would be notable to bookkeepers well before self-driving truck drops solicitations in their
office.
2.6 New AI and Digital Technologies in Accounting profession
Artificial intelligence and computerized advances have the ability to make a progressively
proficient and exact workforce that costs 40% less and can be effectively scaled to fulfil
developing need. Budgetary estimate could be consequently refreshed for day by day changes
in deals and a game-plan prescribed to improve the top line and meet development targets.
Customers may even have the option to stroll into the workplace and request that their
computerized associate demonstrate the most recent income projections by item. Welcome to
a world empowered by AI and computerized innovations.
2.6.1 AI as a Tool for Business Enablement and Productivity Enhancement
As per an overview by PwC, business pioneers trust AI will be key later on with 72%
believing AI to be a business advantage. Artificial intelligence will enable associations to
lessen costs, increment commitment for workers and customers and improve client support.
The effect of AI advancements on business is anticipated to expand work profitability by up
to 40% and help individuals utilize their time. With the rise of intellectual robotization, fund
and bookkeeping jobs will change. Jobs comprising of manual yet routine undertakings, for
example, accounting, finance, AR/AP board, and passage level expense, including centre
administration jobs regulating those staff are in danger of being computerized (Hoffman
2019). Monetary arranging, examination and budgetary detailing will encounter a move

toward higher-ability places that creates attention towards making an interpretation of the
hidden investigation to help complex business choices.
2.6.2 Robotics Process Automation
Robotics Process Automation (RPA) is a programming computerization instruments with a
lot of capacities that enables clients to make their own robots (bots) to deal with high-volume,
low intricacy, and repeatable errands quicker and with more precision and lower cost than
people. Run of the mill RPA arrangements can give cost reserve funds going from 20%-60%
of FTE costs. In any case, most in-advertise RPA arrangements right now just copy errands
performed by people and still require escalated human programming to set up. While still
constrained in what it can do, there has been critical advancement to wed conventional RPA
with AI to make subjective computerization that is prescient, mindful and self-mending.
What could this mean for money and bookkeeping? A rethinking of whole business forms.
For instance, an ordinary combination procedure requires a focal group first to gather data
and layouts physically from different business groups. The focal group at that point surveys
the merged outcomes and solicitations follow-up help on huge fluctuations. Notwithstanding
robotizing the manual union, a subjective RPA arrangement could get to and merge the
information from the frameworks utilized by the different business groups straightforwardly.
This proselytes the focal group work from a receptive, dismantle arranged group to a push-
situated and worth included accomplice (Dunis et al. 2016). It is still early days, yet this is
actually the sort of chance professional accountants ought to get ready for and utilizing.
RPA arrangements accessible today are regularly isolated into three seller classes:
 Enterprise sellers commonly offer the most elevated complexity and regularly require
brought together arrangement and more elevated levels of venture. Instances of big
business sellers incorporate BluePrism and Automation Anywhere, the two of which
presently offer a few components of intellectual mechanization over their
conventional RPA contributions.
 Desktop Work area merchants are commonly utilized for computerizing tedious
undertakings on autonomous work areas yet don't take into consideration co-
appointment between situations. In any case, they are generally simple to send and
require lower speculation. A case of a work area seller would be WinAutomation.
 Hybrid merchants fall some place in the middle. They offer a decent harmony among
abilities and intricacy and can be scaled up or down for big business use or littler scale

move outs as required. Instances of cross breed sellers incorporate WorkFusion and
UiPath.
2.6.3 Advanced Analytics
Analytics isn't new to PAs. The account elements of PAs as of now influence examination to
give money related knowledge. For instance, PAs in money related revealing jobs help
organizations comprehend what affected their monetary outcomes and why. PAs in money
related arranging and examination jobs utilize chronicled information and presumptions to
demonstrate and figure monetary outcomes.
Presently, with progressions in AI calculations, AI has empowered propelled business
examination. Prescient examination utilizes information, calculations, and AI systems to
envision future results. Prescriptive examination makes this one stride further by directing
forecasts into activities (i.e., how might we get it going). Right now advanced analytics isn't
as adult as prescient investigation, however Gartner gauges the prescriptive investigation
programming business sector will reach $1.1 billion by 2019 and 35% of associations will
utilize some type of prescriptive examination by 2020.
The job of account must advance to accomplish more than report verifiable key execution
pointers (KPI) and make occasional conjectures. Organizations are moving quicker than at
any other time and the speed to knowledge for account must quicken too. The utilization of
Big Data by cutting edge investigation arrangements can help create further comprehension
of new market patterns, distinguish new KPIs for execution the board, and improve exactness
and practicality of the gauging procedure (Kochenderfer 2015). "Increasingly compelling
estimating and a more profound comprehension of how markets are probably going to
advance can give account pioneers more trust in their figures, preparing to better joint effort
with different business pioneers and expanded trust in the fund group," said Andrej
Suskavcevic, CAE, President and CEO of Financial Executives International and Financial
Executives Research Foundation.
To adjust to this information driven world, PAs should create abilities in data innovation that
enable them to comprehend and interface with data frameworks. There will likewise be a
heavier accentuation on understanding information relationships and patterns, which will
require strong comprehension of measurable techniques, for example, relapse investigation,
test size assurance, and theory testing. Information contextualization (i.e., realizing what

inquiries to pose about the information) will be as significant as playing out the real
examination. An insignificant comprehension of coding will end up basic to most callings,
including PAs. To dissect huge measures of information, it will never again be adequate to
use devices, for example, Excel. It will be useful for PAs to figure out how to interface
straightforward with databases by means of programming dialects, for example, SQL, R, and
Python as opposed to depending absolutely on information researchers for each solicitation.
PAs will likewise need to see how to use self-serve information revealing and perception
devices, for example, IBM Cognos BI, SAP BusinessObjects, Microsoft Power BI, and
Tableau to make reports and dashboards to introduce new bits of knowledge to the board.
2.7 Why Accountants Must Embrace Machine Learning
There is as of now much dread and promotion around AI and its effect on bookkeepers. In
Gartner's Hype Cycle of Artificial Intelligence, most of AI applications are climbing and
peaking the Peak of Inflated Expectations—implying that desires are high and numerous
advancements are as of now neglecting to meet those desires. In any case, this doesn't imply
that AI will leave. It implies that we're beginning to push through the publicity and make
sense of practical applications for AI—some of which will be valuable to bookkeepers and a
considerable lot of which will be utilized by the associations we serve.
2.7.1 The Spectrum of Artificial Intelligence
Some portion of the test with a rising innovation is that there is regularly a misty meaning of
what innovation is, and what it isn't. As sellers' promoting divisions try to use the publicity to
drive deals, they frequently start alluding to innovation trendy expressions in a free sense and
cause extra obscuring of the definition. To help appropriately make way for breaking down
the ramifications of AI on bookkeepers beneath are the expansive AI innovation
classifications are examined:
 Machine Learning: the capacity of the PC to perceive and apply designs, determine its
own calculations dependent on those examples, and refine those calculations
dependent on input.
 Deep learning: the capacity of the PC to distinguish connections and affiliations, and
apply those in comparable conditions (this mostly what our mind does).
 Machine reasoning: the capacity of the PC to apply its "understanding" of
information, connections, rules, and so forth., to "think" however the ramifications of
a specific arrangement of data and give some examination or translation.

 Natural language preparing: the capacity of the PC to "comprehend" human discourse.
 Computer vision: the capacity of the PC to "see" pictures and "perceive" individuals,
things, exercises, and states (for example upbeat, pitiful, moving, and so on.) in those
pictures.
Of these classes, AI has the broadest accessible applications and its usefulness can most
enormously enhance the resources of a bookkeeper, so this article will concentrate on AI.
2.7.2 Accounting Already Experienced Machine Learning
AI is great at "inductive thinking"— where dependent on a lot of existing information focuses
or models, a PC can make sense of what the "rules" are to decide an outcome. Return a stage
to your insights class in college and you may recall procedures like direct relapse, the
estimation of the connection, and unwavering quality of different information focuses. At an
exceptionally fundamental level, these are the sorts of examinations that AI calculations are
applying to anticipate results (Appelbaum et al. 2017). With the computational intensity of a
PC behind it, AI can process a large number of information focuses about a given
arrangement of circumstances to make sense of which ones are significant, and which are not,
and afterward apply the induced standards to another comparative arrangement of
information to anticipate results.
Amazon's, Kindle's, and Netflix's proposals to clients are an extraordinary case of this.
Netflix can utilize your evaluations of different shows in its library and information focuses,
similar to type, chief, entertainers, and so forth., to anticipate whether you will like another
show. Arouse accomplishes something comparative for books. Amazon utilizes item viewed,
other customers' buy history, and complimentary things to those in your truck to recommend
extra items you may like. Regardless of whether you understand it or not, you've just had AI
applied to attempt to foresee what you may like.
2.7.3 The Impact of Biased Data on Inductive Reasoning
Since inductive thinking "learns" from existing informational indexes, it is essential to
comprehend whether the informational indexes that are utilized to "educate" AI calculations
have inalienable predispositions. An oversimplified case of this is on the off chance that you
just watch blood and gore films on Netflix and rate them all high, and you likewise happen to
watch other low-spending motion pictures on Netflix in light of the fact that you can't get
them on another stage, Netflix will presumably foresee that you just like awfulness and low-

spending motion pictures. Netflix doesn't have a clue about that you really like a wide
assortment of motion pictures—it simply doesn't approach that information.
As it should be obvious, there is the potential for positive and negative effects from one-sided
information. On the off chance that the one-sided information speaks to a result that you
need, at that point utilizing the information focuses from that one-sided information is a
positive effect. Then again, if that one-sided information causes AI to give examinations that
will bring about a negative effect, at that point the best possible shields must be set up to
counteract or distinguish the negative effect. Or on the other hand stated in a progressively
well-known way: we should guarantee that inner controls are executed to deal with the
dangers related with a negative effect from the use of AI.
2.7.4 Machine Learning Implications for Auditors
There is a high potential for AI to give expanded examinations to evaluators. AI is simply
one more apparatus in the reviewer's belt of Computer Assisted Auditing Tools and
Techniques (CAATTs). Rather than testing information, evaluators can drive a substance's
whole record through robotized examination (Tirgul and Naik 2016). This, incidentally, is not
machine learning or AI; this is an ability that as of now exists in instruments like IDEA and
ACL. These devices can play out an assortment of examinations, structured by people, and
afterward give arrangements of special cases to the reviewer to assess. AI possibly becomes
the most important factor as the evaluator affirms the exemption or nullifies that special case
and the machine figures out how to "look" at the examiner's decisions and attempt to
recognize extra information focuses about the positives or negatives to apply to extra
exemptions it distinguishes (Li, K., Niskanen, Kolehmainen and Niskanen 2016). Along these
lines it figures out how to all the more likely distinguish special cases.
The information inclination chance in this application is that if a reviewer inaccurately clears
things that ought to be affirmed as special cases, AI would begin to clear different things that
ought to be exemptions. So an audit procedure must be set up to guarantee that cleared
special cases truly are not exemptions (West and Bhattacharya 2016). The opposite is
likewise valid for affirmed special cases that ought not be exemptions. In a further developed
application, a lot of exchanges could be given to an AI apparatus and AI would distinguish
the patterns in exchanges and have the option to recognize what "ordinary exchanges"
resemble. Utilizing this strategy, it would then recognize those exemptions that does not

coordinate with the standard as special cases. This use of AI is additionally dependent upon
information inclination since its image of an ordinary exchange depends on the arrangement
of information given (Jiang et al. 2017). On the off chance that the informational index that
was utilized additionally happened to have a high rate of deceitful exchange, at that point
those false exchanges might be deciphered to be typical exchanges since they are
exceptionally present in the informational collection that the AI gains from.
There is certainly a future requirement for a human evaluator even as AI begins to enlarge
review methods (Lin 2016). The evaluator job, for both inside and outside examiners, will
change from execution of the methods to structure of the techniques, elucidation of the
outcomes, and observing the viability of the understanding.
2.7.5 Machine Learning Implications for Management Accountants
The ramifications of AI for the board bookkeepers and other expert bookkeepers working in
business and government is much more noteworthy than it is for inspectors. Notwithstanding
AI being applied inside money, it might likewise be applied in different pieces of the
association and the board bookkeepers must guarantee that there are appropriate
administration and inner controls applied to AI all through the association. Inside the
controllership work, AI might be applied to help with the grouping of exchanges (Dirican
2015). Inductive thinking could be applied to the source information of authentic exchanges
to help "anticipate" the characterization of extra exchanges as they are recorded. Since the
results of numerous merchants have a genuinely reliable characteristic arrangement, generally
this is alright. In any case, there are a few sellers that could be set in various common
arrangements relying upon how their item is utilized. Take, for instance, an email pamphlet
apparatus used to give showcasing and special messages to clients and delegated promoting
spend (Dai and Vasarhelyi 2017). Nonetheless, in the event that it were utilized to produce
worker bulletins, it might rather be named representative relations or an IT cost. Human
approval of the grouping for sellers with this hazard might be essential relying upon the
materiality of the potential mistake.
At the point when utilized as a major aspect of budgetary arranging and examination
(FP&A), AI can be utilized to break down information to characterize or refine information
models utilized for gauging (Sterne 2017). The nature of the informational index being
utilized and the danger of inalienable predispositions may again affect the nature of the

expectations gave by AI. FP&A bookkeepers must exercise care because of the effects of the
informational indexes utilized for their models.
2.7.6 Openings in Enterprise Use of Machine Learning
Associations that are investigating AI should likewise address the extra administration and
inward controls contemplations for the related risks. As offices outside of fund look to utilize
AI, FP&A bookkeepers have a major chance to give their information investigation and
demonstrating ability to enable different offices to build up their utilizations of AI. This isn't
a territory that can be tended to by only it. A comprehensive perspective on the information,
procedures, and utilization of data given by AI must be acquired. For each task, bookkeepers
in account and inner review must make certain to comprehend the consistence prerequisites,
and evaluate the plan of controls to relieve AI dangers from one-sided information. Inside
evaluators additionally play a significant on-going job in assessing the plan and adequacy of
the administration and inner powers over AI, and in assessing the viability of the strategies
picked to decrease the danger of negative effects from one-sided information.
2.7.7 Bookkeepers Must Embrace Machine Learning
Computer based intelligence as a bookkeeper substitution legend is extremely only a greater
amount of the publicity that will be refuted. Rather, expanded utilization of AI will enable
bookkeepers to concentrate on giving preferred choice help somewhat over on information
social event and manual investigations. Expanded utilization of AI will likewise expect
bookkeepers to step up and address related dangers with AI through viable administration and
inside controls (O'Leary 2015). Bookkeepers need to take a gander at both how we can use
AI to encourage our job as inspectors and bookkeepers. There is likewise a huge open door
past the account setting to direct different offices in their utilization of AI and help with the
structure of interior powers over their applications.
The corporate administration capacity should likewise be acclimated to address business
procedure arrangement of the dangers displayed by AI innovation. A sub-work concentrated
on information administration ought to be built up to address the two information inclination
hazards just as consistence dangers, similar to security. There are territories that IT can't
address alone since they don't have the hazard and controls skill that bookkeepers do. By
holding onto AI as an apparatus, bookkeepers can move where we're investing our energy

from performing humble information planning and investigations to the drawing of bits of
knowledge from those examinations. Bookkeepers' ability in controls structure and
understanding information inclinations can likewise be utilized to serve different divisions in
the association as the offices look to grasp AI. AI gives an exceptional chance to bookkeepers
and we should grasp it to improve both our vocations and the upper hand it can give to the
associations that we serve.
2.8 Conclusion
This chapter provides an in-depth analysis of studies available related to the chosen filed. It
also helped the researcher to find out the major issues based on which research objectives
were set. The next chapter is research methodology which further explained the process, the
researcher has adopted to perform this study.

Chapter 3: Research Methodology
3.1 Introduction
This part of research exhibits the philosophical assumptions supporting this examination, just
as to present the exploration methodology and the exact systems applied. The part
characterizes the extension and restrictions of the exploration structure, and arranges the
examination among existing exploration customs in data frameworks. The philosophical
assumptions base this exploration originate from the interpretive convention. This infers to an
emotional epistemology and the ontological conviction that the truth is socially developed.
This specific thesis was performed considering the machine learning approach and how it
eases the accounting profession. As the aim was to understand whether artificial intelligence
is the future of accounting profession or not, the researcher has tried to find out how machine
learning approach, a specific segment of artificial intelligence can be useful in several ways.
The fundamental information gathering strategies utilized in this examination study were
secondary data available in Kaggle website.
This part is partitioned into three areas. In the principal, the interpretive position in the field
of data frameworks is analysed. The following area is about the exploration technique. It
depicts the exploration approach followed on the off chance that reviews investigation. The
third part covers the explanations behind choosing associations, information sources, inquire
about investigation sub-units, information accumulation and investigation, and a short
synopsis of the desires from the hypothetical system received.
3.2 Research Design
Since the broad objectives of this study was to evaluate how AI is changing the landscape of
financial institutions as well eases accounting profession, the author tried to focus a specific
aspect. With such aim, the specific focus was to comprehend how AI, specifically machine
learning algorithm can help financial institutions to get desired result without much
involvement of human resource and thus to avoid human errors. One of the specific aspects is
fraud detection in financial transaction. It is evident that to conclude whether a financial
transaction is fraud or not, a manual investigation is time consuming and often shows less
success ratio. On the other hand, involvement of machine learning algorithm is very much
efficient in this particular area. The researcher thus considered a secondary data set available

on Kaggle website related to financial transaction and tried to design a machine learning
algorithm to show how financial institutes can perform such investigation and can predict
whether fraudulent activities are happening or not. Again, distinguishing firms that
misshaped budget reports is a difficult and energizing issue among inspectors, banks and
speculators who depend on financial statements to decide. However, it is hard to banner out
these organizations as cooking the books can take a few structures: concealing losses through
different elements, perceiving income too soon or in any event, getting an unregistered
reviewer to guarantee that the fiscal report is valid and reasonable. Indeed, even after money
related reviews, extortion on budget summaries can remain revealed as misrepresentation
procedures are winding up increasingly modern. While manual investigation is difficult to
catch such distortion, again application of machine learning algorithm is a smart way to deal
such issue.
Further, the application of AI can also reduce human involvement in accounting profession in
some other ways as well. For example, AI can be utilized in Charging of Invoices to Account
Codes. The reason for this task is to naturally charge invoice to accounting codes, lessening
human exertion in the instalment preparing process. It additionally lessens irregularities in the
record charging process by killing the distinctions in judgment of various money officials.
Other than diminishing human exertion and mistake, it is likewise expected that utilizing a
machine learning approach consequently charge invoices to accounting codes would build
information quality, to help downstream investigation of what we are spending on. At
present, on the grounds that distinctive money officials charge similar kinds of spending to
various recorded codes, it is hard to decide the amount we are spending on each sort of
capacity.
Hence, the researcher here tried to show how machine learning algorithm is helpful in these
two specific situations, which further helped the researcher to conclude whether AI can be
treated as the future of accounting profession.
3.3 Data Collection for analysis
As mentioned above, the researcher intended two evaluate two datasets available in the
Kaggle website for this study purpose. The detailed of these two datasets are discussed as
below:

3.3.1 Dataset related to predicting Fraud in Financial Payment Services
This dataset is by and by just one of four on Kaggle with data on the rising danger of
computerized money related extortion, underlining the trouble in getting such information.
The fundamental specialized test it stances to anticipating extortion is the exceptionally
imbalanced conveyance among positive and negative classes in 6 million columns of
information. Another hindrance to the utility of this information comes from the potential
inconsistencies in its depiction. The objective of this investigation is to settle both these
issues by an itemized information investigation and cleaning pursued by picking an
appropriate AI calculation to manage the slant. I demonstrate that an ideal arrangement
dependent on highlight designing and extraordinary angle helped choice trees yields an
improved prescient intensity of 0.997, as estimated by the territory under the exactness
review bend. Urgently, these outcomes were gotten without counterfeit adjusting of the
information making this methodology appropriate to genuine applications. The description of
the dataset is mentioned as below:
3.4 Data Analysis Model
The financial frauds which are taking place are becoming more and more complex and it is
also to be noted that the cyber criminals are adapting to technological changes which makes it
imperative to develop a model which has predictive capability and accuracy with the help of
machine learning. The models which is formulated with the help of Machine learning have
tens of thousands of parameters which are far more superior than an expert system of human
reviewer for identifying any correlations or irregularities.
In case classification is done, there are two types of machine learning algorithms which are
supervised and unsupervised learning. The former is often used for annotated data, for review
of fraudulent activities conducted by humans and also for learning complex patterns which
are followed by such criminals. The complex patterns are learned from the data set which is

available for analysis. The unsupervised learning approach is for unlabelled data and infers
inner data structure by itself.
The scientists of the data have access to range of techniques which can be segregated on the
basis of which problems which such techniques deal with and the same are classification and
regression. The techniques can be used effectively for analysing a transaction or set of
transactions and providing an outcome whether the same are accurate or fraudulent in nature.
The typical supervised machine learning algorithms are used to solve problems which are
related to logistic regression, decision trees, random forests, and neural networks.
 Logistic regression is one of the popular tools which is used for analysing the cause
and effect relationship between different variables of a data set. This method can
create algorithm whether the transactions are accurate or not.
 Decision Trees can be used for detecting frauds and anomalies and the same is done
with the help of created rules related to model customer behaviour.
 Random Forests ensemble multiple weak classifiers into one strong classifier and the
same can be used for creating decision trees.
 Neural networks are an appropriate technique which is derived from workings of the
human brain. These networks help in learning and adapting to patterns of normal
behaviour and identify frauds on a real time basis.
Unsupervised techniques are formulated on basis of clustering algorithms which have similar
data points and the same are used for identification of any anomaly. Algorithms used in the
unsupervised approach are K-means clustering, Local Outlier Factor and One-Class SVM.
 K-means clustering is a method which helps to divide datasets in clusters. The
algorithm works iteratively and allocates data points to the predefined number of
classes (k), based on the characteristics that are present in the data set.
 Local Outlier Factor is a method which allows users to compute local density in data
sets and also identify regions which have a similar density in terms of data. The
concept can be used in an effective manner for distinguishing between lower density
dataset from that of higher density of dataset. These are used for detection of
fraudulent activities.
 One-Class SVM learns a function used for novelty detection. The idea of novelty
detection is to detect rare events, i.e. events that happen rarely. The problem is then
that the usual way of training a classifier will not work.

Machine learning have long reaching positive impacts but the same also have limitations
which also need to be considered before application of the same.
Machine learning models are as good as the data set which is available for analysis as results
are derived on the basis of the same. While financial services have access to huge data sets,
there are relatively few fraudulent transactions associated with the same, which affects the
predictive capability of the system. There are several approaches to dealing with this
problem.
One way for financial firms for overcoming the challenge of imbalance of data is by
artificially creating data points which emulate the under-represented class. This is a case of
oversampling. A popular technique is SMOTE (Synthetic Minority Over-sampling
Technique). This technique uses the k-Nearest Neighbours algorithm to review minority
classes in the dataset and characterise the same. Next, it randomly chooses one of the nearest
neighbour data points and creates an artificial randomly changed data point near to it.
One of the other useful techniques is under sampling. It is fairly simple algorithm which
randomly samples the dominant class to reduce its size. The third approach is known as
Combined Class Method, uses SMOTE to interpolate data sets on the boundary of outliers
and inliers and the nearest neighbour technique to clean the data near the separatrix of both
classes, making datasets cleaner and easier to distinguish. The techniques which are used for
machine learning algorithm are much better balanced and therefore the same helps in taking
major decisions for the business. In addition, the current uses of machine learning in fraud
detection still involve a two-stage process where the first stage is automated, but the second
stage requires manual checking. The system exposes the financial firms to the possibility of
human errors. The employing of Artificial Intelligence has numerous advantages in detection
of financial frauds.
3.5 Data analysis tool
The researcher here used Python to apply the machine learning model. Apart from python,
the researcher can utilize tools like R, however, python is one of the best tools for building
machine learning model because of its efficiency as well as easy to interpret visualization.
3.6 Limitations
There are so many ways machine learning algorithm can be utilized to showcase how AI is
considered as the future of accounting profession. However, the researcher has chosen two

specific areas as mentioned this chapter to avoid data discloser issue as well as because of the
time constraint.

Chapter 4: Data Analysis
4.1 Introduction
This section of this research paper is intended to show the analysis performed with the help of
python. In addition to this, this section of this study will also provide a basic explanation to
the model performed and their efficiency level. At the end of this chapter, the researcher will
provide information which can be investigated further in some future research works.
4.2 How to predict and detect fraud with Machine Learning
4.2.1 Which types of transactions are fraudulent?
It tends to be seen that of the five kinds of exchanges, misrepresentation happens just in two
of 'Move' where cash is sent to a client/fraudster and 'CASH_OUT' where cash is sent to a
trader who pays the client/fraudster in real money. Astoundingly, the quantity of false
TRANSFERs nearly rises to the quantity of fake CASH_OUTs. These perceptions show up,
from the outset, to hold up under out the portrayal gave on Kaggle to the business as usual of
fake exchanges in this dataset, to be specific, extortion is submitted by first moving out assets
to another record which accordingly gets the money for it out.
Figure 2: Transaction type to identify fraud
4.2.2 What determines whether the feature isFlaggedFraud gets set or not?
Things being what they are, the starting point of isFlaggedFraud is indistinct, standing out
from the portrayal gave. The 16 passages (out of 6 million) where the isFlaggedFraud
highlight is set don't appear to correspond with any illustrative variable. The information is
portrayed as isFlaggedFraud being set when an endeavor is made to 'Move' a 'sum' more
prominent than 200,000. Actually, as demonstrated as follows, isFlaggedFraud can remain

not set regardless of this condition being met. Will oldBalanceDest and newBalanceDest
decide isFlaggedFraud being set? The old is indistinguishable from the new parity in the
starting point and goal accounts, for each TRANSFER where isFlaggedFraud is set. This is
probably on the grounds that the exchange is stopped. Curiously, oldBalanceDest = 0 in each
such exchange. Be that as it may, as demonstrated as follows, since isFlaggedFraud can
remain not set in TRANSFERS where oldBalanceDest and newBalanceDest can both be 0,
these conditions don't decide the province of isFlaggedFraud.
Figure 3: Fraudulent feature
4.2.3 Looking at Account Types

Figure 4: Accounting type
From the dataset, it seems that fraud transactions only occur when the transaction type1 is CC
(Customer to Customer). Since the dataset is sample of the population, I would have
resampled the data to see if this phenomenon held. Since I do not have access to the
population, I will assume that transaction only occur when transaction type1 is CC. This also
means that the datasets fraud and valid don't need to be subset. However, since all relevant
observations have type1 = "CC", the type1 column is no longer necessary.
4.2.4 Looking at Transaction Types
Figure 5: Transaction Type
From the dataset, it seems that fraud transactions only occur when the transaction type is
CASH_OUT or TRANSFER. Since the dataset is sample of the population, I would've have
resampled the data to see if this phenomenon held. Since I do not have access to the
population, I will assume that transaction only occur when transaction type is either
CASH_OUT or TRANSFER.
4.2.5 Looking balances before and after the transaction
From the summary statistics on the errorBalanceOrg, it seems that a large proportion of the
data have an error of 0 or close to zero. This is indicated by the fact that the most negative
error is -7.450581e-09 or −7.450581x10−9 which is very small and close to 0, and the 3rd

quartile is 0 (that is, about 75% of the data is between -7.450581e-09 and 0). However, there
are some large errors, the largest error being 10,000,000. On the other hand, for valid
transactions, a large proportion of the data have large errors. For instance, about 75% of the
data haver errors exceeding 52,613.43 (the first quartile). The largest error is 92,445,520.
4.2.6 Artificial Neural Networks
In the context of fraud detection the performance of the Neural Network isn't terrible, but it
isn't great either. The loss is performance is very likely due to the phenomenon that Neural
Networks perform worse when the data is imbalanced. When data is imbalanced, Neural
Networks and many other models trained on the data tend to be very biased towards the
*majority class*. In our case, the majority class are valid transactions. This model will be the
benchmark that I will compare other individual models against. The next few models will be
generated from methods that are well-known for handling imbalanced data effectively.

Figure 6: ANN network
4.2.7 Model 2: Random Forest.
A random forest is an algorithm that generates several decisions trees and pools the results of
each tree to make a more robust prediction. Another great thing about Random Forest is that I
can assign weights to each class to reduce the bias of the model towards the majority class, in
this case valid transaction. As expected, the Random Forest performs much better than the
Neural Networks. Instead of crowning this model as the best model, let's try another model
known for performing well in imbalanced datasets.

Figure 7: Random Forest
Clearly, the random forest and the extreme gradient boosted trees performed better than the
neural networks, but which one is better? Instead of comparing all of the metrics from a
purely statistical or mathematical point of view, let's look at them at give some practical
interpretation. First off, let's compare their confusion matrices. Additionally, instead of
comparing the number of correct predictions, let us compare the number of wrong
predictions.
4.3 Discussion
The detection of frauds is considered to be one of the top goals of financial sector and the
same can be addressed with the help of machine learning. As per a report which was
published in 2017 by Nelson, the losses which arose from card-based transactions reached
22.8 billion dollars which is significantly high. This is a problem which is forecasted to get
worse by 2021 and the amount of losses is anticipated to reach a level of 32.96 billion dollars.
Over the years, fraudsters have become more aggressive and moved forward from just taking
pin number from credit or debit cards and using the same in brick-and-mortar stores. The
merchants of leading cards like Visa and Mastercard had taken steps like mandating the
application of EMV - chip card technology. This allows the merchant to have security of pin
for every transaction undertaken by the customer. However, the frauds which are taking place
in online transactions are rising and are predicted to be as high as $32 billion in 2020. In
addition to this, banks need to move forward from chip technology and introduce more
innovative approaches for detection and prevention of frauds.
Anomaly detection-based fraud detection and prevention solutions are the most common
approaches which is available to a bank and the same is regarded to be better than manual
solutions. This type of process requires a much more common machine learning model that is

available on a continuous stream of data which is incoming. The model is trained to have a
baseline sense of normalcy for the transactions which are of banking origins, application of
loans, or information for opening a new account. The software can then notify the supervisor
regarding any irregularities in the data and the same can be verified for the purpose of
detection of frauds. The monitor can accept or reject this alert, which signals to the machine
learning model that its determination of fraud from a transaction, application, or customer
information is correct or not.
The above procedures would be training the machine learning model to “understand” that the
deviation which have been identified was either fraud or a new acceptable deviation. This
kind of baseline could also be established from interactions which is conducted with different
banks and also by assessing the nature of transactions which are undertaken by such banks. In
addition to clients who own banks accounts, fraud can also be caused by merchants or
issuers, and their transaction information can be used to train a machine learning model to
recognize transactions processing properly. This would usually involve pricing but could also
involve the omission of unpaid merchandise.
The application of Artificial Intelligence can be used for lot of positive purposes and the
same is used by large banks for the detection of frauds and also the same assists the banking
officials in maintaining investments from clients. The Artificial Intelligence is used by banks
for detecting frauds and also for scoring transactions considering the risks which are
associated with the transactions and data available regarding the customers. The banks also
have the option of using machine learning-based fraud detection solutions so that the data are
analysed in an effective manner. The model which is applied can be used for detecting or
identifying frauds from one or more transactions. This type of solution can be implemented
by a data scientist in an effective manner and can be regarded as one of the most effective
solution available to the institutions.
Data science dashboards are created for datasets which re using machine learning model once
the same has been implemented. This process allows the banking officials and data scientists
to develop an understanding of the conclusion and how the same has been reached and if
there exist ant correlations.
Machine learning models which are used for detection of frauds can be applied for
development of predictive and prescriptive analytics software. Predictive analytics can
provide a distinct process for detection of frauds by processing the data with a pre-trained

algorithm to score a transaction on its fraud riskiness. Prescriptive analytics considers the
assumptions which are made from the correlations of a predictive analytics engine and uses
the same to give recommendations for what is to be done if fraud is detected. It is to be noted
that both Predictive and Prescriptive analytics requires the data set and same level of training
for implementation. The data experts who are operating in banks needs to consider all the
data available and classify the same as either legitimate or fraudulent and the run the same
through machine learning model. This allows the machine learning model to identify fraud
data sets and make necessary learning for the same. For example, a transaction which is of
fraudulent nature may be for a product that the owner of the account has never purchased or
would likely purchase. Additionally, the geographical location of the person who undertakes
the transaction may not line up with where the owner of account was at the time of purchase.
The software can effectively detect such kind of transactions as the same becomes more and
more trained so it will be more sensitive to those data points and in future would be able to
flag it if the same is located. The banks also have past records which has data labelled as
frauds or suspicious and the same is considered to be important as experts requires such data
when training the machine learning regarding such frauds or suspicious data. The software
can distinguish between the data set after being trained and therefore it is quite useful for the
decision-making process. Banks can apply predictive analytics-based fraud detection
software to detect fraud across multiple channels involved in payment processing. The
software can be used for mobile and online based payment system as well and thereby
making the entire process much more efficient.
In context of recurring financial services payments, the process which is followed would be
remaining the same for detection of frauds. However, the data analysed would involve
changes to payment information and when they were done processing. These types of
payments are charged automatically and in case of any problem the holder of account is
notified. A criminal might attempt to make changes to their billing information by making it
someone else’s, which a predictive analytics application will easily recognise. E-Commerce
payments often are processed through a third-party payment processing system which the
merchant has an tie-up with. For example, crowdfunding website Patreon uses Stripe for
processing payments, which a bank using predictive analytics software could recognize in a
simple manner. The software could make comparison between the processing data for a given
transaction with an established baseline for how Stripe is supposed to process payments. This
allows the software to identify when a transaction uses a third-party service so as to decrease

false positives. This is an effective means for covering such transactions but there is still the
threat for transactions which are of online nature and the same still possess a threat to the
banks.
This challenge is due to the various applications which are available on a smart phone.
Predictive Analytics software could also detect anomalous user by analysing the application
usage, such as accout login from a phone that the account owner has never even seen. This
can be discovered with smartphone geolocational data and also from any personal data which
is stored in the phone. This may enable the software to detect a fraudulent login and prevent
fraud transactions.

Chapter 5: Conclusion and Recommendations
5.1 Conclusion
The enhancement of technology would also lead to application of sophisticated techniques by
criminals for committing frauds. In such cases, banks and financial institutions should
consider the power of AI to protect their businesses and make improvements in the overall
customer experience. These conceivable outcomes are progressive. Artificial intelligence has
the ability to change the arrangement of bookkeeping and money related administrations,
however it must be embraced where it's required most. The business ought to be available to
new potential outcomes and spotlight on the business issues it enables its clients to tackle so
as to survey where AI will include generally esteem. In the interim, pioneers in these callings
must concentrate on the regions where human judgment, relationship building, versatility and
understanding are imperative. This will assist them with seeing where they ought to be
bolstered, and which humble undertakings AI can remove their hands.
By concentrating on these regions these callings can move towards a promising future where
people are allowed to concentrate on the most significant assignments, helped and
empowered by keen machines. Today, mechanization is one type of AI that is profiting an
expanding number of records payable offices.
Applying machine learning to fraud detection enables financial firms to identify genuine
transactions in comparison to fraudulent transactions in real time, and with efficiency.
Through a combination of supervised and unsupervised methods, models are capable of
learning and recognizing new patterns that criminals might be using for committing frauds.
Given the huge sums of money at stake, and the relentless threat posed by criminals, the
application of better technology is the key to make improvements. While organizations may
not be in a position to move to advanced data analytics immediately, but steps needs to be
taken in this respect by examining the current data set, identifying data requirements, and
developing the expertise necessary to begin as soon as possible. The detection of frauds is
considered to be one of the top goals of banking sector and the same can be addressed with
the help of machine learning. As per a report which was published in 2017 by Nelson, the
losses which arose from card-based transactions reached 22.8 billion dollars which is
significantly high. This is a problem which is forecasted to get worse by 2021 and the amount
of losses is anticipated to reach a level of 32.96 billion dollars. Over the years, fraudsters
have become more aggressive and moved forward from just taking pin number from credit or

debit cards and using the same in brick-and-mortar stores. The merchants of leading cards
like Visa and Mastercard had taken steps like mandating the application of EMV - chip card
technology. This allows the merchant to have security of pin for every transaction undertaken
by the customer. However, the frauds which are taking place in online transactions are rising
and are predicted to be as high as $32 billion in 2020. In addition to this, banks need to move
forward from chip technology and introduce more innovative approaches for detection and
prevention of frauds.
5.2 Recommendations
There is no doubt that AI is the main edge of man-made brainpower (AI). It's a subset of AI
where machines can learn by utilizing calculations to decipher information from our general
surroundings to foresee results and gain from victories and disappointments. As machines
penetrate bookkeeping undertakings to assume control over the more everyday and tedious
errands, it will free up bookkeepers and clerks to invest more energy utilizing their expert
learning to examine and decipher the information to give suggestions to their customers.
5.2.1 AI will drive advancement in bookkeeping
When bookkeeping programming organizations wiped out work area support for cloud-based
administrations, bookkeeping firms had to adjust to life in the cloud. Correspondingly,
bookkeeping offices and firms will be compelled to embrace AI to stay aggressive since
machines can convey continuous bits of knowledge, upgrade basic leadership and sling
effectiveness.
5.2.2 Bookkeeping assignments that machines can figure out how to do
Instead of wipe out the human workforce in bookkeeping firms, the people will have new
partners—machines—who will match with them to give increasingly productive and
powerful administrations to customers. Presently, there is no machine trade for the passionate
insight prerequisites of bookkeeping work, however machines can figure out how to perform
excess, repeatable and in many cases amazingly tedious assignments.
5.2.3 Flaws of human mistake
While human instinct is its own psychological marvel, with especially propelled capacities in
versatility and adaptability, it has its cutoff points. The human cerebrum is always impeded
by its own irregularities and inclinations, with things like accessibility predisposition and
affirmation predisposition demonstrating to be expensive in ventures and basic leadership of
different sorts.

The approach of AI and AI is really ready to help human basic leadership, instead of supplant
it totally, so toss out the dread of apathetic robots assuming control over the world (at any
rate for the time being). AI is here to computerize and advise the tedious and excess
undertakings bookkeepers do on an everyday premise, which in this manner saves their
opportunity to concentrate more on rewarding and top to bottom examination.
5.3 Future direction
5.3.1 The open door in AI
AI is the utilization of factual models and calculations that really reflect subjective qualities
like example acknowledgment and logical, explicit learning. The most dominant capacities of
AI exist in its capacity to process enormous datasets, its flexibility in gaining from complex
and always evolving examples, and its relentless consistency. The way that computerized
reasoning never gets drained, joined with its total absence of inclination and littler edge for
blunder, makes it an innovation that is boundlessly versatile in numerous enterprises.
In bookkeeping explicitly, ML's help of basic leadership transforms into offering
bookkeeping experts faultlessly referential information driven bits of knowledge and a blend
of monetary and non-money related investigation. Simulated intelligence additionally outfits
bookkeepers with the instruments to unravel present and contemporary issues they face, for
example, the conveyance of solid, cleaner and less expensive information, notwithstanding
the previously mentioned capacity to give bookkeeping experts a chance to apportion their
opportunity to critical thinking, prompting, procedure creating, and driving, rather than the
unremarkable undertakings of securing and arranging information.
5.3.2 The cutoff points of AI
Since we've built up why ML is so competent and incredible in the best possible conditions,
it's a great opportunity to uncover the catch: ML calculations and procedures are just as
natural as the information they're utilizing. In the event that the datasets being contribution to
these models are fragmented, deficient or are loaded with their own predispositions, the bits
of knowledge that ML models will spit pull out will have similar issues. In case you're
utilizing ML and other AI procedures to gather results that need high degrees of certainty,
this is an issue.
Likewise, only one out of every odd errand is proper for AI at this time. While the potential is
perpetual, the substances of the stage at present enable it to just execute undertakings with a
level of repeatability. This enables the stage to perceive designs, sum up its learnings and

apply them appropriately. The yields of ML calculations are prescient and intriguing in
nature, which means not all assignments can be dealt with in this structure.
5.3.3 Critical thinking, as once huge mob
The open door inserted in the current advanced change of money and bookkeeping is
enabling individuals and AI to work as one and depend on one another to sustain and
contribute in the zones where their qualities exist. The mind-desensitizing, dreary details of
calculating in bookkeeping is dealt with, and all that is left is for bookkeeping experts to
handle the errands their psychological motors are designed for.
AI has demonstrated that arranging and parsing information is never again an assignment for
the human mind. Machines have at last demonstrated they can improve, and on the off chance
that we let them detect the examples and furnish us with experiences on the most proficient
method to use them successfully, we'll be better for it.

Reference
Al-Htaybat, K., von Alberti-Alhtaybat, L. and Alhatabat, Z., 2018. Educating digital natives
for the future: accounting educators’ evaluation of the accounting curriculum. Accounting
Education, 27(4), pp.333-357.
Anderson, B. and McGrew, D., 2017, August. Machine learning for encrypted malware
traffic classification: accounting for noisy labels and non-stationarity. In Proceedings of the
23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
(pp. 1723-1732). ACM.
Appelbaum, D., Kogan, A., Vasarhelyi, M. and Yan, Z., 2017. Impact of business analytics
and enterprise systems on managerial accounting. International Journal of Accounting
Information Systems, 25, pp.29-44.
Ayoubi, S., Limam, N., Salahuddin, M.A., Shahriar, N., Boutaba, R., Estrada-Solano, F. and
Caicedo, O.M., 2018. Machine learning for cognitive network management. IEEE
Communications Magazine, 56(1), pp.158-165.
Cao, J., Xu, H. and Li, S., 2019. Exploration of Training Measures for Applied
Undergraduate Finance and Accounting Professionals.
Dai, J. and Vasarhelyi, M.A., 2017. Toward blockchain-based accounting and assurance.
Journal of Information Systems, 31(3), pp.5-21.
Dirican, C., 2015. The impacts of robotics, artificial intelligence on business and economics.
Procedia-Social and Behavioral Sciences, 195, pp.564-573.
du Chenne, S., 2019. Industry 4.0–fight or flight for accounting professionals?. Professional
Accountant, 2019(35), pp.6-7.
Dunis, C.L., Middleton, P.W., Karathanasopolous, A. and Theofilatos, K. eds., 2016.
Artificial Intelligence in Financial Markets: Cutting Edge Applications for Risk
Management, Portfolio Optimization and Economics. Springer.
Fisher, I.E., Garnsey, M.R. and Hughes, M.E., 2016. Natural language processing in
accounting, auditing and finance: A synthesis of the literature with a roadmap for future
research. Intelligent Systems in Accounting, Finance and Management, 23(3), pp.157-214.

Hays, O., 2019. Future Professionals, in Their Own Words: A Virtual Roundtable of
Accounting Students. The CPA Journal, 89(9), pp.26-35.
Hoffman, C., 2019. Accounting and Auditing in the Digital Age.
Issa, H., Sun, T. and Vasarhelyi, M.A., 2016. Research ideas for artificial intelligence in
auditing: The formalization of audit and workforce supplementation. Journal of Emerging
Technologies in Accounting, 13(2), pp.1-20.
Jiang, F., Jiang, Y., Zhi, H., Dong, Y., Li, H., Ma, S., Wang, Y., Dong, Q., Shen, H. and
Wang, Y., 2017. Artificial intelligence in healthcare: past, present and future. Stroke and
vascular neurology, 2(4), pp.230-243.
Kochenderfer, M.J., 2015. Decision making under uncertainty: theory and application. MIT
press.
Kokina, J. and Davenport, T.H., 2017. The emergence of artificial intelligence: How
automation is changing auditing. Journal of Emerging Technologies in Accounting, 14(1),
pp.115-122.
Li, K., Niskanen, J., Kolehmainen, M. and Niskanen, M., 2016. Financial innovation: Credit
default hybrid model for SME lending. Expert Systems with Applications, 61, pp.343-355.
Lin, T.C., 2016. Compliance, technology, and modern finance. Brook. J. Corp. Fin. & Com.
L., 11, p.159.
O'Leary, D.E., 2015. Twitter mining for discovery, prediction and causality: Applications and
methodologies. Intelligent Systems in Accounting, Finance and Management, 22(3), pp.227-
247.
Pannu, A., 2015. Artificial intelligence and its application in different areas. Artificial
Intelligence, 4(10), pp.79-84.
Popenici, S.A. and Kerr, S., 2017. Exploring the impact of artificial intelligence on teaching
and learning in higher education. Research and Practice in Technology Enhanced Learning,
12(1), p.22.
Sterne, J., 2017. Artificial intelligence for marketing: practical applications. John Wiley &
Sons.

Sun, T., 2017. Accounting Information Systems outputs: XBRL, AI and in-memory
technologies. In The Routledge Companion to Accounting Information Systems (pp. 108-
119). Routledge.
Sutton, S.G., Holt, M. and Arnold, V., 2016. “The reports of my death are greatly
exaggerated”—Artificial intelligence research in accounting. International Journal of
Accounting Information Systems, 22, pp.60-73.
Tirgul, C.S. and Naik, M.R., 2016. Artificial intelligence and robotics. International Journal
of Advanced Research in Computer Engineering and Technology (IJARCET), 5(6), pp.1787-
1793.
Torresen, J., 2018. A review of future and ethical perspectives of robotics and AI. Frontiers
in Robotics and AI, 4, p.75.
West, J. and Bhattacharya, M., 2016. Intelligent financial fraud detection: a comprehensive
review. Computers & security, 57, pp.47-66.
Woods, M. and Linsley, P., 2017. Future research in accounting and risk. The Routledge
Companion to Accounting and Risk, p.296.

Appendix
Code:
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline
import matplotlib.cm as cm
from random import seed,sample
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.neural_network import MLPClassifier
from sklearn.metrics import classification_report,confusion_matrix,accuracy_score,
roc_curve, auc,\
precision_score
from sklearn.ensemble import RandomForestClassifier
from xgboost.sklearn import XGBClassifier
df = pd.read_csv('fraud_detection.csv')
df = df.rename(columns={'oldbalanceOrg':'oldBalanceOrig',
'newbalanceOrig':'newBalanceOrig', \
'oldbalanceDest':'oldBalanceDest', 'newbalanceDest':'newBalanceDest'})
print(df.head())
print('\n The types of fraudulent transactions are {}'.format(\
list(df.loc[df.isFraud == 1].type.drop_duplicates().values)))
dfFraudTransfer = df.loc[(df.isFraud == 1) & (df.type == 'TRANSFER')]

dfFraudCashout = df.loc[(df.isFraud == 1) & (df.type == 'CASH_OUT')]
print ('\n The number of fraudulent TRANSFERs = {}'.\
format(len(dfFraudTransfer)))
print ('\n The number of fraudulent CASH_OUTs = {}'.\
format(len(dfFraudCashout)))
print('\nThe type of transactions in which isFlaggedFraud is set: \
{}'.format(list(df.loc[df.isFlaggedFraud == 1].type.drop_duplicates())))
dfTransfer = df.loc[df.type == 'TRANSFER']
dfFlagged = df.loc[df.isFlaggedFraud == 1]
dfNotFlagged = df.loc[df.isFlaggedFraud == 0]
print('\nMin amount transacted when isFlaggedFraud is set= {}'\
.format(dfFlagged.amount.min()))
print('\nMax amount transacted in a TRANSFER where isFlaggedFraud is not set=\
{}'.format(dfTransfer.loc[dfTransfer.isFlaggedFraud == 0].amount.max()))