Download as pdf or txt
Download as pdf or txt
You are on page 1of 31

Page |1

“Investigation on Deep Learning in Auditing


Theory, Practice & Scholars’ Assertion”
Course: Advanced Audit and Assurance (ACCT 3203)
By
Group ‘E’
Sl. No. Name ID
1 Arafat Molla B170201094
2 Shamim Ahmed B170201098
3 Md Ridoy Hossain B170201109
4 Md. Masudur Rahman B170201123
5 KM Masum Billah B170201132
6 Farjana Munni B170201143
7 Ummay Habiba Urme B170201159
8 Md. Asikur Rahman B170201161

Submitted to
Dr.Taslima Akther
Associate Professor
Department of Accounting & Information Systems
Jagannath University, Dhaka-1100

Submission Date: 31, December 2020


Page |2

Contents
Abstract ................................................................................................. 3
Deep Learning ..................................................................................... 3
Personal View on Deep Learning …………………………………………………….4
Flowchart……………………………………………………………………………………….…..7
3) Automation through Audit Sub-Processes/Audit as a Production
Line(Croce, Rossini, and Basili 2020) ............................................... 12
Example .......................................................................................... 16
Input Data ....................................................................................... 16
Output Features .............................................................................. 16
Sentiment, emotion, entry, topic, concept, keyword, auditors. ... 16
Applicable Audit Procedures ........................................................... 16
Inspection, analytical procedures, confirmation. .......................... 16
Speech recognition: ........................................................................ 16
Input Data ....................................................................................... 17
Output Features .............................................................................. 17
Applicable Audit Procedures ........................................................... 17
Input Data ....................................................................................... 18
Output Features .............................................................................. 18
Applicable Audit Procedures ........................................................... 18
Judgment support: .......................................................................... 18
Implementation............................................................................... 19
Factors to be considered to implement Deep learning in
Auditing (Alvin A Areans, 2014) .................................................. 21
Summary of Deep Learning Related Some Significant Article & Journal:
............................................................................................................. 27
References ............................................... Error! Bookmark not defined.
Page |3

Abstract
Deep learning is a new addition in the sector of Artificial Intelligence. It serves a great service
in analyzing and depicting data of various ground especially Auditing all over the world. In
this paper, we show the theoretical analysis, procedures, related technology on deep
learning. We also address how Audit Sector become benefited using deep learning technique.
Some factors and challenges also showed here to mark the difficulties of it. We also make a
survey on the benefactors to make importance of using deep learning. An online investigation
is provided to show the scholars opinion to draw attention on emphasis. More information
may add in this paper to enrich it.

Deep Learning
Deep learning is a class of machine learning algorithms that uses multiple layers to
progressively extract higher-level features from the raw input. For example, in image
processing, lower layers may identify edges, while higher layers may identify the concepts
relevant to a human such as digits or letters or faces.(Alom et al. 2019)

Deep learning is an artificial intelligence (AI) function that imitates the workings of the human
brain in processing data and creating patterns for use in decision making. Deep learning is a
subset of machine learning in artificial intelligence that has networks capable of learning
unsupervised from data that is unstructured or unlabeled. Also known as deep neural learning
or deep neural network. (Marshall Hargrave, Nov 24,2020)

Deep learning (also known as deep structured learning) is part of a broader family
of machine learning methods based on artificial neural networks with representation
learning. Learning can be supervised, semi-supervised or unsupervised.(Bengio, Courville,
and Vincent 2013)
Recently, due to the accelerated improvement in data storage and the computational capability
of a modern computer (e.g., cloud computing), a DNN trained with a large volume of data can
represent more and more complex functions. Compared to a traditional neural network, a DNN
has more consecutive hidden layers and more neurons within each layer. This structure allows
the neural network to identify high-level and abstract data features from
Page |4

Figure 1.1 A Simplified Deep Neural Network

(Adopted from Nielsen, 2015)

Human brains contain about 100 billion interconnected neurons. Each neuron receives input
signals (e.g., the visual signal of a cat) from other connected neurons. If the combination of
these signals exceeds a certain threshold (or activation level), the neuron transmits an output
signal to other neurons. Humans’ deep layers of neurons and their complex interconnections
result in a “thinking machine,” the human brain.

Although the idea of artificial neural networks dates back to the 1950s, such networks could
not be called real artificial intelligence until recent advances in computational power and data
storage enabled the development of deep neural networks that model the structure and thinking
process of the brain.

Personal View on Deep Learning


Deep learning is a type of machine learning (ML) and artificial intelligence (AI) that imitates
the way humans gain certain types of knowledge. Deep learning is an important element of
data science, which includes statistics and predictive modeling. It is extremely beneficial to
data scientists who are tasked with collecting, analyzing and interpreting large amounts of
data; deep learning makes this process faster and easier.
Page |5

At its simplest, deep learning can be thought of as a way to automate predictive analytics.
While traditional machine learning algorithms are linear, deep learning algorithms are stacked
in a hierarchy of increasing complexity and abstraction.

To understand deep learning, imagine a fresher student whose first word is Audit report. The
fresher student learns what an Audit report is -- and is not -- by pointing to objects and saying
the word Audit report. The Auditor says, "Yes, that is an Audit report," or, "No, that is not an
Audit report." As the fresher student continues to point to objects, he becomes more aware of
the features that all Audit report possesses. What the fresher student does, without knowing it,
is clarify a complex abstraction -- the concept of fresher student -- by building a hierarchy in
which each level of abstraction is created with knowledge that was gained from the preceding
layer of the hierarchy.

How deep learning does work?


Computer programs that use deep learning go through much the same process as the toddler
learning to identify the dog. Each algorithm in the hierarchy applies a nonlinear transformation
to its input and uses what it learns to create a statistical model as output. Iterations continue
until the output has reached an acceptable level of accuracy. The number of processing layers
through which data must pass is what inspired the label deep. (Searchenterpriseai, 2017)

Deep learning has evolved hand-in-hand with the digital era, which has brought about an
explosion of data in all forms and from every region of the world. This data, known simply as
big data, is drawn from sources like social media, internet search engines, e-
commerce platforms, and online cinemas, among others. This enormous amount of data is
readily accessible and can be shared through fintech applications like cloud computing.

However, the data, which normally is unstructured, is so vast that it could take decades for
humans to comprehend it and extract relevant information. Companies realize the incredible
Page |6

potential that can result from unraveling this wealth of information and are increasingly
adapting to AI systems for automated support. (Hargbrag, November,2020)
Examples of Deep Learning at Work
Deep learning applications are used in industries from automated driving to medical devices.

 Automated Driving: Automotive researchers are using deep learning to automatically


detect objects such as stop signs and traffic lights. In addition, deep learning is used to
detect pedestrians, which helps decrease accidents.
 Aerospace and Defense: Deep learning is used to identify objects from satellites that
locate areas of interest, and identify safe or unsafe zones for troops.
 Medical Research: Cancer researchers are using deep learning to automatically detect
cancer cells. Teams at UCLA built an advanced microscope that yields a high-
dimensional data set used to train a deep learning application to accurately identify
cancer cells.
 Industrial Automation: Deep learning is helping to improve worker safety around
heavy machinery by automatically detecting when people or objects are within an unsafe
distance of machines.
 Electronics: Deep learning is being used in automated hearing and speech translation.
For example, home assistance devices that respond to your voice and know your
preferences are powered by deep learning application.
Page |7

Flowcharts:
Here are presented some flowcharts about Deep learning and deep learning in auditing:

Flowchart-1: Training Deep learning model:


Page |8

Flowchar-2: Payroll Function


Page |9

Flowchart-3: Deep Learning Audit Process

How deep learning is related with auditing?


The Need of Deep Learning for Audit Decision Making (CPA, Oct/Nov2020)
The increasingly developed world necessitates openness to adopt modern data
intensive technologies. Business applications (such as ERP systems), RFID
readers, sensors, cloud storage, social media, remote communication tools (such
as Skype, live streams, pod-casts), and other front-line technologies have been
integrated into the business daily life. It contributes heavily to the production and
maintenance of massive amounts of unstructured or semi-structured big data
(National Research Council, 2013). As a supplement to traditional structured
financial data, unstructured data contains information from various perspectives
and sources, facilitating business to explore the status of their products, services,
and operations. This is because big data provides evidence that is more reliable
and makes auditors less client data-dependent. Consequently, examining and
extracting meaningful patterns from big data offer insights for audit decision
making. However, big data analytics is not easy due to the following reasons:
(1) The vast majority of big data is semi-structured or unstructured, requiring
human experts’ efforts of labeling and classification.
(2) The volume of the data is too large to be processed manually.
(3) Big data is usually generated on a real-time basis, which requires timely
responses.
P a g e | 10

(4) Big data is complex as it has a variety of data types and comes from different
sources.

Thus, to understand the data, one must have a related background of knowledge
or skills. A survey conducted by AICPA (2014) shows that big data analysis is
regarded as one of the top challenges in the future by a quarter of 180 CPA
participants. Therefore, to make big data useful and usable for decision making
auditors, who usually lack professional data mining and information system
knowledge and skills, need an efficient and effective approach to automate audit
procedures (Sun and Vasarhelyi, 2017). With deep learning, they could simply
use a DNN pre-trained and tested by deep learning specialist along with their own
professional accounting judgment and enjoy its benefits.

How deep learning can be used in the Auditing procedures?

Some audit tasks are tedious and complex. The automation of such tasks will
significantly enhance effectiveness and efficiency of audit work. Trained with
sufficiently large samples about how auditors make decisions under different
circumstances (which can be realized by providing different values of data
attributes), a DNN enables auditors to automate many structured or semi-
structured tasks that have been conducted manually for decades, like checking
inventories, processing paperwork, reviewing contracts, and drafting audit
reports. Even for certain risk assessment activities requiring professional
judgment (also called unstructured audit tasks), deep learning provides a new way
to support audit decisions. For instance, items in a financial statement or other
P a g e | 11

financial records can be scanned and automatically linked to related evidence,


such as images of inventory captured by the webcam, shipping documents, sales
invoices, bank confirmations, auditor working papers, and other supporting
documents that have been identified and classified by deep learning systems.
Furthermore, a list of risky items or even recommended responses can be offered.

1. Deep Learning in Language Analysis (Hussien Issa,Ting Sun,MicklosA,


2016/13(2))
Deep learning in conjunction with linguistic analysis is able to automatically
analyze text, including both HTML/text documents and webpages. Unlike the
prevalent “bag-of-words”-based text mining method that requires time-
consuming data preprocessing steps,6 deep learning does not require a human
user to preprocess the text. The returned value of the textual analysis includes the
authors, keywords, concepts, relationships among those concept, involved
concepts and, more importantly, the attitude (sentiment) and other emotions (such
as anger, joy, disgust, sadness, etc.) for targeted phrases, entities, or keywords.

a) The extracted metadata could constitute informative attributes for audit


analytics. This task is conducted by learning from the significantly large
amounts of textual data (for example, AlchemyLanguage API is trained on
billions of webpages) using a complicated computational process and
running it through the deep hierarchical neural network involving multiple
hidden layers and nodes. In fact, the trained text analysis model is capable
of understanding the meaning and even the context of the given text, which
in turn can be used for pattern identification (IBM 2016). With this
technology, the machine will extract attributes from conference call
transcripts, management discussion and analysis sections of 10-Ks, and
other text data, and the extracted attributes are informative factors for audit
analytics (Sun and Vasarhelyi 2016; Sun et al. 2016).

b) Deep Learning in Natural Language Classification

Besides understanding the meaning of the text, deep learning can be applied to
classify it. To do this, the training text must contain labels indicating what
category the text file belongs to. The trained deep learning model interprets the
intent behind text and returns a corresponding classification with associated
confidence levels. This technique is able to function in a number of auditing
scenarios. For instance, we can develop a model to classify earnings conference
call transcripts into two groups: “fraudulent” and “non-fraudulent.” To train the
model, each conference call transcript in the dataset is labeled either as
“fraudulent” or “non-fraudulent.” The trained model will be able to classify the
P a g e | 12

future transcripts into “fraudulent” and “non-fraudulent” categories with varying


confidence levels. The output from the classification process can subsequently be
used to trigger a follow-up audit action. For example, transcripts classified as
“fraudulent” may warrant a more careful examination by auditors who possess
the skills to identify problematic items. Similarly, auditors could rely on the
language classification model to analyze forum posts, comments, and
conversations from social media like Twitter and Facebook to obtain
supplemental audit evidence. To illustrate, tweets can be classified as events,
news, or opinions and the categories can be further classified as criticism or
praise.

2) Deep Learning in Speech Recognition

Speech, in the form of streamed or recorded audio like phone calls, interview
conversations, and management presentations, presents another type of audit
evidence. Deep learning speech recognition techniques bridge the gap between
the spoken word and its written form. It analyzes the grammar and language
structure of the speech and transfers the audio signal to transcript. This is a
difficult task, as speech may contain errors, accents, and environmental noise.
The voice interactivity can transform speech into searchable and analyzable audio
data.

3) Automation through Audit Sub-Processes/Audit as a Production


Line(Croce, Rossini, and Basili 2020)

The inclusion of technology has dramatically changed the nature of audit work,
however, surprisingly little work has been published in the line of production
engineering (Parker and Lewis 1995) to rationalize and systematize work. Before
devices that make decisions can be easily incorporated into the audit production
line, substantive organization and formalization must be performed. Internally,
large CPA firms have progressively developed methodologies that they consider
proprietary in this direction, which they consider to be part of their competitive
advantage.
As obstacles to the wider rationalization and formalization of the assurance
processes, many intrinsic issues must be considered:

 Bill-by-hour: Although the rigid bill-by-hour approach has been


progressively replaced by more comprehensive pricing strategies, the
incentives for automation and reduction of the labor involved are significantly
less than in other industries.
P a g e | 13

 Rigidity of the standards: The standards being derived from processes when
technology was very different, has led to anachronistic manual processes.
Since the Sarbanes-Oxley Act of 2002, they have been aggressively enforced
by tight examination of anachronistic rules.

 Formalization of audit steps: It is possible to conceive that certain parts of the


audit could be standardized and automated through the adoption of audit data
standards (AICPA 2015a, 2015b, 2015c, 2015d), standardized audit
planning, formalization of assertions, adoption of quantitative schemata of
weighting evidence, closer integration with corporate monitoring processes,
etc.

With regard to the last point, it helps to revisit the AI-enabled auditing process
from a production line perspective, analogous to production assembly lines. In
such a process, the output of one phase becomes the input of the subsequent one.
Below is a general audit process, comprised of seven phases (Louwers, Ramsay,
Sinason, and Strawser 2015). AI can help automate this process and transform it
into a high-efficiency and highly effective audit line production process. The
proposed automated phases are presented below:

Planning Procedures

Pre-Planning Phase: This phase involves acquiring initial knowledge of the


client and their industry. AI can collect, aggregate, and examine Big Data from
various exogenous sources. Next, AI will incorporate client's organization
structures and their operational methods, as well as their accounting and financial
systems. AI then estimates the initial risk level associated with that client.

Contracting Phase: Using the output from the previous phase (i.e., initial risk
level), AI estimates the number of hours the engagement would require and
calculates audit fees. Subsequently, AI refers to a database of previously analyzed
contracts, and automatically generates a client-specific engagement letter. Both
auditor and client sign the AI-prepared contract.

Understanding Internal Controls and Identifying Risk Factors: This step is


important for the planning of all aspects of the audit engagement. Using text
mining and image recognition techniques, AI analyzes client-provided
flowcharts, narratives, and filled questionnaires. Rather than physically taking a
tour, or a walkthrough, drones can be used to capture video footage, which can
be analyzed to identify any anomalies. At this stage, AI uses pattern recognition
P a g e | 14

and visualization methods to identify risk factors. Finally, all this information is
aggregated to identify fraud and illegal-act risk factors.

Control Risk Assessment: This phase involves the examination of the client's
internal control system design and implementation. An AI-based continuous
control monitoring system examines the complete population of records to
identify any control violations and reports them. In the case of a high number of
violations, a ranking system can be implemented to prioritize the identified
violations based on their level of riskiness (Issa 2013; Issa and Kogan 2014; Issa,
Kogan, and Brown-Liburd 2016). Moreover, AI runs process mining (Jans et al.
2014) on the complete population to ensure that the internal control system is not
only properly designed, but also configured and implemented correctly. Logs are
generated automatically to ensure the integrity of the data and to prevent the
tampering with such audit evidence. This step can be repeated as frequently as
desired, which allows for a minimal response time if any control violations are
found.

Substantive Tests: The difference between traditional auditing and AI-enabled


auditing is most pronounced in this phase. Data provenance and data quality are
examined as they are collected, eventually in real time. Rather than running a
periodical test of details on a sample of the transactions, AI can examine 100
percent of the population on a continuous basis. The same idea applies to the test
of details of balances. This continuous and comprehensive test of details
decreases the likelihood of an abnormal record passing undetected. Moreover, by
running continuously, the time it takes to identify such an anomaly is significantly
decreased. The concept of exception prioritization that was discussed in the
previous phase can also be applied to the substantive tests to help address large
numbers of exceptions riskiness (Issa 2013; Issa and Kogan 2014; Issa et al.
2016). The incorporation of pattern recognition, visualization, benchmarks, and
outlier detection methods on top of analytical procedures can greatly increase
audit effectiveness.

Evaluation of Evidence: This phase will be included in the previous phase, due
to the importance of ensuring data quality prior to running the substantive tests.

Audit Report: The final step in the audit process is the issuance of a verdict
based on the findings from the previous steps. In traditional audits, the auditor
issues a categorical opinion (clean, qualified, adverse, etc.) to the client company.
P a g e | 15

How deep learning can be used to improve the audit procedures?


The automation of some tedious and repetitive audit processes would
significantly enhance audit effectiveness and efficiency. Given a sufficiently large
sample of how auditors make decisions under various circumstances, a deep
learning system allows auditors to automate many tasks that have traditionally
been conducted manually, such as checking inventories, processing paperwork,
reviewing contracts, and even drafting audit reports. (Ting
Sun;Vasarhelyi;Miklos A, June,2017)
Deep learning can be used to improve the audit procedures in 4 ways:
1. Text understanding
2. Speech recognition
3. Visual recognition
4. Structured data analysis
Text understanding:
A large amount of textual data is generated and disseminated during a company’s
operating processes such as regulatory filings, transcripts of conference calls,
press releases, earnings announcements, MD&A, business contracts, news
articles, and social media messages. Textual data provides information on
multiple aspects of a business from various perspectives. For example MD&A
contains management’s perspective on the company’s current financial situation
and future prospects, analysts’ reports include retrospective analysis of past
events and forecasts of future earnings and cash flows and social media posts can
include advertisements, product reviews, and news announcements.
This textual analysis is better suited to unprepared content (the Q&A section of
conference calls, Face book status updates, blog posts) than prepared content
(press releases, the presentation section of conference calls). Compared to
prepared content, unprepared content may be filled with slang, idioms, and other
linguistic clues reflecting the cognitive process of the speaker, which may provide
clues to potential risks. Exhibit 1 provides examples of textual documents,
possible output features, and audit tasks for which deep learning is applicable.
(Druz, Wagner, & Zeckhauser, 2015)
P a g e | 16

Exhibit 1
Input, Output, and Applicable Audit Procedures for Deep Learning–Based
Textual Analysis
Example
Regulatory filings, press releases,
Input Data earnings announcements, MD & As,
business contracts, comment letters
from SEC, new articles, analysts’
reports, e-mail, disclosures on
company websites, social media
messages.
Output Features Sentiment, emotion, entry, topic,
concept, keyword, auditors.
Applicable Audit Procedures Inspection, analytical procedures,
confirmation.
Speech recognition:

To acquire background information about a client’s business and industry


environment and to collect audit evidence, auditors interview management,
internal auditors, employees, predecessor auditors, bankers, legal counsel,
underwriters, analysts, or other stake-holders. The language that subjects use and
how they respond to questions over the course of the interview can be just as
important as the answers themselves because they may indicate deception.
Although public accounting firms offer deception detection training to help their
auditors identify verbal red flags, the information processing of interviews is a
daunting task because interviewees exhibit myriad verbal behaviors. It is difficult
and inefficient for auditors to analyze all oral responses or transcribe them into
text manually and even transcribed text documents can still be a chore for auditors
to analyze.
The speech recognition function of deep learning can now transcribe and
translate speech on a real-time basis regardless of noise or the various accents of
speakers, enabling analysis of the text and extraction of emotion, risk factors, and
other insights directly. Other audio documents such as conference calls, phone
calls, and project meetings could be automatically processed by deep learning
technology. Exhibit 2 depicts possible input and output data, as well as
applicable areas for deep learning–based speech recognition. (D.Pickard,
Schuetzler, JosephValacich, & Wood, 2017)
P a g e | 17

Exhibit 2
Input, Output, and Applicable Audit Procedures for Deep Learning–
Based Speech Analysis

Example
Input Data Interviews, conference calls, phone
calls, project meetings, presentations.
Output Features Risk of deception, sentiment, emotion,
entity, topic, concept, keyword,
speaker, suggested follow-up actions.
Applicable Audit Procedures Inquiry, inspection.

Visual recognition:
Certain routine, manual audit procedures can now be automated with the
visual recognition function of deep learning. For example, deep learning
algorithms can identify the content (model, quantity, condition of inventory) of
an image from a video filmed by a drone in a company’s warehouse. Deep
learning systems are able to extract a series of predefined numerical attributes
describing the content of the video, attach searchable tags accordingly, and save
both the attributes and the images to the auditor’s data warehouse. Furthermore,
deep learning video analytics are capable of recognizing human faces, detecting
objects, and identifying concepts and types of scenes on a nearly real-time basis,
and the video processing speed is extraordinarily high; Clarified, for instance,
offers deep learning software that can analyze a 3.5-minute video clip within 10
seconds. The internal control observation procedure could eventually be
automated using this technology to analyze videos recorded by drones in offices
or at large worksites.
Exhibit 3 shows the input video/image data, output features, and audit procedures
to which the image and video analysis function of deep learning could apply.

Exhibit 3

Input, Output, and Applicable Audit Procedures for Deep Learning–Based


Image and Video Analysis
P a g e | 18

Example

Input Data Inventory counting and other control activity;


interviews, video taken in office, warehouse or
store.

Output Features Object, human face, concept and type of scene.


Applicable Audit Observation, inquiry, inspection.
Procedures

Structured data analysis:

The data analytical process must be well planned to balance human


resources and technology as well as to ensure that auditors strengthen their
professional skills and judgment by concentrating on high-risk items identified
through deep learning. In addition, regulators should start considering guidance
for the use of data analytics technology. In fact, the Data Analytics Working
Group of the IAASB released a Request for Input, “Exploring the Growing Use
of Technology in the Audit” in September 2016 and is seeking advice on
determining whether new or revised international standards or guidance are
necessary. Furthermore, the PCAOB has set up a research project on analytics in
audits and Steven Harris, a PCAOB board member, issued a passionate plea for
disruptive change in auditing and the need for analytics. (Technology and the
Audit of Today and Tomorrow, Apr. 20,2017)
Deep learning serves two major functions in supporting audit decision making:
1. Judgment support
2. Implementation

Judgment support:
In addition to conducting repetitive and mechanical tasks, deep learning provides
a new way to support audit judgment and improve audit quality. Financial
statements can be scanned and the financial statement items automatically linked
to related supporting evidence such as video clips, press release, news, tweets and
interviews as well as corresponding data attributes ( the overall sentiment and the
topic of a news) that have been extracted by deep learning models. For example,
P a g e | 19

auditors can select data attributes in order to predict fraud risk; the selected
attributes are then combined with traditional financial or non financial data fields
to develop a new deep learning prediction model. Deep learning performs as an
appropriate prediction algorithm in this case because, by introducing the extracted
attributes, the number of predictors is much larger than what a traditional
machine-learning algorithm could process. For each assertion, the output of the
model could be the predicted risk level or suggested follow-up tests, depending
upon the nature and the label of the training data.
Implementation
Auditors can leverage deep learning technologies without being experts by using
cloud-based services. But to improve the predictive performance of deep learning
algorithms for audit automation and judgment support, it is important that
auditors and machine learning specialists make a cooperative effort to develop
auditing-specific training datasets to establish deep learning models that are
designed specifically for audit tasks. It is worth noting that retrieving the auditing-
specific data is easier today than it was several years ago. Various data services
allow the users to search for financial and nonfinancial information in databases
containing millions of records in seconds. (Audit , n.d.)
Summery:
The application of deep learning to audit procedures by illustrating how the
capabilities of deep learning for text understanding, speech recognition, visual
recognition, and structured data analysis fit into the audit environment. Based
on these four capabilities, deep learning serves two major functions in
supporting audit decision making: information identification and judgment
support. The application proposes a framework for applying these two deep
learning functions to a variety of audit procedures in different audit phases.
Finally, the paper discusses the challenges faced by the accounting profession,
regulators, and educators when it comes to applying deep learning. (Sun, 2019)
The application of deep learning to improve audit efficiency and effectiveness is
especially relevant to facilitating repetitive audit procedures and supporting audit
judgments. Deep learning can add value for routine tasks that involve massive
amounts of data and require significant effort for auditors to solve, such as text
analysis, speech recognition, and parsing images and videos. It can also be used
to reduce manual work by automating some substantive procedures, such as
confirmation and examination. Furthermore, these competencies will allow
P a g e | 20

auditors to perform tasks, such as examining all corporate contracts that are
currently cost prohibitive or too complex for unaided human minds. (CPA,
Oct/Nov2020)
Why auditor should use deep learning?
In a word, accuracy. Deep learning achieves recognition accuracy at higher levels
than ever before. This helps consumer electronics meet user expectations, and it
is crucial for safety-critical applications like driverless cars. Recent advances in
deep learning have improved to the point where deep learning out performs
humans in some tasks like classifying objects in images.
There are two main reasons it has only recently become useful in auditing:
1. Deep learning requires large amounts of data. For example, preparing an
audit report requires large number of data, thousands of business
transactions and also need internal access & control.
2. Deep learning requires substantial computing power. High-performance
GPUs have a parallel architecture that is efficient for deep learning. When
combined with clusters or cloud computing, this enables development
teams to reduce training time for a deep learning network from weeks to
hours or less. That’s why an Auditor can effectively & efficiently perform
auditing in an organization, and prepare an accurate audit report for the
organization within short time. (Deep Learning and the Future of Auditing:
How an Evolving Technology Could Transform Analysis and Improve
Judgment., Jun 2017)
There are several reasons why an Auditor uses deep learning for performing
audit.

1. Deep learning performs with large data.


2. Takes long time to train.
3. Provide high accuracy.
4. Hardware depends on GPU (Graphical Processing Unit).
5. Deep learning performs an noteworthy matrix multiplication.
6. Rules based, data-driven process that utilizes neural networks.
7. Relationships are found in data, no hard-coded output.
P a g e | 21

8. Deep learning method can easily find the error.


9. Provide accurate result.
For the above reasons, an Auditor can choose deep learning in auditing process.
What are the factors to be considered to implement deep learning in
Audit Sectors?

Factors to be considered to implement Deep learning in Auditing


(Alvin A Areans, 2014)

The following factors should receive due consideration while planning(Approach


n.d.) Size of the company and nature of its operations.
 Accounting system, internal control and adherence to standard.
 Environment in which the company operates.
 Previous experience with the client
 Knowledge of client’s business.
 Capability of Using Software
 Budget of Auditing of the Company
 Interest of Users
 Size of data
 Manpower

Again following Factors also should follow before amplification of deep learning
in Auditing
1. Complexity of the Audit: The scope of work and reporting responsibilities is
analyses in order to determine the complexity of audit.
2. Environment in which the entity operates: This enables the Auditor to
understand various operational aspects of audit, e.g. extent of computerization,
nature of internal controls, general attitude of personnel, etc.
3. Previous experience with the client: By analyzing the previous year's audit
working papers and other relevant files, the Auditor should pay particular
attention to matters that required special consideration and decide whether
they might affect the work to be done in the current year.
4. Knowledge of the client's business: This is required to establish the overall
audit plan. The Auditor will be able to—
P a g e | 22

(a) Identify areas of special audit consideration,


(b) Evaluate the reasonableness both of accounting estimates and management
representations, and
(c), make judgments regarding the appropriateness of accounting policies and
disclosures.
5. Discussion with Client: The Auditor can discuss his overall plan and certain
audit procedures with the client to improve the efficiency of the audit and to
coordinate audit procedures with work of the client's personnel. The overall
audit plan and the audit program, however, remain the auditor's responsibility.
The above factors can be shown in this way

Chart 1 (Mihret, n.d.)


P a g e | 23

Is any deep learning usages in Audit Sectors Bangladesh?


AUDIT SERVICES MARKET IN BANGLADESH (ICAB, 2019)

Bangladesh has a rapidly growing market for audit services. In addition to


performing the statutory audit function, most audit firms provide a range of
advisory services including tax, valuation, systems design, and due diligence
advisory. The regulatory framework governing audit function in Bangladesh is
robust. There exist a number of legislative provisions requiring companies to
have statutory audits by a chartered accountant or a firm of chartered accountants.
These legislative provisions principally emanate from the Companies Act 1994,
the Securities and Exchange Rules 1987 along with its significant amendments in
2000, the Banking Companies Act 1991, the Insurance Act 1938, the Income Tax
Act 1922, the Income Ordinance 1984, and the Nationalization Orders of 1972
and 1973. Hence, all joint stock companies registered in Bangladesh are required
to have a statutory audit once every year by a chartered accountant. Relevant
provisions of these Acts and Ordinances solely empower the members of one
professional body, the Institute of Chartered Accountants of Bangladesh (ICAB),
to carry out statutory audits. As on 30 June 2005, there were 171 public
accounting firms in Bangladesh. Sole trader ship firms still occupy a significant
share of the audit services market although they mainly serve the small client
segment. Of the 171 firms, 110 were sole trader ships and 61 were partnerships.
Majority of these partnership firms (47) have only two partners firms while the
highest number of partners that can be found in a firm is 7. Most Bangladeshi
audit firms operate from a single office, predominantly located in Dhaka.
However, some firms have multiple offices. The number of audit firms operating
from a single office is 147 while 23 firms have two offices (e.g., in Dhaka and
Chittagong or in Dhaka and London) and only 1 has three offices. A total of 266
members of the Institute are in practice while 494 members are not in practice.
Most of the members who are not in practice generally occupy influential
positions in business, commerce, and government. As at the same date, 127
members are working abroad. At present, none of the international Big Four audit
firms directly operate through a named branch office in Bangladesh although they
do in other South Asian countries, e.g., in India, Pakistan, and Sri Lanka.
However, some Bangladeshi audit firms have official affiliations with
international Big Four and non-Big four firms.

Practice of Deep learning in Bangladesh (ICAB, 2019)

Deep learning could be applied various audit procedures to enable audit


automation & improve decision-making. This document is the perceptions of
professional’s use of deep learning technology-an emerging form of artificial
intelligence (AI) in the accounting and auditing industry of developing country
P a g e | 24

such as Bangladesh. It finds that there is an unexplored and untapped market for
AI in Bangladesh, which can be utilized if proper policy is taken by authorities.
In the current business environment, the development of data-intensive
technologies (e.g., ERP systems, sensors, cloud storage, remote communication
tools) facilitates the production and maintenance of large amounts of data, which
necessitates a new data environment and serves as a motivator for audit
automation. Leading accounting firms have leveraged deep learning, a cutting-
edge use of artificial intelligence, to conduct audit tasks. For example, KPMG
applies IBM Watson’s deep learning–powered systems to analyze banks’ credit
files for commercial mortgage loan portfolios, and Deloitte has allied with Kira
Systems to review contracts, leases, invoices, and tweets. The adoption of deep
learning within the accounting profession is still, admittedly, at an early stage. To
accelerate the wider use of this technology, it is necessary to create economies of
scale by integrating its cognitive capabilities in the areas of textual analysis, voice
recognition, image and video parsing, and judgment support into the audit
process. With breakthrough innovations in the accounting industry, auditing
practices are ever evolving. It is not only enhancing auditing quality but also
eliminating time-consuming and manual labor of professionals. As more
resources and tools are available to accounting professionals, they can save time
by eliminating dreary processes that give them more opportunities in other
business strategies. Keeping up with the revolutionary changes in all the sectors
other than the accounting industry, it is important to explore the prospects of AI
in auditing to facilitate auditing practices with more ease and simplicity.
However, Bangladesh is yet to adopt artificial intelligence and other modern
technological means in auditing practices extensively. Artificial intelligence and
other technological systems are being used at a small scale across the firms in
Bangladesh. Thus, this study aims at exploring the current scenario of the
applications of artificial intelligence in Bangladesh. Due to limited resources and
time, only Dhaka-based audit firms have been taken into account for collecting
primary data. This study explores and concludes the current scenario of the
auditing industry of Bangladesh in terms of using innovative technologies in daily
practices.

Scenario of AI in the Auditing Industry of Bangladesh(Ferdousi 2012)


There are over 170 accounting firms in Bangladesh and each come with their own
working strategy and corporate environment. Since only 2 firms have been taken
into consideration for the purpose of this study, it cannot be concluded that the
results of the study depict the entire scenario of AI in auditing. However, both the
surveyed firms (KPMG and ACNABIN) are MNCs, which gives us a rational
idea that AI is being rigorously implemented on multinational accounting firms
in Bangladesh. Since 61.1% of the respondents were aware of the use of AI in
their respective firms, it can be presumed that it has been 3 years or more since
P a g e | 25

AI had been adopted in the auditing practices of surveyed firms. 92.6% of the
respondents agreed that their firm was using personalized audit software for
auditing. Although it is difficult to ascertain how extensively AI is being
implemented in the auditing industry of Bangladesh, it can be concluded that
auditing practices are computerized.
Perception of Professional Accountants on AI in the context of Auditing
Practices across Firms in Bangladesh(Karim 2012)
AI has made tedious and highly manual auditing tasks a lot simpler and quick.
Grass-root accountants were previously hired primarily for the purpose of
computing framework and data input for auditing. But now since AI can process
the 10 hours’ worth effort of hundreds of accountants in just minutes with a single
click, it does intimidate the accountants. Most of them also agreed or strongly
agreed that AI has made most of the audit work computerized and auditors only
required to review them. So it can be concluded that most of the audit work is
heavily computerized, that poses a threat to grass-root and mid-level auditing
professionals. Not only that, most professionals thought that they were under the
pressure of skill transformation due to the advent of AI. Hence, auditors will be
at risk if they do not familiarize themselves with AI software’s and their
implementation.
Problems in Implementing AI in the Auditing Industry of
Bangladesh(Ferdousi 2012)
Although the use of AI is in practice across multinational accounting firms in
Bangladesh, it still is a long way until AI can be applied extensively in all audit
and accounting firms to completely transform the industry. Our research finds
some drawbacks in the use of AI. For instance, many accounting practitioners
agreed that AI incurs huge cost when it comes building, obtaining and
maintaining systems and also training beginners. Bangladesh having a low
economy still does not possess the financial capability and resources to be able to
fully implement existing technologies in par with other countries. Besides, as
discussed before, it poses a huge threat to young professionals, as their jobs are
more at stake as audit procedures continue to become more and more simplified.
Another problem is that it poses a huge threat to data security when it comes to
digitalized data processing. Bangladesh having weak digital infrastructure
directly impose a threat to information security in the audit industry. However,
over the period these problems will be solved with government and private sector
interventions.
This study aimed to understand about the current perception of AI on the auditing
industry of Bangladesh. Results reveal that 3 factors dominantly impact the use
of AI in Bangladesh. These are comprehensible idea of audit practitioners about
AI, computerization of audit work and skill transformation of auditors. Since
P a g e | 26

artificial intelligence is new to Bangladesh, more research should be conducted


on AI and accounting industry of Bangladesh in order to forecast future uses for
it in the country.
P a g e | 27

Summary of Deep Learning Related Some Significant Article & Journal:


Author & Objective of the Responde Methodology Model used for the Findings Important factors identified Implication/Recomme
Year Title of the study nt study ndation
(From paper
Reference)
(Alom et al. A State-of- Depth review of 174 Artificial Supervised, Individual Individual advanced techniques Developed or update
2019) the-Art deep learning and Citation Intelligence’s Unsupervised, and techniques should for training large-scale deep method should be
Survey on its applications & Application Reinforcement be considered learning models and the used in deep learning
Deep over the past few 880 Specially DNN, Learning (RL), as because of their recently developed method of Technique
Learning years Readers CNN, and RNN, d well as their needs generative models
Theory and Unsupervised applications in
Architecture Learning different domains
s Techniques, were reviewed
(Barredo- Plausible Allowing inexpert 3 citation Adversarial Image Plausible data The GAN framework may serve GAN model
Arrieta and Counterfact users for a much & 18 learning classification, and should use as well to check underlying framework should
Del Ser uals: greater readers framework hybridized an input rather than biasing phenomena present in a use to draw a good
2020) Auditing understanding of oriented towards attribute GAN subtle data so that model and/or a dataset. Indeed, result for betterment
Deep the capabilities and the multi-objective model with a the user of the a subtle bias of the dataset
Learning pitfalls of a model synthesis of multi-objective image could (more blond women) was
Classifiers used in a real adversarial meta-heuristic realistically shown to be crucial for the
with application samples aimed at search engine imprint in the adversarial success of male and
Realistic explaining the input to the mode female counterfactuals,
Adversarial performance limits informing the user about the
Examples of Deep Learning risk of not tackling it
classification
problem
(Sun and Embracing The usefulness of 16 Artificial Natural Language Big Information The lack of efficient and Natural Language
Vasarhelyi textual data the information citation Intelligence Understanding from textual may effective technology solutions Understanding from
2018) analytics in provided by & 103 Application from IBM draw better result hampers the full use of text data IBM Watson , Tensor
auditing various textual data readers Watson , Tensor due to large volume of data Flow should use to
with deep in auditing and Flow analysis big data
learning introduces deep
learning, an
evolving Artificial
Intelligence
approach
P a g e | 28

(Croce et al. Auditing To explain a new 3 Quantitative The proposed Nonlinear While NLP systems become KDA Method should
2020) deep dimension model Citations evaluation of the approach applies learning method more pervasive, their use for better result
learning for better outcome & 27 automatic LRP to a Kernel- is more effective accountability gains value as a
processes readers generation of based Deep than NIP system focal point of effort.
through epistemologically Architecture Epistemological opaqueness of
kernel- transparent and (KDA) that non- linear learning methods,
based linguistically fluid redistributes the such as deep learning models,
explanations for prediction value to can be a major drawback for
explanatory
neural inferences training entries their adoptions
models
(i.e., annotated
landmarks)
(Afroze and Perception Exploring the 2 Five point Likert Questioner Model 3 factors Although the use of AI is in Computerized
Aulad 2020) of perceptions of Citations scale and a dominantly affect practice across multinational especially AI should
Professional accounting &5 questionnaire the use of AI in accounting firms in Bangladesh, be use broadly to
Accountants professionals about readers survey over Audit Bangladesh. it still is a long way until AI can analyses data for
about the the application of Firm within Dhaka These are be applied extensively in all Auditing
Application AI in their city comprehensible audit and accounting firms to
of Artificial profession in idea of audit completely transform the
Bangladesh practitioners industry.
Intelligence
about AI,
(AI) in
computerization
Auditing of audit work and
Industry of skill
Bangladesh transformation of
auditors.
(Ferdousi Challenges The existence of 5 Interviews and Supply and Challenges of Without using modern It is urgent to
2012) of the challenges of Citations audit report Demand theory: Technology based technology, accountability of implicate the deep
Performanc performance audit & 16 analysis Demand audit: insufficient audit will fall and foreign learning and AI to go
e Audit in in the public sector readers Side(auditee and incompetent business will set off. forward.
the of Bangladesh organizations & manpower,
Implementa Supreme Audit unable to recruit
tion Phase : Institution) consultant when
Supply needed, lack of
Bangladesh
Side(capacity awareness and
Perspective
building and absence of pre-set
recruitment of goals among the
consultants and auditees.
P a g e | 29

independent
variables)
(Marcus Deep Ten concerns for 540 Individual Criticism Deep learning has 10 special characteristics are: New Dimension
2018) learning: A deep learning, and Citations Assertion and obtained new 1.Deep learning thus far is data Challenges should
critical suggest that deep 1.9K criticism the dimension and hungry mark up carefully
appraisal learning must be readers development of among 10 2.Deep learning thus far is
supplemented by deep learning since character is shallow and has limited
other techniques if 2012 to 2018 described here. capacity for transfer
we are to reach 3.Deep learning thus far has no
artificial general natural way to deal with
intelligence. hierarchical structure
4. Deep learning thus far has
struggled with open-ended
inference.
5.Deep learning thus far is not
sufficiently transparent
6.Deep learning thus far has not
been well integrated with prior
knowledge
7.Deep learning thus far cannot
inherently distinguish causation
from correlation
8. Deep learning presumes a
largely stable world, in ways
that may be problematic.
9. Deep learning thus far works
well as an approximation, but
its answers often cannot be
fully trusted
10. Deep learning thus far is
difficult to engineer with
Another
References
Alvin A Areans, R. E. (2014). Auditing and Assurance Service(anintegrated approach) 15
edition. Michigan: Pearson.
Audit . (n.d.). Retrieved from seekedgar: http://www.seekedgar.com
CPA. (Oct/Nov2020). Deep Learning and the Future of Auditing. CPA Journal.
D.Pickard, M., Schuetzler, R., JosephValacich, & Wood, D. A. (2017). Next Generation
Accounting Interviewing: A Comparison of Human and Embodied Conversational
Agents as Interviewers. Journal of Information Systems.
Deep Learning and the Future of Auditing: How an Evolving Technology Could Transform
Analysis and Improve Judgment. (Jun 2017). CPA Journal.
Druz, M., Wagner, A. F., & Zeckhauser, R. J. (2015). Tips and Tells from Managers: How
Analysts and the Market Read between the Lines of Conference Calls. National Bureau
of Economic Research.
Hargbrag, M. A. (November,2020). Deep Learning. https://www.investopedia.com/terms/d/deep-
learning.asp.
Hussien Issa,Ting Sun,MicklosA. (2016/13(2)). Research Ideas for Artificial Intelligence in
Auditing: The Formalization of Audit and Workforce Supplementation. American
Accounting Accociation a strategic and imerging technology, 1-20.
ICAB. (2019). Deep Learning Practice in Bangladesg. ICAB, 4-15.
Marshall Hargrave. (Nov 24,2020). Deep Learning. Investopidia Web-Journal.
Mihret, D. (n.d.). Dessalegn Mihret. Retrieved from researchgate:
https://www.researchgate.net/profile/Dessalegn-
Mihret/publication/233390627/figure/fig1/AS:300151492104192@1448573100109/Fact
ors-influencing-internal-audit-effectiveness_Q640.jpg
Sun, T. (2019). Accounting Horizons. Retrieved from meridian.allenpress.com:
https://doi.org/10.2308/acch-52455
Technology and the Audit of Today and Tomorrow. (Apr. 20,2017). PCAOB/AAA
AnualMeeting.
Ting Sun;Vasarhelyi;Miklos A. (June,2017). Deep Learning and the Future of Auditing: How an
Evolving Technology Could Transform Analysis and Improve Judgment. CPA, Vol.87
Issue 6, p 24-29.6p.

Plausibl Afroze, Dilruba, and Anisha Aulad. 2020. “Perception of Professional Accountants
about the Application of Artificial Intelligence (AI) in Auditing Industry of Bangladesh.”
Page |1

Journal of Social Economics Research 7(2):51–61. doi: 10.18488/journal.35.2020.72.51.61.


Alom, Md Zahangir, Tarek M. Taha, Chris Yakopcic, Stefan Westberg, Paheding Sidike, Mst
Shamima Nasrin, Mahmudul Hasan, Brian C. Van Essen, Abdul A. S. Awwal, and Vijayan
K. Asari. 2019. “A State-of-the-Art Survey on Deep Learning Theory and Architectures.”
Electronics (Switzerland) 8(3).
Approach, A. N. Integrated. n.d. AUDITING AND ASSURANCE SERVICES A LV I N A . A R E
N S.
Barredo-Arrieta, Alejandro, and Javier Del Ser. 2020. “e Counterfactuals: Auditing Deep
Learning Classifiers with Realistic Adversarial Examples.” Proceedings of the International
Joint Conference on Neural Networks. doi: 10.1109/IJCNN48605.2020.9206728.
Bengio, Yoshua, Aaron Courville, and Pascal Vincent. 2013. “Representation Learning: A
Review and New Perspectives.” IEEE Transactions on Pattern Analysis and Machine
Intelligence 35(8):1798–1828. doi: 10.1109/TPAMI.2013.50.
Croce, Danilo, Daniele Rossini, and Roberto Basili. 2020. “Auditing Deep Learning Processes
through Kernel-Based Explanatory Models.” EMNLP-IJCNLP 2019 - 2019 Conference on
Empirical Methods in Natural Language Processing and 9th International Joint Conference
on Natural Language Processing, Proceedings of the Conference 4037–46. doi:
10.18653/v1/d19-1415.
Ferdousi, Nusrat. 2012. “Challenges of Performance Audit in the Implementation Phase :
Bangladesh Perspective.” 1–88.
Karim, A. K. M. Waresul. 2012. “Audit Pricing, Audit Concentration, and Big-4 Premium in
Bangladesh.” SSRN Electronic Journal (October 2006):1–43. doi: 10.2139/ssrn.1613454.
Marcus, Gary. 2018. “Deep Learning: A Critical Appraisal.” ArXiv 1–27.
Sun, Ting, and Miklos A. Vasarhelyi. 2018. “Embracing Textual Data Analytics in Auditing with
Deep Learning.” International Journal of Digital Accounting Research 18(December 2017):49–
67. doi: 10.4192/1577-8517-v18_3.

You might also like