INTERNAL AUDIT ANALYTICS AND AI

INTRODUCTION

Artificial Intelligence (AI) is set to be the key driver of transformation, disruption and competitive advantage in today’s fast-changing economy. We have made an attempt in this article to showcase how quickly that change is coming, the steps that Internal Auditors need to take to get going on the AI highway and where our Internal Audits can expect the greatest returns backed by investments in AI.

1.0 Artificial Intelligence Defined

While there are many definitions of Artificial Intelligence / Machine Intelligence, the easiest to comprehend is about creating machines to do the things that people are traditionally better at doing. It is the automation of activity associated with human thinking:

(A) Decision-Making,

(B) Problem-Solving, and

(C) Learning.

A more formal definition would be, ‘AI is the branch of computer science concerned with the automation of intelligent behaviour. Intelligence is the computational ability to achieve goals in the world’.

1.1 Common AI Terms and Concepts

Machine Learning (ML): This is a subset of AI. Machine Learning algorithms build a mathematical model based on sample data, known as ‘training data’, in order to make predictions or decisions without being explicitly programmed to perform the task.

* Unsupervised ML – Can process information without human feedback nor prior data exposure;

* Supervised ML – Uses experience with other datasets and human evaluations to refine learning.

Natural Language Processing (NLP): A sub-field of linguistics, computer science, information engineering and artificial intelligence concerned with the interactions between computers and human (natural) languages, in particular how to programme computers to process and analyse large amounts of natural language data. NLP uses ML to ‘learn’ languages from studying large amounts of written text. Its abilities include:

(i)   Semantics – What is the meaning of words in context?

(ii) Machine Translation – Translate from one language to another;

(iii)  Name entity recognition – Map words to proper names, people, places, etc.;

(iv)  Natural Language Generation – Create readable human language from computer databases;

(v)  Natural Language Understanding – Convert text into correct meaning based on past experience;

(vi)  Question Answering – Given a human-language question, determine its answer;

(vii) Sentiment Analysis – Determine the degree of positivity, neutrality or negativity in a written sentence;

(viii) Automatic Summarisation – Produce a concise human-readable summary of a large chunk of text.

Neural Network (or Artificial Neural Network): This is a circuit of neurons with states between -1 and 1, representing past learning from desirable and undesirable paths, with some similarities to human biological brains.

Deep Learning: Part of the broader family of ML methods based on artificial neural networks with representation learning; learning can be supervised, semi-supervised or unsupervised. Deep Learning has been successfully applied to many industries:

(a) Speech recognition,

(b) Image recognition and restoration,

(c) Natural Language Processing,

(d) Drug discovery and medical image analysis,

(e) Marketing / Customer relationship management.

Leading Deep Learning frameworks are PyTorch (Facebook), TensorFlow (Google), Apache MXNet, Chainer, Microsoft Cognitive Toolkit, Gluon, Horovod and Keras.

1.2 AI Concepts – Context Level (Figure 1)

1.3 Artificial Intelligence Challenges

Many challenges remain for AI which need to be managed effectively:

(1) What if we do not have good training data?

(2) The world is biased, so our data is also biased;

(3)   OK with deep, narrow applications, but not with wide ones;

(4) The physical world remains a challenge for computers;

(5) Dealing with unpredictable human behaviour in the wild.

2.0 Global Developments

There has always been excitement surrounding AI. A combination of faster computers and smarter techniques has made AI the must-have technology of any business.

At a global scale, the main business drivers for AI are:

(i)   Higher productivity, faster work,

(ii) More consistent, higher quality work,

(iii) Seeing what humans cannot see,

(iv) Predicting what humans cannot predict,

(v) Labour augmentation.

2.1 Global Progress on AI – A few examples

2.2 The Internal Audit Perspective

Robotic Process Automation (RPA) is a key business driver for AI in audit in the sense that it has the potential to achieve significant cost savings on deployment. The goal of RPA is to use computer software to automate knowledge workers’ tasks that are repetitive and time-consuming.

The key features of RPA are:

*    Use of existing systems,

*    Automation of automation,

*    Can mimic human behaviour,

*    Non-invasive.

The tasks which are apt for RPA are tasks which are definable, standardised, rule-based, repetitive and ones involving machine-readable inputs.

Sample listing of tasks for RPA:

2.3 Case Study of Application of RPA for Accounts Payable process (Figure 2; See following page)

2.4 AI Audit Framework for Data-Driven Audits
(Figure 3; See following page)

3.0 Internal Audit AI in Practice – Case Study

RPA Case Study from India:

A leading automobile manufacturer had the following environment and challenges:

(a) Millions of vendor invoices received as PDF files;

(b) Requirement for invoice automation, repository build, duplicate pre-check;

(c) Manual efforts were fraught with errors;

(d) PDF to structured data conversion was inconsistent;

(e) Conclusion: A Generic RPA tool was needed.

The solution proposed entailed:

(1) Both audit analytics and RPA being positioned as one solution,

(2) Live feed to the PDF files from diverse vendors,

(3) Extract Transform Load jobs were scheduled for the PDF files,

(4) Duplicate pre-check metrics were built and scheduled,

(5) Potential exceptions were managed through a convenient and collaborative secure email notification management system plus dashboards,

(6) Benefit – 85% reduction in effort and 10x improvement in turnaround time.

4.0 How you can get started on using AI in your Internal Audits

You can get started on your AI journey in Internal Audit by bringing your analytics directly into the engagement. With AI in Audit the efficiency, quality and value of decision-making gets significantly enhanced by analysing all data pan enterprise as one.

Some of the steps you can take to get along in your Audit AI journey are listed below:

(I)     Integrate your audit process / lifecycle;

(II)   Collaborate with clients on a single platform;

(III) Make every audit a data-driven audit;

(IV) Use data analytics through all phases of projects;

(V)   Use RPA where manual work is an obstacle;

(VI) Use Audit Apps where process is well-defined;

(VII)     Augment audits with statistical models and Machine Learning;

(VIII) Evolve to continuous monitoring and Deep Learning.

(Adapted from a lecture / presentation by Mr. Jeffery Sorensen, Industry Strategist, CaseWareIDEA Analytics, Canada, with his permission)