45
1 Document Classification: KPMG Confidential © 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity. All rights reserved. Use of Predictive Analytics in business and internal Audit Sudharshana Balasubramanian Director Advisory Services 16 th April 2019

Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

1

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Use of Predictive Analytics in business and internal Audit

Sudharshana Balasubramanian

Director – Advisory Services

16th April 2019

Page 2: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

2

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Synopsis of the presentation

Game Changers and KPMG’s Global D&A Survey results

Case studies using Predictive Analytics with client benefits

KPMG’s Global Data and Analytics Platform

Predictive Analytics and Machine Learning1

2

3

4

Page 3: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

3

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Top 10 breakthrough technologies – MIT Technology Review Predicting Preemies – every year over 15 million babies are born

preterm

Custom Cancer Vaccinations– Conventional Chemotherapies

replaced with Predictive custom medicines

Robot Dexterity– Robots are teaching themselves to handle the

physical world using predictive data analytics

ECG on your wrist–to continually monitor heart condition with

wearable devices using supervised machine learning

Smooth-talking AI assistants: will be able to perform

conversation-based tasks through supervised machine learning

Page 4: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

4

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Types – Machine Learning

Machine

Learning

Dimensional

Reduction

Clustering

Unsupervised

Learning

Classification

Regression

Supervised

Learning

Reinforced

Learning

Meaningful

Compression

Big Data Visualization

Structure Discovery

Feature Elicitation

Recommender Systems

Targeted Marketing

Customer Segmentation

Robot Navigation Learning Tasks / Decision Trees

Skill AcquisitionReal-Time Decisions

Image Classification

Fraud Analytics

Diagnostics

Customer Retention

Advertisements and

Popularity Prediction

Revenue, Cost and

Expense Prediction

Market ForecastingGrowth Prediction

https://youtu.be/f_uwKZIAeM0

Page 5: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

5

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

What is Machine Learning?

Supervised Learning Unsupervised Learning Reinforced Learning

—Algorithms apply what has been learned

in the past to predict future events on

labelled and classified data.

—Learns by comparing with intended

output and finds errors to modify the

model accordingly.

—Algorithms train information that is

neither classified nor labelled.

—Studies how systems can infer a function

to describe a hidden structure from

unlabeled data

—Algorithms that interact with the

environment by producing actions and

discovers errors or rewards.

—Automatically determines ideal behavior

within a specific context to maximize

performance

Key Features Key Features Key Features

Database Marketing Information Extraction

Pattern RecognitionOptical Character

Recognition

Cluster Analysis Anomaly Detection

Multivariate AnalysisGenerative Topographic

Map

Error Driven LearningDistributed Artificial

Intelligence

Temporal Difference

Learning

State-Action-Reward-

State-Action

“Machine Learning is the science of getting computers to learn and act like humans do, and improve their learning over time in autonomous

fashion, by feeding them data and information in the form of observations and real-world interactions.”

Techniques Techniques Techniques

Linear regression Random Forests

Naïve Bayes

Classification

Support Vector

Machines

Gradient Boosting Artificial Neural Network

K-Means Clustering Local Outlier Factor

Principal Component

AnalysisSelf Organizing Map

Expectation –

Maximization AlgorithmHierarchical Clustering

Brute Force Monte Carlo Methods

Direct Policy Search Q Learning

Page 6: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

Predictive D&A around the world

Page 7: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

7

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Game changers in Prediction

IBM: Developed “predictive attrition

program”, to predict employees most likely

to resign with 95% accuracy using AI.

Gro Intelligence: Agricultural information

group uncovers trends it hopes can counter

looming food shortage.

https://www.ft.com/content/e6530830-2b9f-

11e9-9222-7024d72222bc

Australian Securities and Investment

Commission: Analyze large amounts of

speech and text to identify patterns to

detect / predict misconduct and improve

regulation

German Trains: Sensors and analytics are

making predictive maintenance

work, for engineers and carmakers –

https://www.ft.com/content/9fb0d378-6ad4-

11e6-ae5b-a7cc5dd5a28c

Rolls-Royce: “There are massive savings if

you can pre-emptively perform maintenance

on these aircraft during a stop”.

NYPD: New York Police Department use

Patternizr, to validate and refer “hundreds of

thousands” case files using OCR.

Page 8: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

8

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Game changers: Using data analyticsMedical Industry: Developed a FDA

cleared platform for Patient Monitoring

using Predictive Analytics for trials.

https://hitconsultant.net/2019/04/03/mic-

fda-cleared-patient-monitoring-

funding/#.XKnaUuZlLIU

Law firms’ sifts and summarizes data with

speed and precision to replace routine tasks –

contract reviews using Natural Language

Processing.

FDA: Federal Govt of US leveraged

predictive analysis to discover patterns and

associations to identify the occurrence of

food based infections

https://intellipaat.com/blog/7-big-data-

examples-application-of-big-data-in-real-life

China seeks glimpse of citizens’ future with

crime-predicting data and analytics

Companies and police develop technology to

stop criminals before they act

https://www.ft.com/content/5ec7093c-6e06-

11e7-b9c7-15af748b60d0

Insurance: Employ Anti-Fraud arsenals

through advanced data analytics, as CAGR

spiked to 64% in 2019, compared to 16% in

2016

Sports analytics: Xebia developed a

predictive fitness model and logistics

operations management tool for athletes in

Special Olympics.

Page 9: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

Implementation of Predictive Analytics

Page 10: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

10

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Types - Data analytics

Probability depiction of future outcomes and trends on analysis of

historical data / categorical data

Used by FEW COMPANIES

Tools used: SAS, Python, R, SPSS

3

Critical Analysis of a problem statement to derive solutions

Used by MANY COMPANIES

Tools used: Python, R, Alteryx

2

Preliminary interpretation of Raw Data to provide an accurate

assessment

Used by ALMOST EVERY COMPANY

Tools used: ACL, Tableau, Shiny R

1

Descriptive

Data AnalyticsDiagnostic

Data Analytics

Predictive

Data Analytics

Prescriptive

Data Analytics

Illustration of possible actions to mitigate risks and guide future

decisions

Used by VERY FEW COMPANIES

Tools used: CPLX, LINDO, SAS programming

4

1

2

3

4

Page 11: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

11

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Predictive Data AnalyticsPredictive Data Analytics leverages the use of statistical models and forecasts techniques to understand a future state to answer the question “what

could happen?”.

Techniques Limitations When to Use

— Regression

— Neural Networks

— Depends on availability of labeled data –

training data.

To make an informed decision on probable

results with cluster/regression/hidden

analysis

Predictive

Data

Analytics

Identify Outliers through an

algorithm rather than random

guessing of Audit samples

In-build auditor-on-demand

to provide greater assurance

with the limited resources

Enhance the First and Second

lines of defense with

embedding predictive analytics

Leverage Flexible, Low-Cost

Storage Technology

Data Models and Algorithms

Growing Volumes and Types of

Data

Enablers Outcomes for Auditors

Limitations

Page 12: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

12

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Explanation of Predictive AnalyticsPredictive Analytics is a branch of advanced analytics that enables to make predictions about unknown future events. It incorporates

techniques such as, data mining, modeling, machine learning and artificial intelligence to analyze current data.

Historical DataPredictive

AlgorithmsModel

New Data Model Predictions

Model Building

Model Prediction

Key Feature of Predictive Analytics

https://hbr.org/video/5299994733001/the-refresher-regression-analysis

The technology that learns from past data to predict future

outcomes

Page 13: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

13

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

“the technology that learns from experience (data) to predict

likelihood of… Manipulation or Major Outliers... to provide:

1. ESTIMATED values with confidence intervals; and

2. CLASSIFICATIONS on High, Medium and Low risks.

Use of Predictive analytics in

Page 14: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

Detailed Case Studies Predictive Data Analytics

Page 15: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

Case Study Predictive Cost Analytics:“Spotting expenses Outliers from huge loads of data”

Page 16: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

16

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Executive Summary (1/4) – Project overview

1. Perform comprehensive analysis

for 5 years Cost and expenses

2. Reconcile cost information from

varied data bases

3. Establish relationship amongst

major cost centers and create

predictive algorithm

4. Compare inventory consumption

by type across various periods and

suppliers

5. Develop methods to establish the

acceptable range of the cost to

identify outliers.

1. Reviewed the overall cost for year

2012-2017

2. Performed cost cross-validation

between Materials-services

Requests with Payments and GL

3. Integrated information from GL,

Work Schedule, Information

Management System, Electronic

Data, and Report.

4. Identified key cost drivers for all

major well cost centers

5. Developed a predictive well cost

algorithm to identify outliers

1. Identified major cost centers

contributing to the majority total well

cost

2. Observed mismatches in the

overall reconciliation from varied

data bases

3. Developed acceptable range for

cost centers per asset

4. Identified wells outside the

acceptable range or with abnormal

expenses

Preliminary Scope

of Work

01

OurApproach

02

Results

03

Page 17: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

17

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Review of Databases (2/4) – Access to various data centersDuring this assignment, we reviewed and analyzed the data from the below mentioned

reports/data sheets for the last 5 years

Monthly operational reports Various Information

Management System

Daily Reports

KPI and BSC workings Sub-LedgersGeneral Ledger from ERP

Payments Desk Top Requests Invoice

Page 18: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

18

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

c

Internal Audit

Data Analysis

Integrate Procurement and

Consumption Data spread across many

systems using Clustering techniques

Establish the Correlation plots

amongst the cost elements to

establish degree of relationships

Use Regression Techniques

to detect the abnormality and

spot outliers

Establish both detailed & big-

picture perspectives of Purchase &

Consumption using Descriptive &

Visual Analytics

Cost Saving

Opportunities

“the goal is to

turn data into

information and

information into

insights which

leads to better

decisions.”

Approach (3/4)

Page 19: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

19

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Summary of Findings

Value added (4/4)

Predict Abnormalities: Built predictive

algorithms to identify abnormal cost

centers and specific outliers in individual

cost elements with threshold limits

amounting on 5 years of data.

In-built three-way matching: Spotted

major irregularities amongst: General

Ledger, Payables and Inventory records.

Identified major Cost Drivers:

Identified 12 major cost centers (out of

140) that contributed to 80% of the total

cost & established one to many and

many to one correlation factors.

On-going monitoring: Developed

Interactive dashboards to stakeholders

on discrepancies in three-way matching,

irregular well costs and expenses

outliers, which are accessible even in

hand-held devices.

Page 20: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

Case StudyPredictive Maintenance:“Understand how likely an equipment needs maintenance

and preempt failure”

Page 21: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

21

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Using Predictive Risk Analytics (1/2)Problem statement

How to use Predictive Risk Analytics to gain insight into:

Maintaining large-scale refinery infrastructure in energy industry is a

challenge, and achieving optimum asset performance even more so:

— Optimize asset management through effective preventative and

predictive maintenance

— Improve reliability by avoiding asset failures (which require corrective

maintenance)

— Identify chronically problematic assets with a high ratio of repair-to-

replacement costs

Predictive Maintenance: “To predict the probability of failure in the

next cycle based on operational / maintenance data and other

data sets”

I want to

understand

how likely a

machinery /

equipment is

going to

need

maintenance

in the next

cycle and

which

equipment is

critical for

maintenance

requirements

Page 22: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

22

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved. 22© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Document Classification: KPMG Confidential

Dashboard (2/2)

Devices with high

propensity of

failure identified

using the output

of statistical

models

Detailed analysis

of each of the key

service

parameters which

are indicative of

the failure

Key indicators

which are used

by the model

to predict

failure

Page 23: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

Case Study Detect Financial Statement Risks Analysis

Page 24: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

24

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Computation of over 40 + financial measures

Page 25: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

25

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Establishing the intricate relationship between metrics

Page 26: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

26

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Cash Coverage Ratio

Interest Coverage Ratio

Interest Coverage

Days Inventories Outstanding

Total Debt to Total Shareholder's Equity

Debt to Equity

Return on Common Equity

Equity Multiplier

Capitalisation Ratio

Long term Debt to Equity

Return on Assets

Inventory Turnover

Total Asset Turnover

Operating Profit Margin

Net Profit Margin

Free Cash Flow to OCF Ratio

Gross Profit Margin

Long term Debt to Total Assets Ratio

Debt Service Coverage Ratio

Operating Return on Assets

Capital Expenditure Coverage Ratio

Operating Cycle

Net Working Capital to Sales

Inventory to Net Working Capital

Working Capital Turnover

Currrent Ratio

Quick Ratio

Short Term Debt Coverage Ratio

Operating Cash Flow to Sales

Accounts Receivable Turnover

Average Collection Period

Fixed Assets Turnover

Total Debt to assets

Applying Machine Learning to group the financial metrics

Page 27: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

27

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Predictive data analytics to spot potential anamolies

Page 28: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

Case StudyVulnerable IT assets for Cyber Defense“Intrusion Detection System for Cloud Systems”

Page 29: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

29

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Network log

Executive Summary – Assignment Overview

Network packet sized Service requests Network logNetwork packet

sizedService requests

Intrusion Attackers or Non-Intrusion Service seekers

Network Log

Packet Size

Service Request

Observed Sequences

Intrusion Attackers

Non Intrusion Attackers

Unobserved States or Hidden

States Generating the

Observations

Real World Scenario: Intrusion Detection System for Cloud Services

Cloud Systems suffer from lot of security vulnerabilities and it is necessary to detect and alert intrusion attacks in advance.

Problem: Normal Service activities and intrusion attackers activities, both generate the data points. But, there is a different pattern for intrusion attackers and normal

service users.

Observed Data Points: Network Logs, Packet Size and Event Logs

Page 30: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

Case Studies Predictive Data Analytics from an IA perspective

Page 31: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

31

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Case Studies

First and Third Party Fraud1 Cognitive Analytics4

“Time off Work” using PA6Lease Accounting and IFRS3

Predictive Auditing2 Behavioral Model to Predict Renege5

Following are case studies that made immense strides within their business with the use of Machine Learning and Predictive Data Analytics.

Internal Audit Risk Management

Internal Audit and Risk Management

Manage Demand using PA Consumer Preferences7 8

Page 32: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

32

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

International Case studies

— Analyze the patterns of the

consumers for a leading bank to

identify fraudulent consumers

— Develop a predictive model to

identify potential future fraudsters

— Highlight first and third hand

fraudulent events within the data

source

Engagement Goals Benefits to the Auditor

First and Third Party Fraud

— Deployed Social Network Analysis

system, to highlight individual entities

that would potentially perpetrate

large scale frauds – which were

taken-up for audits.

— Reduced the annual costs by

decreasing investigation time to

highlight “high-likelihood” cases

— Develop robust auditing platform to

identify trends of emerging risks –

moving away from control testing

— Use of both proactive and

predictive auditing

— Analyze associations and attributions

between different departments

processes

— Incorporate group company set of

criteria within the platform

— Enhancing efficiency & effectiveness,

channeling auditing resources to

more value adding activities

— Proactive auditing on the go, moving

from hindsight to foresight

— Predictive mapping trends to forecast

and identifying emerging risks

— Prevent incidents from occurring

Engagement Goals Benefits to the Auditor

Predictive Auditing for a larger Asian BankIA IA

Page 33: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

33

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

International Case studies

— Analyze the real-time sensor data to

monitor the health of the aircraft

during flight

— Predictive model to assess the

maintenance or replacement of the

aircraft parts

Engagement Goals Benefits to the Risk Manager

Cognitive Analytics for Aircraft Maintenance

— Improved the risk categorization

through consolidated information for

all the parts required for aircraft to be

replaced

— Risk assessment was conducted on

the list of parts with “no fault” from

contractor, yet identified as faulty in

the current system – built a key risk

indicator profiles.

— Highlight aircraft parts due for

maintenance in advance in order to

saving procurement costs for new

parts

— Automated processing lease account

documents

— Extraction of key facts: contractual

parties, length of contract etc.

— Digitized, indexed and searchable

document corpus

— Classification of lease documents

categories

— Additional information like

geographical overview, notification on

expiring contracts etc.

Engagement Goals Benefits to the Auditor

Leasing Accounting / IFRS 16

— Manual processing is time-

consuming and can be (partially)

automated using cognitive

technology

— System is trained by leasing experts

and available around the clock for

IFRS 16.

— All contracts details centrally available

and linked to the respective original

contract

— Use of Natural Language Processing

in predicting potential leasing

clauses in contracts.

RM IA

Page 34: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

34

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

International Case studies

— Identify the leading designs in the

fashion industry, to help predict the

upcoming trends

— Analyze real time sales of stores, to

determine items to reorder or

abandon

— Incorporate e-commerce statistics, to

gain valuable insights on preferences

from browsing history and customer

interaction

— Ability to gauge the “Risk Factor” of

upcoming season trends for the

organization

— Real time response initiative to

changing trends and delay mass

production to reduce losses

— Focused decisions based on country /

demographic

Engagement Goals Benefits to the Risk Manager

Consumer Preferences

— Predict the influx of patients in a

Health Center, to plan for the

appropriation of staff and medical

supplies for cardiovascular

examinations

Engagement Goals Benefits to the Auditor

Predictive Analytics to Manage Demand

— Identify the baseline to maintain and

manage inventory depending on

the influx of patients

— Plan and assess the organization’s

staff appropriation for different health

centers within a geographical

region

— Ability to schedule medical

procedures and surgical treatments at

an optimal level

IA/RM IA/RM

Page 35: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

35

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

International Case studies

— Identify the key drivers that influence

the candidate joining/not-joining the

organization

— Descriptive analytics on the renege

— Predictive model to analyze the

probability to accept an offer and join

the company

— Identified 6 key driver stages to

gauge a candidate joining/not-

joining the organization

— Insights provided on each stage the

probability of the candidate to join

the organization

Engagement Goals Benefits to the Risk Manager

Behavioral Model to Predict Renege

— Predict the time required for

claimants to return to work to

pinpoint cases that need additional

support

— Analyze the additional time off work

required by claimants

Engagement Goals Benefits to the Risk Manager

“Time off Work” using Predictive Analytics

— Identified high risk claims (over 90

days required) with 90% confidence

— Cost reduction through earlier

assignment of project

managers/occupational

therapists/temporary workers to

maintain business continuity

— Process improvement opportunities

and policy reconsideration strategies

— Work force planning and business

continuity model.

RM RM

Page 36: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

Challenges in implementing Predictive Analytics

Page 37: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

37

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Issues and mitigation optionsActions to

avoid issues

Frequently

observed issues

Lack of skilled resources Hire and / or train to use case need

Extensive data detective workPut data ‘under governance’ enable high quality data

for top use cases

Purchasing the ‘wrong’ technology

Understand use cases prior to purchasing the ‘right’ tools

are not always ‘new’

Confusion on services offered

Creation of a service catalog with defined business

alignment

Inability to describe value generated

Identify high priority use cases Outline value tracking

mechanisms

‘Lack-of’ or slow adoption

Demonstrated executive commitment Engage

stakeholders early

Page 38: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

38

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

How I imagine my data lake to be… What it’s really like

Copyright Dale Williams data lake memes

Page 39: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

© 2018 KPMG International Cooperative (“KPMG International”). KPMG International provides no client services and is a Swiss entity with which the independent member

firms of the KPMG network are affiliated.

KPMG AI in Control Methodology

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Integrity

Track the lineage and provenance of raw data,

training data, model experiments, ongoing

changes made by SME’s. Model training incl

changes by stored in a immutable ledgers.

Strategy & Governance Understand & Design Model & Train Evaluate & Assure Deploy & Evolve

Explainable

Models that can explain the knowledge learned.

Can provide explanation in business terms on how

decisions were made. Interpretations are gained

or inferred from explanations.

Free from prejudice

Models as well as the training data that must be

free of bias, are inclusive and avoids unfair

treatment of certain protected groups. Be certain

that the models incl the trainer comply with

policies & regulations

Agile and robust

Models are interoperable between various

runtimes, providers, or frameworks. Consumable

from apps & processes. The models, ground truth

and feedback are safe & secure from harm or

adversarial attacks.

— Identified Business / LOB

owner

— Defined key business

objectives

— Modeled measurement

metrics (business)

— Model Capability mapped to

biz reqs

— Involved persona metadata

— Regulatory mandates and

requirements

— Explainability result template

— Features comply with

policies

— Features comply with biz

reqs

— Model usage restrictions

— Need to know, groups, users

etc.

— Model SLA’s

— Required skills and support

to manage and maintain

— Identified raw & training

data sources

— Pre-processing performed

on data

— Training data provenance

— Data SME’s Involved?

— Training data metadata

— Training data descriptive

data

— Training data outliers

— Define explainability schema

— Population sampling methods

— Sampling size

— Training data completeness

— Test data completeness

— Features list

— Data usage guidelines

— Confidential features?

— Model weights transferable

— Model metadata

— Model trainer profile

— Training methodology

— Techniques and algorithms

applied

— Feature changes

— Model metadata

— Knowledge representation

ontology

— Type of explainability

— Technical vs business lingo

used

— Bias detection techniques

— Bias remediation

techniques

— Model weights verification

— Training data protection

— Training data access

— Frameworks and libraries

used

— Tooling used

— Experiment setup and

config

— Model experiments reports

— QC & assurance on model

— Model evaluation reports

— Feature change log

— Model evaluation metrics

and scores

— Human validation reports

— Handling unexplainable

events

— Redundant coding

— Inclusiveness listing

— Model drift evaluation

— Simulation on new data

— Approved framework and

runtime

— Concept drift audit log

— Intermediate representation

— Framework/runtime

vulnerability testing

— Feedback & usage logs

— Experiment iterations

from feedback

— Model

improvement/change log

— Continuous testing &

monitoring of explainability

— Training vs usage data

differences

— Model serving access

protection

— Model serving monitoring?

— Cloud vs on-premise

— Feedback retention policies

— Data stationarity checks

Page 40: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

KPMG’s D&A Tools: e-Audit, Ignite, Clara and Sofy

Page 41: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

41

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved. 41© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Document Classification: KPMG Confidential

Reinforcement Learning

Supervised Learning

Unsupervised Learning

Knowledge-Based Systems

Natural Language Generation

Natural Language Processing

Deep Learning

Data Components Algorithms and Tools Human-in-the-Loop

Training

Automation

Acceleration

Enhanced

Insights

Text/Semi-

Structured

Speech

Image

Structured

Data

Ignite employs advanced analytical techniques and algorithms to train computers how to

use data from a wide variety of sources and formats to enhance, accelerate, automate,

and augment decisions that drive growth and profitability.

41© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Document Classification: KPMG Confidential

Ignite Methodology

Page 42: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

42

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

KPMG Clara MethodologyScaling the base Stabilizing the core Innovating the future

We are on a journey

Data & Analytics

Knowledge &

Collaboration

KPMG Clara

Validation &

Integration Predictive

Analytics

Prescriptive

Analytics

Machine

Learning

Cognitive

Analytics

Growing global solutions ecosystem

Page 43: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

43

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

KPMG Sofy

Access

Management

Controls

Management

3rd party Risk

Management

Policy & Regulation

Management

Risk

Assessment

Continuous Controls

Monitoring

Process

Insights

My Tasks

Spend

Analytics

VAT

Analytics

VAT

Compliance

Transfer

Pricing

Customs

Manual Journal Entry

IFRS 16

Data

Governance

Data Quality

Monitoring

Tax Management

Data ManagementGovernance, Risk & Compliance

Strategy & Operations Performance

Finance Excellence

Risk

Management

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Page 44: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

44

Document Classification: KPMG Confidential

© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved. 44© 2019 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with KPMG International Cooperative (“KPMG International”), a Swiss entity.

All rights reserved.

Document Classification: KPMG Confidential

– IT’S OUR IMAGINATION

It’s not technology that limits us today

Page 45: Use of Predictive Analytics in business and internal Audit · Data Analytics Identify Outliers through an algorithm rather than random guessing of Audit samples In-build auditor-on-demand

Document Classification: KPMG Confidential

kpmg.com/socialmedia kpmg.com/app

This approach note is made by KPMG, the United Arab Emirates member firm of the KPMG network of independent firms affiliated with KPMG International

Cooperative (“KPMG International”), and is in all respects subject to the negotiation, agreement, satisfactory clearance of KPMG’s client and engagement

evaluation process, and signing of a specific engagement letter or contracts. KPMG International provides no client services. No member firm has any authority

to obligate or bind KPMG International or any other member firm vis-à-vis third parties, nor does KPMG International have any such authority to obligate or bind

any member firm.

© 2017 KPMG Lower Gulf Limited and KPMG LLP, operating in the UAE and member firms of the KPMG network of independent member firms affiliated with

KPMG International Cooperative (“KPMG International”), a Swiss entity. All rights reserved.

The KPMG name and logo are registered trademarks or trademarks of KPMG International.

Sudharshana Balasubramanian

+971 52 559 0100

[email protected]

Thank you