4. Common Challenges
According to a KPMG study, Audit departments are challenged by:
• Disparate systems supporting different business models (e.g. T&E)
• Establishing the definition of an “exception”, addressing “false positives” and “false
negatives”
• Bridging the gap on what the audit population is (e.g. Benford’s)
• Relying on intuition rather than data to support audit risk assessment (e.g. defining a
manual JE)
“Data analytics will likely be unsustainable without linkage to, or integration with, an audit
work plan and the related audit objectives.”
5. Our Challenges
1.Data acquisition – understanding and processing the data; need to start
with client-provided data as a base and then become more independent
as you get comfortable with the data
2.Finding the right resources – BI, Auditor, Business Analyst?
3.Bandwidth
4.Technology needs
5.Over-dependence by auditors – analytics are just the beginning of the
audit dialog
6. What we can do
• Understanding process is critical to provide valuable analysis
• Right sizing the analytics for the size of the organization and risks being
assessed
• Continuous improvement on analytics effectiveness
7. Audit Data Analytics Lifecycle
Planning
Brainstorming
Session
Communicate scope
& objectives
Understand business
context
Fieldwork
Knowledge sharing
Integrate DA
documentation
Reporting
Integrate analytics
Feedback on use of
analytics
8. Program
Management
Establish development
methodology (e.g. Agile)
Business process driven
Audit Data Analytics Key Elements
Access
Data
Acquisition
Tools
Understand business
processes
Identify data sources
Establish data acquisition
approach (direct connection,
backup restoration, system
canned reports, etc.
Evaluation of development
tools
Excel
SQL
ACL
R/Python
Tableau / QlikSense
Understand what data is
captured by the source
system
Examine the data
quality, integrity, and
completeness
Design testing approach
based on the data
obtained
Data Source Project Management
10. When Benford Analysis Is or Is Not Likely Useful
When Benford Analysis is Likely Useful Examples
Sets of numbers that result form mathematical combination of
numbers
AR (number sold *price), AP (number bought * price)
Transaction-level date – no need to sample Disbursement, sales, expenses
On large data sets – The more observations, the better Full year’s transactions
When Benford Analysis is Not Likely Useful Examples
Data set is comprised of assigned numbers Check numbers, invoice numbers, zip codes
Numbers that are influenced by human thoughts Prices set at psychological thresholds($1.99), ATM withdraws
Accounts with a build in minimum or maximum Set of assets that must meet a threshold to be recorded
11. Key Analytics
The Wharton School has published basic data analytical tests that can assist in re-
focusing efforts in planning and executing audits in areas that could indicate incentives
for management to manipulate results.
• These tests fall into the following areas:
– Dupont Analysis
– Revenue & Expense Recognition Management
– Discretionary Accruals & Expenditures
– Fraud Prediction – Beneish M-Score
14. Revenue Recognition Red Flags
Potential red flags that identify potential changes in revenue recognition
policies:
• Unusual seasonally-adjusted quarterly (monthly) trends
• Growth in Revenue
• Growth in Accounts Receivable
• Unusual trends in Ratios
• Days Receivable and Accounts Receivable/Revenue
Then, we will try to find what happened
• Do earnings management incentives exist?
• Is there anything unusual in the Revenue Recognition policy
15. Year-over-Year Growth Trends
Due to seasonality need to compare to same quarter / month of the prior year
• YoY Revenue Growth
• YoY Growth in AR
Benchmarks
• Time-series: is growth unusual in one specific quarter for the firm?
• Cross-sectional: is growth unusual for the industry in a given quarter?
18. Fraud Prediction
• Fraud prediction models examine companies that have been caught committing fraud to model
how they differ from companies not caught
• Uses statistical techniques to chose a small set of ratios
Advantages
– Specifically tailored to characteristics of fraud firms
– Model parameters are fixed and don’t have to be re-estimated for each company
Disadvantages
– Models based on companies that were caught with large frauds
M-Score is based on eight ratios
– Higher M-Score means higher likelihood of manipulation
– Uses comparisons between current year and prior year
Editor's Notes
Statistics
Using standard deviation to ID unusual Jes
Benfords
Subject matter knowledge
What is the data telling us in the context of the business process
Data Presentation
1. Need to understand how the data is going to be used whether it is Excel, Tableau, or something else.