analytics case studies · nuix* enterprise discovery server (neds) and metadata is exported so that...

15
Analytics Case Studies Presented by Tasia Livaditis Senior Director, Corporate Analytics Office of the Chief Knowledge Officer Australian Taxation Office 30 November 2011 ATO Analytics Case Studies 2 Overview Debt Risk Models Pg 3 Lodgment Risk Models Pg 8 Evidence Process Capture (Text Mining) Pg11 Income Tax & GST/VAT Refund Fraud Pg 16 Network Risk models Pg 20 Business Process Improvement Pg 28

Upload: others

Post on 16-Oct-2020

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Analytics Case Studies · Nuix* Enterprise Discovery Server (NEDS) and metadata is exported so that the user can determine the origin of the evidence. ... This presentation was current

Analytics Case Studies

Presented by

Tasia Livaditis

Senior Director, Corporate Analytics

Office of the Chief Knowledge Officer

Australian Taxation Office

30 November 2011

ATO Analytics Case Studies 2

Overview

Debt Risk Models Pg 3

Lodgment Risk Models Pg 8

Evidence Process Capture (Text Mining) Pg11

Income Tax & GST/VAT Refund Fraud Pg 16

Network Risk models Pg 20

Business Process Improvement Pg 28

Page 2: Analytics Case Studies · Nuix* Enterprise Discovery Server (NEDS) and metadata is exported so that the user can determine the origin of the evidence. ... This presentation was current

ATO Analytics Case Studies 3

Debt Risk models

Propensity to Pay

Capacity to Pay

ATO Analytics Case Studies 4

Propensity To Pay Model

The „propensity to pay‟ score predicts the probability that the taxpayer will

pay all outstanding liabilities in full within a 90 days.

The score is determined by taxpayer data such as:

Amount of Debt

Age of Debt Case

Recent payment behaviour

Recent debt cases & effort required to collect

Recent lodgment behaviour

Other compliance area activities

Demographic information

Page 3: Analytics Case Studies · Nuix* Enterprise Discovery Server (NEDS) and metadata is exported so that the user can determine the origin of the evidence. ... This presentation was current

ATO Analytics Case Studies 5

Capacity To Pay (Company) Model

The „capacity to pay‟ score assesses the financial capacity of the taxpayer

to pay the debt. It predicts the likelihood of company insolvency in the next

12 months.

Key attributes:

Financial ratios associated with Income, Liabilities, Working Capital,

Earnings, Owner‟s Equity (based on updated versions of Altman‟s Z

score)

Company demographics

Trends in profit/losses over last 2 years

Debt/Lodgment behaviour over last 2 years

ATO Analytics Case Studies 6

Capacity To Pay (Individual) Model

The „capacity to pay‟ score assesses the financial capacity of the taxpayer

to pay the debt. It predicts likelihood of individual bankruptcy in next 12

months.

Key attributes:

Total Income / expenses associated with Employment, Interest,

Dividends, Capital Gains etc

Account history (payments, liabilities)

Social indicators including Unemployment, Child Support and

Pensioner indicators, Spouse indicators, and number of Dependents

Number of links to companies with compliance actions

Page 4: Analytics Case Studies · Nuix* Enterprise Discovery Server (NEDS) and metadata is exported so that the user can determine the origin of the evidence. ... This presentation was current

ATO Analytics Case Studies 7

Use of Debt Risk Model outputs

The Risk Analytical Models are run weekly across all active Receivables

Management System (RMS) cases.

Capacity to Pay

▼Unable ▼Long Term ▼Short Term

►Unlikely

►Likely

►Willing

U/U (0.9) U/LT (0.8) U/ST (0.7)

W/U (0.3) W/LT (0.2) W/ST (0.1)

L/U (0.6) L/LT (0.5) L/ST (0.4)

Pro

pen

sit

y t

o

Pay

Risk category

combination

ATO Analytics Case Studies 8

Lodgment Risk models

Propensity to Lodge

Risk to Revenue

Page 5: Analytics Case Studies · Nuix* Enterprise Discovery Server (NEDS) and metadata is exported so that the user can determine the origin of the evidence. ... This presentation was current

ATO Analytics Case Studies 9

Lodgment Risk Models

Aim to predict „Revenue that would arise’ if a large number of apparently

overdue returns, both Income Tax and Activity Statements, were lodged by

a taxpayer.

Key Attributes:

Previous lodgment behaviour (timing, amounts) Known income & economic activity (third party data) demographic information Whether in the “instalment” system (leading to possible credits)

ATO Analytics Case Studies 10

Good revenue outcomes

Australian Financial Review 10 July 2006

Page 6: Analytics Case Studies · Nuix* Enterprise Discovery Server (NEDS) and metadata is exported so that the user can determine the origin of the evidence. ... This presentation was current

ATO Analytics Case Studies 11

Text Mining

ATO Analytics Case Studies 12

Text mining – an overview

With the dramatic growth of hard drive capacity, the average data size of a

case received from Access visits is currently around 250GB. Some cases

even receive terabytes of data.

Processing and analysing such large amounts of data using manual

processes is challenging and time consuming.

Text mining aims to replace these manual processes by enabling text data to

be cleansed, classified, prioritised and extracted, offering a considerable

advantage in efficiency.

Page 7: Analytics Case Studies · Nuix* Enterprise Discovery Server (NEDS) and metadata is exported so that the user can determine the origin of the evidence. ... This presentation was current

ATO Analytics Case Studies 13

Text Mining in the Evidence Capture Process

IT forensics seize evidence

or copy it on site during

Access / Warrant visit.

Evidence is brought back to

the ATO and registered in the

Evidence Room.

Evidence is loaded onto the

Nuix* Enterprise Discovery

Server (NEDS) and metadata is

exported so that the user can

determine the origin of the

evidence.

Auditors and Investigators use

ATOnet to search, classify

and analyse documents.

ICT Trusted Access Forensics

use NEDS to do conversion,

extraction, copying and

Indexing.

The data is then transferred

to ATOnet via the Forensic

Network.

Text Miner cleanses data

for relevancy e.g. removes

file formats.

Text Miners help the Auditors and Investigators by classifying

and prioritising documents. This is an ongoing process and

saves time.

Advanced text mining can also be undertaken for larger and more

complex cases.

2 3 4

6

Working together

Evidence Room

NEDS NEDS

Auditors/Investigators

* Nuix - a data analysis and evidence management tool

ATO Analytics Case Studies 14

Text mining procedures

Electronic documents

Document

Classification

Relevant documents

Rule based template

developed using

case knowledge

Templates

Data extraction using templates

(Text analysis)

Data extracted

in CSV format

(Excel)

Explore

Important Concepts

and their relationship

Concepts and their relationship

Page 8: Analytics Case Studies · Nuix* Enterprise Discovery Server (NEDS) and metadata is exported so that the user can determine the origin of the evidence. ... This presentation was current

ATO Analytics Case Studies 15

Whole document set

(4476 documents)

Documents that are

identified as relevant :

661 documents Case relevant documents

but not found in the search:

5 documents

Case irrelevant documents

but in the search results :

153 documents

The document classification model built using text mining is over 98%

accurate in identifying the relevant documents. Most of the documents

of interest were identified, missing only 5 relevant documents. Only

153 irrelevant documents had to be manually discarded rather than

nearly 4000!

D1

D2

D3

D4

Dn-1

Dn

Input documents

Text Mining

Income Tax & GST/VAT Refund

Fraud Detection models

Page 9: Analytics Case Studies · Nuix* Enterprise Discovery Server (NEDS) and metadata is exported so that the user can determine the origin of the evidence. ... This presentation was current

ATO Analytics Case Studies 17

Income Tax Forms Processing

ATO Analytics Case Studies 18

Analytical Models for IT Forms Processing

Page 10: Analytics Case Studies · Nuix* Enterprise Discovery Server (NEDS) and metadata is exported so that the user can determine the origin of the evidence. ... This presentation was current

ATO Analytics Case Studies 19

Outputs

ATO Analytics Case Studies 20

Network Risk Models

Page 11: Analytics Case Studies · Nuix* Enterprise Discovery Server (NEDS) and metadata is exported so that the user can determine the origin of the evidence. ... This presentation was current

ATO Analytics Case Studies 21

Network Model Use Cases

Purpose Number of Network

Vertices

Number of Network

Links/Arcs

Filter low-risk entities from

database of high-risk entities

200 16,000

Using watch-list of known

high-risk entities, discover

related unknown high-risk

entities

300 5,000

Find non-agent returns that

appear to have a common

“guiding mind”

6,000 returns from 1,200,000 20,000,000

Find high risk company

structures

1,200,000 18,000,000

ATO Analytic Case Studies www.ato.gov.au

Coding Networks in a SQL Database

(coy -> ind)

(ptr -> ind)

(trt->ind unk)

(coy -> (coy -> ind))

Page 12: Analytics Case Studies · Nuix* Enterprise Discovery Server (NEDS) and metadata is exported so that the user can determine the origin of the evidence. ... This presentation was current

ATO Analytic Case Studies www.ato.gov.au

Coding Networks in a SQL Database

ATO Analytic Case Studies www.ato.gov.au

Associated Entities

One Degree of Separation

Two Degrees of Separation Three Degrees of Separation

Four Degrees of Separation Five Degrees of Separation

Companies

Government

Individuals

Partnerships

Suoerannuation

Trusts

Triangle = non lodged

Circle = lodged

Size = Ind … Large

Page 13: Analytics Case Studies · Nuix* Enterprise Discovery Server (NEDS) and metadata is exported so that the user can determine the origin of the evidence. ... This presentation was current

ATO Analytic Case Studies www.ato.gov.au

Risk Differentiation for Networks

ATO Analytics Case Studies 26

Business Process Improvement

Page 14: Analytics Case Studies · Nuix* Enterprise Discovery Server (NEDS) and metadata is exported so that the user can determine the origin of the evidence. ... This presentation was current

ATO Analytics Case Studies 27

Business Process Improvement Examples

Predicting if an audit activity is likely to result in an objection being raised.

As case information is updated, the case is rescored and an alert raised if it

becomes high risk. This alert allows auditors to respond early to prevent

unnecessary objections.

Replacing manual review of accounts to support financial reporting

requirements such as bad and doubtful debt provisions and performance

measurement of collection treatments.

Detecting when an employer has missed a payment or made a payment to

the wrong cost centre for Income Tax Withholding and notifying the service

officer that this has occurred and recommending the steps to take to

correct the problem.

Mapping of internal processes such as provision of advice or tax return

processing to follow progress with our service standards; what the workflows

are, where the time delays occur.

ATO Analytics Case Studies 28

Questions?

© COMMONWEALTH OF AUSTRALIA 2011

This presentation was current in November

2011

Page 15: Analytics Case Studies · Nuix* Enterprise Discovery Server (NEDS) and metadata is exported so that the user can determine the origin of the evidence. ... This presentation was current

ATO Analytics Case Studies 29

Taxpayers

Candidate

Population

Ranked

Candidates

Cases

Results

Risk

Prioritisat

ion

Resourc

e

Allocatio

n

Demand

Manage

ment

Risk

Identifica

tion

Risk

Treatme

nt

Develop

ment

Model &

Treatme

nt

Strategy

Coverage

& Revenue

targets

Modelling

Operationalise

Analytics

Siebel Work &

Case Mgmt

Audit Case Selection: End to end process