LEVERAGING ANALYTICS IN COMPLIANCE … AND KYCLEVERAGING ANALYTICS IN COMPLIANCE … AND KYC LAURENT COLOMBANT FRAUD & COMPLIANCE SPECIALIST SOUTH WEST EUROPE ROLAND THEYS ... in various

Copyr i g ht © 2012, SAS Ins t i tu t e Inc . A l l r ights reser ve d .

LEVERAGING ANALYTICS

IN COMPLIANCE … AND KYC

LAURENT COLOMBANT

FRAUD & COMPLIANCE SPECIALIST

SOUTH WEST EUROPE

ROLAND THEYS

DIRECTOR FRAUD & COMPLIANCE

SOUTH WEST EUROPE


AGENDA

1. The context of Compliance

2. Why one should use Analytics?

3. What are Analytics?

4. Use Cases

5. The case of KYC

6. Demo

7. Conclusion


1. THE CONTEXT OF ISSUES IN COMPLIANCE


NOTES COMPLIANCE CONTEXT

The increasing pressure of the regulator emphasize on the obligation for the financial institutions to be more precise and

effective in the Anti-money Laundering field. Institutions should be able to analyze and understand their risks, to put

measures in places to counter those risks, and be able to justify how their detection model was configured.

AML systems in place generate tons of false positives. Due to the obligation of managing every single alert, it is becoming a

real nightmare for those institutions. So some financial institutions are tempted to take the approach to determine how many

employees they can enroll on these compliance activities and to set up their detection rules and thresholds accordingly. In

all cases they fear a visit from the national authorities as they suspect they might have difficulties justifying their approach.

The main reason of this ineffective workload is that the different AML functions (Customer screening, payment screening,

transaction monitoring) are monitored through outdated and inefficient business rules that will raise the same alerts again

and again on a booming quantity of transactions, devices and channels.

Smart analytics can solve the problem:

• To fully understand your risks and set up clever detection models and thresholds

• To use analytical models to better manage false & true positives

• To ensure you can justify your choices in front of the regulator


AML CONTEXT

Context

• Booming volumes

• Regulatory pressure

Directives

Channels

Devices

Data


• Context


• Booming quantities

AML Issues

• Outdated IT solutions

• One size fits all rules

• False positives

• Moving target

• Budget-based

• Challenge

• Understand

• Efficiency

• Justify

AML ISSUES


KYC Issues

• Often paper based

• Multiplication of data

sources to exploit to

achieve certainty and

accuracy

• Time-consuming

• Speed has negative

effect on customer

experience

• Ongoing regulatory

changes

• Spiraling costs

• X levels downs to gain

certainty

KYC ISSUES

KYC & identity resolution

Offshore leaks

DBs & External registrars

Sanctions & PEPs

Networks & Siloes


Context

• Booming volumes


Issues

• Outdated and one size

fits all rules

• False positives

• Moving target

• Budget-based

Challenges

• Understand

• Efficiency

• Justify

INTRODUCTION CONTEXT, ISSUES, CHALLENGES … AND THE SOLUTION

• Explore & fully understand your data

• Use Analytics to optimize efficiency

(segments, thresholds, detection, etc.)

• Justify your model

Solution


2. WHY SHOULD ONE BE USING ANALYTICS?

Wikipedia:

Analytics is the discovery, interpretation, and communication of meaningful patterns in data. Especially valuable in

areas rich with recorded information, analytics relies on the simultaneous application of statistics, computer

programming and operations research to quantify performance.

Organizations may apply analytics to business data to describe, predict, and improve business performance. Specifically,

areas within analytics include predictive analytics, prescriptive analytics, enterprise decision management, retail analytics,

store assortment and stock-keeping unit optimization, marketing optimization and marketing mix modeling, web analytics,

sales force sizing and optimization, price and promotion modeling, predictive science, credit risk analysis, and fraud analytics.

Since analytics can require extensive computation (see big data), the algorithms and software used for analytics harness the

most current methods in computer science, statistics, and mathematics.


FOUNDATIONS ARE KEY TO COPE WITH EVOLUTIONS

Without the Data & Analytics layers, one does not have the necessary foundations to

extend its solution over time to tackle new patterns and improve detection accuracy.

AML KYC CDDSanctions UBOs

… …


ANALYTICS LIFT CURVE

Advanced analytics push the boundaries

FR

AU

D

POPULATION

█ Advanced analytics with Risk-

Scored Networks

█ Advanced analytics

█ Random samples

█ Business rules


3. WHAT ARE ANALYTICS ?

Fraud Analytics Using Descriptive, Predictive, and Social Network

TechniquesIt is estimated that a typical organization loses about 5% of its revenue to fraud every

year. More effective fraud detection is possible, and this book describes the various

analytical techniques your organization must implement to put a stop to the revenue

leak.

• Examine fraud patterns in historical data

• Utilize labeled, unlabeled, and networked data

• Detect fraud before the damage cascades

• Reduce losses, increase recovery, and tighten security


UNSUPERVISED

When no targets exist

Examine current behavior to

identify outliers and abnormal

transactions

ANALYTICAL METHODS

Use when a known target is

available

Use historical information to

predict suspicious behaviors

similar to previous patterns

Rule and analytic based

network scoring

Automatically risk score while

building relevant networks

SUPERVISED NETWORKS

Clustering, K-means, anomaly detection, (neural

networks), etc.

Linear & logistic regressions, decision tree, neural

networks, etc.


NOTES ANALYTICAL METHODS

Unsupervised machine learning is the machine learning task of inferring a function to describe hidden structure from

"unlabeled" data (a classification or categorization is not included in the observations). Since the examples given to the

learner are unlabeled, there is no objective evaluation of the accuracy of the structure that is output by the relevant

algorithm—which is one way of distinguishing unsupervised learning from supervised learning and reinforcement learning.

Supervised learning is the machine learning task of inferring a function from labeled training data.[1] The training

data consist of a set of training examples. In supervised learning, each example is a pair consisting of an input object

(typically a vector) and a desired output value (also called the supervisory signal). A supervised learning algorithm analyzes

the training data and produces an inferred function, which can be used for mapping new examples. An optimal scenario will

allow for the algorithm to correctly determine the class labels for unseen instances. This requires the learning algorithm to

generalize from the training data to unseen situations in a "reasonable" way (see inductive bias).

Social network analysis (SNA) is the process of investigating social structures through the use

of network and graph theories.[1] It characterizes networked structures in terms of nodes (individual actors, people, or things

within the network) and the ties, edges, or links (relationships or interactions) that connect them. Examples of social

structures commonly visualized through social network analysis include social media

networks,[2] memes spread,[3] friendship and acquaintance networks, collaboration graphs, kinship, disease transmission,

and sexual relationships.[4][5] These networks are often visualized through sociograms in which nodes are represented as

points and ties are represented as lines.

https://en.wikipedia.org/wiki/Machine_learning

https://en.wikipedia.org/wiki/Supervised_learning

https://en.wikipedia.org/wiki/Reinforcement_learning

https://en.wikipedia.org/wiki/Machine_learning

https://en.wikipedia.org/wiki/Supervised_learning#cite_note-1

https://en.wikipedia.org/wiki/Training_set

https://en.wikipedia.org/wiki/Inductive_bias

https://en.wikipedia.org/wiki/Network_theory

https://en.wikipedia.org/wiki/Graph_theory

https://en.wikipedia.org/wiki/Social_network_analysis#cite_note-1

https://en.wikipedia.org/wiki/Social_media


https://en.wikipedia.org/wiki/Internet_meme


https://en.wikipedia.org/wiki/Weak_ties

https://en.wikipedia.org/wiki/Collaboration_graph

https://en.wikipedia.org/wiki/Kinship

https://en.wikipedia.org/wiki/Disease_transmission

https://en.wikipedia.org/wiki/Sexual_network



https://en.wikipedia.org/wiki/Sociogram


ANALYTICS FOR MATCHING

15

Match Codes Creation

Name Match Code (95%

Sensitivity)

Match Code (90%

Sensitivity)

Match Code (85%

Sensitivity)

John Q Smith4B7~2$$$$$$$$$$C@B$$$$$$$

$Q

4B7~2$$$$$$$$$$C@P$$$$$$

$$Q

4B&~2$$$$$$$$$$C@P$$$$$$

$$$

Johnny Smith4B7~2$$$$$$$$$$C@B7$$$$$$

$$

4B7~2$$$$$$$$$$C@P$$$$$$

$$$

4B&~2$$$$$$$$$$C@P$$$$$$

$$$

Jonathan

Smythe

4BR~2$$$$$$$$$$C@B&~2&B$

$$$

4BR~2$$$$$$$$$$C@P$$$$$$

$$$

4B&~2$$$$$$$$$$C@P$$$$$$

$$$

Match code generation process:

• Data is parsed into its components (Given Name,

Family Name, …)

• Ambiguities and noise words are removed (e.g. 'the')

• Transformations are made (e.g. 'Jonathon' 'John')

• Phonetics are applied (e.g. 'PH' 'F')

• Based on the sensitivity selection, the following occurs

• Relevant components are determined

• Only a certain number of characters of the

transformed relevant components are used


DATA DATA QUALITY, INTEGRATION AND ENTITY RESOLUTION

It’s all about dataPrecision

Complete

ness

Consist

encyStructure

Unicity

Validity

• End to end data management capabilities (cleansing, enrichment and quality improvement) added to a flexible and extensible data models

• Data quality is monitored over time

• Input data are profiled to easily detect, report, alert any anomaly and take immediate action (alert, reject, correction, enrichment)

Expected benefits of Data Management and Data Quality

Limit manual intervention and guaranty the relevance

of the alerts generated

Connect to internal and external data sources (Open

Corps, National Registrars, D&B, BVD, Bloomberg,

Reuters, other providers)

Accelerate the loading of the Data Model

Manage the entire data quality life cycle : a

continuous process for alerts accuracy

Ease the addition of new data

Provide an end to end consistent solution, from

the source systems to the alerts dashboard


SOCIAL NETWORK

ANALYSISENTITY RESOLUTION

John Smith 13/01/1980 18 Queen Street JN 12 34 56 A

John Smith 13/01/1980

John Smythe JN 12 34 56 A

J. Smith 18 Queen Street13/01/1980 0208 123 45676

Smith 0208 123 4567613/01/1980

Dis

pa

rate

da

ta s

ou

rce

s


ENTITY RESOLUTION





Smith 0208 123 4567613/01/1980

SOCIAL NETWORK

ANALYSIS


ENTITY RESOLUTION





Smith 0208 123 4567613/01/1980

SOCIAL NETWORK

ANALYSIS


SOCIAL NETWORK

ANALYSISNETWORK BUILDING

More entities are

resolved and link

together to form a

network


Everything is connected to

everything else…

But not all connections have

equal significance…

We may have lived at the same

address, but was it at the same time?

We may work for the same company

but there are 5,000 employees?

What if they only employ 5 people?


EXPLOITING HYBRID TECHNIQUES TO HIGHLIGHT INTERESTING SUB-NETWORKS

Reducing the Super Cluster

SocialNetworkAnalysis

NetworkAnalytics

NetworkScoring

BusinessRules

Analytics

AnomalyDetection

PredictiveModeling


3. USE CASES


ANALYTICS USE CASES

in various Fraud & Compliance domains

Payments Fraud

AML Peer Groups

modeling

AML False Positives

optimization

VAT Carrousel Tax Fraud

for government

KYC

Procurement Fraud

F

R

A

U

D

A

M

L


Raw Data K-Means (3 Clusters)

AML OPTIMIZATION

Use Case: Peer Group Modeling

Automated Peer Groups modeling

Raw Data K-Means (3 Clusters)

Result:

- Better detection granularity


• Incumbent AML Solution producing 13,000

alerts for 1 scenario over 12 months

• After optimisation 30% False Positives

Reduction with 100% SAR capture

AML OPTIMIZATION

Use Case: False Positives Reduction

Results:

- Better detection accuracy

- Better thresholds management

- Less false positives & negatives


FRAUD USE CASE

Belgian Tax Ministery

VAT Carrousel Fraud Eradication

Issues

• high velocity fraud

• 600.000 tax-payers

• 5.000.000 VAT returns

• 24.000.000 Intra-community transactions

Results

• VAT carrousel fraud reduced by 98% (from 1.1 Billion € to 29 Million €)

• Ultra-early detection

• SAS hybrid approach provides a high accuracy model

• (80% true positive rate)

SAS NET = Network Enrichment Tool

Build on SAS SNA

Search and visualize 3rd party data:230 Million Compagnies

135 Million Active Compagnies

98 Million Inactive Compagnies


5. THE CASE OF KYC : UBO AND CUSTOMER ONBOARDING


Various sources of information

IDENTIFICATION OF ULTIMATE BENEFICIARY OWNERS

Who’s who

Internal DBs

- Customers

- Alerts

- Cases

Sanction & PEP lists

National Registrar

International Chambers of Commerce

International Data

Providers (D&B, BVD, OpenCorps, Kyckr, etc.)

ICIJ

• Wikileaks

• Panama Papers

• Bahamas Papers

100.000

5.000

10.000

300.000

200.000

500.000• Many tick boxes

• Disparate DBs contain the information

• Data quality issues that prevent easy identification

• What about the GDPR which will make this even

more difficult to manage?

Great data management and exploitation

techniques are key

• Is this prospect really new?

• Does he have ties with other customers?

• Are there existing AML alerts or cases?

• What about the Credit loans DB?

• Sanctions lists / PEP lists

• What does the National Registrar say?

• Do I need to check iinformation from the

international chambers of commerce

• Do I need data providers to identtify UBOs &

Officers

• …


Various sources of information


Who’s who

Internal DBs

- Customers

- Alerts

- Cases

Sanction & PEP lists

National Registrar

International Chambers of Commerce

International Data

Providers (D&B, BVD, OpenCorps, Kyckr, etc.)

ICIJ

• Wikileaks

• Panama Papers

• Bahamas Papers

100.000

5.000

10.000

300.000

200.000

500.000

Data management must be combined with

exploration techniques such as

- Fuzzy Matching & Entity resolution

- Social Network Analytics

- Vizualisation


• Using national company data is just a start

• Using external international data is the next step companies ‘Out of Business’

‘non-corporate companies’, one-man companies

companies in ‘Sovereign Tax Havens’ (First criminal link)

companies in ‘BRIC countries’ (Brazil, Russia, India, China)

companies in ‘MINT countries’ (Mexico, Indonesia, Nigeria, Turkey)

‘Alternative Linkage’ (Minority Linkage <50%, one-man companies, franchise)

• Using ICIJ data brings you a step beyond

• Linking all this information is key for good

exploitation and investigation efficiency

ISSUE APPROACH AND RESULTS



ICIJ – Panama Papers – Bahama Papers – Wiki Leaks


• Bribery

• Corruption

• Illegal trade

• Money laundering

• Non-declared taxes

• Higher costs to society

• Economically not a level playing field

If you don’t find it, one will probably do it for you…


ISSUE APPROACH AND RESULTS


3rd party

ownership data(BvD, D&B, Open corporates...)


Company Confidential - For Internal Use Only

Copyright © 2016, SAS Insti tute Inc. Al l r ights reserved.36

AML END-TO-END SOLUTIONSa

nd

bo

x/M

od

el D

evel

op

men

t

Man

agem

ent

Info

rmat

ion

Rep

ort

ing

Ale

rt a

nd

Cas

e M

anag

emen

tSo

urc

e D

ata

Alert Generation

EnrichmentCase Management

Customers Accounts Trans ListsExternal

Data

Dat

a P

rep

SegmentationPeer

GroupingEntity

ResolutionNetwork Analysis

Text Analytics

Data Assessment

Ongoing Model

Monitoring

Exploration

Au

dit

Tra

il

Scenario Admin / ESP

Operational Dashboard

Regulatory Reporting

Simulation

Champion Challenger

Model Risk Documentation

Ad Hoc

Risk MgmtPortfolio Reports

Country Reporting

IntelligenceLoopback

SWIFT

SWIF

T

Gateway


CUSTOMER DUE DILIGENCE

37

A flexible framework is necessary to adapt to every client risk based approach

Attributes

Rules

Categories

Score

Rating

Categories:

Entity Type

Product/Service Type

Country Type

Transaction Type

Customer Risk Rating

Customer Risk Score

Entity Score (40)

Rule (20)

Attribute (20)

Rule (20)

Attribute (20)

Country Score (25)

Rule (25)

Attribute (10) Attribute (15)

Product Score (10)

Rule (10)

Attribute (10)

Transaction Score (25)

Rule (25)


CLIENT

ONBOARDINGWITHOUT ANALYTICS

1. Search external database

2. Identify all possible matches and

select the right ones

3. Retrieve all UBO’s and Officers

4. Investigate them

• 25% ? 10% ? Less ?

• How many levels down ?

• How much time do you have ?

• What level of certainty ?

• When will you be exhausted searching entities

where you find nothing…

5. Calculate the CDD risk score


CLIENT

ONBOARDINGWITH ANALYTICS

1. Search external database

2. Identify all possible matches and select the right

one

3. Retrieve all UBO’s and Officers

4. Use analytics to know where to search

• Fuzzy match all UBOs and officers with sanctions lists

• Fuzzy match them with the database of existing customers

• Check whether alerts and/or cases do exist for any of them

• Check whether any of the linked entities are compromised

with Panama Papers and the likes

5. Calculate the CDD risk score


6. DEMO

Call to an external data provider's API

Give the name of the prospect you’reonboarding

Call to an external data provider's API

Give the name of the prospect you’reonboarding

Retrieve the list of possible matches

Select the one that most likely matches

Select the one that most likely matches

Review the Company Details

Select your search options

Select the min. shareholding %

Review additional information about this company

Previous names

All known Shareholders automatically retrieved

All known Officers automaticallyretrieved and

matched againstSanctions lists

All known filings

Officers and shareholders are

automatically included in the social

network (sanction list matches in red!)

Known ownership information indicated by the link thickness

Inactive officers marked 'grey'

Associated household 'BOB J TAYLOR'

Customer 'BOB J TAYLOR' has a SAR filed against him

Officer 'Robert Michael TYLER' matches customer 'BOB J TAYLOR'

Scorecard details

Ongoing due diligence provides a detailed

scorecard

Workflow supporting enhanced due diligence

investigation


7. CONCLUSION


- Advanced Analytics

- Data lab approach

- White box

- Non-Levenshtein screening

- Optimization

- Risk scored networks

2.

3.

Standard AML package

- Transaction

- Screening Watchlists

- CDD

Data analysis (ETL/realtime)

Compliance

level

Core

Unique SAS value

Data Quality & Performance Detection efficiency

& data coverage

STEPPED APPROACH FOR ADDED VALUE IN COMPLIANCE

ROI

• Better detection

accuracy

• Increased

regulatory

compliance

• Reduced risk of

fines

• Manpower

management


Contact details

LAURENT COLOMBANT

FRAUD AND COMPLIANCE SPECIALIST

SOUTH WEST EUROPE

[email protected]

+33 6 16 41 62 43

ROLAND THEYS

FRAUD AND COMPLIANCE DIRECTOR

SOUTH WEST EUROPE

[email protected]

+32 474 94 00 19

Documents

LEVERAGING ANALYTICS IN COMPLIANCE … AND KYCLEVERAGING ANALYTICS IN COMPLIANCE … AND KYC LAURENT COLOMBANT FRAUD & COMPLIANCE SPECIALIST SOUTH WEST EUROPE ROLAND THEYS ... in various