fraud risk management in practice - université libre de...
TRANSCRIPT
Atos, Atos and fish symbol, Atos Origin and fish symbol, Atos Consulting, and the fish itself are registered trademarks of Atos Origin SA. August 2006
© 2006 Atos Origin. Confidential information owned by Atos Origin, to be used by the recipient only. This document or any part of it, may not be reproduced, copied,
circulated and/or distributed nor quoted without prior written approval from Atos Origin.
Fraud Risk Management In Practice
Olivier Caelen – Business Analytics Competence Center
17 April 2012
ULB Séminaire – Questions Actuelles d'Informatique
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 2
Agenda
ATOS Worldline
FRM Strategy
FRM Infrastructure
Staffing
Q&A
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 3
Agenda
ATOS Worldline
FRM Strategy
FRM Infrastructure
Staffing
Q&A
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 4
Subsidiary of
an IT Company
Specialized in critical
electronic transaction
processing
End to end
service provider
Atos Worldline is…
Hi-Tech Transactional Services
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 5
65% 24% 11%
Hi-Tech Transactional Services
Electronic PaymentseCS
Customer, Citizen &
e-Community Services
Financial Markets
Global Turnover = 844m€
4800 employees
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 6
Electronic Payments
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 7
ATOS Worldline in Electronic Payments
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 8
Agenda
ATOS Worldline
FRM Strategy
FRM Infrastructure
Staffing
Q&A
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 9
FRM Strategy
Awareness
Technical infrastructure
EMV
3D-Secure
Analysis
Transaction scoring
Case investigation
(Mass-) blocking
Cardstop
Chargeback
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 10
FRM Strategy In Practice
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Breakdown by Fraud Type (Credit - Issuing)
INTERNET
MOTO
CF
NRI
LST/STL
Counterfeit Lost/Stolen Internet
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 11
FRM Detection Strategy
Centralization
Independence
Embedded flexibility
Issuer Acquirer
Counterfeit Lost/Stolen Internet
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 12
FRM Detection Strategy In Practice
Skimming Detection
Skimming Abuse
2002
Skimming Abuse
2004
Abuse
2007
Skimming
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 13
Agenda
ATOS Worldline
FRM Strategy
FRM Infrastructure
Staffing
Q&A
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 14
FRM Infrastructure
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 15
Daily Average: 170.000 people
use their Credit Card.
Daily Average: 17 Fraud Cases.
Inte
llig
en
t
Filt
er
Expert Driven Fraud Detection
Translate Domain
Expertise into Rules
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 16
Expert Driven Fraud Detection - Approach
N_Trans
Tot_Amt
x
x
x x
x
x
x
x
xoo
o
oo
o
ooo
oo
o
o
o
o
o
o
o
o
o
o
o
o
50 1000
1000
2000
3000
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 17
Expert Driven Fraud Detection - Approach
» If N_Trans > 80 and Tot_Amt > 2000 then ‘x’
N_Trans
Tot_Amt
x
x
x x
x
x
x
x
xoo
o
oo
o
ooo
oo
o
o
o
o
o
o
o
o
o
o
o
o
50 1000
1000
2000
3000
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 18
Expert Driven Fraud Detection – Advantages
and Limitations
» Advantages
› Fast development
› Easy to understand
› Explain why an alert was generated
› Exploit Domain Expert knowledge
» Limitations
› Ask 7 experts, get 7 opinions
› Hard boundaries
• From Tot_Amt = 10 to Tot_Amt = 1.990 changes nothing
• From Tot_Amt = 1.990 to Tot_Amt = 2.010 changes everything
› Space is ‘boxed’
› Humans have difficulties thinking in more than 3 dimensions
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 19
Expert Driven Fraud Detection –
Limitations, less easy
N_Trans
Tot_Amt
x
x
x x
x
x
x
x
xoo
x
ox
o
oox
oo
o
o
o
o
o
o
o
o
x
o
o
o
50 1000
1000
2000
3000
x
xx
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 20
Expert Driven Fraud Detection –
Limitations, less easy
x
x
x
N_Trans
Tot_Amt
x
x
x x
x
x
x
x
xoo
o
o
oox
oo
o
o
o
o
o
o
o
o
o
o
o
50 1000
1000
2000
3000
x
xx
If (N_Trans > 80 and Tot_Amt > 2000) or (N_Trans < 40 and Tot_Amt < 1500) then ‘x’
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 21
Expert Driven Fraud Detection –
Limitations, less easy
N_Trans
Tot_Amt
x
x
x x
x
x
x
x
xoo
x
ox
o
oox
oo
o
o
o
o
o
o
o
o
x
o
o
o
50 1000
1000
2000
3000
x
xx
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 22
Expert Driven Fraud Detection –
Limitations, hard
N_Trans
Tot_Amt
x
x
x
x
x
x
x
o
o
o
o
o
o
o
o
o
o
o
o
o
50 1000
1000
2000
3000
x
x
x
x x
xx x
xx
x
x
x x
x
x
xx
x
x
oo
o
o
o
oo
o
o
o
o
o
oo
o o
o
x
x
x
xo
o
o
o
o
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 23
Expert Driven Fraud Detection –
Limitations, hardest
» Think of the previous shape in 6 dimensions.
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 24
The Fraud Detection Problem
» Fraud behaviour is complex
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 25
Data Driven Fraud Detection
Daily Average: 170.000 people
use their Credit Card.
Daily Average: 17 Fraud Cases.
Inte
llig
en
t
Filt
er
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 26
Data Driven Fraud Detection – Data Mining
» ‘Data’› Large volumes› Complex, high dimensional
» ‘Mining’› Actionable information› Obscure, difficult to uncover
Extracting actionable information from large volumes of data
» In our case: Use historical information as examples to build a model that distinguishes between fraud and genuine examples. Score new information with the model to estimate the probability of fraud.
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 27
Data Driven Fraud Detection – Advantages
» Advantages› Optimal model based on all available cases
› Robust, i.e. their response is smoother than rules
0.0 5.00
0.5
0.0
-5.0
1.0
0.0 5.00
0.5
Rules
0.0
-5.0
1.0
Neural Network
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 28
Data Driven Fraud Detection – Advantages
» Can model complex shapes
» Scale naturally to high dimensional problems
Examples Model
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 29
Data Driven Fraud Detection – Limitations
» Limitations› Need examples
› Development needs specialised
• Resources (data management, data mining, domain knowledge, …)
• Software
• Time
› For some modelling technologies, no explanation why the alert was generated is
available
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 30
Conclusion
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 31
Agenda
ATOS Worldline
FRM Strategy
FRM Infrastructure
Staffing
Q&A
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 32
Staffing
» The Business Analytics Competence Center
» A small team of highly skilled professionals in
› Data Management
› Data Mining
› Reporting
» Our Mission: Turn company data into value for us and for our customers
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 33
Agenda
ATOS Worldline
FRM Strategy
FRM Infrastructure
Staffing
Q&A
HIGH-TECH TRANSACTIONAL SERVICESFRM In Practice – 34
Thank you for your attention