More than a Numbers Game Building Better Strategies that Leverage Unique Kinds of Data Dennis D’Ambrosio Senior Fraud Operations Analyst KeyBank Michael Holloway Senior Fraud Operations Analyst KeyBank © 2014 Fair Isaac Corporation. Confidential. This presentation is provided for the recipient only and cannot be reproduced or shared without Fair Isaac Corporation’s express consent. Lamar Shahbazian Sr. Director FICO Congratulations! Bank: Hello, this is your bank. The check you deposited is counterfeit and your account is now overdrawn $2,400. It can’t be, I won the Clearing house sweepstake. Image Source: http://www.bellasavvy.net/archives/49380 2 © 2014 Fair Isaac Corporation. Confidential. Check Fraud Despite declining check usage, check fraud is still ever present Source of Check Fraud Losses Trend in Noncash Payments Checks Percentage of Total Losses Debit Counterfeits 50 Forgeries 75 45 40 % of Total Losses Billions 35 30 25 20 15 10 50 25 5 0 0 2000 2003 2006 2009 2012 1Q06 1Q07 1Q08 1Q09 1Q10 1Q11 The 2013 Federal Reserve Payment Study. *RDIs include NSFs, closed accounts, refer to maker, stop payments, uncollected funds, remotely created checks, no account found, government reclamations, and other return loss reasons. 3 © 2014 Fair Isaac Corporation. Confidential. 1Q12 1Q13 Check Printing Past and Present Past ► MICR Ink required ► Limited 4 check printing resources © 2014 Fair Isaac Corporation. Confidential. Present ► Print anywhere ► Image exchange networks ► Mobile Deposit Check Detection Past and Present ► Past ► Feel the paper ► Look for the perforation ► Look for discoloration ► Hold it up to the light ► Present ► No Challenges… Paper to handle ► Black & white images 5 © 2014 Fair Isaac Corporation. Confidential. Image Analysis ► Image Fraud Detection ► Takes check fraud detection beyond the MICR line ► Signature and check stock analysis Address Payee Signature characteristics are extracted, weighed and compared using a neural network and fuzzy logic. Crossings Curves and Loops General design 6 © 2014 Fair Isaac Corporation. Confidential. MICR Data Upstrokes Enclosed Areas Data Elements ► Signatures and check stocks are digitized and scored ► ASV Result Codes = analysis outcomes: Signature OK, different, new, etc ► ASV Match Rates: 0–100 confidence score ► APIA Result Codes = analysis outcomes: OK, validation failed, new, etc ► APIA Match Rates: 0–100 confidence score ASV RESULT 10 7 © 2014 Fair Isaac Corporation. Confidential. ASV MATCHRATE 00 APIA RESULT 15 APIA MATCHRATE 08 Strategy Design Why Use Decision Trees? ► Run large volumes of data and “find the nuggets” ► Implement existing rules ► Profiling fraud against non-fraud allowed us to look for improvement ► Visual ► Easy 8 to explain © 2014 Fair Isaac Corporation. Confidential. Strategy Design Decision Trees Help Verify and Suggest Potential Improvements ► Image analysis output = result codes and match rate scores ► Rules drive alerts based on scores, account type, transaction data ► Strategy Trees validate rules and performance 9 © 2014 Fair Isaac Corporation. Confidential. What’s Next? ► System ► New 10 Upgrade—new analysis engine and risk indicators Strategy Design © 2014 Fair Isaac Corporation. Confidential. Wire Fraud 11 © 2014 Fair Isaac Corporation. Confidential. Wire Fraud “Wire Fraud is difficult to detect; comes in many different colors and flavors.” David Pollino Senior Vice Presitdent Enterprise Fraud Prevention Officer at Bank of the West 12 © 2014 Fair Isaac Corporation. Confidential. Wire Transfer Fraud Detection ► KeyBank implemented enterprise fraud detection application 2011 ► Moved from static rules to behavioral based application ► Over-compensated by implementing additional rules ► And the result was… Image from: www.obviouslyopinionated.com 13 © 2014 Fair Isaac Corporation. Confidential. Detection and Volume Challenges White-boarded rules ► White-boarded rules to eliminate unnecessary alerts from Q2 2012 ► Improvement but monthly volume was still too high Alerts Goal: Continue to reduce alerts while maintaining frauds identified Frauds 14 © 2014 Fair Isaac Corporation. Confidential. Looking for Needle in a Stack of Haystacks Less than 50 Frauds out of 960,000 Transactions ► Decided to pursue more extensive rule analysis using decision trees ► Built a transaction file of frauds and non-frauds ► Initially met via webex, then opted to have two day “bootcamp” ► Interactively dissected ‘pockets’ of fraud ► Constructed several rule-set options in a few hours ► Learning: Risks unique to specific channels became evident 15 © 2014 Fair Isaac Corporation. Confidential. Balancing Volume with Quality of Alerts ► Implemented new rules in Q3 and Q4 2013 ► Reduced alerts created by 39% Q3 over Q2, further reduced alerts by 45% Q4 over Q3 2013 Alerts ► Fine tuning in 2014 has continued to reduce alerts and increase detection rates 16 © 2014 Fair Isaac Corporation. Confidential. Frauds Stellar Results While alerts went down, averted losses went up ► New rules led to richer alerts for operations team ► Majority of all frauds were detected ► Ongoing analysis for continued optimization 17 © 2014 Fair Isaac Corporation. Confidential. Alerts Averted Next Steps ► Continue to evaluate rule effectiveness of rules on an ongoing basis ► Most recent rule project leveraging FICO® Analytic Modeler Decision Trees Professional software is for Mobile Deposit Fraud detection ► Card 18 rules on-deck © 2014 Fair Isaac Corporation. Confidential. Conclusion ► FICO’s analytic tools have transformed our fraud detection methodology; we use data to drive insights and empirically based decisions ► Our operations teams have a greater confidence in alerts; our rules look at several factors such as customer segment, channel, business logic in addition to score ► This 19 approach has yielded improved false positive rates and a reduction in fraud losses © 2014 Fair Isaac Corporation. Confidential. Thank You! Dennis D’Ambrosio Michael Holloway Lamar Shahbazian dennis_d'ambrosio@keybank.com michael_w_holloway@keybank.com lshahbazian@fico.com © 2014 Fair Isaac Corporation. Confidential. This presentation is provided for the recipient only and cannot be reproduced or shared without Fair Isaac Corporation’s express consent. Learn More at FICO World Related Sessions ►Product Showcase: Analytic Modeling in the Cloud Products in Solution Center ►FICO® Analytic Modeler Decision Tree Professional Experts at FICO World ►Lamar Shahbazian ►Jeff Dandridge Blogs ►www.fico.com/blog 21 © 2014 Fair Isaac Corporation. Confidential. Please rate this session online! Dennis D’Ambrosio dennis_d'ambrosio@keybank.com 22 © 2014 Fair Isaac Corporation. Confidential. Michael Holloway michael_w_holloway@keybank.com Lamar Shahbazian lshahbazian@fico.com