Is It Fraud or New Behavior? Two Analytics to Tell the Difference
Is It Fraud or New Behavior? Can you tell the difference? Tom, a tech entrepreneur, purchases an expensive snowboard in Aspen, Colorado a transaction that looks odd compared to his historical behavior patterns. Is it the first sign his card has been compromised by a fraudster? Or is this just new behavior by a legitimate cardholder? An instant decision with more at stake for your bank than fraud prevention Reducing false positives legitimate transactions referred or declined as suspicious has a direct, immediate impact on your cardholders perceptions of service quality. And while Tom will appreciate being alerted if fraud is underway, too many inquiries will make him wonder 2014 Fair Isaac Corporation. All rights reserved. 2
Does this bank even know who I am?!!! 2014 Fair Isaac Corporation. All rights reserved. 3
Two ways to turn Tom s perception around New analytics help banks protect his account and customer experience by: 1 Understanding more about what Tom typically does 2 Understanding what Tom doesn t do (but is likely to in the future) 2014 Fair Isaac Corporation. All rights reserved. 4
1 Understand in detail what the cardholder typically does Tom doesn t typically purchase expensive snowboards in Aspen A fraud model would likely score Tom s snowboard transaction as suspicious and might decline the authorization. Trained on months or years of transactional data, it would detect high risk based on not only the unusual purchase type and location (Tom lives in New York), but historical behavior patterns of other cardholders with similar profiles. But what if the fraud model could take into consideration not only historical behavior patterns but the most current behavior going on in the production environment? 2014 Fair Isaac Corporation. All rights reserved. 5
1 Understand in detail what the cardholder typically does Adaptive models are sensitive to short-term changes in purchasing behavior An adaptive model, working in conjunction with the fraud model, refines detection by comparing Tom s transaction to other recent activity both known fraud and non-fraud data. It turns out it s not just Tom with a new-found interest in snowboarding. The adaptive model has seen lots of sporting goods purchases like this recently. Cases generated on other cardholders with similar profiles were determined to be non-fraud. 2014 Fair Isaac Corporation. All rights reserved. 6
1 Understand in detail what the cardholder typically does Favorite ATM and cash withdrawal amount Favorite gas station, spend amount and frequency Favorite Wi-Fi Wi-fi cafe Here s another way to understand Tom s typical behavior: Like all consumers, he s a creature of habit Favorite travel city He may have a favorite ATM and specific amounts he typically withdraws at certain times of the and hotel day or week. Favorite stores and online vendors for certain types of purchases. Favorite travel destinations and hotels. FICO s Behavior Sorted Lists keep track of these favorites as they change over time They add power to traditional cardholder profiles by providing a more detailed view of typical transactional patterns. That improves your ability to distinguish between normal and suspicious behavior patterns. Here s how 2014 Fair Isaac Corporation. All rights reserved. 7
1 Understand in detail what the cardholder typically does How Risky? 2:00 am $200.00 ATM withdrawal Behavior Sorted that would Lists generally update favorites look suspicious with each is clearly transaction legitimate for some individuals POS Online banking ATM Mobile payments Advantages: Megan: Approve The 1 transaction ATM_77 is in-pattern: F1=3.2 2 taking place at one of Megan s favorite ATMs 2 ATM_318 F2=9.2 1 (near the restaurant where she works) at a favorite time of day (after her shift) Index Entity Frequency Rank 3 ATM_291 F3=0.3 4 4 ATM_54 F4=2.7 3 Frank: Contact The transaction is out-of-pattern: not matching any of Frank s favorites Tracks virtually any type of entity (zip codes, cell phone numbers, bank accounts, countries, etc.) Allows patterns of favorites to evolve over time Rankings change with frequency Measures variance of in-pattern behaviors Enables creation of complex variables, such as what percentage of last 5 transactions had a rank of 3 or above? X 2014 Fair Isaac Corporation. All rights reserved. 8
1 Understand in detail what the cardholder typically does Fraud risk is lower if even some aspects of the transaction match cardholder favorites When Sarah purchases an expensive toy during a business trip to New York City, the transaction goes through without a hitch. One of Sarah s favorite purchase categories in Phoenix, where she lives, is high-end children s toys and clothing. Even with no favorites match, there is a way to anticipate new cardholder behavior Remember our snowboarder Tom? He s never purchased snowboarding or skiing equipment, and he s never been to Aspen before. Still, fraud analytics predict that this behavior is likely Flip the page to find out about innovative analytic technology being developed by FICO 2014 Fair Isaac Corporation. All rights reserved. 9
2 Understand what the cardholder doesn t do (but is likely to) People with similar characteristics tend to behave in similar ways Tom shares characteristics with some other cardholders. POS ATM Sports enthusiast Artsy Home improver Tech lover Eco conscious FICO s Collaborative Profiles generate archetypes from these global behavioral patterns ACH Online Mobile Transactional data streams from any number of sources Any mix of data combined into unstructured document Advanced algorithm discovers archetypes by finding similarities in customer behavior When this new technology, currently under development at FICO, is incorporated into fraud detection, fraud models will be able to accurately predict new customer behavior. That includes transactional behavior never before seen on Tom s account, but which is nevertheless likely based on the archetypes he fits. Here s how 2014 Fair Isaac Corporation. All rights reserved. 10
2 Understand what the cardholder doesn t do (but is likely to) 35.2% 1.3% 22.6% 40.5% 0.1% 38.7% 0.2% 25.1% 38.4% 0.2% Actual customer behavior is mapped to archetypes in real time With each transaction, analytics update the customer s allocation of behaviors across archetypes. Actual customer Tom Real-time updates from transaction streams 34.9% 0.5% 27.2% 42.1% 0.8% Dynamic mapping of Tom s behavior Is Tom making the transaction? Or is it a fraudster? Current distribution 34.9% 0.5% 27.2% 42.1% 0.8%! How wide is the difference?! Distribution if transaction goes through 40.2% 0.2% 25.6% 37.9% 0.3% The wider the change in archetype distribution, the riskier the transaction Is this Tom buying the snowboard? Or is it a fraudster who has stolen his card or data? Let s find out 2014 Fair Isaac Corporation. All rights reserved. 11
2 Understand what the cardholder doesn t do (but is likely to) 34.9% 0.5% 27.2% 42.1% 0.8% Distribution if transaction goes through 40.2% 32.4% 0.2% 0.5% 25.6% 33.5% 37.9% 49.3% 0.3% 0.5% It s probably Tom, since buying What if he d bought a sound system? a snowboard is something he s statistically likely to do Tom also strongly matches Home Improver, which has a propensity for electronics. He has a strong allocation to Sports Enthusiast, which has a high propensity to spend on this kind of equipment. The fraud model incorporates that information and lowers the fraud risk score. X 34.9% 0.5% 27.2% 42.1% 0.8% On the other hand, he s not likely to buy a painting from a gallery in Soho Tom has a very low allocation to the archetype, Artsy, with its propensity to purchase art and high fashion. Distribution if transaction goes through 28.2% 20.5% 29.6% 40.3% 0.3% 2014 Fair Isaac Corporation. All rights reserved. 12
1 + 2 Combine for even more valuable customer insights What impressed us most was that, even as an industry leader, FICO has not become complacent, but continues to Start with the benefits Behavior Sorted Lists bring to fraud management Higher detection rates, lower false positive rates, happier customers. Later gain additional benefits with Collaborative Profiles. These new analytic techniques will work together to help you do even more to protect customer accounts and deliver a superior customer experience For example, there might be a big change in Behavior Sorted Lists but not in Collaborative Profiles. This may be a sign that the customer is traveling, or moving or has changed jobs. Gradual rather than abrupt changes in both analytics may indicate that a longer-term change, such as the adoption of a more health-conscious lifestyle and/or rising prosperity, is underway. Such insights will be valuable beyond fraud management. In fact, you will be able to innovate. CEB TowerGroup 2012 2014 Fair Isaac Corporation. All rights reserved. 13
Understand not only who your customer is, but who they re becoming Do that, and you ll nail fraud detection and deliver a customer experience that s nothing less than thrilling Learn more: Download the white papers on this topic: Insights #69: Is It Fraud? Or New Behavior? Insight #7: How Can Fraud Models Combat New Tricks? Check out our blog: Banking Analytics Blog For more information www.fico.com North America toll-free Latin America & Caribbean Europe, Middle East & Africa Asia Pacific +1 888 342 6336 +55 11 5189 8222 +44 (0) 207 940 8718 +65 6422 7700 info@fico.com LAC_info@fico.com emeainfo@fico.com infoasia@fico.com FICO and Make every decision count are trademarks or registered trademarks of Fair Isaac Corporation in the United States and in other countries. Other product and company names herein may be trademarks of their respective owners. 2014 Fair Isaac Corporation. All rights reserved. 3045K 1/14 PDF