I believe most people who own credit cards have experienced fraud or false positives (a non-fraud transaction declined), and so have I. Three months ago, I experienced overseas credit card fraud with an amount around $3,000 USD; my bank is still investigating the case. What gave me a deeper impression was the complicated procedures of talking with the bank to apply for a refund of the fraud. I have submitted 10 different kinds of paper proof to the bank and had at least six calls with the bank’s representative. One kind of paperwork the bank requested was the police investigation record, which means I had to go to the police station to report the fraud case, so I went to the police station twice . . . You would think that I must be getting annoyed with my bank for these procedures full of hassles; I don’t feel this way, however, since I know the bank can do little about it — they are just like me, the victims of the increasingly sophisticated fraud activities empowered by technologies and the internet.
What is more interesting is that I was happening to conduct research about AI’s impact on fraud management just at the same time as I encountered credit card fraud. From a victim’s perspective, I got a deeper understanding of how AI can potentially transform the fraud management industry and address the pains of vulnerable customers as well as banks. Thanks to my colleagues Andras Cser and Danny Mu, who are the coauthors of the report, we published “Artificial Intelligence Is Transforming Fraud Management” earlier this month, and I do believe it will give both AI and fraud vendors as well as financial institutions lots of inspiration on how to envolve their existing fraud detection mechanisms.
We identified and summarized several key insights:
1. AI is a trend in the fraud management industry. With the development of technologies and the penetration of the internet, current fraud activities are way smarter and at lower costs than before. The overseas credit card fraud case I have experienced is a typical example showing that criminals have sophisticated fraud tricks that can go cross-border easily. Also, with the exponential growth of customers’ digital transaction data, the traditional rule-based fraud detection models are having increasing difficulty meeting the requirement. AI can augment the existing rule-based models and strengthen human fraud analysts significantly, which can improve accuracy and efficiency while reducing costs.
2. Fraud management use cases have common technical requirements for AI algorithms. However, each use case prioritizes these technical requirements differently: For instance, transaction monitoring requires the highest level of response time, training data availability and quality, error rates and precision, explicability, and ease of model building, compared to other use cases such as fraud investigation and reporting.
3. Select the right technology for each fraud use case. AI includes different algorithms such as supervised learning, unsupervised learning, knowledge graphic, etc. Different algorithms are suitable for different fraud detection use cases. You can match each AI algorithm to common use cases. Case in point, for the use case of transaction monitoring that requires exceptional performance and accuracy, the supervised learning algorithm dominates. The use case of reporting requires visualization tools. Clustering algorithms based on unsupervised learning is a good fit here.
For all these key insights, we have deeper analysis and comprehensive charts and real-world examples to support them. Please feel free to read my report, and ask questions by emailing email@example.com if any.
Finally, I would like to wish everyone a Happy New Year and hope 2020 will have less fraud with the power of AI.