Beware fraudsters: Evolutionary AI is going to catch-you
By Nate Greenhunt, Client Solutions Executive
With the rise in bank fraud, fraudsters are targeting bank customers in an attempt to fool them into revealing their confidential banking details, sometimes even using advanced AI algorithms such as evolutionary computation of their own. Scams like ATM card skimming and mobile SIM card swap have already bilked thousands of bank customers worldwide. Now, their scams are carried out in the form of vishing. The fraudster commits this scam entirely through a phone call and before the victim realizes, his or her money is already taken. Fraud takes other forms too, online and through text messaging. Today, professional fraudsters are even using computational intelligence to make smart decisions with the help of artificial intelligence and machine learning to calculate who to victimize. These scams are easy to arrange, replicate and execute.
A bank will traditionally look at the transactional data and master data for fraud. Additionally, they will examine third party data such as known fraud schemes including “hot” lists for people, products, location or watch list data. When a bank is looking for potentially fraudulent payments, they will need to look at multiple data sources and combine multiple analytics and algorithms. Simple or traditional rules-based tests and common algorithms don’t perform as well in today’s world because fraudsters have grown more sophisticated and can evade these.
In today’s world, banks cannot just look at a single data source or algorithm to solve fraud challenges. Multiple data sources and algorithms need to be considered to get ahead of the fraud curve. As we catch more fraud, fraudsters get more creative and find new algorithms. With so much data and the advanced algorithms available to banks, fraud units will need to get more creative with AI and machine learning, and even AutoML, to uncover deeper fraud schemes and scams, find new pre-indicators of fraud on accounts and identify the genome of a fraudster. Advancements in evolutionary algorithms are enabling us to look at data from a diversity of viewpoints and linkages to uncover hidden patterns and fraud scams.
Surrogate modeling uses creativity to identify fraud
The key factor in driving innovation and defining a strategy to detect and prevent fraud is creativity. Businesses need systems capable of harnessing creativity and converting large amounts of data into actionable and customizable solutions to stop fraud at the door. Deep learning and evolutionary algorithms are transforming artificial intelligence and machine learning and demonstrating how banks can make a difference in the real world.
Evolutionary computation is a suite of algorithms, inspired by biological evolution, that allow the efficient search of large and complex spaces. Evolutionary computing uses rapid and repeated generations of tests to arrive at viable solutions to the most daunting fraud problems, where traditional machine learning falls short. These technologies are making it possible to discover and even prevent frauds in process. A surrogate model allows for the discovery of indicators of behavior. For example, evolutionary methods can identify indicators of vishing scams and suggest steps to minimize fraud. Additionally, evolutionary algorithms can be used in the vishing scam example to more effectively investigate and hunt down the perpetrators of fraud.
We are applying these principles today on recent historical data at financial institutions to reveal hidden insights and block fraudsters. At a large global bank, we have collaborated with their team to create a credit card fraud analytics platform. This resulted in $60M in reduced fraud losses, a reduction of over $100 million in merchant loss, and hundreds of thousands fewer false positives.
What we have learned by stopping fraud
Moving businesses toward more AI-native architectures and embedding AI and analytics into their operations will help reduce fraud. Below are the ways evolutionary computation can be applied to greatly improve detection and intervention:
- Improve predictions: Evolving deep network architectures and hyper-parameters automatically design evolutionary AI solutions that are more efficient and effective, reducing data scientist’s time from weeks to hours.
- Drive impactful outcomes: Iteratively improve decision prescription systems with a goal to catch fraud earlier in the process, before it becomes a large problem.
- Facilitate effective experimentation: Augment or replace processes using principled AI and ML learning-based approaches to stay one step ahead of fraudsters who are using their own technology solutions to defraud.
- Insights are more difficult to glean from the data: Scams run by sophisticated fraudsters leave as little data behind as possible; this makes them difficult to track. Evolutionary algorithms detect insights on smaller data or even sparse data to develop prescriptive solutions that enable them to more effectively identify the perpetrators.
- Nuanced analysis gives more information to humans to make quicker, better-informed judgments: These new, deeper insights provide more context and detail to investigators reviewing fraud cases and reduce false positives. This makes every action and reaction against fraudsters much more potent.
A new approach to data science
Evolutionary computation builds digital surrogates capable of uncovering fraud schemes and scams, even on small data sets. When we talk about a minimum viable data set, it is important to note that evolutionary systems are able to predict and prescribe even in the presence of smaller datasets. Thus, we can look at smaller data sets and still find insights and fraud schemes that are typically more challenging to find to get ahead of the fraud curve.
The issues around identifying targeted analytics for fraud are complex due to the nature of how it is committed. Not surprisingly, lawbreakers want to hide their tracks. This means that data scientists need to evaluate all the combinations of data sets and algorithms, and that’s a lot of data to analyze.
A typical bank needs many PhDs, business analysts, data scientists and fraud specialists to do this. Evolutionary AITM generates complex models and discovers novel decision-making strategies automatically, by taking advantage of massive data sets and distributed computing capacity. By reducing the amount of time data scientists and analysts spend looking for patterns and anomalies, real-time detection and prevention becomes a reality.
Getting ahead of the fraud curve
Preventing fraud requires a more proactive strategy. Looking at historical data will only identify fraud that mimics the past, not new types of fraud. Conventional approaches and wisdom are outdated and not sufficient to catch constantly evolving fraudsters. Evolutionary computation looks at data in new ways so fraud schemes and scams are found quicker and, in many cases, interrupt unlawful transactions in process.
Cognizant’s Evolutionary AI team develops models that can get ahead of fraudsters. We have deep insights into complex challenges and quickly understand what matters most and least in your data. Feel free to comment below if you have any questions for me or click here to learn more about Cognizant Evolutionary AI.
About the Author
Nate is a Client Solutions Executive at Cognizant. He has over 16 years of experience in cognitive systems, artificial intelligence, decision support systems, large scale system modernization and transformation, and continuous improvement programs.
His area of expertise is in banking and financial services. Nate holds a Bachelor of Science, Master of Science and Master of Business Administration from Rutgers University and is a Certified Fraud Examiner. Currently, Nate is enrolled in Harvard’s Business Analytics Program.