Since the onset of the COVID-19 pandemic, fraud attempts have nearly tripled
, with a dizzying assortment of new scams emerging as cyber criminals continue improving their game. But with a fresh outlook on risk scoring transactions and the emergence of updated deep-learning technology, new methods of mitigating fraud risks have arrived.
Artificial intelligence and machine Learning are not new technologies for banks and credit unions working to fight fraud. Models are programmed with a variety of inputs to analyze both monetary transaction data and non-monetary data (such as a login or password change), and then instantly calculate a score for each transaction based on assumed fraud risk. Financial institutions can set thresholds and automatically decline any transaction that scores as too high a risk.
These models are a significant improvement over the fraud detection tools that preceded them, but they often assign too many transactions with a mid-level score because they detect something pointing to potential fraud, but without full confidence. This area of mid-range scores is considered the “zone of uncertainty.”
Transactions in the zone of uncertainty present a challenge for FIs because, while there is still a chance of fraud, there is a harder tradeoff between catching fraud and impacting genuine customers.
So, the larger the zone of uncertainty, the more inaccurate the model and the larger the tradeoff between catching fraud and declining genuine customers. As accuracy improves to provide more extreme scores (both high and low), the zone of uncertainty decreases.
Minimizing the zone of uncertainty
As real-time payments expand, digital transaction volumes grow and customer demand for the instantaneous movement of money accelerates, minimizing the zone of uncertainty will become the new standard for FIs defending against scams, account takeover, card and payment fraud attacks.
One way to do this is by using updated deep learning technology to deploy multiple layers of neural networks to learn from large data sets. Going beyond input and output processes of traditional machine learning models, such technology can automatically discover the most important data elements to retain and how to compare past behavior with any new transaction. This allows them to remember events over different time periods and compare recent patterns against long-established behaviors.
By automatically discovering and retaining the most relevant data, updated deep learning technology can build its own logic for decision-making when separating fraudulent and genuine behaviors. This also includes greater detection of low-volume, high-value fraud, as well as high-volume, low-value fraud.
For example, if someone usually buys their groceries in small increments throughout the week, but the same account is seen purchasing a large amount of groceries soon after another recent purchase, traditional ML models could identify these patterns as long as the algorithm had been told specifically what to measure. With updated deep learning technology, the model can learn what is important to measure and will recognize and flag the high-value transaction combined with the abnormal frequency of the purchase.
By extracting and analyzing more nuances from the data, such a system can more accurately separate fraud from legitimate transactions. Transactions that previously would have obtained mid-level scores are now more likely to have a low- or high-level score. This reduces the zone of uncertainty and allows FIs to more confidently set thresholds to stop fraudulent transactions.
PWC’s Global Economic Crime and Fraud Survey 2020 found that 56 percent of consumers experienced fraud in the past 24 months. The Federal Trade Commission saw a 45 percent increase in identity theft and fraud reports in 2020 alone. As the threat of losing money to fraud attacks increases each year, customers look to their banks to protect them.
At the same time, customers expect hassle-free transactions. When conducting their daily banking business, like paying bills or making purchases, they don’t want to be overly questioned. Therefore, financial services providers and payment processors must strike that delicate balance of preventing fraudulent transactions while enabling customers to freely use their accounts.
By utilizing updated deep learning technology, FIs and payment processors have a new weapon to wield and can successfully decrease that all-too-familiar zone of uncertainty that comes with detecting, fighting and preventing fraud.
Dave Excell is founder of Featurespace.