The window financial institutions (FIs) have to determine “good” customers from “bad” lasts milliseconds.
As fraudsters steal their unwitting victims’ online identities, intercept SMS messages, mask device locations to commit payments fraud, banks and other firms need to be able to spot “signs” hidden in the eCommerce deluge that can separate genuine transactions from fraudulent ones.
It’s a $40 billion problem, that, as Dave Excell, founder of Featurespace, told Karen Webster, needs deep learning networks and a range of automated advanced technologies and models to construct the best lines of defense against the fraudsters.
Two new patents, leveraging those advanced technologies, can help FIs pinpoint behavioral changes and identify high-risk behavior — stopping fraud and financial crime before it happens.
Filing For Patents
Featurespace said Monday (July 12) it had filed those two global patents, aimed at transforming network architecture and risk scoring to protect customers and accounts.
Drilling down into the mechanics of the patents, Excell delved into the first one, for Automated Deep Behavioral Networks. That architecture, which was invented and launched by Featurespace six months ago, is geared toward monitoring card-related activity and card payments.
He explained that the patent covers a “deep neural network architecture” that can extract meaning from a range of transactional data and the sequences in which those transactions occur.
He told Webster that over the past few years there’s been increasing adoption of deep learning technology, that mines and extracts meaning from images, text, video and voice that can be transcribed and analyzed.
The approach deviates from traditional models and fraud prevention architecture used in financial services, he said.
“Many detection systems work by requiring human experience and expertise to ‘guide’ the models in terms of understanding what ‘represents’ fraud or ‘represents’ good customer behavior,” Excell explained.
But the algorithms tied to the Automated Deep Behavioral Networks, he said, are able to learn those distinctions naturally from the data itself (and thus lessening the reliance on humans).
With that approach, he said, Featurespace has been able to see a 38 percent increase in the fraud the firm has been able to detect — while maintaining the same customer satisfaction levels, which indicates no additional friction is being introduced into the process.
In the case of extracting meaning from transactional sequences, he said, the models can examine how someone spends money or interacts with an online website.
Applying the technology within that domain set discovers unique attributes within that set, especially when looking at low-frequency, high-value transactions (that may have the biggest financial impact vis a vis fraudulent activities).
Given the inexorable movement of transactions toward real-time status, he said, automated, predictive modeling can help senders and receivers feel more confident about the security of money movement.
“Rather than needing to have a human in the loop,” he said, “the model can be there to automatically capture all of those transactions as they’re taking place.”
The second patent is for the company’s Behavioral Anomaly Score, which identifies anomalies in individual customer behavior — without having to have prior knowledge of context for that behavior.
The company said Monday that the Behavioral Anomaly Score lets FIs see just where a consumer’s behavior has “changed” and, in real time, construct models to track further changes, gauging risk and stopping fraud before it makes inroads into someone’s accounts.
Uncovering The Fraud Signals
The solution, Excell said, is to examine the sequence of activities that an individual is performing when making transactions.
“We can identify a behavior that is out of context for that individual and use it as a signal to identify fraud that is taking place that we historically have not seen,” said Excell. Thus, the fraudster who is attempting to mask themselves behind a legitimate persona in the midst of account takeovers would be quickly found out as, for example, an elderly consumer all of a sudden is spotted trying to make a large bitcoin purchase.
Against that backdrop, the solutions help financial institutions act in a proactive way — messaging the legitimate customer to verify that anomalous transaction, with a nod to the fact that they may be targeted by a potential scam (or, in other cases, anomalous behavior might simply be tied to, for example, buying birthday presents or items to celebrate a holiday).
Excell maintained that it’s always important to have a feedback loop with the customer to determine what a normally representative level of risk as humans are just being humans and acting on their own free will.
Along the way, said Excell, the FI is able to cement its relationship with the customer, increasing the individual’s trust in the enterprise.
The ingenuity of the fraudsters is a constant in eCommerce, said Excell, “and these kinds of techniques and models that are the best ways to stop them.”
We’re always on the lookout for opportunities to partner with innovators and disruptors.
Learn More