Machine learning takes the fast track
Among the multitude of new technologies entering banking’s sphere, machine learning likely will quickly come to the fore.
Imagine some poor bank employee sitting at a desk, peering at multiple computer screens. On those screens come waves and waves of charts, transaction lists, voice recordings, geographical data points, and social media posts, all constantly changing—and all focused on a single bank customer.
The bank employee’s job, at the moment: Decide whether that customer actually used his own credit card to buy a big screen television, or if that transaction, which just occurred, is fraudulent.
And he only has a moment to decide. Seconds later, all that particular data will be erased and replaced with waves and waves of more data associated with yet another customer and yet another transaction somewhere that might or might not be suspicious.
A tall order, indeed, yet not so far-fetched in these days of near-real-time transactions, and ultra sophisticated cyber crooks using ever-increasing technology. Making it even more intense are regulatory mandates for banks to stay on top of suspicious activity monitoring, as well as internal imperatives to mitigate potential losses, avoid false alerts, and improve customer relationships.
The standard transaction monitoring practices using traditional analytic software, backed by human oversight, may no longer be enough to meet anti-fraud work in particular.
Increasingly, the challenge will be met by banks through the use of machine learning, a subset of artificial intelligence. Through software, all those waves and waves of data can be processed much faster and with more insight than any single human could possibly do. Elements of this are already in use, particularly in regard to fraud.
“We do have some software in our bank that has the ability to learn based on customer patterns and activities and transactions and use of devices and types of credit instruments, to learn about our customers’ activities, to anticipate when something looks unusual,” says Peter Graves, chief information officer, Independent Bank Co., Grand Rapids, Mich., in an interview with Banking Exchange. “That’s where you get this fraud pattern. When that pattern is broken there is a prediction that this has a high probability of being some kind of a fraud transaction because it is not what we would normally expect.” He declined to name the specific solution for competitive and security reasons.
Other bankers and analysts interviewed by Banking Exchange say much the same thing.
Sridhar Rajan, robotics and cognitive automation lead for Financial Services at Deloitte Consultants, puts it this way: “Machine learning uses technologies that can self-learn with little to no human intervention. They get better at what they need to do, the more information you feed them. The whole idea for machine learning is that those technologies don’t require constant human intervention to get better and better … They eventually mimic human judgment at high speed, high scale, and low cost.”
Put more simply, “Machine learning is a branch of artificial intelligence that utilizes data and algorithms to train software logic instead of programming that logic through explicit rules,” says Brad Stewart, senior vice-president, head of product, AI Enterprise Solutions, Wells Fargo.
Again, the fraud area is particularly suited for machine learning application, says Chris Nichols, chief strategy officer, CenterState Bank, Winter Haven, Fla.
