What could possibly be worse for banks and retailers than fraud? According to Dr. Akli Adjaoute, CEO, Brighterion, it’s declines.
“What good is a system if it saves you $1 on fraud but loses you $100 in false declines?” Adjaoute asked in a recent episode of the “AI Myths” podcast with Karen Webster. “You hear about the cost of fraud, but declines — that is, false positives — are just as costly. It’s a nightmare for FIs. Every time you do that to a customer, you’re making him feel frustrated, because he’s trying to make a legitimate transaction.”
Indeed, Adjaoute said, financial institutions (FIs) lose more than money when the cards they issue get declined. They also lose trust. They have put a customer in an embarrassing situation, said Adjaoute, and with so many other cards to choose from, the customer will likely reach for one of those instead the next time.
The sweet spot, he said, is to reduce declines while offering the right service to each customer – but that’s no easy task. Customizing fraud protection for individuals is too big a job for people and too complex for older artificial intelligences (AIs), which apply the same logic to everybody — even though, as Adjaoute noted, everyone shops and spends differently.
It takes a powerful, modern AI, backed by ongoing machine learning, to better protect and serve their customers using personalized Smart-Agents. In “AI Myths,” Adjaoute explains why legacy AIs aren’t enough, and why it’s worth the effort for FIs to put a good AI to work on their behalf.
According to Adjaoute, antiquated fraud detection methods are the biggest anchor holding FIs back. Many FIs still depend on rules, statistics, neural networks and data mining to protect their customers. It’s not that these tools can’t catch fraudsters; it’s that turning them up high enough to do so also translates into higher decline rates, which is when the system erroneously interprets a legitimate customer’s activity as fraud.
Rules are rules, said Adjaoute, and there’s no flexibility or context. If criterion X is met, then the transaction must necessarily be fraudulent. But that’s not always the case. People spend differently, so different logic must be applied to their actions. What may look sketchy coming from Adjaoute may not look sketchy if his wife or son did the same thing.
As for neural networks, these consider attributes such as type of merchant and category to weigh whether a transaction is risky. But Adjaoute said these have the same problem as rules: Neural networks, too, apply the same logic to all customers across the board. That, he said, is where declines come from.
“The value of real AI tech,” said Adjaoute, “is that every single person is different. This allows FIs to know how you spend and your behavior — what is personalized and specific to you? That way, they can protect you well — and also serve you well with the kinds of products you want to see.”
A Good AI Makes Dollars and Sense
More data from more sources makes a smarter AI, as the greater system learns from smaller, individual interactions. Adjaoute called it “collective intelligence,” in which aggregated data across the portfolio of cardholders can combine to protect the entire system.
Cross-channel data creates a richer portrait of the consumer, so the FI can not only reduce declines, but also detect anomalies — again, at the individual level, as anomalous behavior varies just as widely as a person’s “usual” behavior.
Adjaoute said a truly smart AI can even predict life events based on changes in spending patterns. If someone just graduated, is getting married or is about to have a baby, good AI can figure that out and use the knowledge to better serve the customer.
Imagine an FI’s AI has identified a recent college graduate who is now working and moving toward marriage. That’s a customer who’s going to need a loan. Adjaoute said the AI creates the opportunity for the FI to get there and make the offer before someone else does.
On the flip side, imagine the AI discovered a previously trustworthy customer was no longer making payments like he used to. Perhaps the customer lost his job. Maybe he will have a new one before long. A blind legacy system could never predict that, but a good AI can support default prediction, showing whether someone who’s struggling to pay right now will ultimately come through to address that credit.
Imagine the customer satisfaction and loyalty the FI could create by continuing to serve that customer in his time of greatest need. That’s a much longer-term payoff than simply getting this month’s installment paid on time. Thus, said Adjaoute, it’s possible for FIs to make a profit and make their customers happy.
To sum it up, said Adjaoute: “When it’s personalized, you know better, you see better, you decline less and you protect better.”