How Machine Learning Is Setting ‘Payments On Fire’
There’s no need to panic, but machine learning is eating the world. Advances in data collection, storage, retrieval and data science are transforming the way many organizations do business. Feedzai’s CEO Nuno Sebastiao recently spoke to Glenbrook’s George Peabody for the “Payments on Fire” podcast about how machine learning is transforming fraud and risk management operations at leading consumer enterprises.
While machine learning is turning out to be incredibly useful, Nuno insists it’s not a silver bullet. Maximizing this technology’s outcomes requires carefully selecting applications and working in combination with human analysis.
Read on for some highlights from Nuno’s candid conversation:
The Promise Of Machine Learning
Nuno pointed out that the nature of commerce has now shifted. There is a proliferation of new channels, as well as transaction and payment methods. Companies are doing business with customers from multiple countries at the same time (e.g. China and the U.S.), and via multiple methods (e.g. phone line, app or website).
As a result, it’s become difficult to make sense of all this information and make decisions accordingly – e.g. Can I trust this order from an American customer at a China-based IP address? According to Nuno:
“The promise of machine learning is the ability to digest all of that information and still make sense of it once you determine what you’re looking for, i.e. your business goals.”
In the past, security software relied on broad, rules-based systems that had to be constantly managed and operated best in only single, narrowly defined channels. He explains:
“If you think about self-driving cars, what was there before was essentially the same. Imagine if you had to have rules for every single little road, for every single little cross-junction, every single traffic light, as opposed to telling the system that the general behavior for whenever you see a junction or whenever you see a red light, it needs to behave like that.”
This, in a nutshell, is what machine learning offers, implementing a generalized decision-making process by learning behavior rather than hard-coding specific cases. That being said, machine learning alone is not sufficient for every possible application. Today, the technology really performs well only in specific use cases. Nuno illustrates this by comparing to a self driving car:
“I’ll use the self-driving car example again, today you can only do it safely for some aspects, like driving on a highway, but not to drive on general roads or all the roads out there. Whereas any average driver can drive on all the roads. That (self driving) is the promise of machine learning, but there is also the care that we all need to take; it is not a one size fits all solution for all problems out there.”
Making Humans Successful
In enterprise deployments of machine learning, the machines help, but humans take the final call. For this reason, Feedzai is built using the concept of a ‘whitebox’ explanation. Instead of a black box spitting out a decision, it brings human judgment into the equation by presenting the information with plain English explanations of the underlying reasons. According to Nuno, that element of human judgment can make a huge difference in rooting out fraud without overwhelming the system with false positives.
“We basically say this is our opinion, it’s called a whitebox score; this is what we think of this transaction, and here’s why. It’s this semantic layer that basically can produce what we call human friendly outputs that explain and justify the underlying machine logic because you still need people to look at this.”
No one knows your industry or customers better than you, and businesses can make better decisions when they know why a transaction has or has not been approved by the system. Retailers are less likely to reject sales coming in from a new region, and banks won’t shut out legitimate parties from opening accounts.
The Need For Agility
As the notifications on your phone constantly remind you, business never sits still. Unfortunately, neither do threats. Nuno discussed the demands of maintaining a solution that can handle shifts in how you engage customers.
“What we see in our clients is the channels keep on changing – they keep on coming up with new ways of selling, [like] selling on social channels as opposed to just selling on mobile.”
For security solutions to stay ahead of breaches and fraud tactics, they need to be agile, taking into account both immediate factors and long-term trends. The systems must be based in models designed to evolve, rather than requiring time-consuming and expensive processes of rebuilding and retraining. And it is essential that these systems be data agnostic – any new channel is just a new data source that can be added to the risk engine, not something that requires its own separate model implementation.
Feedzai therefore analyzes data through multiple models based on the past year, month, week and even the last five minutes. Nuno explains:
“What we have is the ability to have multiple models that work together to make a decision.
We have models that are based on a year’s worth of data, and they are good to pick up things like seasonality trends. So if it’s summertime, how does it compare to summertime last year?
Then we have other models that look at, for instance, a month of data, and they’re good for us to understand how Monday compares against two Mondays ago. Or, how does the first Monday of the month compare against the first Monday of last month?
Then we have models that looks at a week’s worth of data, then a day’s worth of data, and even as little as one hour’s or five minutes’ worth of data.
Each of them pick up different signals, and it’s the combination of all of them that yields the best result”
Through this ensemble approach, the solution adapts to the constantly changing landscape of financial technology and fraud. Working across multiple platforms and channels, each machine learning model presents its own decision. Those decisions are then synthesized into a general recommendation.