How to separate the wheat from the dangerous chaff in online lending? The good borrowers from the frightful fraudsters?
In the latest installment of Data Drivers, Karen Webster, Joe DeCosmo, chief analytics officer at Enova Decisions and Kevin King, director of Product Marketing at ID Analytics, told the story of why online lending and digital data analytics have to mesh well to weed out the bad apples — in only three data points, of course.
Data Point Number One: 66 percent
This is the percent of consumers who fail to make it all the way through a credit application online, walking away before the digital ink is dry.
King noted that the number “caught me off guard” upon digging into the data surrounding application abandonment, since most of his time is spent thinking about how to improve the underwriting process once an application has been completed. But, he added, figuring out how many people consider applications and then somehow back off “was pretty striking.”
The bulk of efforts, he said, remains focused on digital payments, where companies spend time and effort developing and then driving consumers toward digital channels. But at the same time, once the loan application process gets underway, a wave of data is requested — from social security numbers to addresses and the like. Consumers, said King, see that data request deluge and say “not right now, and maybe never.” Fraud data comes into play here, he said, in helping firms stanch that exodus.
One way is through “confirmed fraud,” which is paramount in building systems to combat malfeasance. He stated that IDA (shorthand for ID Analytics) has about 4.5 million confirmed fraud applications, where “we know the fraudster committed this name, this social, this address to try to get fraudulent access to credit.”
But he said there is also another form of fraudulent data that falls under the heading of identity “that goes together and helps you solve fraud problems” and solve abandonment of applications, too. He said that it is possible, with real and pertinent knowledge of identity, to take an application with 20 fields and shrink it down to three fields. That truncating can come with, for example, a name and the last four digits of a social security number, thus sparing the consumer undue friction.
Said DeCosmo of the 66 percent abandonment rate: “I was floored. I do not come from eCommerce and conversions. I come from lending and approval rates.” Much of the application data, he said, can be filled in with verified information, which in turn helps change the workflow and make the application process smoother — and gives the fraudster less opportunity to do, well, fraud.
But really determining who is the bad guy is both “art and science to the process,” said King.
Enova Decisions has adapted its workflow process to also add analytics on the backend to follow up with abandoned applications through targeted outreach to would-be consumers.
The end result, said DeCosmo, has been that Enova has seen increases in conversion rates on the order of 10 percent to 20 percent, depending on the lending product in focus.
Data Point Number Two: 66 Percent
No, you are not seeing double. In this case, the 66 percent refers to the higher frequency, relatively, of fraud attempts that are tied to the subprime lending market, as calculated by ID Analytics.
For fraudsters targeting subprime borrowers, there is an aspect to the business itself that is appealing, where not all of the players in the subprime space are as well defended as the national institutions against fraud.
The subprime market also serves as a testing ground, he said, especially in synthetic fraud (creating an identity from whole cloth). In fact, DeCosmo noted that synthetic fraud is one of the key areas keeping risk managers at the top 20 to 30 banks “up at night … it’s tough to stop, and you do not know how much of it [there] is.”
The profile of the subprime fraud attempt, said King, is one where they may be testing the limits of the data they have pilfered and “taking the identity for a test drive … they may ask for small loan amounts that keep them under the radar.”
Conversely, those fraudsters looking to really bilk the system have laden applications (and many applications) with the same pieces of data. “If I see the same information 20 times on credit applications,” he said, “in 24 hours, that is really high velocity.”
Concurred DeCosmo, “you need to have some strong [velocity] checks” that can uncover and point to the spike in data repetition. “The patterns change every day of every week.”
Data Point Number Three: 29 percent
This is a “hopeful statistic,” said Webster, as it is the percentage of savings in underwriting costs that are attributed to decision flow based on a U.K. brand of Enova Decisions. The improvement comes with knowing “when to pull different types of data,” starting with fraud checks and then moving onto credit, said DeCosmo.
Paying for data you don’t need can be expensive and delay the application process, he said. “You want to be able to access risks in an orderly way,” he said. Data such as working capital can be pulled as needed, say, as from a primary credit report backed up by a secondary one. “Some of it is common sense, but if you are new to [digital lending] these kinds of decision-making and waterfalls are critical.”
Said King, not having the right screens up front, along with digital decisioning, “adds fuel to the fraud fire.”