Big Data • Big Analytics • Big Insight

February 29, 2012

The Profitability of Failure

Robert Gelber

Banks have lost at their own game. The long list of failures tells the story of stunted loans and bad bets. Instead of waiting for credit to flow again, however, a number of startups think they can do a better job at underwriting by tapping the power of machine learning algorithms.

Machine learning is gaining unprecedented attention these days due to the buzz around big data analytics. The technology’s ability to draw information from disparate data sources—all the while learning from experience–has companies like Wonga jumping into the world of alternative lending.

The company began as a same-day cash outfit, and claims to be the Internet’s first fully automated creditor. It took just days after startup for a borrower to default on a micro-loan, but this was all part of the plan.

For the first year, SameDayCash was feeding data to the model that would eventually run Wonga. The service did more than just check for a credit scores. It pulled as much data about its borrowers as possible, including online profile information.

After learning from the SameDayCash project, which at one point saw a 50% default rate, Wonga began issuing short-term loans. In one year, it issued 100,000 loans amounting to £20 million. The company earned £15 million through interest payments, while the banks missed out again.

Even though the company has a 95 percent satisfaction rate, there are some vocal critics about how the company operates. Mainly, the Representative 4214% APR printed on the front page of their website (Wonga says the number is closer to 360% APR) has proven controversial. 

On that note, some even joked that an individual could be on the hook for more than the U.S. national debt after seven years of non-payment. To Wonga’s defense, the website simply does not allow anyone to borrow for a year.

Wonga is not alone in the field of micro lending, other companies like Zestcash, Kabbage, and OnDeck all run similar, small lending operations that rely on machine-learning to vet borrowers.

One area most of these companies have in common though, is an inability to provide solid, long-term loans. Almost all of them only offer short-term, unsecured, subprime loans. To move into the realm of long-term prime loans, the companies have to be willing to burn precious capital while their algorithms learn what makes a good home loan.

As the banks sort through the mess of bad paperwork and failed loans from the housing crisis, machine learning may get a chance to prove itself as the only way to run risk analysis.

Related Stories

Research Spawns “Anomalator” to Fight Financial Collapse

Financial Firms See Big Data’s Big Picture

Netuitive Banking on Predictive Analytics