I read Michael Brooks’ article, Doing Science the Wonga Way, with great interest.
I have had the algorithm Wonga uses to distinguish between applicants explained to me, and it is fascinating. It makes the most of the fact that the internet is replete with thousands of pieces of information about us that, in aggregate, paint a reasonable picture of who we are. More importantly for Wonga, they also paint a picture of how creditworthy we might be. It doesn’t take this lightly: I’ve been told it uses 6-8000 data points about the each of the people it checks.
The problem was that I only had the algorithm explained to me after its accuracy was seriously put in to question.
Wonga has a weekly survey of people who they consider to be good customers, and they brag about them to journalists. When the Guardian‘s Amelia Gentleman interviewed Errol Damelin, the chief executive of Wonga, he and his team had a chance to show that their model worked. When they put names of potential customers through their high-tech filters, the system ought to tell them whether they would be good customers. They would then only lend if it would be responsible to. After all, Wonga says that it turns away two-thirds of applicants.
Rather than the “web-savvy young professionals” that the company says that it lends to, one of the “good customers” on their weekly survey was Susan, an unemployed former nurse dependent on disability benefits. She uses the loans she receives from Wonga to buy food when she is short of cash. In fact, at the time of the Guardian interview, she had taken out 6 loans with Wonga, nearly double the amount of payday loans the average customer takes out (3.5).
We have two options here. Either we can assume Wonga purposefully targets people who are not median income, employed and web-savvy, unlike what they say, or their algorithm doesn’t work as well as they say.
In the same interview with the Guardian, John Morwood, Wonga’s communications director, said:
Sometimes we will make loans to people on significant benefits, but it is not something we do very frequently. It is very infrequent. I’m not going to say it doesn’t happen.
Dr Brooks is correct to say that the company has enjoyed some fantastic and enviable funding from several organisations. Last time I looked, Wonga were the beneficiaries of £3.7m from Balderton Capital in 2007, £14m from Accel Partners (also investors in Facebook) in 2009, then £73m from Oak Investment Partners, Meritech Partners and the Wellcome Trust.
I can’t be certain, but my assumption is that at least some of these backers are interested in Wonga as an example of good science put into action by business, and aren’t particularly interested in funding legal loansharking.
But Wonga’s algorithm clearly doesn’t alter the fundamentals of their business as much as they claim. Even with their flashy, investor-attracting scientific background, they still lend to people whose custom they admit they ought not to take.
Wonga itself is either misusing its own system to justify lending to people who should be served by less expensive lenders such as credit unions (which I think payday lenders should be obliged to advertise to low-income customers), or its algorithm needs a lot more work than it says.
As it stands, if the system confuses repeat borrowers who are unemployed and on benefits to buy food for people who are middle class, have bank accounts, are in full time employment and need the cash for minor financial shocks here and there, then there is a major issue.