The mass collection and sale of personal data is an increasingly sensitive topic for consumers and business leaders. Promising technologies such as artificial intelligence and machine learning depend on gathering lots of data. But as people have seen their data misused by some tech firms everyone within the industry is struggling to locate the line between reasonable data collection and invasion of privacy.
The whereabouts of that dividing line depend on what’s at stake. New research suggests that people are surprisingly comfortable sharing their data as long as it’s being put to good use, especially in the world of lending. A recent Harris Poll found a surprising consensus among American adults — 71% of those surveyed said they would be willing to share more personal data with a lender if it led to a fairer loan decision. (The survey was commissioned by my company, ZestFinance.)
More significantly, the Harris Poll shows that the majority of consumers feel unfairly treated by the current system. More than 80% of African-Americans and Hispanics – and 7 in 10 of all adults – say they wish there were a better way to prove themselves to lenders. Why is this? Broadly speaking people seeking a credit card, auto loan, mortgage, or any number of products are judged by one simple number: their credit score.
Credit scoring began as a way to more easily conduct business as the country became more populous and urban, meaning it was less likely a store manager would know all of their customers by name. Retailers then started combining files about people into regional credit bureaus which, over time, have been consolidated into three national bureaus — Experian, Equifax, and TransUnion — who rely on the methodologies of Fair Isaac Corp., developer of the ubiquitous FICO scoring system. As a group, they made great strides in applying scientific rigor to standardizing credit files and reducing the kind of subjective bias that consumers once faced.
But in today’s era of large-scale data collection and unlimited computing power, the traditional credit score is showing signs of age and mathematical limitations. Traditional scoring models only consider items directly related to credit actions – that is, how much of your credit lines you’ve used or if you’ve had a late payment in the past six months. Consider some of the things credit scoring excludes: your job history, your roots in your town and thousands of other bits of information that are left untapped. It’s no wonder 70% of people polled say it’s difficult to get financial institutions to see them as anything but a number between 300 and 850.
And many people believe that three-digit perception holds them back. More than 60% of Gen Xers and Millennials told Harris the current scoring system sets them up to fail. One-third of renters say credit scores have kept them from buying a home and most folks who get loans said they don’t know why they received their interest rate. Even worse, almost 20% of Americans are unbanked or have no credit history at all.
There is evidence that people with no credit histories can still be good risks. In California, Oakland Community Check Cashing, a nonprofit servicing unbanked residents, failed to collect on just $9,900 of $1.8 million payday loans in 2017 – a default rate well under one percent. Improving credit evaluation can only help bring these people into the financial system.
The nonprofit Center for Financial Inclusion has done extensive research about the benefits of improving banking access. Its conclusion: to get more people into the banking system, lenders need to embrace alternative data. Alternative data simply means anything that’s not included in calculating a traditional credit score. It could include timely utility payments, history of holding down a job, even large debts you had years ago (outside of the traditional credit score models) that you successful paid off.
Bringing that added data into the credit scoring decision creates a more nuanced picture of borrower risk, helping banks spot worthy borrowers further down the credit spectrum (and flagging troublesome borrowers who look good on paper).
The Omidyar Foundation found that in the six biggest emerging economies, using more data and machine learning techniques can get as many as 580 million unbanked people into the financial mainstream. That’s astonishing, especially in countries where there is comparatively less consumer data available. If ML credit models can improve the lot of half-a-billion people in the developing world, imagine what machine learning can do in the data-rich U.S.
