Credit scores have an enormous impact on people’s lives and opportunities. This makes it all the more worrying that the precise systems and methods used to calculate them are closely guarded secrets, known to just a handful of influential agencies. More concerning is that what we do know is that traditional credit scoring often gets it wrong: people who are creditworthy find themselves turned down for arbitrary reasons, while less reliable applicants discover ways to “game” the system.
The truth is that credit scores are an extremely narrow metric, showing just a tiny sliver of the full picture. You need better ways to understand your customers; an approach that doesn’t open you up to errors and fraud. A model that takes into account much more than just one number; a number that’s often misleading and inaccurate, at that.
The trouble with credit scores
Much like risk models, our existing credit scoring models are due a serious upgrade. They rely excessively on a handful of financial indicators such as existing debt and past credit history, without taking into account any contextual information that explains the real story behind those figures.
What this means, in many cases, is that people who have borrowed more in the past have better scores than those who don’t. Clearly, this is irrational: if someone has spent a lifetime being financially cautious and living within their means, that hardly makes them a worse candidate for a carefully considered loan or mortgage than someone who regularly splashes huge sums of money on their credit card, but just about managed to repay it by the skin of their teeth in the past
It’s also a deeply unfair system for anyone who spent much of their life unbanked or underbanked, who recently moved to the country, or who has made several credit inquiries simply to compare rates, not realizing that each one has a detrimental effect on their credit rating. Plus, the system tends to favor people in full-time employment and disadvantage freelancers, the self-employed, and anyone working in the casual or gig economy — even if the freelancer has several years of decent, sustained income to show for themselves.
How can we create alternative credit models?
But if credit scoring is so inadequate, what else can lenders do to figure out whether someone applying for credit or a loan is worth the risk?
This is where alternative lending data and alternative credit models come in. Alternative data in this context refers to any external data sources you wouldn’t usually draw on to assess creditworthiness, but which gives you a useful insight into how potential borrowers really behave with money. This could range from a person’s monthly spending (and what they actually spend their money on) to the way they conduct themselves online.
For example, you could look at a person’s history of rent payments, cellphone contracts, and utility bills to get a clear sense of how organized and reliable they are. Do they pay the agreed amounts on time, in full, every month?
Alternative data is fast gaining traction because it’s easier than ever before to access, too. So much of daily life takes place online, and as banking and finance infrastructures improve, organizations can work together to share their digitally-captured data quickly and easily. New regulatory frameworks provide clear boundaries to ensure user rights are protected, too. Many people are keen to share extra information in support of their applications because it strikes them as a fair and accurate representation of how they behave with money. This is especially true when you compare this to the old ways of doing things — the confusing, often opaque ways credit scoring agencies use financial indicators from a person’s credit history.
Meanwhile, a person’s digital footprint contains plenty of useful clues, too. Social media is where many people conduct their day-to-day life and it can be very revealing about the kind of person they are. Whether they act in a way that is impulsive or irresponsible, for example. Whether they are accountable to those around them and well-respected by colleagues.
A word of warning, though: when using this type of data, you need to pay close attention to nuance and context to ensure that you aren’t replacing one flawed, distorted view of a person’s creditworthiness with another. There are also serious ethical considerations, not to mention regulations when it comes to using social media and other data to judge credit applicants. What’s more, failing to examine your algorithms and datasets with a critical eye can hard-bake damaging biases into the model. This leads to results that do a disservice both to you and to your customers. You can learn more about preventing data bias here.
Final thoughts: it’s all about the data
As we’ve seen here, you need access to the most relevant, complete, accurate data to make a truly informed decision about whether or not to extend credit to an applicant.
You need to be sure that the data you use, both internal or external, meets rigorous quality standards. You need to be able to rely on external data providers to comply with relevant privacy legislation. You might need to incorporate many types of structured and unstructured data, formatted in a range of ways that are not entirely compatible with one another.
All these issues are much easier to manage when you’re using a robust machine learning platform that lets you automatically connect to pre-vetted and approved data sources. Not only will this ensure that you’re working with carefully chosen datasets, but the right tools will also handle much of the time-consuming cleaning and harmonizing work. That means you can get straight to what’s important: using the data to build models that give you real, deserved confidence in your lending decisions.