Credit scorers are using artificial intelligence to analyze alternative consumer data, but some in the industry are wary
The editorial content below is based solely on the objective assessment of our writers and is not driven by advertising dollars. However, we may receive compensation when you click on links to products from our partners. Learn more about our advertising policy.
The content on this page is accurate as of the posting date; however, some of the offers mentioned may have expired. Please see the bank’s website for the most current version of card offers; and please review our list of best credit cards, or use our CardMatch™ tool to find cards matched to your needs.
Credit scoring firms, credit reporting agencies and lenders have begun to embrace artificial intelligence (AI) and machine learning to help assess potential borrowers’ creditworthiness. Experts believe AI has a key role to play in helping “credit invisibles” – the estimated 26 million Americans with little or no recent credit history – get access to loans and credit cards through better evaluation of alternative data.
AI is perhaps best known for powering new technologies that only a few years ago would have seemed like the stuff of science fiction. It’s being used for self-driving cars and digital voice assistants such as Apple’s Siri and Amazon’s Alexa, and it can put card sharks and chess masters to shame.
But AI is also revolutionizing the financial technology industry in various ways. Machine learning – the ability of a computer to evaluate reams of data without being programmed to do so by a human – is used in fraud detection, investment planning, customer service and lending. Credit scoring firm FICO’s success in using AI for fraud protection suggests there’s a long-term role for machine learning in other areas.
“I don’t think it’s the flavor of the month,” said Ethan Dornhelm, principal scientist at FICO. “Machine learning algorithms are here to stay.”
An automated alternative data analyst
FICO used algorithms that learn from and adapt to data in real-time to build its Falcon fraud management program. Dornhelm said the firm now sees “considerable promise” in machine learning to help analyze consumer data such as address history and utility and cellphone bill payments. FICO has been at the forefront of alternative data, having introduced its FICO Score XD in 2015.
“The single biggest challenge in assessing the creditworthiness of populations that aren’t what we call ‘traditionally scorable’ is really a lack of data about those consumers,” Dornhelm said. “If you’re not traditionally FICO-scorable, it’s because you don’t meet minimum scoring criteria.”
The use of alternative data in credit scoring has been a hot topic in recent years, and it’s not without controversy. Proponents believe nontraditional data can expand consumers’ access to credit by giving lenders a more complete picture of any individual borrower. Skeptics say it can harm consumers because it relies on data that’s often inaccurate or incomplete, or inadvertently favors certain race or ethnic groups, among other concerns.
Democrats in Congress have on several occasions introduced legislation designed to encourage the use of alternative data, but those efforts have stalled. However, federal regulators are intrigued – in February 2017, the Consumer Financial Protection Bureau announced it was seeking public comment on how alternative data could help credit invisibles.
Dornhelm stressed that AI can’t help consumers who lack the bare minimum of sufficient data on which to assess their creditworthiness.
“If a consumer doesn’t have a FICO score today because all that shows in their credit file is one collection account from seven years ago, and we have no recent information on them, AI can’t solve that,” he said.
But AI can help credit scorers with active credit files make quick sense of other nontraditional data, such as telecom and utility bill payment history.
“You could almost think of it as an automated analyst that can, in very short order, detect the most important patterns inherent in these new data sources and allow us to make sure we’re capturing those as we build these new algorithms,” Dornhelm said.
Other firms such as ID Analytics – which also uses machine learning for fraud prevention and credit risk assessment – see a role for AI in helping consumers with spotty credit histories.
“We can find people who do have some negative information in their credit files, but given everything else that we see about them and the way they behave, we can say even though this person’s a little behind the ball in certain aspects of life, he still has a desire to pay,” said Stephen Coggeshall, chief analytics and compliance officer at ID Analytics.
Not so fast
The use of AI in credit scoring has encountered some resistance in the financial world, primarily due to its complexity and concerns about regulatory compliance. The successful implementation of AI and machine learning to evaluate data in any given application requires a certain level of expertise. Coggeshall said ID Analytics’ progress with AI has been earned through years of hard work from PhD-level machine learning scientists.
FICO’s Dornhelm said it can be challenging to square machine learning with traditional credit scoring models honed for decades by human experts. Algorithms can be relatively difficult to control, and they can spit out unfavorable interpretations of data. That’s less likely to come from humans – for instance, when credit scoring experts painstakingly assemble a picture of someone’s creditworthiness using a time-tested formula.
“A machine learning-based score may actually encode a pattern such that it rewards people who have more credit card debt than less, which is counterintuitive to what 25 years of expertise in this area shows,” Dornhelm said. “More importantly, it would make for an unpalatable experience for consumers who are trying to improve their scores over time and are being told you have to rack up more credit to do so.”
Introducing new technology into a heavily regulated practice such as credit reporting without running afoul of authorities can be tricky as well. Coggeshall noted that many businesses have stuck to traditional credit scoring models because it’s easier to explain credit denials to prospective borrowers – a legal requirement.
“When someone is denied credit, you have to come up with justifiable reasons why these complicated nonlinear models have made that decision,” he said. “That’s a substantial barrier.”
Cracking the ‘black box’
Despite the challenges, experts are optimistic that AI will eventually become more transparent and help expand credit access to more consumers.
“Where I see things headed is a continued improvement in cracking the ‘black box’ that is most machine learning and AI algorithms today, to make them more interpretable – to really understand what’s going on under the hood,” Dornhelm said.
Coggeshall said resistance to AI is beginning to fade as more of ID Analytics’ competitors embrace it. That’s good news for consumers who fall short under traditional credit scoring criteria – Coggeshall said alternative data can help score “virtually everybody in the United States.”
Of course, unscorable consumers don’t have to wait until the day humans and robots work together in perfect harmony to become creditworthy. There are other ways to build credit, such as secured credit cards or becoming an authorized user on someone else’s card account.
But if you’re ready to jump into credit without taking baby steps, fear not – the robot cavalry is coming.