Is the FICO Score Obsolete?

Seeking Inclusivity and Fairness Through AI

Parent and child sitting on a couch looking at a computer in a cozy looking house.

Great news: Eighty percent of U.S. residents have never defaulted on their loans or credit lines.

Not-so-great news: Prime credit is out of reach for most of them.

What this means: Better credit assessments could help lenders approve many more borrowers than they do today — and with better long-term repayment success.

What will it take to make this happen? Could artificial intelligence widen the path to mortgages and home ownership? Some industry newcomers are working on it.

Credit Sector Shakeup

For decades, the FICO Score® from the Fair Isaac Corp. has gated the path from an applicant to a loan approval. FICO scores tend to work well for people with long credit histories. This is why financial advisers often recommend leaving your first credit card open forever. The scores aren’t so helpful, though, for young people, recent immigrants, hard workers and good savers with low-to-moderate incomes.

The industry knows change is necessary. By developing new scoring models, Fair Isaac is working to “meet evolving user needs.” The FICO 9 model, for example, factors in rental payments. This can help renters with limited credit histories. But many peoples’ rental payments go unreported.  

A rival model is the VantageScore, developed by credit reporting companies Equifax, Experian, and TransUnion. The VantageScore can work with a credit profile that’s at least one month old (FICO needs six months).

Artificial intelligence (AI) technology is bringing still more options, and considering much more information than either FICO or the VantageScore. AI constantly improves through machine learning, so an AI vetting process should get fairer and more accurate with time.

The leader on this scene is Upstart, created by former Google employees in 2012. Based in San Mateo, California and Columbus, Ohio, yet devoted to a remote hiring model, Upstart partners with local banks and credit unions across the country.

Artificial Intelligence in Credit Assessments

Upstart matches hopeful borrowers with credit unions and area banks, and equips these institutions with its software. Upstart’s AI model digests some 1,600 data points to identify creditworthiness factors. These data points include things like:

  • The applicant’s work history.
  • The applicant’s professional achievements, plans, and qualifications.
  • The applicant’s prior banking transactions and loan application data.
  • The applicant’s credit experience.
  • The cost of living where the borrower will use the funds.

Assessments are done in just a few moments. The AI model approves 86% more near-prime applicants (people with scores from 620 to 660) than the FICO models do. Obviously, that’s welcome news for applicants with traditional credit scores in the 600s, who have struggled in vain to get loans and credit approvals.

But the banks also stand to benefit. This is because banks that adopt AI technology in their creditworthiness assessments can approve more applicants who will repay their loans as promised.

In other words, banks using AI can make more money.

Applying for a mortgage? has tips on strengthening your credit profile. Get the Facts on Credit Scores — and How to Improve Yours.

Lenders Embrace a New Method

Upstart has been steadily winning over lenders. To name some highlights:

  • In February 2021, Associated Bank, with more than 200 branches, decided to integrate Upstart’s software into its own web presence.
  • As of May 2021, the Ohio-based Telhio Credit Union also works with Upstart. Telhio likes the all-digital, customer-focused model, and was ready to move beyond a dependence on traditional scoring methods and debt-to-income ratios in loan applications.

How, you might be wondering, does the federal government feel about all this? It’s looking good on that front. The risk of regulatory interruptions are low, as the Consumer Financial Protection Bureau (CFPB) issued Upstart a No Action Letter affirming it “will not bring a supervisory or enforcement action” considering the potential consumer benefits from AI. One of those benefits is the potential of AI to improve the fairness of the loan approval process.

Busting Bias

Image of the word Approved written on a glass board in green by a hand holding a green pen.

A congressional task force recently held hearings to examine machine learning and racial bias. Dave Girouard, Upstart’s CEO, testified: “The concern that the use of AI in credit decisioning could replicate or even amplify human bias is well-founded.”

We have noted this before on A 2019 study we’ve noted, Consumer Lending Discrimination in the FinTech Era from the National Bureau of Economic Research, finds “lenders charge Latinx/African-American borrowers 7.9 and 3.6 basis points more for purchase and refinance mortgages respectively, costing them $765M in aggregate per year in extra interest.”

AI bias? Fair housing advocates are addressing the problem of algorithmic discrimination. Read more at Black Homeownership and Housing: Amid Persistent Discrimination, New Potential for Change.

So, can machine learning improve the situation? Evidence shows it can. Algorithmic discrimination is still an issue, the study shows — but 40% less so in AI than in person-to-person assessments.

Upstart runs fairness tests on every loan application it manages. The company then turns the test results in to the Consumer Financial Protection Bureau on a quarterly basis.

So far, the results are encouraging. In 2020, the algorithmic assessment model approved at least 30% more African-American applicants than are approved through the industry standard. Moreover, the AI approvals issued notably lower interest rates along with the approvals. Among Hispanic applicants, 27% more received approvals than with the standard norm. Again, approvals came with comparably low rates. Upstart is now creating a Spanish-language assessment tool.

Mortgages Next

So far, Upstart is focused on debt consolidation and personal loans, yet its vetting methods are applicable to a range of loan types, including mortgages. Consider the way SoFi® (Social Finance, Inc.), another California company applying AI to credit, expanded from its student loan focus to an array of services that includes user-friendly mortgage applications. There is no minimum FICO score with SoFi, either.

Speaking of SoFi, it’s earned the Office of the Comptroller of the Currency’s conditional approval for a national bank charter. The Federal Reserve and the FDIC are currently working with SoFi on the rigorous process of licensing the bank. Assuming all goes well with the final approval, SoFi will have authority to accept its members’ deposits, and use money it holds to extend loans. This would cut out the interest charges SoFi currently pays to established banks.

So, when can we expect Upstart to enter the mortgage arena? We’ll be on the lookout, and we’ll keep our readers posted.

What’s Next? AI at the Big Banks?

It’s likely that AI-driven assessments could be copied and adopted by large banks that can afford the technical teams to do this. Meanwhile, local and regional banks can benefit by collaborating with innovators and taking early-adopter positions.

No matter how the changes play out, regulators need to know how machine learning works. And financial AI should only be used in tandem with a system that ensures its fairness.

Supporting References

Tom Taulli for Forbes: Upstart – Can AI Kill the FICO Score? (Aug. 13, 2021).

Upstart Holdings (blog): Upstart CEO Testifies About AI in Credit Underwriting (July 25, 2019).

Upstart Press Release (via BusinessWire):  Telhio Credit Union Selects Upstart for Personal Lending(Jul. 21, 2021).

The Value Gap via FICO Scores Leave Out People on the Margins, Upstart CEO Says.  Can AI Make Lending More Inclusive — Without Creating Bias of its Own? (Jul. 23, 2021).

Jon Markman for The Street: Bank-Loan Disruption Boosts Upstart Stock (updated Aug. 12, 2021).

Anna Irrera with Mark Potter (ed.) for Reuters: Fintech Startup SoFi Gets Preliminary Approval for U.S. Bank Charter (Oct. 28, 2020).

Photo credits: Ivan Samkov, via Pexels, and InspiredImages, via Pixabay.