You apply for a loan. The answer comes back: No. Perhaps you wanted to buy a car. Maybe you applied for a credit card, needed money to start a small business or hoped to refinance debt at a lower interest rate. Whatever the reason, the decision matters. Access to credit can determine whether someone buys a home, replaces a failing car, handles an emergency or gets a business off the ground.
But who—or what—made the decision? For generations, lenders have used mathematical systems to help answer a deceptively simple question: How likely is this person to repay the money? Credit scores turned parts of our financial histories into numbers that could be compared with those of millions of other borrowers. Today, artificial intelligence and increasingly sophisticated algorithms are potentially taking that process much further.
Instead of examining only a conventional credit score and a handful of financial variables, newer underwriting systems can analyze larger collections of data and discover relationships that traditional models might miss. That could make lending faster and, in some cases, give people who have been poorly served by traditional credit scoring a better chance of obtaining credit. But it also introduces a troubling possibility: a computer may participate in a decision that profoundly affects your life using a process you cannot see and may not understand.
The Credit Score Was Already an Algorithm
It is easy to imagine algorithmic lending as something invented during the recent AI boom. It wasn’t. Credit scoring has been algorithmic for decades. Information about existing accounts, repayment history, delinquencies and other factors has long been transformed mathematically into predictions about credit risk. What is changing is the amount and variety of information computers can analyze and the sophistication of the models doing the analysis.
The Federal Reserve reported in 2025 that advances in data availability and computing have expanded the potential use of what the financial industry calls alternative data. Some of this is straightforward financial information, such as cash flow, income and expenses. Other potential sources are more unconventional and can include educational or professional information and elements of a person’s digital footprint.
That creates an intriguing possibility. Someone with little conventional credit history may actually be financially responsible. A traditional credit file may know very little about that person, while cash-flow information could show regular income, responsible spending and an ability to manage monthly obligations. Used carefully, more data could make the system fairer rather than less fair.
When More Data Can Help
Millions of financial decisions are difficult precisely because the traditional system does not see everyone equally well. A young adult may have a short credit history. An immigrant may arrive in the United States with years of financial responsibility that are barely visible to American credit bureaus. Someone who has avoided borrowing may have relatively little information in a conventional credit file.
The Federal Reserve has described how alternative financial data, particularly cash-flow information, may help lenders evaluate some consumers who have no credit score or whose traditional score does not fully reflect their ability to repay. It has also noted evidence that some uses of alternative data can expand access to credit. That is the optimistic case for smarter underwriting.
Instead of asking only, “What has this person borrowed before?”, a lender might gain a fuller picture of how someone actually manages money. But once we accept the principle that more information can improve a prediction, a much harder question appears: How much information should a lender be allowed—or willing—to use?
When the Data Gets Personal
There is an enormous difference between examining someone’s monthly income and analyzing a digital footprint. The more variables a model considers, the greater the possibility that it will discover statistical relationships that are difficult for a consumer to understand and perhaps difficult for the lender itself to explain.
A particular behavior might correlate mathematically with repayment risk without having an obvious connection to whether an individual deserves credit. Some nonfinancial data may also correlate with characteristics protected under fair-lending laws.
Federal Reserve officials have warned that machine-learning systems can perpetuate or amplify biases already present in training data and that apparently neutral variables can sometimes correlate with protected characteristics. The concern becomes greater when a model is opaque and the institution cannot readily explain how particular variables influenced its decisions.
An algorithm does not need to be deliberately discriminatory to produce troubling results. It can learn patterns from a world in which disparities already exist. The machine may be mathematical. The data came from society.
The Black Box Problem
This leads to one of the central problems of AI-assisted lending: explain ability.Imagine a traditional lending officer telling you that your application was rejected because your income was insufficient relative to your existing debt. You may dislike the decision, but you can understand it.Now imagine a highly complex model examining hundreds or thousands of variables and concluding that you represent too much risk.You ask why.And nobody can give you an intelligible answer. That is sometimes described as the black box problem: a model may be capable of producing a prediction even when explaining exactly how it arrived there is difficult.
In American consumer credit, however, “the algorithm decided” is not an adequate explanation. The Consumer Financial Protection Bureau has made clear that creditors using complex algorithms still must comply with the Equal Credit Opportunity Act and Regulation B requirements concerning adverse-action notices. A lender cannot escape its obligation to provide specific reasons simply because its model is complicated or difficult to interpret. That principle is enormously important. If an institution cannot explain why its machine rejected you, the sophistication of the machine does not eliminate your rights.
If You Are Rejected, Ask Why
For consumers, this is where an abstract debate about artificial intelligence becomes very practical. If a lender rejects an application, federal law generally requires the lender to provide the principal reasons for the decision or tell the applicant how to obtain them. If the decision was based on a credit report, additional information must be provided, including the credit-reporting company involved and, in applicable circumstances, information about the credit score used. The explanation cannot simply amount to “you failed our system.”
CFPB rules state that the reasons must relate to factors actually considered or scored in making the decision. The agency has also emphasized that creditors cannot simply select a convenient generic explanation from a standard checklist when it does not accurately describe why the adverse action occurred. That gives consumers something extremely valuable in an increasingly automated financial system: the right to ask a human institution to account for what its machine has done.
Can an Algorithm Be Fairer Than a Human?
There is another side to this story that shouldn’t be ignored. Human beings are hardly perfect decision-makers. A loan officer can have prejudices. People can make inconsistent judgments, be influenced by irrelevant impressions or treat two similar applicants differently. A carefully designed algorithm could theoretically apply the same criteria consistently to everyone and discover creditworthy borrowers whom traditional methods overlooked. So the real question is not whether humans are good and algorithms are bad.
It is whether the system—human, algorithmic or some combination of both—uses relevant information, complies with fair-lending laws, can be tested for harmful outcomes and remains accountable for the decisions it makes. Algorithms can potentially expand opportunity. They can also automate unfairness at enormous scale. Both possibilities can be true at the same time.
A Machine Can Calculate Risk. It Cannot Decide What Is Fair.
We are entering a period in which increasingly consequential decisions may be assisted by machines. Should you receive a loan? How much should you be allowed to borrow? What interest rate should you pay? How risky are you compared with someone else? Algorithms can help answer these questions because they are extraordinarily good at finding patterns in data. But there is another question hidden beneath all of them: What information should society permit a machine to use when judging a person? That isn’t a mathematical question. It is a human one.
The future of credit will almost certainly involve increasingly sophisticated technology. That technology may make lending faster, more accurate and more accessible to people overlooked by older systems. But sophistication should not mean invisibility. If an algorithm participates in deciding whether you deserve credit, you should still be able to ask the oldest and simplest question in the world: Why? And someone should be able to answer.
