A US consumer with a FICO score of 720 walked into a credit union in early 2025 and was offered a mortgage rate one hundred and forty basis points cheaper than the same borrower would have received at 670. That fifty-point swing, worth more than a hundred thousand dollars over a thirty-year loan, is the working stake for credit scoring algorithms in US life today. This piece explains what those algorithms actually do, who builds them, what is changing, and where US consumers and businesses meet them.
What credit scoring algorithms actually are
A credit scoring algorithm is a statistical model that turns a person’s history of borrowing and repayment into a single number meant to predict default. The two scores that matter for nearly every US consumer credit decision are FICO and VantageScore. FICO, built by Fair Isaac Corporation, is the older of the two and dominates US mortgage underwriting because Fannie Mae and Freddie Mac require it. VantageScore is the joint product of the three US bureaus, Experian, Equifax, and TransUnion. Both scores run from 300 to 850, and both are updated with new versions every few years.
The current US workhorses are FICO 8, which dominates card and auto underwriting, and the newer FICO 10 and FICO 10T, which were released to lenders in 2020 and 2024 respectively. FICO 10T includes trended data, meaning balance and payment patterns over the prior twenty-four months rather than just current snapshots. VantageScore 4.0, released in 2017, uses machine learning techniques and was the first major score to include rent and utility payment data when available. VantageScore 4.0 reaches roughly thirty-three million US adults that FICO 8 does not score because of thin or stale credit files.
Each score reads the same three-bureau data, but each weights inputs differently. Payment history is the largest weight at roughly thirty-five percent. Amounts owed and utilization sit at roughly thirty percent. Length of history, new credit, and credit mix make up the remainder. The exact weights are proprietary and shift between versions. The Consumer Financial Protection Bureau’s research reports, on the CFPB research reports page, track how these weights and the resulting score distributions have moved over the past decade.
The CFPB rule on adverse action notices
When a US lender denies a credit application, federal law requires that the lender provide an adverse action notice that lists the specific reasons for the denial. The CFPB has been explicit, in a 2022 circular and in supervisory actions since, that the requirement applies even when the underlying decision came from a machine learning model that is hard for a human to interpret. The lender owns the explanation, not the vendor.
That CFPB position has shaped US credit scoring algorithm design in two visible ways. First, US lenders that buy or build complex models have invested heavily in reason code generation, the systems that turn a model’s output into the four or five specific factors that drove a given decision. Second, several US lenders have walked back the most opaque deep learning techniques in favor of gradient-boosted trees or interpretable additive models, which are easier to explain. The 2022 circular and the CFPB’s later commentary frame the regulatory expectations that every US lender now plans around.
The Federal Reserve and the OCC reinforce the same position through SR 11-7 model risk guidance. The combined effect is that no US lender can hide behind a vendor’s algorithm. If a model denies credit to a US consumer in a way the lender cannot explain, the lender is the one the regulator calls.
Alternative data and the thin-file problem
Roughly forty-five million US adults have thin or no credit files at the major bureaus. For those consumers, the conventional FICO and VantageScore engines either do not produce a score or produce one based on so little data that lenders discount it. Closing that gap is the largest single fairness story in US consumer credit.
Three products lead the alternative data response. UltraFICO, launched in 2019 and refined since, lets a consumer opt in to share checking and savings account cash flow data with FICO, which then incorporates the data into the score. Experian Boost, launched in 2019 and expanded since, lets a consumer opt in to add utility, telecom, and certain streaming service payment history to their Experian file. VantageScore 4.0, by design, weights rent and utility data when bureaus carry it. Rent payment reporting has expanded sharply in the past five years, with most large US property managers now reporting to one or more bureaus.
For US small businesses, the alternative data story is cash flow lending. Companies like Bluevine, Pipe, and the bank-owned platforms underwrite on accounting platform feeds and bank account history rather than on the business’s bureau score. McKinsey’s financial services research on the McKinsey financial services insights page has documented the rise of this approach as a complement to traditional small business scoring.
Federal Reserve research, summarized through the Federal Reserve payments and consumer page, has tracked the consumer access gains and the operational difficulties of integrating alternative data at scale. The headline finding is that alternative data can move several million thin-file US consumers into the scoreable pool without raising default rates at the lenders that use it.
Where US consumers and businesses meet these algorithms
The first contact point is the mortgage application. Fannie Mae and Freddie Mac require a specific FICO version, and the score sets the rate sheet a US consumer sees. A fifty-point move can swing a US mortgage rate by more than a hundred basis points, which is real money over the life of a loan.
The second contact point is credit cards and auto loans. Issuers and auto lenders use a mix of FICO and VantageScore, often combined with the lender’s own internal model that adds bank data and product data. A US consumer who applies for a card at one issuer and gets denied may be approved at another because the internal models disagree about the same bureau record.
The third contact point is small business credit. Banks pull both the personal credit of the principal and a business credit score from Dun and Bradstreet, Experian Business, or Equifax Business. Cash flow lenders skip the bureau entirely and run their own predictive models on accounting and bank data. The TechBullion fintech news hub covers the US small business credit market in detail, and the payments coverage hub tracks the cash flow platforms.
The fourth contact point is rental housing and insurance. Several large US property managers and several US insurance carriers use credit-based scores in their decisions, which is one reason that score improvement has effects beyond pure credit costs. State rules vary, and a few US states limit the use of credit data in insurance pricing.
What changes in US credit scoring over the next two years
Three patterns are visible. First, Fannie Mae and Freddie Mac are moving toward FICO 10T and VantageScore 4.0 for conforming mortgages, a multi-year transition the Federal Housing Finance Agency announced in 2022 and refined in 2024. The shift extends scoring to roughly five million more US adults and is the largest single change to the US mortgage scoring stack in three decades. Second, the CFPB’s open banking rule, when final, expands cash flow data inputs across consumer credit, which makes UltraFICO style products mainstream rather than opt-in extras. The TechBullion AI in financial services explainer traces the related model adoption pattern. Third, the National Institute of Standards and Technology AI Risk Management Framework adoption inside US lenders will raise the documentation bar on every model in production. NIST AI RMF is voluntary, but US bank supervisors have begun citing it as the benchmark for model risk practice. The next twenty-four months of FHFA implementation, CFPB rule activity, and NIST adoption will decide whether the score that drives so many US household financial outcomes becomes more inclusive and more transparent, or simply more expensive to operate.
