Two numbers tell you how much the US credit scoring system has changed in the last 15 years. In 2010, 81.6% of American adults had a scorable credit record. By 2020, that figure was 87.5%, or about 225 million people, according to the Consumer Financial Protection Bureau’s June 2025 technical correction to its credit invisibles estimate. The share of credit invisibles — adults with no credit record at a nationwide reporting agency — fell from 5.8% to 2.7% over the same decade. The population that credit models are now trying to serve has never been larger. And for the first time since the 1990s, the algorithms that score them are being rebuilt.
How US credit scoring got to a single dominant model
For most of the last 30 years, the US credit scoring question was settled. FICO published its first consumer credit score in 1989 and became the de facto standard through the 2000s, when Fannie Mae and Freddie Mac adopted Classic FICO as the required score for every mortgage they purchased. By 2015 the FICO Score was used in roughly 90% of US consumer lending decisions. A lender looking for competitive pressure on FICO’s pricing had nowhere to go: the Enterprise-required score was FICO, and most underwriting stacks were built around it.
The alternative, VantageScore, was launched in 2006 as a joint venture of Equifax, Experian, and TransUnion. It was designed to score consumers that FICO’s older models struggled with — thin-file consumers, people without traditional credit products, and anyone whose data did not fit the cleanest cohort FICO’s training set assumed. For 15 years VantageScore grew inside auto lending, credit cards, and fintech underwriting, but it was not accepted by Fannie or Freddie, which meant the mortgage market stayed locked to Classic FICO.
The break came in 2018, when Congress passed the Credit Score Competition Act, requiring the Federal Housing Finance Agency (FHFA) to establish a process for validating alternative credit score models for the Enterprises. FHFA finished that process in October 2022, validating both FICO 10T and VantageScore 4.0 as approved models. The FHFA’s own framing at the time — noting that the new models consider additional sources of data including rent payment history and can score consumers more accurately — was the clearest official signal that the Classic FICO era had an expiration date.
What the credit scoring map actually looks like in 2025
The most consequential change of 2025 is policy, not technology. On July 8, 2025, the FHFA announced that lenders could immediately use VantageScore 4.0 for mortgages sold to Fannie Mae and Freddie Mac, ending nearly two decades of mortgage-market lock-in to Classic FICO. VantageScore’s own estimate was that roughly 5 million additional Americans — including veterans and rural-community prospective homebuyers — would benefit from the model change, and that $1 trillion in incremental mortgage activity was in play over time.
| Metric | Value | Source |
|---|---|---|
| US adults with a scored credit record, 2020 | 87.5% (225.3 million) | CFPB (June 2025 update) |
| Share who were credit invisible, 2020 | 2.7% (7.0 million) | CFPB |
| Share with a scored record, 2010 (corrected) | 81.6% (191.3 million) | CFPB |
| Corrected credit-invisible estimate, 2010 | 5.8% (13.5 million) | CFPB |
| VantageScore usage, 2024 | 42 billion scores (+55% YoY) | VantageScore |
| Financial institutions using VantageScore, 2025 | 3,700+ | VantageScore |
| Additional consumers scorable under VantageScore 4.0 | 33 million vs traditional models | VantageScore |
| Incremental mortgage activity estimate | $1 trillion | VantageScore |
The adoption numbers tell the same story. VantageScore has publicly disclosed that 42 billion of its scores were used in 2024, a 55% jump year over year, and that more than 3,700 US financial institutions now consume the model. That is a decisive shift in distribution: VantageScore is no longer a fintech-specific or auto-lending-specific model, and the FHFA decision moved it into the mortgage stack too.
What the algorithms actually do differently
The FICO and VantageScore models most US lenders were using in 2015 — FICO 8 and VantageScore 3.0 — were logistic-regression-style scorecards trained on bureau data at a specific point in time. They scored from a snapshot of the consumer’s credit file. The newer models do two things those older models do not.
The first is trended data. FICO 10T and VantageScore 4.0 both use longitudinal credit-file data — the consumer’s balances, payment behaviour, and utilization ratios across multiple months — rather than a single snapshot. That lets the score distinguish a consumer whose balance is steady from one whose balance has been creeping up, even when both end the month at the same number. The predictive lift from adding trended data has been shown to be meaningful in head-to-head industry benchmarks.
The second is alternative data. VantageScore 4.0 in particular was designed to pull in rent, utility, and telecom payment history where bureau data is incomplete. The explicit goal was to score the thin-file and no-file populations that Classic FICO either flagged as unscorable or scored conservatively. VantageScore’s public numbers say the 4.0 model can score about 33 million more consumers than traditional models — which maps closely to the population of thin-file and “insufficient unscored” adults the CFPB’s updated research identifies. The reason the FHFA decision matters is that these are the consumers who, in practice, have had the hardest time accessing the mortgage market and the credit that sits alongside it.
The third shift, less visible but just as important, is modelling technique. FICO 10T uses gradient-boosted decision trees under the hood, a departure from the logistic-regression scorecards of older versions. VantageScore has published work on machine-learning techniques within its own development pipeline. Both companies still have to explain each score to the consumer under the Fair Credit Reporting Act, which means the production models are interpretable by design — but the underlying architectures are considerably more flexible than what the market was using 10 years ago.
Where the challenger data sits
A separate layer of the credit scoring market sits below the two big scorecards: the alternative-data companies that either feed the bureaus or underwrite directly for lenders. Nova Credit brings international credit files to US lenders so new-to-country consumers can be scored. Petal and the issuers it powers underwrite credit cards from cash-flow data in a bank account. Plaid and similar cash-flow providers make the bank-data inputs themselves usable at scale. Experian Boost and UltraFICO add utility, telecom, and bank-balance signals into the bureau file. Socure and SentiLink, which show up more often in fraud stacks, also feed into thin-file underwriting by confirming identity before any scoring happens.
For a fintech founder, the practical question is rarely “FICO or VantageScore” — it is which combination of scorecard-plus-alternative-data produces the best approval rate at the lowest loss rate for the specific product. For consumer-lending products outside the mortgage channel, this is where most of the interesting work has been happening, including in the P2P and direct-lending stacks that survived the 2015 peak by building internal credit engines rather than relying solely on the bureau score.
What it means for fintechs and lenders
For lenders that still run everything through Classic FICO, the 2025 mortgage change has a specific operational implication: underwriting stacks need a parallel path for VantageScore 4.0 before the Q4 2025 transition date, and teams need to test what happens to their approval, pricing, and loss models when the score distribution shifts. A move from Classic FICO to VantageScore 4.0 is not a drop-in swap because the score ranges are not identical and the underlying risk predictions are not calibrated to the same loss curves.
For fintechs, the opportunity is sharper than it has been in a decade. Products that depend on scoring thin-file consumers — new-to-country, young, gig economy, or post-bankruptcy — have a much bigger addressable market once mortgage channels open to the newer models, because the same underwriting improvements tend to flow downstream into auto, card, and personal-loan decisioning. The companies competing in neobanking are particularly exposed: the economics of challenger banks in the US depend heavily on how cheaply and accurately they can score the customers they already have.
For operators in credit-heavy verticals, the hardest question is model governance. Running two scorecards in parallel means maintaining two sets of overrides, two fair-lending reviews, two vendor contracts, and two sets of adverse-action letters. Most US lenders with real volume have been building out the capability to run multiple models simultaneously for the last 18 months; the ones that have not are now scrambling.
The bottom line
Credit scoring in 2025 is a category in transition. 87.5% of US adults have a scorable credit record, up almost six percentage points in a decade. The CFPB’s own reanalysis has shrunk the credit-invisible population by roughly half from the 2015 headline figure, but the consumers who remain invisible are disproportionately Black, Hispanic, and low-income — which is why the FHFA’s July 2025 decision to accept a model that can score 33 million additional consumers carries more than technical weight. The plumbing of the US credit system is now, for the first time in a generation, genuinely contested. FICO still has most of the market. VantageScore has most of the growth. The quiet part is that both models are much smarter than they were five years ago, and the lenders who know how to read what each one is actually telling them are the ones who will capture the next cycle’s approvals.
Last updated: June 17, 2026
