Data-Driven Credit Scoring: Alternative Data in US and EU Lending

Utility, rent and payroll data are moving into credit decisions in the US and EU. What changes for under-banked borrowers, lenders and the privacy rules involved.

2026-11-08 · 8 min read

Key takeaways

  • Alternative data can score part of the roughly 45 million US adults whose bureau files cannot be scored
  • Rent, utility and cash-flow data land in different files, so the lender's chosen scoring model decides whether they count
  • In the EU, automated creditworthiness decisions trigger GDPR rights and staged high-risk duties under the Artificial Intelligence Act
  • Fees, dispute duties, proxy-discrimination risk and unstable signals all have to be priced in

The shift from bureau files to behaviour data

For most of the past four decades, a US consumer's borrowing options were decided by three files and a three-digit number. The Fair Credit Reporting Act, which defines what counts as a "consumer report" and a "consumer reporting agency", still underpins that system (15 U.S.C. § 1681a). What has moved is the boundary of what a lender is willing to read.

The gap the industry is trying to close is measurable. The Consumer Financial Protection Bureau's Data Point on credit invisibles, built on December 2010 credit and census data, found that about 26 million US adults — roughly one in ten — had no credit record with a nationwide credit reporting company, and that a further 19 million, about 8.3% of adults, had records a widely used scoring model treated as unscorable. Together that is about 45 million people who may be denied credit because their files cannot be scored (CFPB, Data Point: Credit Invisibles).

The same analysis shows how unevenly that falls. Almost 30% of consumers in low-income neighbourhoods were credit invisible and an additional 15% had unscorable records, against 4% and 5% respectively in upper-income neighbourhoods (CFPB, Data Point: Credit Invisibles). The CFPB restated those findings in a follow-up policy report, which also notes that the credit invisible and unscorable together make up almost 20% of the adult population (CFPB, Who are the credit invisibles?). That distribution is why alternative data credit scoring is discussed as an access question, not only as a modelling one.

What "alternative" means in practice

Traditional bureau files are built from credit accounts and collections: cards, instalment loans, mortgages, student loans, plus public records. Alternative data is everything else that demonstrates ability and willingness to pay, usually collected through open banking or a voluntary opt-in.

What alternative data actually includes

Source What it shows Who reports it Main caveat
Rent payments Housing payment discipline, often up to 24 months of history Property managers, rent platforms, third-party reporting services Cash, money order and most peer-to-peer app payments are excluded
Utility, telecom, insurance and streaming bills Recurring household payment behaviour Opt-in services such as Experian Boost Often reported to only one bureau, and only for on-time payments
Bank-account cash flow Deposit consistency, balance stability, overdrafts Open-banking aggregators in the US and account information services in the EU Requires account linking and exposes sensitive transaction data
Employment and payroll records Income stability and tenure Payroll and income-verification providers Consent-based; harder to verify for gig and informal work

A detail that trips up product managers: data being in a file is not the same as a score reading it. Rent payments are incorporated into some newer scoring models but not into older versions still used in mortgage underwriting, so a borrower can hold a rental tradeline and still be scored by a model that ignores it (myFICO).

The US position: opt-in data, shifting rules

Experian Boost is the best-known consumer-facing version of opt-in data. It is free, requires the consumer to link a bank account, and reports qualifying utility, telecom and streaming payments to Experian's file only; Experian states that most people get an instant increase in their FICO Score averaging 13 points (Experian Boost). The scoping matters: a lender sees the effect only if it pulls Experian data or a score built on it.

On the cash-flow side, US scoring vendors have combined bureau scores with linked bank-account data, while European lenders have built comparable affordability models on account information services under the payments rules — the same account-linking pattern that underpins embedded finance (our piece on embedded finance APIs). Both substitute observed behaviour for bureau history, and both depend on the borrower linking accounts.

The legal frame has not changed, but the enforcement posture has. In December 2024 the CFPB proposed a rule that would have treated many data brokers as consumer reporting agencies under the FCRA, imposing accuracy, permissible-purpose and dispute duties on a wider set of firms. The Bureau withdrew that proposal on 15 May 2025, stating that rulemaking was not necessary or appropriate at the time and that it was revising its interpretation of the FCRA (Federal Register).

For lenders and vendors, the practical read is less about the withdrawal than about classification. Whether an alternative-data provider is a consumer reporting agency is decided under the statutory definition in the FCRA — a person that regularly assembles consumer information for the purpose of furnishing consumer reports to third parties (15 U.S.C. § 1681a(f)) — as implemented through Regulation V (12 CFR part 1022). That call is now made in supervision and litigation rather than by rule. The cautious approach is to build FCRA-type controls anyway: documented permissible purpose, accuracy checks, an accessible dispute route and correction timelines.

Two notice duties always apply. Under Regulation B, a creditor must generally notify an applicant of adverse action within 30 days and state the specific reasons (12 CFR 1002.9). If a consumer report was used, the FCRA separately requires telling the consumer and identifying the reporting agency (15 U.S.C. § 1681m). In a thin-file decision, "insufficient credit history" is usually not an adequate reason on its own; the lender has to name the factor that drove the outcome.

Disputes are the operational tail of all this. Under the FCRA's dispute procedure, the reinvestigation duty is triggered when the consumer notifies the agency of an inaccuracy; the agency must then reinvestigate free of charge and complete the process within 30 days, extendable by no more than 15 additional days in defined circumstances (15 U.S.C. § 1681i). Detection sits with the consumer, and each dispute consumes bureau and furnisher time. A lender adding a new data source should budget for that, not treat it as an exception.

The EU position: creditworthiness, privacy and high-risk classification

The EU has no single equivalent of the US tri-bureau system. Depending on the member state, credit information sits in private bureaux or public credit registers, and much affordability assessment relies on bank transaction data shared through account information services. Three rulebooks now shape how alternative data can be used.

Consumer Credit Directive II

Directive (EU) 2023/2225 requires the creditworthiness assessment to be carried out in the consumer's interest to prevent over-indebtedness, and credit to be granted only where the result of that assessment is positive. Member states had to transpose it by 20 November 2025, with application from 20 November 2026 (Directive (EU) 2023/2225). National supervisors read the text as barring a decision based exclusively on the consumer's credit history, which pushes lenders toward verified income and expense data (Malta Financial Services Authority circular). Where the assessment involves automated processing, consumers get a right to human intervention and a meaningful explanation.

GDPR

Article 22 gives consumers the right not to be subject to a decision based solely on automated processing that produces legal or similarly significant effects, unless an exception applies, while Articles 13 to 15 require information about the logic involved (Regulation (EU) 2016/679). For alternative-data products, the work sits in lawful basis, transparency that survives a real-person test, and data minimisation.

Artificial Intelligence Act

Annex III, point 5(b) of Regulation (EU) 2024/1689 classifies systems intended to evaluate the creditworthiness of natural persons or establish their credit score as high-risk, with fraud detection carved out (Regulation (EU) 2024/1689). Article 113 stages the dates on which those obligations begin to apply, and the Commission has put forward changes to that timetable through its digital-omnibus proposals, so lenders and vendors should confirm the version in force (European Commission).

Risks and costs

Proxy discrimination. Utility payments, address history and device data can encode race, ethnicity, disability or immigration status. In the US, fair-lending law applies whatever the data source is, and disparate-impact analysis normally needs proxy testing precisely because protected attributes are not collected. In the EU, special-category restrictions under the GDPR add a second layer.

Consent that is genuinely revocable. Opt-in data can be withdrawn. If a payment history disappears mid-relationship, a borrower's file can end up worse than it started, while the lender's decision may already be priced.

Data quality and disputes. Reporting services differ in which bureaux they feed and how far back they reconstruct history: myFICO's rundown of the market shows RentReporters and Rental Kharma reporting to Equifax and TransUnion, LevelCredit to all three bureaux, and PaymentReport to Equifax and TransUnion (myFICO). A service that also reports late payments can therefore work against the borrower, and correcting an error means a consumer-initiated dispute and the reinvestigation clock described above.

Model risk. Cash-flow signals are volatile: a balance that looks thin between paydays, a seasonal income gap, or a job change can swing an outcome. How large that swing is depends on the model, the lender's cut-offs and the rest of the file, and comparable published magnitudes are scarce — so we describe the mechanism rather than a number. What is clear is that models trained in an easy-credit period tend to overestimate repayment later, which makes cohort-level outcome monitoring and drift checks a core control rather than an optional extra.

Cost, on both sides. Consumers often pay to have rent reported. A myFICO fee rundown published in January 2023 lists RentReporters at a $94.95 sign-up fee plus $9.95 a month (or $7.95 if paid annually), Rental Kharma at a $50 setup fee plus $8.95 a month, LevelCredit at $6.95 a month with a one-time $49.95 charge for 24 months of backfill, and PaymentReport at $2.95 a month (myFICO). Providers change prices, so confirm current terms directly. Lenders pay per pull, per verification and per integration build.

Cheaper underwriting is not cheaper credit. Better data improves the accuracy of a decision; it does not lower the interest rate of the loan that follows. The Federal Reserve's latest G.19 Consumer Credit release puts the average APR on all commercial-bank credit card accounts at 21.19% for the third quarter of 2026 (Federal Reserve, G.19).

Fragmented visibility. A boost applied to one bureau's file may be invisible to a lender that pulls another, and a lender's own affordability model may ignore reported tradelines entirely.

Access without affordability. Wider eligibility can increase borrowing. In the EU this is explicitly the regulator's concern, which is why the Consumer Credit Directive II frames the assessment as protecting the consumer from over-indebtedness rather than only protecting the lender.

Bottom line

Alternative data extends the credit file; it does not change a borrower's income. For lenders and product managers, the sensible framing is a governed additional data source: consent and revocation handling, FCRA-type dispute duties in the US, GDPR and Consumer Credit Directive II duties in the EU, documented fairness testing, and stability monitoring on the scores it feeds. For consumers comparing loan options, three questions do most of the work: which bureau or file the data lands in, whether the reporting service charges a recurring fee, and whether the lender's scoring version actually reads the data. Providers named here are examples, not recommendations, and the effect on any individual outcome is never assured.

This article is for information and education only. It is not investment advice and does not take into account your personal situation. Read the full disclaimer.

Follow the shortlist

Following payments, banking and market-infrastructure change? Our global ETF and stock research watchlist is tracked daily. Premium members get the full list, the reasoning behind every entry and updates when entries change.

See premium access

Information and education only, not investment advice.

Related