Credit bureaus were not created to exclude anyone. Their practical purpose was to let lenders share repayment behavior, rather than assess every applicant from zero. The underlying idea was sound: consistent repayment across several creditors is meaningful evidence. Combining that evidence across thousands of lenders creates useful infrastructure.
The weakness is structural. The infrastructure centers on a limited range of behaviors: borrowing from formal institutions, receiving payroll formally, and holding checking or savings accounts. It emerged when formal work and banking were treated as nearly identical to economic participation. That was never true for large populations in many markets, and in Latin America it was untrue for most people.
The Gap Is Not Random
To see who sits outside the bureau, consider how its files arise. Files grow through inquiries and trade-line reports showing balances and repayment. Yet reporting requires a product first, such as a card, personal loan, mortgage, or car payment. Without earlier credit, there is neither a file to report to nor one to consult.
This creates a particularly high entry barrier for several groups. Young adults starting work have no history because they have never used credit. Informal workers may lack payslips, complicating income checks for lenders that treat employment documents as a bureau-adjacent input. Internal migrants may have records elsewhere but no local footprint. Recent immigrants face the same issue across borders: a strong record in one country does not move into another bureau system.
Bureau exclusion therefore clusters among younger, recently urbanized, informally employed, and recently migrated people. This does not mean they are uniformly low-income. Many earn real income, handle genuine obligations carefully, and are creditworthy. Their missing element is a data trail in the format bureaus require.
Brazil as a Case Study
Brazil's bureau infrastructure is among the most technically advanced in any emerging market. Serasa Experian and SPC networks cover hundreds of millions of accounts. Expanded in 2019, Cadastro Positivo introduced positive payment records into a model that had previously been mainly negative-only. That represents genuine progress.
Still, about 35 million Brazilian adults lack any usable credit file. Depending on methodology, some estimates are higher once files too thin for a dependable score are included. Most are concentrated in the informal economy, about 40% of Brazil's workforce according to IBGE labor force estimates. They also include many younger urban residents who work but only recently started earning formal income.
This does not mean Brazil's bureau system failed. It achieved its intended purpose. The gap remains because it was not built for activity centered on prepaid mobile service, cash, utility bills, and informal peer transfers. Closing the gap calls for a different data architecture, rather than a better version of the current one.
The Circularity Problem
Thin-file exclusion is especially frustrating because it is circular. A new borrower cannot prove creditworthiness without credit, while most lenders require proof before granting it. The person may be creditworthy behaviorally, but the system cannot observe that. They are excluded not for being risky, but for being unseen.
Earlier responses included secured cards, credit builder loans, and microfinance requiring intensive field checks. These products help, but they are slow. Borrowers must spend months proving creditworthiness in bureau terms before reaching mainstream products. For someone needing funds immediately, that delay is often impractical.
Alternative data takes another route. Rather than asking borrowers to build a bureau-style record, it examines behavior already present: phone-plan management, utility-payment regularity, and mobile-money flows. Where bureau information exists, this does not replace it. It runs alongside it when the bureau path is empty.
The Lender Incentive Problem
It is tempting to call thin-file exclusion a moral failure by lenders, but that is not useful. Lenders manage risk using the information at hand. When that information cannot support a lending decision, declining is rational. Criticizing lenders for relying on imperfect available data will not alter their behavior.
Behavior changes when available information changes. A credible, validated score with real predictive value for repayment can shift a thin-file borrower from "too uncertain to approve" to "scorable with an alternative methodology." The lender is not asked to accept greater risk, but given a way to measure risk previously hidden.
That is the business case for alternative data scoring, separate from the financial inclusion argument. The claim is not "lend to thin-file borrowers because it is right." It is "we can identify which thin-file borrowers are creditworthy, and the market is substantial." The two aims align, but the commercial case stands independently.
What Changes and What Does Not
Alternative scoring does not turn every risk into an approval. Poor behavior across several signals, such as late utility bills, irregular mobile recharges, and no stable geolocation, indicates poor credit risk, which a calibrated model should show. The aim is not universal approval, but preventing creditworthy people from systematic exclusion simply because they lack a bureau file.
For the foreseeable future, bureaus will probably remain the main credit scoring infrastructure. They contain years of validated predictive history, sit deeply within lender workflows, and reach much of the adult population across developed and developing markets. Alternative scoring need not displace them. Its role is to extend coverage to groups bureaus were never built to reach.
For the 35 million Brazilian adults without usable files, and hundreds of millions like them across Latin America, this gap is tangible. It can prevent a creditworthy person from borrowing to launch a business, purchase a refrigerator, or handle an unexpected cost. Closing it is an engineering problem. We work on it every day.