How to Underwrite Thin-File and First-Time Borrowers

You need credit history to get credit. You need credit to create credit history.

Traditional underwriting reads a record of borrowing and repayment. When a consumer has no reported history, lenders struggle to assess risk and price the uncertainty. Consumers with limited credit histories face reduced access and higher borrowing costs because of that observational gap, rather than based on true observed risk.^[2]

Bureaus are excellent at answering how someone handled reported debt, yet have no insight into cash flow, income, rent, utilities, and daily transaction activity for a consumer, and bureau information is even more inaccessible for business credit. CFPB research found more than 20% of Buy-Now-Pay-Later users held active loans that the major bureaus did not record.^[5] Globally, coverage is more fragmented still: as of 2019 only 117 of 191 measured economies had a private credit bureau, with adult coverage ranging from 1.2% to 100%.

Despite this, financial behaviour is digitising fast. 79% of adults globally hold a financial account, 86% own a mobile phone, and 61% of adults in low- and middle-income economies made or received a digital payment in 2024.^[1] The underwriting data target has moved to evaluate the history that already exists.

Evidence is scattered across banks, wallets, employers, landlords, platforms, and the borrower’s own documents. Your job is to gather it, verify it, and reason over it at a cost that works for the ticket size.

Four segments account for most of the observability gap, with each needing a distinct evidence strategy.

None of these describe bad borrowers, but rather those current lending data pipelines cannot resolve. If your approval funnel drops applicants at the bureau lookup step, this is the population you are declining.

Every data source answers a narrow question well and a broad question badly. The discipline that separates good alternative-data programmes from bad ones is holding both columns in view: what the source reveals, and what it leaves open.

Payroll feed proves employment at one employer, not total household income. A wallet statement proves activity on that wallet, not the cash trade beside it. Identifying the limits of your current credit policy.

Five approaches are in production across emerging markets today. They are not alternatives to each other. Most strong thin-file programmes run three or four at once.

1 · Graduated limits

Issue a small first line, then expand it on observed repayment. This is the core mechanic of digital microcredit — Kenya’s M-Shwari is the canonical example. It works because the lender manufactures its own history. The failure mode is scaling limits on repayment alone: across roughly 30,000 East African digital credit customers, aggressive limit increases without a capacity check raised default rates.^[14]

2 · Cash-flow underwriting

Read inflows, outflows, volatility, buffer, and recurring obligations from bank or wallet transactions. This is the closest available measure of ability to repay for a borrower with no tradelines, and the evidence for layering it on top of the bureau is strong.^[7]

3 · Embedded distribution

Lend where the earning already happens. Platform partnership is inherent to business model and can be challenging for standalone microfinance organizations. Regardless, insight as a payment processor sees deposits. A marketplace sees a seller’s receipts. A payroll provider knows income. A mobile-money operator sees wallet flows. The advantage is having observed the borrower’s economic behaviour before they ever applied, which collapses the information asymmetry at the moment of decision.

4 · Documents and self-reported income, reconciled

Where APIs, bureaus, and digital identity are weak the evidence arrives as paper and screenshots. Bank statements, payslips, utility bills, proof of residence, tax filings, invoices from informal self-employment, asset documents, employer letters. Self-reported income is not worthless here; it is a hypothesis. A declared salary gets reconciled against account inflows. A claim of freelance work gets checked against recurring customer payments. An asset offered as security gets verified independently.

There is nothing primitive about this. The loan officer is triangulating several incomplete sources. The problem has always been cost. The Federal Reserve notes that lenders have long used financial information beyond the bureau on larger loans, but that manual analysis was too labour-intensive for small-dollar consumer lending.^[2] Automating the reading and cross-checking is what makes richer underwriting viable at a $200 ticket. That is the whole argument for document capture and extraction in this segment, and we have measured the lift it adds over a bureau score. Document signals added up to +7.0 Gini across 8,000 microloans.

5 · Conversational underwriting

After the documents are read, gaps remain: an irregular deposit, a seasonal trade, a household obligation that never appears in a statement, the actual purpose of the loan. A short structured interview resolves them, and the answers get recorded as evidence attached to the file rather than as a note in someone’s head. This is how a good credit officer has always worked in a thin-file market. What changes is that an assistant can now run the interview consistently, at volume, and write the answers back into the underwriting record in a reviewable form.

Keep the decision itself out of the conversation. The interview gathers and verifies; the policy decides. We have written separately about why the decision logic belongs in code.

Regulators draw a sharp line between financial alternative data and nonfinancial signals such as education, smartphone characteristics, geography, and online behaviour which carry fairness, and privacy risk.^[2,8]

An absent data stream is not negative evidence. Someone may have a thin digital footprint because they are privacy-conscious, older, rural, or simply use cash.

Encode that in the policy. Missing evidence should lower your confidence in the file — a smaller first limit, one more document, a human review. It should not be scored as though the borrower had missed a payment. Systems that conflate the two decline exactly the population the programme was built to reach.

The markets that made the most progress did not build better bureaus. They built identity anchors, payment rails, and consent frameworks that make financial evidence cheap to move.

India built consented data portability on top of Aadhaar.^[9] Brazil paired positive reporting with Pix and Open Finance, which the central bank linked to a 10.4% average fall in personal loan rates.^[10] Kenya reached 84.8% formal inclusion through mobile money.^[11] The UK and Singapore made bank and government records portable by API.^[12]

Credit data availability, market by market

Select a market to see what its bureau covers, which rails exist, and what evidence a lender can actually obtain there.

India

474M consents fulfilled

Aadhaar · Account Aggregator · UPI · GSTN

Bureau: Four licensed bureaus with real depth on formal borrowers, and very little on first-time, rural, and self-employed applicants.

• Consented bank statements (AA)
• GST returns
• UPI transaction history
• EPFO payroll records
• Utility and telecom bills

What works: The Account Aggregator makes bank data portable with the borrower’s consent, so cash-flow underwriting does not depend on screenshots or scraping. For micro-merchants, GST filings verify turnover independently of any bureau.

The most underused source in almost every market is a wallet statement, a QR settlement report, an invoice, because it needs no integration, only reliable reading. And an identity anchor is what turns a document into evidence: without one you have a PDF, not a verified record.

Thin-file consumers lack portable financial evidence.

References

1. [1]

   The Global Findex Database 2025

   Account ownership, digital payments and formal borrowing across low- and middle-income economies, data year 2024.

   World Bank, 2025
2. [2]

   Consumer & Community Context: credit invisibility and alternative data

   Distinguishes traditional credit-account history from financial alternative data such as cash flow, rent, utilities and telecom.

   Federal Reserve, October 2025
3. [3]

   Technical correction and update to the CFPB credit invisibles estimate

   Revises the credit-invisible population and shifts more consumers into the unscored category.

   CFPB, June 2025
4. [4]

   National Survey of Unbanked and Underbanked Households

   14.2% of U.S. households underbanked, 4.2% unbanked.

   FDIC, 2023
5. [5]

   Consumer Use of Buy Now, Pay Later

   Over 20% of BNPL users held active loans not recorded by the major bureaus.

   CFPB
6. [6]

   Identification for Development (ID4D) Global Dataset

   ~800 million people without official ID; 2.8 billion without a government digital ID.

   World Bank ID4D
7. [7]

   Advancing the Credit Ecosystem: Machine Learning & Cash Flow Data in Consumer Underwriting

   Bureau, cash-flow and combined models compared against subsequent credit performance; the combined model was most predictive overall and in every subgroup.

   FinRegLab, 2025
8. [8]

   Interagency Statement on the Use of Alternative Data in Credit Underwriting

   Regulators identify cash-flow data as promising and flag accuracy, fairness and privacy risks in other alternative data.

   Federal Reserve et al., 2019
9. [9]

   Account Aggregator ecosystem statistics

   304M+ linked accounts and 474M+ fulfilled consents as of May 2026.

   Sahamati
10. [10]

    Pix, Open Finance and Cadastro Positivo

    Pix reaching ~80% of the adult population; positive reporting associated with materially lower personal loan rates.

    Banco Central do Brasil
11. [11]

    FinAccess Household Survey

    Formal financial inclusion at 84.8%; exclusion down to 9.9%; mobile money at 82.3%.

    KNBS / CBK, 2024
12. [12]

    Open Banking impact and adoption reporting

    User connections, cumulative API calls and payment volumes for UK Open Banking.

    Open Banking Ltd.
13. [13]

    Credit market indicators for Gen Z borrowers in India

    Credit-eligible share versus formal credit penetration for under-27 consumers.

    TransUnion CIBIL
14. [14]

    Digital credit, automated limit increases and default in East Africa

    Limit expansion without a capacity check raised default rates across roughly 30,000 digital credit customers.

    CGAP

Frequently asked questions

Related reading

Kita · Field guide · September 2026