Alternative Data in Creditworthiness Assessments

Define the dataset and assessment purpose

Alternative data is information outside the conventional bureau and application fields commonly used in credit decisions. Depending on product and jurisdiction, it may include verified rent payments, utility payment history, cash-flow patterns or other permitted indicators. The term is not a legal authorisation to collect arbitrary digital traces. Start with the lending question: does the proposed dataset establish income reliability, existing obligations or the ability to meet the stated instalments? Record which borrower population it covers and whether missing data have a meaningful interpretation. People with little formal banking history may be systematically absent from a vendor's dataset. Do not code an unavailable attribute as proof of elevated risk merely to create a numeric score. Product amount, payment timing and applicable law determine what further evidence is necessary.

EU creditworthiness and personal data safeguards

Directive (EU) 2023/2225 Article 18 provides for creditworthiness assessment using adequate, relevant and accurate financial information, with safeguards on the types and uses of information. Alternative datasets require separate consideration of their relevance, source, accuracy and compatibility with applicable data protection law. Article 19 addresses credit databases and should not be interpreted as a blanket right to access all data sources. National implementing measures generally apply from 20 November 2026, subject to national transition, so creditors must identify current rules for the agreement date. A permitted data source may still be too incomplete or misleading for a sound individual assessment. Maintain a jurisdictional inventory indicating which data categories are allowed, prohibited or require further review, and preserve the legal source and effective date behind each rule.

UK CONC and proportionate evidence

FCA CONC 5.2A requires relevant firms to undertake a reasonable creditworthiness assessment considering repayment and affordability risks. Alternative information may be useful where traditional bureau data is limited, but it does not eliminate the need to assess the customer's financial circumstances appropriately. Verify how a payment record relates to disposable resources and the proposed obligations. An on-time utility history may show past behaviour without demonstrating that the consumer can afford a substantially larger loan. A proportional process can use different sources depending on the risk and known facts. Document any statistical estimate, why it was appropriate for the borrower, and when contradictory information triggered further inquiry. Vendor marketing descriptions such as comprehensive affordability signal do not replace legal review.

US consumer reporting and nondiscrimination

US ECOA and Regulation B restrict discrimination and constrain evaluation using prohibited bases. Where alternative information is compiled by a consumer reporting agency or otherwise triggers Fair Credit Reporting Act requirements, permissible-purpose, accuracy, dispute and notice obligations may matter. Regulation B section 1002.9 requires specific principal reasons for certain adverse actions, including decisions relying on unconventional variables. Complex scoring does not excuse an inability to give a legally adequate explanation. Evaluate source providers for their actual regulatory roles instead of assuming alternative data always falls outside consumer reporting law. A dataset can introduce proxies for protected characteristics or unequal coverage; test and remediate problems under applicable fair-lending requirements. Product-specific obligations and state laws must also be mapped.

Data provenance and error controls

Each alternative field needs a name, unit, source, permission or legal basis, observation period and documented transformation. Rent payments may be reported by only some landlords; transaction feeds may duplicate transfers; mobile or behavioural information may reveal sensitive traits unrelated to repayment capacity. Test what happens when a source is stale, a provider changes categorisation or an applicant disputes an observation. Validate whether alternative indicators add useful information beyond conventional inputs and whether their value is consistent across relevant customer groups. Reconcile a model's implied expenditure against other known obligations. Suppress attributes with no defensible assessment purpose rather than retaining them in case future models find predictive correlations.

Vendor oversight and consumer correction

Contracts with data suppliers should address provenance, quality testing, outages, incident notification, retention, permitted reuse and the ability to correct records. Keep a reproducible link from underlying data version to its transformed underwriting feature and the final decision. Provide a route for disputes and corrected information where required by the governing regime. Monitor approval differences, repayment stress and customer complaints for evidence of invalid proxy variables or inaccurate records. Compliance review should cover both automated approvals and rejections. If the lender cannot explain why the data influenced a denial or cannot demonstrate lawful collection, it should reconsider the field before relying on it. A regulator-ready file should show how the dataset improved a relevant assessment, not just that a vendor claimed a higher predictive score.

Frequently asked questions

Related research

Creditworthiness and Affordability Assessment Compliance ↗

AI Models in Creditworthiness Decisions ↗