Credit Scores Were Always an Approximation
The credit score — in its FICO incarnation in the United States and its various equivalents across European markets — was never a precise measurement of a borrower's creditworthiness. It was always an approximation: a summary statistic derived from a limited set of observable behaviors (payment history, credit utilization, credit age, credit mix, and new credit inquiries) that correlated with future repayment probability at a population level, even if it was noisy and imprecise at the individual level.
The reason credit scores were developed and persisted is pragmatic. Before the computing infrastructure to analyze large, heterogeneous datasets existed, reducing a borrower's credit profile to a single number was the only way to make credit decisions quickly and consistently at scale. The score was a lossy compression of a complex reality, but it was better than nothing, and it was consistent enough to be used reliably across millions of credit decisions.
We are now in an era when the pragmatic constraints that gave rise to the credit score no longer bind. Computing infrastructure can analyze thousands of variables in milliseconds. Machine learning models can identify complex, non-linear patterns in heterogeneous data that linear regression models cannot detect. And the data available to characterize a borrower's creditworthiness — behavioral data, transaction data, telemetry from connected devices, social graph signals, employment platform data, rental payment history — is orders of magnitude richer than the inputs available to traditional credit scoring models.
The credit score era is ending not because credit scores are wrong, but because something far more accurate and equitable is now possible. And the transition to AI-powered credit decisioning is creating substantial investment opportunities in European lending.
The Thin-File Problem: Who Credit Scores Fail Most Severely
The practical consequence of credit scores' dependence on credit history is that they systematically fail to serve populations with limited or no credit history — the so-called "thin file" borrowers. Thin-file borrowers include recent graduates, immigrants who are new to a market, young adults entering the workforce, self-employed individuals with variable income patterns, and people who have primarily used cash rather than credit throughout their adult lives.
The scale of this problem in Europe is significant. Approximately 25 to 30 percent of the adult population in major European markets has limited or no credit file at traditional credit bureaus. Many of these individuals are genuinely creditworthy — they have stable income, prudent spending habits, and strong motivation to maintain a good financial reputation — but they are invisible to traditional credit scoring models. As a result, they are either denied credit entirely or offered credit at interest rates that reflect the lender's uncertainty about their risk profile rather than their actual creditworthiness.
Open banking data is transformative for this population. When a lender can access 12 months of bank transaction history through a consented open banking connection, they can observe a borrower's income patterns, expenditure habits, savings behavior, financial commitments, and cash flow stability directly — without relying on a credit score that doesn't exist. The predictive power of this transactional data for credit risk is, for thin-file borrowers, dramatically superior to any inference that can be drawn from traditional credit bureau data.
Alternative Data Sources: The Signal Universe Expands
Beyond open banking transaction data, a growing universe of alternative data sources is being applied to credit decisioning in ways that improve accuracy for all borrowers — not just thin-file ones. Each data source adds incremental predictive power, and the combination of multiple data sources, processed through modern machine learning architectures, creates credit decisioning systems that are substantially more accurate than any single-source model.
Rental payment data. Rental payments are often the largest regular financial commitment for younger borrowers, but historically this data was not captured by credit bureaus. Several credit reference agencies and fintech lenders in Europe are now incorporating rental payment history — accessed through bank transaction data or direct reporting from landlords and property management platforms — into credit assessments. For younger borrowers who have been renting for several years, rental payment history is a powerful predictor of credit behavior that substantially improves decisioning accuracy.
Employment and payroll platform data. Employment verification — confirming a borrower's employment status, tenure, and income level — has traditionally required manual document collection and verification. Payroll platform APIs and employment verification services can now provide this data in real time, with cryptographic verification of accuracy. For lenders assessing income stability as a key credit factor, real-time verified employment data is significantly more valuable than the payslip copies that borrowers provide in traditional application processes.
Telco and utility payment data. Regular payment of phone bills, energy bills, and water bills is a strong predictor of creditworthiness that has historically been available only to credit bureaus that collected this data from utility companies. Open banking access to bank transaction data now makes this information accessible to any lender with a consented open banking connection, without requiring direct data sharing arrangements with utility providers.
The Explainability Requirement: AI Act Implications for Credit
The EU's AI Act classifies credit scoring and creditworthiness assessment systems used by financial institutions as high-risk AI systems, subject to specific transparency, explainability, and oversight requirements. This regulatory classification has significant implications for the design and deployment of AI credit decisioning systems in Europe.
High-risk AI systems under the AI Act must, among other requirements: provide meaningful information to affected individuals about the basis for automated decisions affecting them; implement human oversight mechanisms for significant adverse decisions; maintain detailed documentation of training data, model architecture, and validation methodology; and undergo conformity assessment before deployment in certain contexts.
These requirements are technically achievable. The AI credit decisioning companies that are building for the European market are implementing explainable AI techniques — SHAP value attribution, counterfactual explanation generation, feature importance visualization — that can produce human-readable explanations for credit decisions without materially compromising predictive accuracy. But these capabilities must be built into the system architecture from the beginning; they are not easily retrofitted into systems designed without explainability in mind.
Founders building AI credit decisioning systems for the European market who treat AI Act compliance as a design constraint from day one — rather than a regulatory obligation to be addressed after launch — will have a meaningful competitive advantage. The regulatory burden of AI Act compliance is real, but it is front-loaded: systems designed with explainability, data governance, and human oversight built in will face significantly lower ongoing compliance costs than systems that must be retrofitted.
SME Lending: The Most Underserved Segment
While consumer credit decisioning is the application that attracts the most media attention, the most economically significant opportunity for AI-powered credit decisioning in Europe is in SME lending. As noted elsewhere in this report, the European SME finance gap represents approximately €400 billion annually — a structural market failure driven primarily by the cost and inaccuracy of traditional SME underwriting rather than a fundamental absence of creditworthy SME borrowers.
SME credit decisioning is more complex than consumer credit decisioning because small business credit risk involves both the financial health of the business itself and the financial circumstances and behavior of the business owner. Integrating business financial data — revenue trends, cash flow patterns, accounts receivable aging, supplier payment history — with personal financial data for the business owner, in a model that weights these inputs appropriately for different business types and sizes, is a genuinely hard machine learning problem.
The companies solving this problem are building valuable intellectual property in the form of trained models that have been validated against historical SME loss data. This trained model represents a competitive asset that becomes more valuable with every additional loan originated and every piece of outcome data collected — a genuine data flywheel that makes first-movers in specific SME lending niches increasingly difficult to displace.
Investment Thesis: The AI Lending Infrastructure Stack
At Elinuse AI Capital, AI credit decisioning is one of the fintech investment themes where we have the deepest conviction. We are looking for companies at multiple layers of the AI lending infrastructure stack: decisioning engines that can be licensed to banks, credit unions, and neobanks; alternative data aggregation and enrichment platforms that provide the raw inputs for AI credit models; explainability and model governance platforms that help financial institutions comply with AI Act requirements; and direct lending businesses that use proprietary AI decisioning to serve underserved borrower segments.
The market timing is ideal. Open banking data access has matured. AI model capabilities have advanced dramatically. The regulatory framework under the AI Act and GDPR is clarifying. And the demand from the 25 to 30 percent of European adults who are underserved by traditional credit scoring is growing as financial inclusion becomes a policy priority across the EU.
Key Takeaways
- 25–30% of European adults have thin or no credit files, making them invisible to traditional scoring — despite being genuinely creditworthy.
- Open banking transaction data is transformative for thin-file borrowers; it enables accurate credit assessment without requiring prior credit history.
- Rental payment data, employment platform APIs, and utility payment data expand the predictive signal universe beyond traditional credit bureau inputs.
- The EU AI Act classifies credit scoring as high-risk; founders who design for explainability from day one have a meaningful compliance advantage.
- SME lending is the largest and most structurally underserved segment; AI models trained on business + owner financial data create valuable, compounding data moats.