Algorithmic credit scoring is the practice of using statistical or machine-learning models to estimate how likely a borrower is to repay. The classic example is the credit score - a model trained on past borrowers that outputs a number summarizing default risk. Modern systems go well beyond traditional credit-bureau data, drawing on bank-transaction history, cash-flow patterns, device and behavioral signals, and other “alternative data” to score people with thin or no credit files.
The appeal is that machine-learning models can find predictive patterns humans miss and can extend credit to people whom rigid rule-based systems would reject. The risk is that the same models can encode bias - if historical lending discriminated, a model trained on that history can learn to reproduce it, and complex models can find proxies for protected characteristics even when those characteristics are excluded as inputs. Opacity compounds the problem: a model can deny someone credit for reasons neither the applicant nor sometimes the lender can readily articulate.
US regulators have made clear that these systems are not outside existing law. The CFPB’s 2023 circular on adverse-action notices held that lenders using complex algorithms must still give applicants the specific, accurate principal reasons for a denial under the Equal Credit Opportunity Act, regardless of how opaque the underlying model is.
Why business readers should care: algorithmic credit scoring is where AI most directly decides who gets access to money. Its promise is wider, fairer access; its hazards are hidden bias and unexplainable decisions. The deciding factor is usually whether the institution can interpret and defend what its model actually did.