PD usually comes from a statistical or machine learning model trained on past loan outcomes. The model takes in application data, bureau attributes and, more and more, alternative data such as cash flow, then outputs a score or a percentage. A PD of 3% means that out of 100 similar borrowers, about three are expected to default within the time horizon.

Credit teams rarely use PD on its own. It sets cut-offs in the underwriting policy, drives risk-based pricing and feeds the expected-loss formula: expected loss = PD × loss given default × exposure at default. Because the model drives real decisions, it also needs ongoing monitoring for drift and governance under model risk rules.