An organization is using a large language model (LLM) to assist in evaluating loan applications,...

Incomplete training data often leads to underrepresentation of certain applicant types, products, or scenarios. In credit and lending, this typically translates into systematic bias : some groups are evaluated on richer historical patterns, while others are evaluated on sparse or unrepresentative information. The greatest associated risk is therefore unfair loan decisions (A), which can manifest as unjustified rejections, inappropriate pricing, or inconsistent risk assessments.

While delays (B), reduced satisfaction (C), or increased manual work (D) may occur, they are secondary operational issues. AAIA highlights that for financial services, the central risks include fairness, discrimination, regulatory compliance, and reputational impact . Incomplete data directly undermines fairness and can violate lending regulations and internal risk appetite.