Business metrics
person concept tool org talk claim — click a node to jump to its page; hover an arrow for the relation
Business metrics are one of five distinct flavors of evals signal used to evaluate LLM-based systems, representing the downstream commercial outcomes that an AI application is intended to drive.
Role in the evaluation stack
According to Dat Ngo of Arize, business metrics sit alongside LLM-as-a-judge, human feedback, golden datasets, and deterministic checks as a core signal type in a comprehensive evaluation framework. He characterizes them as "how do I make more money" in their most direct form — i.e., revenue, conversion, retention, or other financial and product KPIs that ultimately reflect whether an AI feature is delivering value. → source
Significance
Business metrics occupy a distinctive position among eval signals because they are the furthest downstream: while LLM-as-a-judge or deterministic checks assess the quality of individual model outputs, business metrics capture whether those outputs translate into real-world impact for the deploying organization. They serve as a ground-truth anchor for the entire evaluation pipeline — a system may score well on intermediate signals yet still fail to move the metrics an organization actually cares about.