What it means
Model evaluation is the systematic testing of an AI model or system to measure how well it performs on the tasks it is meant to do, using defined test cases, metrics and quality criteria. Good evaluation reflects real use, including difficult and unusual cases.
Real-world example
Before launching an internal assistant for its HR policies, a company builds a test set of 200 real employee questions with approved answers written by HR experts. Each new version of the assistant is scored for correctness, source accuracy and appropriate refusals, and it only goes live if it meets agreed thresholds.
Why it matters for AI adoption
Demos and a few successful examples do not show how a system performs at scale. Evaluation provides evidence for decisions, helps compare models and vendors, and reveals problems before users find them. It needs to be repeated whenever the model, data or prompts change.
Questions to ask
- What test cases and metrics show that this system is good enough for its purpose?
- Who defined what "good enough" means, and is evaluation repeated after every change?