Zuletzt aktualisiert am:
A leading metric (e.g., frequency of out-of-scope predictions, rate of unexplainable decisions) that signals emerging AI risks before they materialize.
A forward-looking measure that indicates rising exposure to particular AI risks - such as an uptick in low-confidence or out-of-distribution inferences, increases in flagged policy violations, or spikes in alert volumes. KRIs enable proactive governance by surfacing potential problems early, triggering risk-mitigation playbooks (e.g., throttling, human review), and feeding into enterprise risk dashboards alongside financial and operational KRIs.
Praxisbeispiel:
A healthcare provider monitors the KRI “percentage of model inferences with confidence < 50%.” When the KRI rises above 5% in a week, it signals data-drift or system issues, prompting data-science teams to investigate input pipelines and retrain the diagnostic model before patient care is impacted.
