COMPEL Glossary / z-score
Z-Score
A z-score is a statistical measurement describing how many standard deviations a data point is from the mean (average) of its dataset.
What this means in practice
A z-score of 0 means the value is exactly average, a z-score of 2 means it is two standard deviations above average, and a z-score of -3 means it is three standard deviations below average. Z-scores are widely used in AI systems for anomaly detection: values with extreme z-scores (typically beyond +/-3) are flagged as potential outliers requiring investigation. Applications include fraud detection (transactions with unusual amounts or patterns), quality control (measurements outside normal tolerance), cybersecurity (unusual network behavior), and model monitoring (detecting when input data distributions have shifted from their training baselines). Z-scores provide a simple, interpretable method for identifying unusual patterns in production AI systems.
Why it matters
Z-scores provide a simple, interpretable method for anomaly detection in production AI systems, applicable to fraud detection, quality control, cybersecurity, and model monitoring. Their interpretability makes them accessible to non-technical stakeholders who need to understand why an alert was triggered. For model monitoring specifically, z-scores detect when input data distributions have shifted from training baselines, providing early warning of potential model degradation.
How COMPEL uses it
Z-scores are used within the Technology pillar's monitoring and observability infrastructure, assessed during Calibrate as part of model monitoring maturity. During the Produce stage, z-score-based anomaly detection is implemented for production system monitoring. The Evaluate stage uses z-score alerts to identify data drift and performance degradation. The Process pillar includes z-score thresholds in operational runbooks for incident response.
Related Terms
Other glossary terms mentioned in this entry's definition and context.