7.3 Module 7 · AI-Assisted Performance Management

Calibration Analytics

Visualise rating distributions across departments, detect manager bias patterns, and compare before-and-after calibration adjustments with statistical indicators.

Distribution Visualiser Calibration Analyser

Distribution Visualiser

Compare the performance rating distribution across departments. Toggle departments to see how each compares to the expected bell curve. The ideal distribution line helps identify skew.

Manager Outlier Detection

Identifies managers who rate significantly higher or lower than their peers. This is not proof of bias, but a signal that warrants investigation during calibration.

Calibration Analyser

See how calibration adjustments change the overall distribution. The before view shows raw manager ratings; the after view shows post-calibration results. Toggle to compare the impact.

Correlation Analysis

Examines whether non-performance factors (gender, tenure) correlate with ratings more than expected. Strong correlations warrant deeper investigation.

Interpretation Guide
Low Correlation (r < 0.2)

Factor has minimal relationship with ratings. Expected for demographic variables in a fair system.

Moderate (0.2 < r < 0.4)

Some relationship exists. Investigate whether legitimate performance factors explain it.

High Correlation (r > 0.4)

Strong relationship that likely indicates systematic bias. Urgent calibration review recommended.

Key Insight: Where Bias Hides

Calibration is where unconscious bias becomes visible. AI can surface patterns that humans miss: a manager who consistently rates women one notch lower, a department where tenure predicts rating more than performance, or a team where everyone gets the same score regardless of output. These patterns do not prove intent, but they demand investigation.

The critical distinction is between detection and decision. AI detects the statistical anomaly. Humans must decide what is fair. A manager who rates highly might have a genuinely exceptional team. A skewed distribution might reflect real performance differences. The value of calibration analytics is not in automating fairness, but in ensuring that every deviation from the norm is a conscious, justified choice rather than an invisible habit.