Definition
Variation in mean visitor risk scores across visitor categories, normalized for actual incident history, to detect whether scoring differences reflect real risk or model bias.
The Measurement Gap
Measurement Gap and Decision
The Measurement Gap
Traditional visitor management systems apply uniform screening protocols or simple rule-based tiers (pre-registered vs. walk-in). There is no score to evaluate for bias because there is no score. AI visitor risk scoring introduces granular risk assessment, and with it, the possibility of systematic bias that would never surface in rule-based systems.
The measurement gap is that organizations deploying visitor risk scoring rarely check whether the scores correlate with actual risk. They assume the model is objective because it is algorithmic. VRSEI closes this gap by comparing score distributions across visitor categories against actual incident rates. If scores differ while incident rates do not, the model has unexplained bias that needs investigation.
The Decision It Supports
VRSEI tells security and compliance leadership whether the visitor scoring model needs a bias audit. High VRSEI (strong correlation between scores and actual risk) means the model is treating categories equitably relative to demonstrated behavior. Low VRSEI means the model assigns different risk levels to categories with similar incident rates, which may create legal, ethical, and reputational exposure depending on which categories are affected.
Formula
Plain Language
Group visitors by category (employer type, visit purpose, host department). Compute the mean risk score and actual incident rate per category. Calculate the correlation between scores and incident rates across categories. Strong correlation means scores track actual risk. Weak or absent correlation means scores reflect something other than risk.
Formal Notation
$$VRSEI = \rho(Mean_score_c, Incident_rate_c)$$
Variable Definitions
| Variable | Definition | Unit |
|---|---|---|
| Mean_score_c | Average visitor risk score for category c | Score (model-specific) |
| Incident_rate_c | Actual security incidents per visitor-visit for category c | Rate |
| ρ | Pearson correlation coefficient across categories | -1 to +1 |
Units: -1 to +1. A value of +1 means scores track actual risk perfectly (fair model). Near 0 means scores have no relationship to actual risk (arbitrary). Negative values mean scores are inversely related to actual risk (actively biased).
Proposed formula, pending validation pilots.
At a Glance
Decision Supported
Is the visitor risk scorer treating visitors equitably, or systematically scoring certain categories based on training data patterns rather than actual risk?