Definition
Proportion of confirmed cross-site deviation patterns that AI flagged before manual identification. Measures whether site-level analytics ever connect into a network view.
The Measurement Gap
A company operates 6 manufacturing sites across 4 countries. Each site's AI deviation analysis identifies trends within that site. But a raw material quality issue affecting an excipient used at 4 sites produces deviations at each site that are investigated independently. Each site's AI identifies the local trend. No system connects the dots: the same failure mode at sites in Ireland, Singapore, New Jersey, and Switzerland, all traceable to a single supplier lot.
Instead, each site finds a local explanation: compression force variability, humidity excursion, operator technique, normal batch variation. Four different root causes for one systemic issue. Four corrective actions, none addressing the actual cause. The pattern goes undetected until a regulatory inspector reviews records across sites, producing a significant GMP finding.
Decision enabled:
When CSDPDR is below 50%, the organization must invest in cross-site data harmonization and deviation taxonomy standardization. Patterns detected by regulators rather than the AI system signal the highest-risk gap.
Formula
CSDPDR = AI_detected_cross_site_patterns / total_confirmed_cross_site_patterns
Where AI_detected counts patterns flagged by AI before manual identification, and total_confirmed counts all cross-site patterns regardless of detection method.
Worked Example:
Eighteen-month retrospective across 6 manufacturing sites. Total confirmed cross-site patterns: 11.
Detected by AI cross-site analysis (3): shared HVAC filter supplier lot affecting 3 sites (month 4), process parameter drift on similar equipment at 2 sites (month 2), shared cleaning agent formulation change at 2 sites (month 3).
Detected by human reviewers during corporate quality review (5): raw material variability at 4 sites (month 9), label defect increase at 3 sites (month 7), yield decline at 2 sites (month 5), environmental monitoring pattern at 3 sites (month 11), equipment qualification gap at 4 sites (month 6).
Detected by regulatory inspection (3): stability trending at 4 sites (FDA inspector), investigation quality decline at 3 sites (EMA inspector), common-cause deviation coded differently across sites (post-483).
CSDPDR = 3 / 11 = 27.3%
The AI catches one in four patterns. Three patterns found by regulators represent the highest-risk failures.
Sign Convention:
Higher is better. Below 50% means the organization depends on manual or regulatory detection for most systemic issues.
Proposed formula, pending validation pilots.