Root Cause Investigation Independence Score (RCIIS)

GMI-PH-031Pharmaceutical OperationsIntegration-Required

Definition

Proportion of AI-assisted investigations where the investigator documented independent verification: physical inspection, an operator interview, or analysis outside the AI's own inputs. Its complement is the anchoring rate.

The Measurement Gap

After deploying an AI deviation analysis platform, investigation cycle time dropped from 23 days to 11 days and closure rate improved from 78% to 94% within 30 days. The quality director credits the AI with transforming investigation capability. But nobody measured whether investigations got faster because they got better, or because investigators stopped investigating.

Before the AI, investigators walked the floor, interviewed operators, reviewed batch records, and cycled through hypotheses. After the AI, the investigator opens the deviation record, sees "probable root cause: worn agitator seal, confidence 87%," and faces a choice between genuine investigation and restating the AI's conclusion. The 94% closure rate suggests the shortcut is winning.

Decision enabled:

When RCIIS drops below 0.5, mandatory independent investigation steps must be documented before AI recommendations can be accepted. When RCIIS drops below 0.8, the organization implements structured investigation checklists requiring evidence of independent analysis.

Formula

RCIIS = investigations_with_independent_verification / investigations_with_AI_recommendation

Companion metric: Anchoring_rate = 1 − RCIIS

"Independent verification" means the investigator documented at least one of: physical inspection, operator interview, data analysis not in the AI's evidence package, or testing of an alternative hypothesis.

Worked Example:

Twelve-month review, 173 deviations with AI root cause recommendations:

62 investigations included documented independent verification (floor inspection notes, operator interviews, independent data analysis, alternative hypothesis testing).
74 investigations restated the AI's conclusion with the AI's evidence and no additional steps.
37 investigations confirmed the AI's recommendation but evaluated no alternative hypotheses (classified as not independent).

RCIIS = 62 / 173 = 35.8%. Anchoring_rate = 64.2%.

Of the 62 independently investigated deviations, 11 (17.7%) found a different root cause than the AI recommended. Extrapolating: approximately 19 of the 111 anchored investigations likely adopted the wrong root cause. Cross-referencing with recurrence data: independently verified root causes had 8.1% recurrence; anchored root causes had 22.6% recurrence.

Sign Convention:

Higher is better. Above 0.8 means most investigations include independent work. Below 0.5 means majority rubber-stamping.

Proposed formula, pending validation pilots.

At a Glance

Unit
Percentage
Trend Direction
Higher
Reporting Frequency
Monthly; per investigation cohort
Category
Quality System & Investigation Effectiveness
← OOS Retest Strategy Scientific Validity Score (ORSSVS)Stability Prediction Retrospective Accuracy (SPRA) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment