Definition
Percentage of AI-flagged reliability trends that result in a meaningful maintenance program action within 6 months of flagging: measuring whether the ML reliability analysis system is generating actionable intelligence or producing a volume of flags that consumes investigation resources without changing anything in the maintenance program.
The Measurement Gap
The measurement gap RAOVI addresses: reliability programs track the flags their AI analysis system generates and the maintenance program changes they implement, but rarely examine the relationship between those two data streams. A reliability manager knows how many flags were reviewed this quarter and knows what interval changes and fleet campaigns were approved, but doesn't have a metric connecting the two: what fraction of the AI's flags actually drove the actions? The flag volume and the action rate are tracked in separate workflows that never produce a combined analysis.
Without this connection, tuning the AI's sensitivity becomes a qualitative judgment. Too many flags and the reliability team is overwhelmed, but how many is too many? RAOVI provides the answer empirically: when RAOVI drops below 20%, the team is spending 80 cents of investigation effort to generate 20 cents of action value. That ratio justifies raising the flagging threshold, regardless of whether the team feels overwhelmed.
The decision RAOVI supports: should the ML reliability system's flagging threshold be raised to reduce noise and improve investigation efficiency? Should the system be retrained to better match current fleet composition and maintenance program content? Should the reliability board review process be redesigned to accelerate action on high-confidence flags? These decisions require knowing the current ratio of flags to actions, which RAOVI makes visible. Without it, the reliability team's intuition about "too many flags" remains undocumentable.
Formula
Plain language: Count all reliability flags generated by the ML system. Track which flags resulted in a concrete maintenance program action within 6 months. Divide actioned flags by total flags.
Notation:
RAOVI = Actioned_reliability_flags / Total_ML_reliability_flags
Variables:
- Actioned_reliability_flags = ML-generated reliability flags that resulted in a documented maintenance program action within 6 months (count)
- Total_ML_reliability_flags = all reliability flags generated by the ML system in the measurement period (count)
What counts as "action":
- Task interval change submitted to and approved by the reliability board
- Fleet campaign initiated (no mandatory removal required; a campaign to inspect or replace on an accelerated basis qualifies)
- Task added to or removed from the approved maintenance program
- Engineering investigation completed with documented findings that influenced a maintenance program decision
What does NOT count:
- Investigations that conclude "no action required". These are "investigated but not actioned." Track separately; they indicate noise.
- "Continue monitoring" decisions without a scheduled follow-up milestone
- Verbal discussions at reliability meetings without documented outcomes
Key diagnostic split: distinguish "investigated but not actioned" from "ignored." Flags investigated and found non-actionable after legitimate analysis mean the AI is generating noise. Flags not investigated at all mean the team doesn't trust the system. Both problems require different responses. RAOVI should be accompanied by an investigation completion rate to distinguish the two failure modes.
Proposed formula, pending validation pilots.