Catalog Merge False-Positive Rate (CMFPR)

GMI-MROSC-018MRO Supply ChainPilot-ReadyPattern O-1: Confidence Calibration

Definition

The fraction of high-confidence AI-recommended item-master merges that prove to be incorrect (legitimately unique parts merged together) as evidenced by downstream wrong-part-issue events, manual reversals, or audit corrections.

The Measurement Gap

Legacy MRO supply chain metrics reward catalog cleanup volume without measuring false-positive consequences. Item-master deduplication rate measures merge volume. Inventory accuracy measures eventual cleanup. Wrong-part-issue rate measures the consequence but rarely traces the cause back to the specific merge that created it. When the AI catalog cleanup tool merges a viton hydraulic seal kit with an EPDM hydraulic seal kit because descriptions, dimensions, and part numbers are similar (but the material difference buried in a specification field was not weighted by the AI) the issued seal kit fails in service contact with the wrong fluid, producing asset damage. A retroactive review of 3,400 high-confidence merges may reveal dozens of false merges, most not yet surfaced as field problems. The merge-rate win looks much less attractive against realized error rate. The decision CMFPR supports: whether AI catalog cleanup recommendations can be applied automatically, require human review by confidence band, or require engineering sign-off for critical-class merges.

Formula

CMFPR = false_merges / total_high_confidence_merges

Variables:

  • false_merge = merge reversed within audit window, OR wrong-part issue traced to this merge, OR engineering rejection after review
  • high_confidence = above the auto-merge threshold set by the catalog cleanup AI
  • surfacing_window = 90 days for general parts, longer for slow-mover or rotating items

Worked example: Last year, 3,400 high-confidence merges. Surfaced false merges: 47. CMFPR = 47 / 3,400 = 0.0138, or 1.38%. By criticality class: critical-class CMFPR = 0.4% (within tolerance); commodity CMFPR = 1.5% (acceptable for low-stakes parts). The headline is dominated by commodity volume; the critical-class number is the operationally important read.

Edge cases: False merges may not surface until parts are issued. 90-day surfacing window minimum, longer for slow-movers. Reversed merges in audit history are unambiguous; field-issue-driven discoveries require linkage from issue records back to the merge. Some false merges will never surface because the merged part is never issued from the incorrect record. These are undercounted by definition.

Data sources: Merge events with original IDs and confidence from catalog cleanup AI log. Reversals from item-master audit log. Wrong-part issues from MATUSETRANS returns and work order rework records. Linkage from wrong-part issues back to specific merge events typically requires custom field augmentation.

Proposed formula, pending validation pilots.

At a Glance

Unit
Percentage
Trend Direction
Lower
Reporting Frequency
Monthly
Category
F - Inspection, Provenance & Master Data Integrity
← Vendor Consolidation Fragility Index (VCFI)Certification Verification Currency Rate (CVCR) →

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