Definition
MBD measures how much an AI-enabled replenishment policy reduces the amplification between actual parts consumption and upstream purchase ordering compared to the pre-AI baseline.
The Measurement Gap
The bullwhip effect (variance amplification as orders travel upstream) has been documented in supply chain literature since the 1990s, yet no standard MRO metric measures it. Inventory Turns can be high while ordering swings wildly. Fill Rate can be excellent while you carry extra safety stock to absorb noise your own system creates. Forecast Accuracy measures prediction quality, not ordering stability. None of these metrics ask whether the variance of your ordering is bigger or smaller than the variance of your consumption, or whether AI is making that ratio better or worse. AI replenishment can quietly automate the bullwhip effect, generating worse ordering volatility than the manual min-max system it replaced, while every legacy metric on the dashboard looks fine.
Decision Supported:
Whether the AI replenishment algorithm is genuinely smoothing demand or amplifying it, and which Lee/Padmanabhan/Whang bullwhip driver (signal processing, batching, gaming) requires algorithmic correction.
Formula
Plain language: Take the ratio of ordering variance to consumption variance under AI, compare it to the same ratio under the pre-AI baseline, and express improvement as a fractional reduction.
Notation:
MBD = 1 − [Var(PO_qty) / Var(Issue_qty)]_AI ÷ [Var(PO_qty) / Var(Issue_qty)]_Baseline
Variables:
| Variable | Definition | Unit |
|---|---|---|
| Var(PO_qty) | Statistical variance of purchase order line quantities for an item over the time window | Each² |
| Var(Issue_qty) | Statistical variance of actual material issues to work orders over the same window | Each² |
| Bullwhip ratio | Var(PO_qty) / Var(Issue_qty); 1.0 means no amplification | Dimensionless |
| Baseline period | Pre-AI window (12–18 months recommended) | Date range |
| AI period | Post-AI deployment window (matching duration) | Date range |
Proposed formula, pending validation pilots.