Blend Give-Away Distribution Index (BGDI)

GMI-PI-016Process IndustriesPilot-ReadyPattern O-6: Autonomous Coherence

Definition

Coefficient of variation of quality give-away across batches within a product grade: detecting whether the AI blend optimizer is achieving low average give-away by tightly targeting specifications on some batches while over-dosing others, rather than uniformly minimizing give-away across the population.

The Measurement Gap

Blend optimization is reported as average give-away. The average margin by which finished product exceeds specification minima on the parameters that matter. Average give-away is the headline metric in blending. It is also gameable. An optimizer can drive average give-away down by producing some batches at essentially zero margin (off-spec risk if a re-test slips) and other batches well above specification (wasting premium components). The average looks good; the distribution is dangerous on one tail and expensive on the other. Average give-away, on-spec rate, and reblend frequency all measure aggregate outcomes that hide distribution shape. Customer complaint rate is a lagging indicator that arrives after low-give-away batches have been shipped. Premium component cost shows up in the monthly cost variance, attributed to crude or component supply variability. The bimodal distribution pattern (clustered low and clustered high give-away) is the signature of optimizer gaming, and it is invisible without explicitly measuring distribution dispersion.

Decision Supported:

Is the blend optimizer producing consistently lean give-away across all batches, or is it averaging out by producing some at off-spec risk and others with excess quality. Wasting high-value components?

Formula

Plain language: For each batch, compute give-away as the percentage by which finished quality exceeds specification minimum. Compute the standard deviation and mean of give-away across the batches in the grade. BGDI is the ratio: standard deviation divided by mean. Low values mean uniform give-away. High values mean the optimizer is averaging out across a wide distribution.

Notation:

give-away_b = ( actual_quality_value_b − specification_minimum ) / specification_minimum × 100

BGDI = σ(give-away) / μ(give-away)

Variables:

Variable Definition Unit
give-away_b Give-away percentage for batch b, on the critical specification parameter percentage
σ(give-away) Standard deviation of batch-level give-away percentages within the grade percentage
μ(give-away) Mean give-away percentage within the grade percentage
BGDI Coefficient of variation: σ/μ dimensionless

Compute per product grade; different grades have different give-away profiles and combining them is meaningless.

Proposed formula, pending validation pilots.

At a Glance

Unit
Coefficient of variation
Trend Direction
Lower; uniform give-away across batches
Reporting Frequency
Per product grade, per measurement period
Category
Safety, Environmental and Regulatory Margin Governance
← Twin Calibration Drift Index (TCDI)Permit Margin Consumption Index (PMCI) →

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