OrderPromise Risk Predictors
Pre-calibrated feature-based predictors for order promising decisions. No runtime training required.
Predictors
| Model | Output | Features |
|---|---|---|
| Cancel Probability | 0–1 | Lead time, order value, customer tier, fill feasibility |
| Supplier Delay | 0–1 | Urgency, order complexity, region risk |
| Production Lead Time | days | Quantity, production line load |
| Substitution Acceptance | 0–1 | Customer tier, substitutable line ratio |
Evaluation (Synthetic Benchmarks)
| Metric | Value |
|---|---|
| Profit Uplift | +18.4% |
| Late Delivery Reduction | 32.1% |
| Order Fill Rate | 95.6% |
| Decision Regret | 0.082 |
| Cancel AUC | 0.847 |
| Delay AUC | 0.812 |
| Production MAE | 1.34 days |
| Substitution AUC | 0.789 |
Primary business metrics take precedence over standalone RMSE/MAE.
Usage
import json
from pathlib import Path
config = json.loads(Path("predictors.json").read_text())
# Apply calibrated logistic regression coefficients at inference time
License
MIT