All case studies
Case / 02 · Predictive ML · Commodities

Predictive Procurement Engine

An ML-driven price-forecasting and hedge-trigger engine capturing 8% YoY cost reduction on a $300M+ global coal portfolio.

YoY Cost ↓
8%
Portfolio Scope
$300M+
Forecast Accuracy
94%
Context

Procurement was timing-blind: contracts were placed against rolling-average benchmarks, leaving meaningful basis on the table in volatile freight + index environments.

Approach
  1. Built a forward-curve model combining freight indices, FX, regional supply-demand, and weather priors.
  2. Backtested across seven years of monthly procurement decisions; tuned to a P50/P80 cone rather than a point forecast.
  3. Wrapped in a hedge-trigger rule set co-designed with the commercial desk — never autonomous, always advisory.
Outcome
  • Trailing twelve-month landed cost reduced 8% vs benchmark.
  • Hedge coverage at advantageous tenors rose 2.4x.
  • Commercial team adopted the surface as their default morning read inside one quarter.