Case study / Applied machine learning
Commodity demand forecasting
A multi-model supply-and-demand forecasting system for one of the country’s leading avocado suppliers, evaluated against real purchasing decisions.
Production context
$100M+ commodity volumeProduction context
3 years beating marketProduction context
Multiple forecasting modelsThe challenge
The operating problem.
Perishable inventory, volatile supply, and uncertain market prices made purchasing decisions materially expensive to get wrong.
The engineering
What was built.
Multiple forecasting approaches were developed, compared, and combined around the purchasing decision so expected demand, changing market conditions, and uncertainty could be evaluated together.
Outcomes
What the system made possible.
- Forecasts tied directly to purchasing decisions
- Uncertainty treated as an input rather than hidden
- A repeatable system that outperformed open-market buying over three years
Have a related problem?
Talk directly with an engineer.
Bring the messy version. We will help identify the smallest credible path to production.