Steady gains in precision
Reliable results
Retail and product-level outcomes
Reduced forecast errors
A CPG company faced month-on-month fluctuations in tea shipment forecast accuracy at the retailer level, and at the SKU level.
Machine learning was needed to capture market dynamics for better forecasting. The company aimed to address missing data, e.g. point-of-sale and promotional inputs, and inconsistent shipment patterns at the retailer level for certain SKUs.
Key challenges
Required accurate forecasts for better promotions and supply chain
Large month-on-month variations at retailer and SKU level
Need for clear shipment patterns at the retailer-SKU level
Need for critical inputs like sales and promotional data
The solution
Feature engineering
Identified key predictors
Analyzed promo impact
Derived high number of features
Feature optimization
Refined features with ‘greedy’ method
Key factors: EPOS, date, week
Tuned for SKU accuracy
1
Model development
Combined with a neural network
Optimized accuracy and bias
Built retailer and SKU models
2
Ensembling and tuning
Integrated models for precision
Used dual-objective function
Adapted to trends
3
Deployment and integration
Embedded into forecasts
Enabled improvements
Delivered insights
Accuracy boost
Demand outpaced forecasts
Significant accuracy
Reliable advancement
Precision gains
Sharp retailer-level accuracy
Improved fairness and accuracy
High SKU-level accuracy
Business impact
Optimized inventory forecasting
Insight-led decisions
Lowered mistake frequency
Scaling AI adoption
Expand AI-driven forecasting to other product categories
Enhancing model precision
Refine algorithms for greater accuracy
Grow supply chain efficiency
Use insights to optimize inventory and reduce waste
