SKU-Level Demand Forecasting at Portfolio Scale
Hierarchical forecasting that gives per-SKU accuracy and portfolio numbers finance and operations can both plan against — reconciled without spreadsheets.
Results
The Problem
Forecasting at the SKU level and forecasting at the portfolio level are different problems, and organizations usually solve them separately — which means the numbers don't tie. Finance plans against an aggregate, operations plans against item-level demand, and the two are reconciled manually in spreadsheets every cycle. At tens of thousands of SKUs, a single statistical model applied uniformly across the catalog fits the high-volume items and fails everything in the long tail.
What We Build
Hierarchical forecasting systems that produce granular, per-SKU models while remaining coherent with aggregate portfolio forecasts. Model selection happens per series rather than globally, so intermittent and seasonal demand patterns are handled by methods appropriate to them instead of being forced through one approach.
Reconciliation is built into the pipeline: bottom-up item forecasts and top-down portfolio targets resolve to a single set of numbers that operations and finance can both plan against.
Outcome
- ◆Per-series model selection across the full catalog, not one global model forced onto the long tail
- ◆Item-level and portfolio-level forecasts that reconcile without manual adjustment
Techniques
- ◆Hierarchical time-series reconciliation
- ◆Per-series model selection
- ◆Intermittent demand modeling
- ◆Automated backtesting and drift monitoring