Transparent, not a black box
Classical statistical methods for the vast majority of SKUs — reproducible, auditable, explainable. Nothing you cannot check against your own history.
Forecasting
Know how much to trust the forecast before you buy against it.
A per-SKU forecast with its own track record. You see how accurate it has been on that SKU, so you know when to lean on it and when to use your judgement.
What it does
Capabilities
Forecasts are only useful to a buyer if they come with a sense of how reliable they are. Every forecast here is scored against what actually sold, per SKU, and that score is shown wherever the number is used.
See Demand Forecasting on your own catalog.
How it's different
Built for buyers who've outgrown spreadsheets.
Scored against reality
Every forecast is compared to realized demand. You see, per SKU, whether the model has earned your trust.
For the technical reader: how demand forecasting works under the hood
- Ensemble of AutoETS, AutoARIMA, AutoTheta, AutoCES, and MSTL; CrostonOptimized for intermittent SKUs.
- Change-point detection: recent regime dominates fitting, full history drives the seasonal profile.
- Method selection per SKU based on rolling realized APE stored in forecast_actuals.
- A daily shadow run evaluates every candidate method on a sample so method selection improves continuously.
- Runs on a dedicated batch worker with parallel SKU processing; integrated with buy-box share and stockout history for demand correction.
Related
Works better with the rest of the purchasing view.
Common questions
The questions purchasing teams ask before a demo.
See the purchasing workflow on a catalog like yours.
A 30-minute walkthrough with a product engineer. Your data stays yours.