Demand forecasting that sees the stockout coming
nexCommerce demand forecasting predicts demand per product per location from your own order, search, campaign and inventory history, then flags low stock and demand surges before they cause failed orders. The replenishment agent proposes quantities and timing; a planner approves. Because it runs inside the commerce platform, it sees signals ERP-only forecasting misses.
Key takeaways
- Per product, per store, per warehouse, per slot: granular enough for grocery and quick commerce.
- Uses commerce signals — searches, carts, campaigns — alongside sales history.
- Proposes replenishment and flags slow movers for promotion; people approve.
- Feeds the search, recommendation and picking agents so the whole system plans ahead.
What does the forecasting agent predict?
Five outputs, each consumed by a different part of the operation rather than sitting in a report nobody opens.
| Output | Horizon | Used by |
|---|---|---|
| Product × location demand | Daily and forward | Replenishment, procurement |
| Stockout risk | Rolling | Storefront availability, assistant substitutes |
| Surge alerts | Campaign and event windows | Fulfilment staffing, slot capacity |
| Slow-mover list | Weekly | Promotion rules, campaign planning |
| Slot demand | Hourly | Delivery capacity planning |
How does stockout prediction work?
It watches the leading indicators rather than the trailing one. Sales history says what happened; searches, carts and a campaign scheduled for next week say what is about to.
- 01Ingest sales, stock, searches, carts, promotions, price changes and calendar events.
- 02Model demand per product and location, accounting for seasonality and promotion effects.
- 03Publish forecasts and risk flags to the console and to the other agents.
- 04Propose replenishment quantities and campaign actions for a planner to approve.
- 05Learn from actuals, and report accuracy per category.
Why forecast inside the commerce platform instead of the ERP?
ERP forecasts see what sold. The commerce platform also sees what customers searched for and could not buy, what they abandoned, and which campaign is scheduled next week. Forecasting where those signals live produces earlier, more specific warnings — and lets the same forecast drive search ranking, substitutes and picking priority.
warehouse picking optimisation softwareDoes it replace ERP demand planning?
No. It produces earlier, commerce-aware signals and proposals; approved quantities sync back to the system that owns purchasing. The two answer different questions, and the ERP keeps owning the purchase order.
Test it against a season you already know the answer to
The honest way to evaluate a forecast is to run it over history you have lived through and see what it would have warned you about.
Can it handle new products?
Yes, using attribute similarity to existing products until enough history accrues for the product to be forecast on its own.
Does it work for multi-warehouse and dark stores?
Yes. Forecasts are per location and per slot, which is what quick commerce needs and what a national-level forecast cannot give you.
Who approves a replenishment proposal?
A planner. The agent proposes quantity and timing with its reasoning; nothing is ordered without a person approving it.