Skip to main content

How does the demand forecast works?

Rewize blends ML patterns + trend/seasonality & reconciles across channels to project future SKU demand.

Paolo Caneparo avatar
Written by Paolo Caneparo
Updated over 2 weeks ago

Two forecasting “experts,” one blended result

Our V2 engine combines:

  • Detail model (ML) – Learns recent sales patterns, product attributes, channel differences, calendar events & promos.

  • Trend/seasonality model – Finds longer-run growth + repeating seasonal cycles in your history.

We weight by horizon: near term leans on recent, data-rich signals; longer term leans on slower-moving trends & seasonality.

Multi-channel reconciliation

Item-by-channel forecasts are rolled up so channel totals = product totals = overall demand. This prevents double counting and gives one source of truth for purchasing.

Short vs long horizon behavior

  • Short windows: more responsive to recent velocity spikes, launches, promos.

  • Longer windows: smoother, anchored to underlying trend + seasonal patterns.

You’ll see this when switching forecast horizons in the product view.

Using the forecast in Rewize

Open a product → Forecast section:

  • Pick History (e.g., 30d, 3m, 1y) to define learning context shown on the chart.

  • Pick Horizon (30d–1y+) to see forward demand. Link them (chain icon) if you want symmetric windows.

  • Toggle Daily / Weekly / Monthly granularity.

  • Include/exclude sales channels to compare channel vs total demand.

Bundles & demand allocation

If bundle SKUs are mapped to components, bundle sales are decomposed into their parts so component demand and future reorders, reflect true consumption.

Data depth & ramp-up

We can produce useful forecasts with only a few months of data (advantage vs tools needing 2+ years). Accuracy improves as more history and channel context accumulate.

Improving forecast quality

  • Connect all active sales channels during onboarding.

  • Keep supplier confirmation + incoming dates current so downstream planning aligns with demand.

  • Map bundles early; unmapped kits hide component demand.

  • Avoid ignoring SKUs you actually replenish.

Troubleshooting perceived errors

If demand looks wrong, check history/horizon selections, channel filters, bundle mapping, and sync completeness before escalating. See Why is the demand inaccurate?

Did this answer your question?