01 / FORECAST LAB · SUPPLY CHAIN

From delivered weight
to a capacity plan.

Explore monthly shipment weights from a real supply-chain dataset. Compare three models, inspect historical error, and convert the forecast into pallet equivalents using an assumption you control.

This is a method demonstration. The data shows completed deliveries – not necessarily all demand the customer wanted.

Perspective
400 kg750 kg1.200 kg
Forecast, next 3 months–Jan–Mar 2015
Capacity need––
Historical WAPE–Jan 2013–Dec 2014
Vs seasonal reference–WAPE

Forecast, next 3 months

…

HistoryActual holdoutForecast
10,324order lines
7,030shipments
90.6%weight coverage
43destination countries

Data foundation

What is tested?

10,324 order lines are consolidated into 7,030 shipments. 6,372 shipments have usable weight (90.6%). The dataset covers 43 destination countries from May 2006 to September 2015; 2015 is incomplete.

A rolling test from January 2013 to December 2014 compares the same month one year earlier, the mean of the latest six months, and a regularised trend+season model. January–March 2015 is held out as a visible check period.

View the data source ↗

Limitations

What the data cannot prove

  • Pallets are not recorded. Pallet equivalents are only weight divided by your kg assumption.
  • Delivered volume is not the same as unconstrained demand; stock-outs, programme decisions and procurement can affect the series.
  • Small segments are lumpy. High forecast error should change the decision, not be hidden.
← Back to RÄVIR Labs