Case study

Integrated Demand Forecast


A comprehensive study by Forecast predicted demand for final box products across the client's Australian and New Zealand manufacturing sites using a sophisticated machine learning model, significantly improving forecasting accuracy and supply chain optimisation.


Our client is the world’s largest privately owned paper, packaging and recycling company. Operating more than 180 facilities worldwide.


As part of a broader Supply Chain Transformation program, the client engaged Forecast to conduct a study to predict demand for final box products across their network of manufacturing sites in Australia and New Zealand.

The benefits associated with improved forecasting across an integrated network of paper manufacture, logistics, board manufacture, box manufacture, and distribution are significant. An accurate demand forecast enables optimisation of the supply chain; production planning, inventory, logistics, waste reduction, and improved DiFoT.


Forecast worked closely with the client’s project team and external advisors to agree the objectives and granularity of the forecasts, which included over 25,000 combinations of product SKUs, Customers, and Sites. Time periods included short term (daily, 10 days), medium term (weekly, 3 months), and long term (monthly, 2 years).

Following a period of data extraction, cleansing, and exploratory analysis Forecast embarked on the development of a sophisticated demand forecasting time-series machine learning model. Developed in Python the demand forecast model tests the predictive strength of multiple time-series algorithms across over 25,000 unique combinations to develop the demand forecasts by manufacturing site.

A detailed report of the study included an Executive Operational Summary, Analytics Summary, Detailed Analytics Report, and Appendix of Key Analytics Techniques. This enabled the sharing of report outcomes with different users within the organisation.


Following a short (8 week) study on the potential application of data analytics for demand signalling, the forecast model achieved above expectation forecasts, with a step-change confidence level improvement compared to the current operations approach.

The next step will involve trials of the demand forecasts across a selection of manufacturing sites, followed by full roll out and productionising the demand forecasting tool within the client’s operating systems.