Forecast Performance

Intro

The COO indicated that their current forecast performance was poor, Sequoia asked to recommend forecast performance KPIs.
Sales are promoted via a catalogue which highlighted a subsection of theorecasting techniques available portfolio.
Demand was highly variable and significantly inflated when a SKU appeared in the catalogue.

We introduced new measures - Sequoia’s Forecast Value Add (COV)® and Bias - to provide a better interpretation of their forecast performance.
There was strong evidence that SKUs had insufficient data to create reasonable forecasts until they had been in a catalogue more than 3 times – this was defined as the steady-state.
Working with a SME from the client’s demand planning team, we created a list of sales uplift factors to explore and employed machine learning and back-casting techniques to identify realistic KPI targets.
Key to the success of this approach was the comprehensive data set that the client had maintained of the layout of each historical catalogue. This allowed us to link sales to not only to when it was in the catalogue, but the specific pages it appeared on and how it was displayed on the page.

Using AI forecasting techniques to mechanise the forecasting of steady-state SKUs provided 25-60% improvement in forecast performance depending on the product range and markets.
Motivated by the impact of data analytics on improving forecast performance, we continued to work with the client to jointly build and deliver a 1-day forecast analytics and 3-day supply chain interactions and optimisation course across their global network.
Just as important was the manhours that were freed up with this new approach, allowing the demand planners to focus their efforts on dealing with NPIs, managing discontinuations and anomalies.