Machine Learning Applied to Improve Forecast Accuracy

Supply Chain Digest Says... The complexity of using causal factors was often sited as a barrier to their use. Perhaps AWS has found a simpler way. What do you say? Click here to send us your comments   Click here to see reader feedback

That’s what we first wondered in late 2022, when AWS announced it was entering the supply chain application arena with a new demand and inventory planning solution. (See Amazon AWS Launches New Supply Chain Applications.)

At the time, AWS said the new platform was an outgrowth of IT development for Amazon’s own needs, including supply chain software.

Amazon also said the new application would leverage machine learning in a way that would give companies greater insight into the current and future states of their inventory.

Since then, not a lot of news until last week, when in a blog post AWS announced that AWS Supply Chain now allows demand planners to input factors (demand drivers) that could influence forecasted demand.

AWS says driver-based forecasting is an adaptive approach that enables machine learning models to capture many influence factors that improve forecast accuracy.

This sounds quite similar to use of what is often called “causal factors” to shape the forecast, things that might include the weather, a big sporting event, competitive activity or other inputs.

SCDigest believes relatively few companies currently use causal factors as part of their demand planning process or technology.

But now AWS says that demand planners can input historical and/or future demand drivers such as price changes and promotional events directly into the AWS Supply Chain Data Lake.

(See More Below)

It is not completely clear, but AWS says its solution allows incomplete or maybe basic use of demand drivers to start with. Then its machine learning algorithms can refine the baseline forecast by correlating historical trends with demand drivers.

AWS Supply Chain measures the relative influence of demand drivers on the forecast with an impact score. These impact scores are displayed on the user interface so demand planners can easily interpret whether a demand driver has as a comparatively higher or lower impact on the forecast. This enables improved forecast accuracy and better decision-making, AWS claims.

The complexity of using causal factors was often sited as a barrier to their use. Perhaps AWS has found a simpler way.

Amazon says customers using its current demand planning system as part of AWS supply chain can get the demand drivers capability at no additional cost.

Any reaction to this AWS tool? Let us know your thoughts at the Feedback section below.