The Case for Machine Learning in Supply Chain Planning

The case for adopting machine learning techniques is logical. It is faster, cheaper, more accurate and is less human input for higher quality output. The case is very obvious — let’s take a look why.

Shaun Phillips

Machine Learning is an application of artificial intelligence in which systems can analyze large volumes of data and make predictions based on correlation and causation without being explicitly programmed. It has been a major technology breakthrough and has been applied across multiple functional areas in organizations including supply chain.  A common application of machine learning is to predict machine failure based upon sensor data such as vibration, operating temperature, noise levels or fluid pressure.  This improves plant utilization and reduces costly supply chain disruptions.

In the past decade or so, there have been some major technology breakthroughs in the areas of storage, data and communications. The invention of the hard drive has allowed us to have write once, read many, fault tolerance and distribute storage at a very low cost. The extension of Moore’s law has allowed us to put up to ten million transistors on one chip. These chips allow us to perform a huge amount of transactions over large amounts of data. This in itself, has led to the proliferation of machine learning applications in supply chain. The case for adopting machine learning techniques is logical. It is faster, cheaper, more accurate and is less human input for higher quality output. The case is very obvious — let’s take a look why.

At its essence, supply chain planning is about balancing supply and demand to achieve alignment with a corporate strategy. To achieve this balance, supply chain planners are bombarded with large volumes of both historical and real time data. With that data, one can identify trends, outliers and exceptions. That data, however, also raises a large number of questions to be answered including:

  • How much or less do I need to sell?
  • How much capacity will I have?
  • How long will it take to deliver?
  • How will this new product launch go?

Having enough of the right kinds of data allows supply chain planners to make informed decisions. The more data, the more informed the decision making process. This is where machine learning comes to play.

There is a very large number of potential applications for machine learning within supply chain planning.  These include data cleansing, product clustering, forecasting with multivariate regression, new product introduction and promotions planning.

Data accuracy is a large problem for decision support systems like supply chain planning.  Machine learning is used to identify missing, rogue or duplicate data points and uses history and historical actions to correct the data.  If a sales order quantity is recorded in a wrong unit of measures (e.g. 1 unit instead of box-of-20), machine learning firstly identifies the outlier and secondly knows to apply a unit conversion to bring the value into an expected range of behavior. This leads to a more automated and more accurate forecasting process.

Machine learning is also useful for augmenting decision-making, especially for new product launches. It looks at the attributes of a product, the geography of a product, the correlating data and provides insights on how a new product launch will behave.  Consider a manufacturer launching a new flavor of juice and needs to accurately predict the market reaction. Machine learning looks at very large number of possibly influencing factors. These may include Sales of similar pack types, sales of similar juices, previous product launches, market size, demographics of target market, future weather patterns.  It identifies which data points influence behavior and by what degree of influence.  It can then determine a future demand profile for the new product launch and life-cycle.

The future of supply chain technologies will be highly automated and highly responsive. Machine learning has the ability to recalculate and process large volumes of real-time data. This data can potentially invalidate the current plan or identify a fresh opportunity for a current plan. Being constantly bombarded with these real-time impacts and opportunities will change the way we do supply chain planning today.

The future for machine learning with supply chain planning will be very different but very exciting. It is definitely a new way of planning. It is a great time to be in the supply chain technologies business.