The self-learning supply chain: Turning manufacturing into a self-adjusting process

By Markus Malinen, Vice-President of Europe, Middle East, Africa and Russia, Quintiq, a Dassault Systèmes brand.

A key challenge for manufacturers is the need to be ready to meet their customers' demands at all times, but without holding stock unnecessarily.

Getting it wrong can mean losing their business to a competitor who can step in to deliver the goods. Unfortunately it's a problem that requires more than simply adjusting production to current demand.

To reduce wastage and protect profit margins, the right products have to be delivered at the right time and prices. To achieve this, manufacturers need raw materials to be available in time, in addition to the right resources in the right quantity to meet production requirements. Balancing capacity and demand is a complex process that requires a smooth workflow and every part in the supply chain to be in sync with one another.

Conventional planning is no longer sufficient to meet the dynamic requirements of manufacturing supply chains. Machine learning technology is a growing necessity due to functionality that allows analysis of real-time data and the ability to quickly respond to deviation in processes.

By allowing the supply chain to learn from business reality, advanced analytics is offering a level of accuracy and insight once thought impossible. This technology learns from recent data, and uses the acquired knowledge as input for planning and optimization. The continuous loop of learning and application forms the basis of an approach known as the Self-Learning Supply Chain.

The cost of falling behind

While advanced analytics is acknowledged to impact business efficiency and profitability, it is not widely used and its integration with the supply chain even less so. To see why, it's important to understand the three stages of analytics:

Descriptive analytics
Analyzing past actions and results.

Predictive analytics
Predicting the future behavior of the manufacturing process and the effects of future choices and actions.

Prescriptive analytics
Making better business decisions informed by predictive analytics and driven by optimization technology.

Gartner reports that 70% of companies surveyed are in the earliest stage of analytics (descriptive), which looks to the past to describe what has happened. Only 30% are in the advanced stages: 15 - 25% practice predictive analytics and 1 - 5% prescriptive.¹

This low adoption of advanced analytics can really affect an organization's competitiveness. Companies that master data and knowledge have supply chains that are more resilient, responsive and far ahead of the competition. Those not using it struggle with inaccurate estimates, inefficient planning and wastage.
To achieve the higher returns from the effective use of data, companies need to move from considering "What has happened?" to "What will happen?" and ultimately, "What actions should we take?"

Not all data is useful

Many manufacturers struggle to obtain and maintain the level of quality required to extract knowledge that allows them to make smart decisions. Low quality data that produces inaccurate knowledge results in unworkable plans, lost opportunities and high costs. No less than 91% of businesses make common errors that result in inaccurate, incomplete and outdated data while 81% struggle to generate meaningful insights because of data inaccuracies.²

A solution is to embed advanced analytics into the supply chain. The self-learning capability enables the supply chain to learn from recent data and replicate the logic and reasoning of the company's best decision-makers to prescribe actions and generate optimal plans that work in the real world.

The three steps of the self-learning process

1. Data capturing
Data is captured from various channels ― for example, order characteristics, process setup times and processing times from the production floor. Outliers in the data are identified and filtered out.

2. Knowledge extraction
The Self-Learning Supply Chain's algorithms identify patterns in the data and process these into knowledge that forms the input for planning.

3. Planning and optimization
Taking knowledge produced by the Self-Learning Supply Chain as input, optimization technology constructs plans that maximize the company's KPIs. The plan is presented for execution in the real world, taking into account uncertainties and inevitable variances. The actual execution data (e.g. setup times, processing times and waiting times) is captured and fed back into the system. This continuous cycle generates knowledge that reflects the reality of operations. As knowledge is continuously refined to reflect the reality of operations, knowledge becomes increasingly precise, and plans become increasingly effective and efficient.

Business value

With knowledge that corresponds to reality, the estimates used in planning becomes more accurate. As operational efficiency improves and planners have reliable decision-making support, customer satisfaction improves, the use of assets and resources is optimized and profitability increases.

The Self-Learning Supply Chain in real manufacturing situations

Climate change, rising population and rapidly changing dietary habits have created much uncertainty for seed manufacturers. They struggle to balance production planning, sales planning and customer satisfaction with consistent revenue generation.

To support seed manufacturers in making smarter decisions, the Self-Learning Supply Chain uses historical and present data to predict seed volume, quality and demand. An example: If supply appears at risk due to seasonality, the manufacturer can take immediate action to secure a greater volume by contracting additional suppliers long before the shortage is predicted to occur.

Another business prone to supply uncertainty is meat processing. Even if the number of animals in consecutive orders is consistent, each animal in those orders can be of a different size and weight. The Self-Learning Supply Chain uses historical data to predict the amount, quality and type of meat available for processing. This empowers planners to create accurate production plans before the animals arrive at the facility.

Now let's take a look at liquid chemical processing. Raw materials are processed in batches to produce the intended chemicals and chemical by-products. The exact yield of these processes and the amount of by-products and waste are difficult to determine in advance, hampering the manufacturer's ability to create an effective sales plan.

Like the two earlier examples, the Self-Learning Supply Chain applies machine-learning algorithms to historical data to accurately predict the yield and the amount of by-products and waste. Manufacturers that want to achieve optimal production to keep costs low while pleasing their customers need highly accurate production and sales planning — two capabilities supported by advanced analytics.

This new technology is instrumental in dealing with uncertainty and the dynamics of frequent real-time changes. Doing so allows manufacturers to transform data flowing through the supply chain into concrete business value.

¹ Jeanne Harris, Thomas Davenport; Competing on analytics: The new science of winning (Gartner)
² The State of Data Quality, Experian, 2013