The self-learning supply chain marks the next major frontier of supply chain innovation. It's a futuristic vision of a world in which supply chain systems, infused with artificial intelligence (AI), can analyze existing supply chain strategies and data to learn what factors lead to supply chain failures. These AI-driven systems then use this knowledge to predict future supply chain problems and proactively prescribe or autonomously execute resolutions. While there is still a way to go before the self-learning supply chain is a reality, recent advancements in AI suggest it is no longer "blue-sky thinking."
The self-learning supply chain of the future marries the benefits of AI with the digital technologies that many companies have already started incorporating into their supply chain disciplines. This digital supply chain transformation is being fueled by several technology advancements: physical "things" incorporating computer technology; readily available big data such as social media, news, events, and weather (SNEW); and computer systems and software becoming more intelligent. These digital technologies are transforming the very nature of the supply chain—which was once built for volume and scale—into an agile, digitally connected framework that leverages a single set of physical assets to support multiple virtual supply chains. These virtual supply chains, sometimes defined as supply chain grids, replace the traditional fixed linear supply chains of the past by providing new flow options that enable accelerated order fulfillment based on near real-time awareness of assets and inventory.
![[Figure 1] Digital Supply Chain Maturity Curve](/media-library/20180330self-learning-fig1.png?id=53611786 "[Figure 1] Digital Supply Chain Maturity Curve")
The path toward the digital supply chain
We predict that the path toward digital supply chain maturity will occur in four stages: visibility, predictive analytics, the prescriptive supply chain, and ultimately in the future, the self-learning supply chain (See Figure 1). As companies move up the maturity curve, their reliance on manual capabilities will be replaced with autonomous capabilities, providing them with significant efficiency gains and cost savings.
Most companies today are in the first stage of digital supply chain maturity: the visibility phase. Currently, there is a huge focus on end-to-end supply chain visibility to help companies better manage constraints. At this maturity stage, visibility is often enabled by various system integrations such as connecting enterprise resource planning (ERP) systems with best-of-breed solutions and customer systems. This type of system integration enables a business to gain an end-to-end view of how product flows through their supply chain.
The next stage of digital supply chain maturity is predictive analytics. This phase leverages predictive analytic algorithms, enabled by big data—such as Internet of Things (IoT) sensor data, SNEW data, and others—to predict where supply chain issues may arise in the future. Predictive analytics, for instance, can be used to analyze real-time data like weather forecasts and port congestion to predict the impact on freighters in route and determine which shipments will be late—even before the captain may know.
The prescriptive supply chain, enabled by supervised machine learning is the next stage of digital supply chain maturity.1 In this stage, intelligent systems will be able to move beyond predicting potential supply chain issues to prescribing the course of action to take to resolve the issue. This technology is already being incorporated into best-of-breed offerings, where prescriptive analytics are used to learn from planners' historical actions. For a shipment that's predicted to be late, for instance, the solution could provide several resolution options (such as swap demand from another resource or purchase from another supplier) and then recommend the best course of action.
The final stage of digital supply chain maturity is the self-learning supply chain, enabled by deep learning. This capability will provide companies—as well as the solution providers that sell it—with the highest level of differentiation in the markets they serve. Deep learning is a form of AI, in which machines learn from machines. As we'll discuss below, this type of AI is already occurring.
The transformational power of deep learning
DeepMind, Google's AI subsidiary, has made significant inroads in deep learning, and its work provides a compelling look at how machine learning will transform the future. Consider that DeepMind has developed software that has mastered the ancient and complex board game of Go.
The first iteration of the software—AlphaGo—was programmed with a dataset of human game strategies. The software studied the gaming strategies and used the knowledge it gained to beat the 18-time human world champion of Go. The most recent version of the software—AlphaGo Zero—was programmed with only the game rules. AlphaGo Zero then developed its own game strategies by competing against itself—millions of times—over the course of three days.
Recently, AlphaGo Zero competed against the original AlphaGo and won 100 times out of a 100. Writing about the achievement in Nature magazine, researchers from DeepMind said, "Humankind has accumulated Go knowledge from millions of games played over thousands of years, collectively distilled into patterns, proverbs, and books. In the space of a few days, starting tabula rasa, AlphaGo Zero was able to rediscover much of this Go knowledge, as well as novel strategies that provide new insights into the oldest of games."
How deep learning will impact the supply chain
Just like the game of Go, supply chain failures (such as missed shipment windows and low order fill rates) are predicated on millions of potential combinations of action and supply chain policies. There are literally millions of combinations of ways that companies can flow product through the supply chain, and larger enterprises receive millions of order lines every day. Additionally, companies must make numerous decisions about strategic concerns such as their network strategy, replenishment method, and transportation mode. All of these decisions have a direct impact on service performance and cost. Furthermore, there are environmental factors—like weather, social sentiment, news, events, competitor activity—that can add complexity to making optimal decisions.
With AI embedded in the self-learning supply chain, machines will be able to examine supply chain strategies to determine where supply chain failures have occurred and why, along with what combination of external factors—such as transactions, loyalty, inventory levels, weather, competitor events, market performance, traffic, or socio-economic events—contributed to the supply chain failure. Machine-learning algorithms will then sift through this data to learn how these factors interact to result in a high probability of a supply chain failure.
In the future, this type of self-learning supply chain will be able to tell a planner that when a certain combination of events occurs at the same time it is predictive of a supply chain failure. The machine will then be able to prevent the failure by moving inventory to a new location, or it will alert the planner to respond to the problem.
The self-learning supply chain of the future
We believe that deep-learning algorithms will drive the supply chains of the future. They will be able to analyze all these combinations of factors, determine which of these items are predictive of a service failure, and build risk mitigation strategies that help organizations "win" by serving customers at the highest level of confidence, at the lowest possible cost. Companies that can do these things—serve customers better than anyone else (that is, faster, with a higher degree of order fill, and on time)—and do it at the lowest cost, will be hard to beat.
Getting to this level of maturity will require reliance on a partner ecosystem that can collect data signals (SNEW and others) to feed into these deep-learning models for real-time insights that can then be used as input to the supply chain plan. While the technology required to support the self-learning supply chain is still being developed, there is a lot of value to be gained in starting to master the early stages of digital supply chain maturity. Companies that embark on a digital supply chain journey now will be well positioned to capitalize on deep learning supply chain capabilities when they are available.
Notes:
1.Supervised learning takes input variables (x) and an output variable (y) and uses an algorithm to learn the mapping function from the input(s) to the output. Common supervised learning frameworks include classification and regression.
