Over the past few decades, globalization has improved the efficiency of our economic system. Goods, labor, natural resources and even services are now sourced from every corner of the world, and delivered nearly anywhere at lightning speed.
At the same time, globalization adds complexity, in terms of the number of sources, diversity of delivery systems, and increased regulatory and geopolitical risks. Globalization introduces more points of failure to those delivery processes, making supply chain management more complicated.
To address these challenges, companies are turning to artificial intelligence. In processing data, AI can predict future demand, improve automation and increase transportation efficiency, giving supply chain managers deeper insight and control over every aspect of the chain. But AI isn’t a magic fix, and leaders should be aware of its limitations.
Software systems have helped address some of the challenges of our modern supply chain. They automate sourcing, purchasing and delivery of resources from around the world to centralized locations that combine them into products. Such systems allow companies to consider a broader range of sources, venues and delivery options, to a degree that’s beyond the ability of humans to evaluate. AI and machine learning enhance these systems, making them more resilient and adaptable to the frequently changing global network of relationships.
Over the past few years, however, the global economy has seen major disruptions to the supply chain. Political tensions, both regional and global, have disrupted the flow of goods and services throughout the world, particularly from and to China, and more recently in Russia. COVID-19-related shutdowns have stressed the global supply chain to a breaking point. If supply chain management systems can’t adapt to the COVID-impacted environment, as well as the ever-changing geopolitical landscape, the global economy could face disastrous consequences, including massive inflation, economic stagnation and long-term harm to regional economies.
One answer lies in the aforementioned AI and machine learning-driven enhancements of supply chain management software. AI has transformed many industries, driving innovation and efficiency by combining domain knowledge and historical data to avoid bad decisions, and identify new solutions to old problems that humans are unlikely to discover on their own. With supply chain management, AI can improve the status quo and make solutions more dynamic and adaptive to changing conditions. However, there are limits to how much AI can help, and the scale of today’s problems might be resistant to the benefits typically offered by data-driven systems.
Before automated systems existed, supply chain management was done using more primitive tools, driven by human knowledge, logic and intuition. At a minimum, new software systems encode this understanding and experience into algorithms that reproduce human behavior at computer scale. They allow companies to manage the complexities of the global economy without relying on manual processes which are prone to human error.
Traditional AI systems add to the process by introducing rules that allow supply chain management systems to be adaptive to changes in the world — encoding human knowledge into complex rule systems that allow static systems to change their behavior based on new facts. For instance, if a natural disaster or war disrupted the flow of resources, a rule-based system could guide the software to avoid the impacted regions and find new sources to replace those held up by the disruptions.
These systems typically use optimization algorithms that encode costs, risks and value forecasts, along with hard constraints that can’t be violated, and soft constraints that are introduced as penalties to steer the optimization away from undesirable outcomes. They combine all of these factors in real time to come up with the optimal solution for sourcing the materials needed to produce products in a timely manner.
Data-driven machine learning and AI systems take this process a step further. They look at historical data of how various phenomena — wars, natural disasters, large price movements and the like — have impacted the variables. Then they make predictions about what’s likely to happen in the future, based on how current conditions predicted behavior in the past. These predictions provide an additional layer of information which can make the algorithms more capable of finding closer-to-optimal solutions to supply chain management decisions.
The main advantage of data-driven machine learning and AI-based systems is that their “reasoning” is quite unlike human logical processes. They’re more likely to uncover novel approaches to finding and delivering resources. Human beings are creatures of habit, emotionally biased, and less than stellar at statistical reasoning. As a result, people are unlikely to switch away from sources and processes that they’ve always used.
Data-driven systems will always consider all options, using historical patterns of behavior to assign risk, cost and value distributions, then use optimization algorithms to combine all that information and determine the supply chain management solution likely to lead to the best outcome.
Where Limits Lie
Like most technologies, AI and machine learning have limitations. There are even ways in which relying on statistical models in uncertain times might lead to suboptimal decision making.
No matter how data-driven and sophisticated your supply chain management algorithms are, they can’t change the laws of physics, chemistry or biology. If a natural disaster or geopolitical crisis has made raw materials, labor or other needed supplies unavailable in sufficient quantities, or if they’ve disrupted the flow of goods and services to the point that they can’t be delivered on time from anywhere, no optimization algorithm can swoop in and save the day. Human-driven decision making will sometimes supersede model-based processes in this regard. And if humans don’t intervene, the software systems might propose solutions that satisfy the constraints but at unacceptable cost or risk.
Another risk to using statistical-based models is that there are sometimes changes in the world that invalidate solutions derived from past data. COVID-19, for example, has changed the nature of supply change management in a fundamental way.
As the impact of COVID-19 ebbs and flows, some of those changes might revert back to old norms, some might stay permanently changed, and some might settle back somewhere between the old and the new. The unstable nature of the global economy throws into question the accuracy and value of models trained on historical data. Some aspects of those models might be resilient to these changes. Others might have reduced accuracy. And still others might be completely invalid and have negative predictive value. The only way to know is to consistently test those models against new data, and evaluate their efficacy continuously until the global supply chain re-stabilizes.
Data-driven AI supply chain management systems typically produce better results than traditional AI- and human-driven systems. However, supply chain managers should retain their authority and sense of responsibility for the execution of the decisions recommended by these systems. Software developers should likewise be more proactive in educating supply chain managers about the nature of the data they’re using, the limitations of the systems, and how to measure the accuracy and usefulness of the solutions produced by the systems over time.
As long as users are informed and knowledgeable about how the systems work, and when they’re likely to fail, these systems can significantly improve the average performance of supply chain management systems, and especially in times of stress. But if humans turn their brains off and let these systems run on autopilot, there are likely to be times when they’ll suggest solutions that are suboptimal and financially harmful.
It’s up to people to use their wisdom and experience to evaluate the available solutions, especially in light of globalization, so that all of the savings generated by these systems won’t be thrown away by one bad decision.
David Magerman is a co-founder and managing partner at Differential Ventures.
Timely, incisive articles delivered directly to your inbox.