The scariest drive of my life took place on a summer’s night in the North Carolina mountains returning from a longer-than-expected hike. The Blue Ridge Parkway is usually a magical drive in the clouds, but it turned frightening as dusk and fog descended, reducing visibility to ten feet. I was torn between twin impulses – speeding up so we could descend out of the fog and slowing down to keep from careening off the mountain to the left or into a ditch on the right. It took intense concentration and a white-knuckle grip on the wheel to emerge safely. CSCOs feel similarly overwhelmed navigating the intense digital fog surrounding AI in supply chain, so I offer five principles for navigation.
The arrival of ChatGPT at the end of November 2022 similarly turned a drive in the clouds into thick fog. Demands to speed up abound. An IBM Institute for Business Value study found that 66% of CEOs feel pressured by their boards to accelerate AI adoption. A Workday study of decision-makers found that 80% feel AI is essential to remaining competitive, so 94% are investing in it.
The pressure is propelled by opportunity – firms that embrace technology experience productivity growth as much as 40% higher than those who don’t. But this opportunity comes with risks; as BCG found, up to 70% of digital transformations fail, so no wonder some are scared at the undertaking. Compounding pressure and risk is low visibility in the fog of hype surrounding complex, rapidly evolving technology. And as AI appears more human-like in its capabilities, emotions ratchet up. Privately leaders admit to feeling upset and overwhelmed, because they don’t understand what AI can do and how to steer through the fog. The impulse to slow down is understandable.
Defining AI in Supply Chain
A definition is a first step in clearing the digital fog. AI is the science of computers mimicking human intelligence to solve problems. This science encompasses many disciplines to improve speed, precision and elegance in decision-making by finding patterns in enormous volumes of data. Examples of the fields are machine learning (including deep learning), optimization, genetic algorithms, robotic process automation, generative AI, and decision management. What can this science do? AI can generate recommendations, predict and surface insights, provide speed and scale, automate processes, and enhance productivity, all capabilities we can apply across supply chains. A definition is a start, but we also need guardrails, so here are five principles for succeeding with AI in supply chain.
First principle: AI should augment humans, not replace them
The capability boundaries AI keeps crossing are nothing short of amazing, from producing creative marketing copy, to complex legal research, to songs, paintings and more. These wonders are possible due to the ability to process data and learn from patterns far beyond the cognitive capacity of humans. These achievements make it easy to forget what machines cannot provide, which I call the 3 C’s: context, collaboration and conscience. Models cannot derive meaning from context, critical in so many areas, like what Zero100 thought leader Kevin O’Marah has termed “machine whispering.” Nor can they work together to solve problems, including addressing issues like sustainability or human rights in supply chains.
This complementarity is why AI should augment humans, not replace them. The most powerful combination is for humans and AI to work together, a belief reflected in a Workday survey of decision-makers, 93% of whom believe in the importance of keeping the human in the loop when AI is making significant decisions.
Second principle: The expert fusion of AI, heuristics and optimization is key
AI can also model problems at scale to produce more precise recommendations, such as greater demand forecast accuracy or better predictions of on-time delivery. Precision is also a benefit of optimization, a field of AI familiar to many in supply chains for its ability to make the best use of resources within constraints to accomplish an optimal solution with a specific objective, such as minimizing costs. But here scale can be a challenge: optimizing a supply network can involve 200 million interdependent variables, slowing even the fastest optimization solver down by hours. Instead some turn to heuristics, a problem-solving model that utilizes a practical solution, or best practice, to produce a quick and feasible course of action good enough for the situation.
So these various mathematical models can offer speed, precision, and elegance, but with trade-offs. However, the newest, fanciest math isn’t necessarily better, no matter what you might hear in the fog. Generative AI is not the right approach to most of our classic supply chain problems, for which heuristics offer unparalleled agility. Deep learning has very specialized applications perfect for the “right” problem but not most, particularly given its greater complexity.
A fusion of methods, like machine learning and heuristics, can “warm start” an optimization model and speed up the ability to solve it, creatively combining the strengths of each approach to achieve an equilibrium of speed, precision and elegance. Keep your hands on the wheel in the fog and remember that the most elegant solution is one that uses the right model for the right problem at the right time, no more, no less.
Third principle: Concurrency amplified by AI is a breakthrough in supply chain management
Supply chains connect many functions across a company and beyond, which is why optimizing one link doesn’t optimize the entire chain. For example, AI can greatly increase the accuracy of forecasts, but we want more than highly-efficient silos. As a trio of Canadian economists argue, “AI can be used to resolve uncertainty, but unless that can translate into aligned decisions all the way down, the fundamental problem—that demand needs to be aligned with supply—hasn’t really been solved. Like the swing of a bullwhip, your own solution has reverberations down the line.” The power of AI on its own is not enough.
The real breakthrough is not from AI but with concurrency, which integrates AI in the workflow to align decision-making across the supply chain for faster response. We want AI for its ability to predict with greater precision, speed, and elegance, and we need concurrency to connect supply chains for better, faster response, no matter what the conditions are. Because AI embedded in concurrency leverages predictions while absorbing the volatility we cannot predict from the inevitable disruptions our supply chains will always face.
Fourth principle: The power of AI must be democratized
For AI to realize its potential it must be unleashed from the lab of data scientists. We will always need experts to explore new ways to apply AI, but empowering supply chain practitioners to adopt it themselves will extend its reach. For this reason, the best solutions require only an understanding of company data and business, not technical proficiency in AI or data science.
So while Workday’s survey found 72% of leaders feel their organizations lack the skills necessary to full implement AI, applying AI doesn’t have to involve diving into the deep end. If solutions are designed for someone with supply chain context and business knowledge, they can “consume” the results of a model without knowing how to build it. Democratizing AI in this way ensures its use, so choose to work with a provider who allows you to start from where you are and evolve.
Fifth principle: Explainability is essential for AI adoption
One additional trade-off of AI is when the speed, precision, and elegance are delivered in black boxes that even the data scientists who constructed them cannot unpack. Lack of explainability is a barrier to adoption, because if you are personally held responsible for a forecast, it can be hard to trust a “machine.” Researchers have found that humans are more forgiving of what they perceive to be error on the part of fellow humans than they are from machines, a trait that can lead to them to develop “algorithm aversion.”
One approach to overcoming this aversion is state-of-the-art techniques that make black box AI models more understandable. For example, tools like a SHAP diagram (SHapley Additive exPlanations) can be surfaced in demand sensing to help a planner see how adding a signal like weather affects predictions. To increase people’s willingness to adopt the value of AI-driven insights, we need to arm them with the information they need to explain their decisions embedded in solutions they can understand.
The opportunity for AI in supply chains is massive
Supply chains have never needed more help, and AI has never been more ready, so the time is now. But as we ramp up our us of AI we need a human-centered approach that amplifies the power of concurrency to drive the most intelligent supply chains on the planet. When AI is embedded across the end-to-end supply chain, expertly fusing the best techniques available, we can reimagine what is possible in our supply chains.
Polly Mitchell-Guthrie is the VP of Industry Outreach and Thought Leadership at Kinaxis, the leader in empowering people to make confident supply chain decisions. Previously she served in roles as director of Analytical Consulting Services at the University of North Carolina Health Care System, senior manager of the Advanced Analytics Customer Liaison Group in SAS’ Research and Development Division, and Director of the SAS Global Academic Program.
Mitchell-Guthrie has an MBA from the Kenan-Flagler Business School of the University of North Carolina at Chapel Hill, where she also received her BA in political science as a Morehead Scholar. She has been active in many roles within INFORMS (the Institute for Operations Research and Management Sciences), including serving as the chair and vice chair of the Analytics Certification Board and secretary of the Analytics Society.