The larger and more complex supply chains become, the more vulnerable they are to disruptions. Pressures such as market volatility, tightening labor, inflation, weather events, and supply shortages have a cascading effect on the global movement of goods.
The good news is that, thanks to the proliferation of sensors, cameras, and digital tools, many of these conditions can be captured in data—a lot of data. Artificial intelligence (AI) gives us a means of understanding that data. This technology can be used to dynamically analyze complex data sets, helping companies predict demand, identify trade-offs, and optimize delivery routes. Together with subsets like machine learning (ML), which utilizes training data and feedback to progressively improve accuracy, AI allows firms to illuminate and even predict supply chain disruptions before they occur. More than anything, AI can help supply chain managers make better, more informed decisions.
While many supply chain managers are aware of the potential of AI, a common stereotype is that AI will replace humans by fully automating data analysis and decision-making. That is not the case. While AI does rely on data analysis to deliver recommendations, often the full story of what is happening in the supply chain cannot be completely captured by the available data. Instead, recognition of many issues and their causes can come only from a human’s prior experience and intuition.
At the Institute for Experiential AI (EAI) at Northeastern University, we have found that human involvement is an important part of getting the most out of artificial intelligence. With a human involved at all stages of the AI “training” and ongoing feedback process, the technology does a better job of increasing efficiency, yielding more ethical outcomes, and providing insights that improve bottom-line performance—better than it would without human involvement. We call this human-centric approach to AI “experiential AI.”
The goal of experiential AI is to augment what humans do best (such as intuitive decision-making under uncertainty, common-sense reasoning, and understanding real-world complexity and subtlety) with what machines do best (such as crunching large amounts of data and documents, performing repetitive/robotic tasks, and operating at speed and scale) to achieve more robust, ethical, and resilient solutions.
Most successful AI applications, in fact, depend on soliciting human input and feedback to ensure accuracy. In order to “train” an AI model or machine learning algorithm to generate the right analysis, you need to provide it with a large set of training data. The best way to get reliable training data is by soliciting input from humans. An experiential AI approach involves soliciting that input in the most efficient and meaningful way possible.
Furthermore, no AI algorithm will be completely error-free. All algorithms require human feedback and confirmation that they are executing efficiently and that their outputs are correct and relevant. The difficult thing is to figure out when an algorithm should seek that feedback and confirmation. Another challenge is to make this act natural and intuitive and to be selective about when to ask for an intervention. Experiential AI can be very helpful in this regard by ensuring that intervention by humans is done efficiently (for both human and machine) and at the right time to maximize learning opportunities for both machine and human. This is an essential ingredient for building trust in the AI technology among human users and operators.
Experiential AI, then, provides a way to get the needed training data, interventions, and human guidance in the context of normal operations so the AI can learn from each interaction. This human involvement also helps AI to generate informed and ethical decisions.
For example, before the COVID-19 pandemic, cutting costs might have sufficed to minimize the impact of problems in the supply chain. Recent pandemic-related disruptions, including staffing shortages and low inventories, however, have shifted companies’ focus from efficiency back to resiliency, which a growing number of economists argue comes at a loss to efficiency, and vice versa. But we believe that efficiency and resilience need not be adversarial. With a human in the loop, AI models can be consistently and naturally modified to deliver better performance, consistent and measurable return on investment (ROI), and long-term adaptability.
Key supply chain applications
There is good reason for supply chain managers to explore how to apply artificial intelligence in their operations. The global management consulting firm McKinsey & Co. estimates that by adopting AI in the supply chain, companies and their customers stand to gain $1.2 trillion to $2 trillion in economic value globally. With such an opportunity on the table, it’s important to survey which areas of the supply chain are most ripe for benefiting from AI. The Institute for Experiential AI sees three core areas of opportunity: transportation and delivery, warehousing and inventory management, and analysis and decision-making.
1. Transportation and delivery. A complex supply chain is not necessarily a resilient one. Each junction in the movement of goods introduces new variables and logistical hurdles. In turn, decision-makers must select from an increasingly complex network of routing and delivery models. As the inputs stack up—think of adding to a growing tower of playing cards—the long-term resilience of the system begins to buckle. The task of supply chain managers then becomes to find and adopt end-to-end solutions that can forecast demand, mitigate risk, and account for multiple variables and distribution routes.
AI makes that possible. Supply chain managers can now use machine learning to process the complex data streams that undergird logistics networks. For example, they can take real-time traffic and global positioning system (GPS) data and use machine learning to identify and select from potentially trillions of delivery routes. They can also use predictive analytics solutions that are enabled by AI to anticipate and plan for demand surges, mechanical failures, shipping updates, or disruptive weather events. AI systems can also monitor news snippets, audio messages, sensor data, text alerts, and other unstructured data and inform decision-makers when a disruption has occurred.
Cold Chain Technologies—a company in the life sciences sector that ships and handles heat-sensitive drugs, pharmaceuticals, vaccines, and biologics—uses AI to monitor, route, and deliver thermal-assurance packages. The company requires transportation solutions that are able to maintain consistent temperatures across the supply chain. (This is critical for transporting COVID-19 vaccines, for example.)
Thermal packaging requires specialized internet of things (IoT) sensors and measuring devices that produce streams of data that algorithms can harness to map real-time conditions in the supply chain. But, as CEO Ranjeet Banerjee explains, the task for supply chain managers is not merely to automate processes, but to forge a path through the technological landscape with human decision-makers at the helm. Value, then, derives from top-level decision-making and human involvement.
“You have to start with the problems, define the use cases, define the value potential, and then come up with a cadence of solutions,” Banerjee says. “But it’s not one-and-done. It’s merely to provide a roadmap of new value.”
2. Warehousing and inventory management. Supply chain leaders have the demanding responsibility of balancing supply and demand. To support that effort, warehouse and inventory managers are turning to machine learning. Machine learning can be used to monitor supply routes, predict lead times, and fulfill orders. In many cases, machine learning can perform these tasks with near or absolute autonomy. However, from a risk management standpoint, it is crucial that the degree of autonomy be customizable so that mission-critical decisions remain in human hands while the ML supplies decision-makers with real-time data.
For instance, inventory managers tasked with balancing warehousing capacities with inbound and outbound deliverables can leverage machine vision to assist in stocking and fulfillment. Computer vision software can monitor the movement of goods and alert managers when supplies are low. The human managers then make the crucial decisions about how to address this low supply. Other tools like automated product classification and AI-powered robotics offer cost-cutting efficiencies that can help optimize the fulfillment process and improve lead times.
3. Analysis and decision-making. Across applications, AI empowers supply chain leaders with sophisticated data tools and end-to-end supply chain visualization. On-the-ground data can be quantified and delivered to AI-enabled systems that can then analyze that data and present it to decision-makers as actionable information. For example, details about how shipping containers are loaded or unloaded can be analyzed by AI to inform decisions about how deliveries should be ordered so that routes are created in the most efficient way possible. AI can also be used by supply chain leaders in the event of a disruption to locate alternative routes, suppliers, or delivery models, saving them time and energy when exploring remedies. Other algorithms and data sets can be used to streamline costs. It’s no surprise, then, that leading firms use data-driven AI to manage carriers, negotiate optimal rates, understand risks, and inform bottom-line financial decisions.
One promising development that is helping drive better decision making across the entire supply chain is the new field of cognitive analytics. Cognitive analytics gives structure to large data sets in forms more relatable to linguistic processing. Such systems can learn from interactions between data and human supervisors to provide detailed, contextualized insights. These insights can be used to connect different areas of the supply chain in a more transparent fashion. And that transparency is key. As Nada Sanders, Distinguished Professor of Supply Chain Management at the D’Amore-McKim School of Business at Northeastern University, points out, successful firms understand that technology that offers transparency between silos in the supply chain is superior to a sophisticated system whose analysis is narrow and deep. In other words, if you only have one very deep technology in one area, then you’ll likely be exposing your operation to variables that would only be visible from a broader, more systemic view.
“When you look at supply chains, the key is to understand that they’re a system; you need to have information transfer, and you need transparency because information flows, products flow,” Sanders says.
Responsible AI in the supply chain
On their own, data analysis and AI can point to bottlenecks, excesses, and oversights in the supply chain. In ideal circumstances, those insights lead to more efficient outcomes. But their true power lies in contextualization—a task generally more suited to humans than AI. For example, AI can fortify and streamline supply chain operations, but these improvements must be carried out in an ethical and responsible way. Having humans in the task loop can make sure that this occurs.
In many applications, algorithms have exhibited latent biases that exclude marginalized people while reinforcing power discrepancies. Facial-recognition tools, for example, have been shown to regularly misidentify people of color. Language models may likewise perpetuate linguistic hegemonies. If these algorithms can run afoul of ethical concerns in social contexts, then they can do the same in supply chains. One widespread example occurs in hiring and recruiting, where AI has been demonstrated to show biases toward privileged groups. Additionally, systems that are automated to select suppliers based on pricing or logistical efficiencies may overlook exploitative labor practices or even sanction regimes that human decision-makers would know to steer clear of.
That is why leading researchers and chief technology officers (CTOs) point to transparency and human-led AI as the only reliable way to secure the responsible use of algorithms. Cold Chain Technologies’ Ranjeet Banerjee acknowledges this, underscoring the value of AI in augmenting, rather than replacing, human intelligence.
“The easy decisions are the ones you automate first,” Banerjee says. “Then you use [automation] to increase the bandwidth of the human. Over time you create a feedback loop, and you see how the actual worked against the prediction, and then you can use the human intervention more thoughtfully.”
It’s crucial to understand that this process is continual. There is no “one and done” ethical AI solution. That means companies may need to upskill or retrain their employees or restructure their organization to secure the promised benefits of AI in supply chains.
Tying insights to the bottom line
As supply chains become even more complicated in response to ballooning data sets, political upheaval, climate disruptions, and increasingly sophisticated algorithmic tools, enterprises will need to look at the wider picture. As Nada Sanders says, it’s not just about logistics.
“It’s money, it’s people, it’s information,” she says. “It’s the linkage of marketing on the demand side and how we sell something, the messaging. They’re all connected, and understanding that system is really where the human element coupled with AI comes into play.”
AI in the supply chain offers scalable levels of visibility, granular oversight of logistics, and dynamic feedback to support human-driven decisions. But these opportunities may require organizational refocusing as companies seek the right tools to measure and quantify outcomes. When it comes to assessing the value of AI and which solutions to focus on, they may need to take a long-term investment approach rather than zeroing in on a few widely used metrics to measure their ROI.
Supply chain leaders may see experiential AI as a means to function at scale, increase the bottom line, and create value for their customers, but with AI and large-scale data analytics still in their infancy, they may not know how to go about implementing it. For the time being, end-to-end AI solutions that dodge the most pressing ethical and technical pitfalls can be found in the B2B market. But it’s also true that many organizations simply don’t need such a comprehensive solution. A better approach for many supply chain organizations is to identify the problem to be solved, measure its scope in the form of data, and then seek out AI experts who can help design, develop, and implement an effective solution that addresses the organization’s specific needs.
