Throughout my career as a supply chain practitioner and leader, there has constantly been a battle between the scientifically perfect model and a business-engineered model that guided the company to decisions it was comfortable making. Perfect supply chain decisions exist primarily in academic literature and textbooks while real-world decisions are impacted by many external factors, such as production outages, order cancellations, labor shortages, quality control, border delays and more. Compounding these concerns are regional and global disruptions that are only becoming more frequent, to the degree that they have made supply chains a conversation starter for the general public.
Supply chain concerns are now CEO and board-level topics due to unprecedented levels of demand and supply volatility, while at the same time, customer expectations have remained, so meeting those expectations is harder than ever. Supply chain planning needs to take advantage of new analytical techniques while catering to a much broader spectrum of concerns than before. I traveled the world early in my career to introduce my company’s suppliers to better inventory modeling, but today inventory optimization is no longer about which demand distribution to assume but rather viewed in the context of growth strategies, cost pressures and trade policy. We are due for a paradigm shift to incorporate the increased complexity and volatility of planning.
Analytical supply chain models incorporating optimization and artificial intelligence and machine learning (AI/ML) have traditionally been developed for specific functional problems in planning silos, in spite of cases where optimizing one silo has negative impacts everywhere else. Even an S&OP process only deals with aggregated concerns and cannot address the details required in various parts of the chain to recover from disruptions. Everything needs to be planned in a way that considers the overall supply chain concurrently, not as disconnected pieces. While the advances in optimization and AI/ML can make our models more intelligent, they also must be delivered at the speed of business and with the appropriate business context to make them relevant. Meeting this challenge requires new data sources but also fusing together the strengths of different methods.
The combination of optimization, AI/ML and custom heuristics must be about enabling real-time, holistic decision support in a way that generates understanding and trust for the overall impact to the supply chain. It cannot be focused on the output of a single problem in isolation. To tackle business problems, the fusion must incorporate rules tailored to the reality of the business and have an ability to simulate multiple potential futures. As problems get more complex, we need methods that can address the scale, and smartly and rapidly reduce the solution space. The fusion needs to provide explainability for comprehending output and empower planners with control over the final results. As one of our customers has said quite unequivocally, “Machine learning will not work without interpretability.” Clarity of action for decision-makers comes from a confidence in the model that is uniquely focused on their business. Supply chain analytics need to be the enablers of confident decisions.
Analytical techniques are now being fused in new ways that harness the benefits of each to create a new generation of advanced decision support. Certain key trade-off decisions are best solved with optimization and others with AI/ML, while custom heuristics provide the framework surrounding these inputs and enable real-time adaptation. Fine-tuning for critical needs by planners must be part of the overall model to avoid them blindly overwriting the final results with negative impacts they don’t understand. Visualizing and interacting with the results are key for decision-makers to champion the output provided.
A new generation of supply chain planning is coming. Supply chain disruptions are not slowing down, different analytical techniques are being brought together in new ways that are specific to each business, and the role of human planners to prepare for all potential futures will only grow. It is a planning future I look forward to.
Note: I give special thanks to Dwayne Smith, Jessie Lamontagne, and Carsten Jordan for some of the thoughts behind this post and the related talk I gave at the recent INFORMS Analytics Conference.