Artificial intelligence could fundamentally change how companies plan their supply chains, from forecasting demand and managing inventory to automating routine tasks. Technology alone, however, cannot solve every planning challenge, as a recent webinar hosted by the MIT Global SCALE Network showed. Researchers from Europe, North America, Asia, and Latin America concluded that AI can accelerate analysis and automate much of planning, but organizations must still address the human, operational, and data challenges that determine whether those capabilities deliver value.
The webinar was moderated by Professor Janet Godsell, director of the UK SCALE center at Loughborough University. The panel included Professor Donato Masi of Loughborough University, Dr. Maria Jesus Saenz of the MIT Center for Transportation & Logistics (MIT CTL), Professor Richard Pibernik of the University of Würzburg and Zaragoza Logistics Center, Dr. Emad Alenany of the Ningbo China Institute for Supply Chain Innovation (NISCI), and Dr. Daniel Prato of LOGYCA.
Faster Information Does Not Always Mean Faster Decisions
Professor Masi explained that while AI has made it easier to gather data, improve forecasts, and monitor supply chains in real time, human decision-making often remains slow. Planners must agree on priorities, balance service levels against cost and cash, and determine who has authority to act. An insight may emerge in minutes, but acting on it can take days or weeks.
To this end, Dr. Saenz presented research from the MIT Digital Supply Chain Transformation Lab involving about 1,600 stock-keeping units (SKUs) at a major fast-moving consumer goods retailer. The research showed that human intervention did not always improve automated forecasts. Its value depended on the product, the level of demand uncertainty, and the forecaster's experience, and adjustments sometimes introduced bias when uncertainty was high. Still, human oversight made planners more comfortable relying on the system compared to if the system were entirely AI-automated. Companies should therefore decide deliberately when humans should intervene and when AI should operate independently.
Automating Routine Work, Focusing on Exceptions
Professor Pibernik described how many organizations manage tens of thousands of products from thousands of suppliers, often with unpredictable demand and limited historical data. His approach combines AI foundation models, specialized agents, and optimization techniques. In one inventory workflow, agents onboard company data and analyze supplier contracts for constraints such as minimum order quantities and lead times, while a forecasting model estimates demand and an optimization system recommends replenishment. Planners receive flagged exceptions with explanations rather than reviewing every product. Pibernik suggested such systems could eventually automate 90% to 95% of the process in suitable applications, and subscription-based tools could make these capabilities accessible to smaller companies.
Why Data Quality Still Matters
Dr. Prato of LOGYCA shared pilots with nine retail companies in Colombia, covering data from more than 700 points of sale. In some cases, traditional statistical methods outperformed newer models, yielding forecasting errors that were substantially lower. Data availability and quality, rather than the AI techniques themselves, set the limits on performance. For example, accuracy reached 70% to 80% for products with predictable demand but dropped for more volatile ones, where the data was harder to use effectively.
Dr. Alenany of NISCI presented an approach that combines supply chain simulation, machine learning, and mathematical optimization to improve inventory planning across multi-echelon supply chains. His key lesson is that organizations must understand the system they are modeling before applying AI, because the structure of the planning problem shapes whether these combined methods produce better decisions.
The Organizational Challenge
The panel repeatedly returned to organizational readiness. Planning responsibilities are often split across departments, decision rights are unclear, and employees may not trust the recommendations. Professor Pibernik noted that companies can explore locally deployed or open-source models to protect sensitive data. Professor Masi cautioned that avoiding AI carries its own risks, particularly as competitors invest. Finally, Dr. Saenz emphasized experimentation and organizational learning, noting that not every organization will adopt new technologies at the same pace.
Revolution or Evolution?
An opening webinar poll asked whether AI represents an evolution or a revolution in supply chain planning. The panel suggested the answer depends as much on organizational choices as on the technology. Some panelists saw revolutionary potential in linking planning activities and connecting strategic decisions with execution, while others stressed building data foundations, governance, and working practices first. Professor Masi identified sales and operations planning (S&OP) as a particularly promising area for AI-driven change. Ultimately, the panel agreed that AI's impact will depend on better data, appropriate applications, effective collaboration between people and technology, and a willingness to rethink established processes.
