Manufacturers need more than insight to compete; they need foresight. Predictive and generative AI are fusing with granular data, agentic interfaces and unified infrastructure to turn static plans into living, breathing operations.
For decades, forecasting in manufacturing has been built around compromise. The traditional approach to demand planning relied on aggregating data into manageable blocks, monthly buckets, product categories, and high-level channels because the computational power simply could not handle granular complexity. Algorithms struggled to accommodate millions of variables, so planners filtered, averaged, and abstracted. The result was a best guess, but one that often carried hidden inefficiencies.
That compromise is dissolving. AI is not only hungry for data; it thrives on detail. By ingesting massive volumes of data down to the level of the stock-keeping unit (SKU), the unique identifier for each individual product variant, as well as the customer and the specific day, modern forecasting models are uncovering the hidden volatility long masked by averages.
Simon Bowes, CVP of Manufacturing at Blue Yonder, explains that this transition is more than technical. It is philosophical. “We used to average lead times, average inventory levels, and treat entire demand channels as homogenous,” he says. “AI has shown us that all that averaging is where a lot of the cost comes from. Now, we can generate precise forecasts at the item and location level. We are not just saying what demand might look like next month; we are anticipating fluctuations over the next few days for specific customers and adjusting accordingly.”
These systems are also moving beyond deterministic outputs. Instead of single-number forecasts, AI is producing probability distributions that reflect a range of possible outcomes. That subtle shift has enormous consequences. It allows planners to make nuanced decisions, such as whether to bias a forecast toward availability or to minimise obsolescence. It enables multi-horizon planning, where short-term forecasts leverage different models than 18-month outlooks. And crucially, it replaces static plans with fluid simulations that adapt to change in real time.
From batch planning to continuous intelligence
Manufacturers have long viewed the planning cycle as linear: demand forecasts lead to supply plans, which trigger production schedules and downstream logistics. Yet, in reality, every stage influences the others. A late shipment, a cancelled order, or a component shortage each introduces friction that ricochets up and down the chain.
Historically, these interdependencies were too complex to model in real time. Planning systems worked in silos, each generating outputs based on stale data transferred via nightly batch runs. This created lag, blind spots, and delays in reacting to disruption.
Bowes believes AI is dissolving those boundaries. “What AI is letting us do is align all of those different planning phases and provide feedback loops across the supply chain,” he explains. “It is no longer a sequence of disconnected handoffs. It is a continuum.”
This is particularly evident in complex sectors like automotive, where planning includes mix optimisation and balancing constraints to determine how many electric, diesel, or sunroof-equipped vehicles to build, followed by slotting and sequencing once orders are received. Previously, each of these stages operated independently. Now, AI enables real-time integration.
“New constraints emerge all the time,” Bowes continues. “It might be a microprocessor shortage this week or a packaging delay next week. You do not want to start the planning process from scratch every time. AI gives us the flexibility to define new constraints on the fly and have the system adapt.”
Agents, not apps
While predictive AI focuses on accuracy, generative AI is revolutionising access. The most significant breakthroughs, Bowes argues, are not just in how data is processed but in how people interact with it. Traditional enterprise systems expect users to navigate complex menus and predefined reports. Generative AI flips that on its head.
“What you really want to do is ask a simple question, which orders are at risk from this port strike? And get a straight answer,” Bowes says. “Generative AI lets us interrogate our data the same way people use ChatGPT.”
That shift in user interface design is being taken further through the development of role-specific AI agents. Rather than overwhelming users with dashboards, these agents act as assistants for specific job roles. The warehouse manager, for example, no longer needs to sift through last night’s performance data. Instead, they start their day with a voice summary that highlights key metrics, delays, and required actions. It is like listening to a podcast on the commute.
“These agents are tailored to the job and the context,” Bowes explains. “They are not just answering questions. They are proactively surfacing the right information before you ask.”
Behind the scenes, this is made possible by scalable cloud infrastructure, specifically Blue Yonder’s data platform built on Snowflake. By consolidating formerly siloed systems into a single data lake, the company can support real-time analysis, eliminate batch delays, and enable these new agentic experiences. Crucially, this architecture supports third-party and customer data without needing to replicate it by pointing applications to external datasets, bringing the app to the data rather than vice versa.
Resilience means predicting the unpredictable
The pandemic and subsequent geopolitical turbulence exposed fragilities in global supply chains that many executives believed were long resolved. AI cannot eliminate disruption, but it can mitigate its impact by simulating responses before they are needed.
A clear example is in container logistics. Most systems rely on promised arrival dates to inform inventory plans. But if a ship is delayed, the system only reacts once the deadline has passed. AI changes this by predicting delays based on upstream signals, weather patterns, port congestion, and customs activity and simulating their consequences. “Not every late container matters,” Bowes says. “Most are covered by safety stock. The question is, which products in that container are going to trigger a problem? And once we know that, what are the options, divert to another port, switch suppliers, or air freight only the critical SKUs?”
Scenario planning, once a luxury for quarterly reviews, is becoming embedded into daily operations. AI continuously evaluates alternative paths, presenting decision-makers with the equivalent of satnav routes: shortest time, least cost, and lowest emissions. The executive does not have to model every possibility. The system does it for them and recommends it.
Sustainability is a constraint, not a consequence
For decades, efficient supply chain planning has coincided with sustainability goals. Lower waste, fuller trucks, and optimised inventories all reduced carbon emissions while cutting costs. That overlap is now diverging. In some cases, the most sustainable route is not the cheapest. Yet both matter.
Bowes sees this as a pivotal challenge. “It used to be that cost and sustainability objectives aligned naturally,” he says. “Now we are building in the ability to run different scenarios where carbon is the primary objective function, not just cost or margin.”
This includes support for formal ESG reporting requirements, particularly as European regulations tighten. AI plays a dual role here: first, in modelling operations that minimise emissions, and second, in compiling the data required to prove compliance. The goal is to enable trade-offs, not dictate outcomes. “Sometimes the sustainable option will cost more. But if you know that, you can make an informed choice,” Bowes adds.
From one-size-fits-all to bring your own AI
Manufacturers have invested heavily in in-house data science teams, many of whom have built proprietary forecasting models or optimisation techniques. Rather than discard this work, Blue Yonder enables customers to import their own models into the platform through a ‘bring your own AI’ strategy.
Bowes believes this is essential. “It avoids an us-versus-them scenario between vendor and customer,” he adds. “If you have built a model that works for your product line, our system can treat it as just another algorithm in the mix-and-match engine.” This flexibility allows manufacturers to evolve toward maturity without losing what already works. AI deployment, he argues, should begin not with technology but with clear outcomes.
“Do not adopt AI for its own sake,” Bowes concludes. “Start by knowing what you are trying to achieve, reduce variability, improve forecasts, or enable real-time IBP. Then find the solution that helps you get there.”
Because in a world where disruption is constant, and supply chains span continents, foresight is no longer a competitive advantage. It is a prerequisite for survival.
