Every manufacturing leader I talk to gives me the same answer about supply chain when I ask what keeps them up at night.
Rather than the tech behind it, it is uncertainty. Geopolitical disruption. Tariff swings that upend a costing model overnight. A tier-two supplier that goes quiet for a few weeks and then goes under. Demand signals that used to be predictable now running six months behind reality.
Supply chain has always been a balancing act. What has changed is the variables doing the disrupting are faster, less visible, and less forgiving than they used to be. A decade ago, a planning team could absorb a bad quarter and course-correct. Today, the same class of disruption can halt a shipment, strand inventory, or break a sourcing plan before anyone downstream even knows there’s a problem.
That is the problem AI was built for. Not because AI is magic, but because it is quite good at doing something humans cannot do at scale: combining external signals with internal data in real-time, and surfacing the pattern before it becomes a crisis.
Where I see the real traction
I spend a lot of time in manufacturing and retail conversations, and two areas come up consistently.
Demand forecasting: Traditional planning models were built for a world where the primary inputs were historical sales data and seasonal patterns. That world still exists, but it now coexists with tariff announcements, supplier fragility, and geopolitical events that move faster than any quarterly planning cycle. The companies doing this well are the ones treating external signals such as trade policy shifts, commodity price swings, and regional instability as first-class inputs to their forecasting models, not footnotes added after the fact. The forecast that only looks backward is already out of date the moment it’s published.
Supplier risk: Every manufacturer has suppliers. Most suppliers have suppliers. A multi-tier supply chain means a disruption at tier three may not surface until it becomes a line-stoppage event at tier one; by which point it’s no longer a risk, it’s an emergency. AI does not eliminate that risk. What it can do is surface the early signals like a shipping delay, a financial filing, a labor dispute halfway around the world and give operations teams time to react instead of scrambing. The value isn’t prediction for its own sake. It’s the extra week or two of lead time that turns a crisis into a manageable problem.
The talent problem
Supply chain AI is unusual in that it requires two things that rarely travel together: deep operational domain knowledge and genuine AI flue
ncy.
Your best supply chain operators know the business. They know what a demand signal looks like when something is wrong. They know what a supplierrelationship looks like right before it breaks. They know what a logistics delay means when a carrier is quietly falling behind on commitments. What they often don’t have is the AI fluency to translate that knowledge into a model.
Your data science team can build models. What they often don’t have is enough operational context to know which signals matter and which are noise.
That gap is real, and it’s growing. Gartner analyzed 35 million job postings over the past three years and found a 387% increase in AI-related roles in supply chain. Companies know they need to close this gap. Most are finding it very hard to hire their way out of it. The organizations moving fastest aren’t trying to build every capability alone. They’re working with partners who bring the technology depth and implementation experience to close that gap now, rather than waiting years for the talent market to catch up.
A word on where to start
I see two approaches in the market.
The first is horizontal: build the data and AI foundation, establish the architecture, then identify use cases. Theoretically sound. Practically very hard to fund, because the ROI is diffuse and distant, and it asks leadership to write a large check on faith before there’s anything to point to.
The second is vertical: start with a specific business problem, build the minimum capability needed to solve it, demonstrate ROI in weeks rather than quarters, and then expand. Supplier risk scoring is a good example. You do not need to boil the ocean. You need the right data for that use case such as shipment records, supplier financials, news and trade signals, the right models trained on that data, and a production deployment that proves the concept before the budget cycle closes.
The foundation matters. It just cannot be the starting point for the conversation. It has to be the thing you build because the first use case worked, not the thing you ask people to trust before anything has worked.
The bottom line
The supply chain problems that feel most intractable right now, such as demand uncertainty, supplier opacity, and logistics volatility, are precisely where AI has something real to offer. Not as a replacement for operational expertise, but as the layer that makes that expertise scale. ele
