Why Your Car Started This Morning — and the Supply Chain Science Behind It
The battery in your car will fail, eventually. The question is whether the replacement will be on the shelf when you need it — at the auto parts store two miles from your house, in the right group size, on a January morning when temperatures have dropped overnight and half the cars in the parking lot are struggling to start. Or on a sweltering August afternoon, when months of accumulated heat have quietly finished off a battery that was already on its way out.
That question, while technical, sits at the center of a forecasting challenge that has become increasingly important for American manufacturing. Addressing it requires an unusual combination: deep technical fluency in machine learning and data systems, and the operational judgment to translate model outputs into decisions that real supply chains can actually execute.
Vishal Singh has spent more than a decade building expertise in this area.
The Problem With Predicting Yesterday
Traditional demand forecasting in manufacturing has a structural flaw that most people outside the industry never encounter: it looks backward. The standard approach asks what sold last January and uses that number, adjusted slightly, to predict what will sell this January. For stable, predictable products, that works reasonably well.
But low-voltage automotive batteries — the kind in nearly every passenger and commercial vehicle on American roads — are not stable or predictable. Their failure rates are driven heavily by external conditions that historical sales data cannot capture: temperature swings, the age distribution of vehicles in a particular region, shifts in consumer behavior at the point of sale, and the invisible inventory dynamics rippling through distribution networks in real time.
“The model isn’t wrong because the math is bad,” Singh explains. “It’s wrong because it’s asking the wrong question. You can’t predict a January cold-snap spike by looking at last January’s cold-snap spike. By the time history tells you what happened, the shortage is already happening.”
His work centers on moving beyond that backward-looking model toward a different approach: forecasting systems that are designed to integrate multiple real-world demand signals simultaneously — ZIP code-level vehicle registration data, live weather patterns, point-of-sale activity, regional pricing dynamics and customer inventory levels — and use machine learning to detect shifts in demand before they surface as shortages.
The technical term is causal forecasting. The practical goal is to help manufacturers position inventory more effectively before demand rises, rather than reacting after service centers begin placing urgent orders.
A Charlotte Problem, a National Scale
Charlotte is a useful place to think about these dynamics. The city sits at a logistics crossroads: a major distribution hub for the Southeast, connected to Midwest manufacturing, Atlantic ports, and an automotive market that spans everything from suburban commuters to commercial fleets. The demand pressures that flow through that network are not theoretical. Across the country, bitter winter cold snaps accelerate battery failures overnight. Summer heat — sustained, weeks-long heat of the kind that bakes the South and Southwest — quietly degrades batteries that won’t show the damage until October. A single weather event in one region can shift demand patterns across an entire distribution network within days.
Those factors are difficult to capture in a spreadsheet based only on historical sales averages. They are better suited to the multi-signal models he has worked on, which treat battery demand as something shaped not only by past sales but also by real-world conditions around the product.
The scale at which these systems operate makes the precision consequential. The low-voltage automotive battery market in the United States serves roughly 280 million registered vehicles. The distribution network that keeps those vehicles supplied spans national retail chains, regional service centers, and commercial fleet operators. Even a modest improvement in forecasting accuracy, applied across a large distribution network, can help manufacturers operate more efficiently and, more practically, reduce the chances that a driver is told the part they need is out of stock.
At that scale, even modest gains in forecasting precision can have a meaningful effect, helping reduce emergency freight runs, limit shortages at service centers, and make the distribution network more resilient when demand shifts.
What AI Actually Means on a Factory Floor
Singh has become something of a critic of how artificial intelligence gets discussed in manufacturing circles. Not of AI itself — his work depends on it — but of the gap between how the technology is marketed and how it actually needs to function if it is going to be trusted by the people responsible for production schedules and inventory commitments.
The models Singh builds are designed to be interpretable: to surface the specific signals driving a predicted demand shift, to explain whether the primary driver is a weather event, a vehicle fleet aging pattern, or a competitor pricing move. That interpretability is not just a design preference. It is, he argues, a prerequisite for deploying AI in high-stakes operational environments where the cost of a wrong call isn’t a delayed email but a missed production window or a regional stockout.
Singh’s approach reflects the view that AI forecasting systems are most useful when they remain connected to human expertise and operational context. His planning experience across industries and geographies has shaped a practical understanding of how forecast outputs are applied in business settings, where accuracy affects inventory, service levels, and day-to-day operations.
The Infrastructure Argument
There is a federal dimension to this work that rarely surfaces in coverage of supply chain technology, but one he finds important to the argument for why forecasting precision matters at a national level.
The White House’s 2021-2024 Quadrennial Supply Chain Review identified resilient supply chains as essential to U.S. economic security, explicitly calling for “improving data and analytical tools to promote supply chain resilience” and noting that supply chain disruptions can propagate into shortages, price spikes, and logistics failures with broad economic consequences. The report specifically highlighted the need to digitize supply chain information to enable “more precise predictions of and responses to potential disruptions.”
Low-voltage automotive batteries sit within that critical infrastructure argument in a direct way. They power not just consumer vehicles but commercial delivery fleets, emergency response vehicles, and the logistics networks that move goods across the country. A regional shortage is not merely a business problem. It can mean a delayed shipment, a grounded fleet, or an emergency vehicle that doesn’t start.
Singh’s framing of his own work reflects this: he describes it not as a competitive advantage for any particular manufacturer but as a contribution to the reliability of a system that most Americans depend on without thinking about it. The forecasting problem is a public infrastructure problem dressed in supply chain clothing.
The Translation Problem
Singh describes forecasting as fundamentally a translation problem. The data exists. The signals are there. The hard part is building systems that make those signals legible to the people who have to act on them — who have to order more stock, shift production, or reposition inventory based on what the model says, knowing that the model is not infallible and that the cost of being wrong is real.
That translation also depends on practical supply chain experience, including familiarity with how operational decisions are made, where implementation challenges can arise, and how model outputs are communicated across operations and finance teams. In practice, this helps connect forecast accuracy with decisions that can be applied inside the business.
In Charlotte, that translation problem plays out against the backdrop of a city that has quietly become one of the more important logistics nodes in the American Southeast — a place where getting the forecast right has consequences that extend well beyond the warehouse.
The Quiet Work
On a cold January morning, or a sweltering August afternoon, when a driver’s car turns over without incident, the forecasting system that put the right battery in the right store at the right time will receive no credit. That is exactly how it is supposed to work.
Singh describes the success condition for his field in terms that reveal something about how he thinks about the work. “You know it’s working,” he says, “when nothing happens. When there’s no shortage, no emergency freight, no call from a service center that’s been out of stock for a week. The value is in the disruption that doesn’t occur.”
For a city that moves as much commerce as Charlotte does — and for a country that runs on vehicles that all share the same basic vulnerability — that quiet reliability is worth more than most people will ever know.
Members of the editorial and news staff of charlotteobserver.com were not involved with the creation of this content. All contributor content is reviewed by charlotteobserver.com staff.