.png){kind=logo}
For years, the decision to shut down a freight lane came down to two numbers: how full the trucks were, and how far they drove.
Everything else a lane costs sat outside the calculation. The driver's hours. The way a sort facility handles a pallet differently from an envelope. The day of delivery speed customers lose when volume gets rerouted. The fuel burned by a truck running half empty on the return leg. Planners knew those costs were real. The tools that made the call did not ask about them.
Uday Dhembare argues that the omission is not a rounding error, and that it explains why so many large networks cannot cut emissions without feeling they are paying a premium to do it.
Dhembare manages a data engineering team responsible for optimizing a continental-scale logistics network. He is the creator and architect of a lane optimization model that decomposes every route into six cost components: truck, fuel, driver, warehouse handling, carbon and delivery speed. The list is not the innovation. Evaluating all six simultaneously, rather than one after another, is.
That structure supports a claim the industry still tends to resist.
"Carbon-inefficient routes are frequently cost-inefficient routes," he says. "Empty miles, suboptimal lane configurations, and poor vehicle utilization waste both fuel and money. When you build systems that see the full picture, cost and carbon savings often converge."
The approach he replaced had already been revised once. The earliest decisions to retire a lane relied on truck volume and driving distance alone. A later two-dimensional cost tool improved on that and stayed wrong in specific, consequential ways. It assumed every affected package lost exactly one day of delivery speed, whatever the package and wherever it was going. It ignored variation in package size in warehouse handling entirely.
Those simplifications carried a cost of their own. Teams pushed back on recommendations built from averages they could see did not describe their operations. Decisions that should have taken days stretched into weeks of alignment discussions.
His framework evaluates each lane using actual volume data rather than simplified proxies. It computes the differentials across all six factors at once, then searches for alternate routes inside the existing network without breaching the limit on how many times a package gets handled. The model, by his account, is capable of evaluating between 500 and 1,000 lanes and identifying a potential cost benefit of $100 million to $200 million across them. Dhembare is precise about what that number describes. It is the size of the opportunity the analysis surfaces, not money already banked.
The more transferable idea is about resolution. Most network analysis runs on band level averages and aggregated metrics, which are convenient and, in his account, actively hide the patterns that drive the largest savings. His model evaluates at the level of individual packages, on the argument that opportunities visible there disappear at every higher level of aggregation.
"You cannot optimize what you cannot measure at the decision-making granularity," he says.
He estimates that 20 to 30 percent of lanes in a typical network run at a net loss once every cost factor is properly accounted for. That figure is his own, drawn from his experience rather than from published industry data, and it is the kind of number that only becomes visible at package level resolution.
One effect he describes is easy to overlook. Retiring a lane does not make its volume disappear. The volume redistributes across the lanes that remain, and often improves their utilization and their economics. Each optimization changes the conditions for the next one, which is why he describes the benefits as compounding rather than flattening out.
Then there is the timescale problem. Network strategy gets set quarterly or annually. Execution happens every day. Strategic intent such as carbon reduction never reaches operational reality, he argues, unless the tools guiding daily decisions have it built in.
His recommendation follows from that. Organizations should "treat carbon as a first-class optimization objective, not a reporting metric," he says, because "when carbon is just a dashboard number reviewed quarterly, it never influences real-time decisions."
Dhembare holds a master's degree in industrial engineering and has worked in large-scale data processing and optimization for supply chains for over 10 years. The model now feeds biweekly network assessments and executive business reviews at his organization, which is a more telling measure of the work than any single output, since it means the six-factor view is what leadership sees by default rather than something requested for a special study.
Adoption was engineered about as deliberately as the algorithm. He built the framework modularly, so the core logic stays constant while parameters tune themselves to local conditions, and produced simplified outputs shaped to what each group of stakeholders needed to see. He says nothing available handled this kind of multi-dimensional evaluation at scale when he started, and he considers the methodology applicable to any organization running a multi-modal transportation network, an assessment he offers about his own work rather than a claim of adoption elsewhere.
He expects the ground to shift underneath all of this. Within five to ten years, he predicts, carbon transparency will move from a competitive advantage to a compliance requirement, and organizations that have already made carbon a decision variable will meet it without rebuilding their systems. That is a forecast, not a schedule. The argument underneath it does not depend on the timing. If a lane's emissions and a lane's economics usually point in the same direction, the case for measuring both is available now, and the companies still treating carbon as a reporting output rather than a decision input will keep paying for information they already have.
