The Last Mile is a Decision Problem, Not a Delivery Problem

When we started Locus in 2015, the on-demand economy was booming. A wave of startups was being built by product specialists, not logistics specialists. And everyone had figured out how to answer “Where is my package?” Nobody was going after the harder question: where should my package be?

That gap is what Locus was built to close.

Logistics is fundamentally a real-world problem. Addresses are imperfect. Traffic is unpredictable. Vehicles break down. A last-minute exception is not the exception. It’s the rule. 

Murphy’s Law runs the warehouse floor, and building a platform that handles all of it, reliably, at scale, is genuinely hard. That’s the problem we picked up. Can we build the most real-world decision-making platform in logistics?

The decisions nobody was automating

Think about what actually happens when you order a pair of shoes online and they show up the next day.

Someone had to decide whether to use their own fleet or a delivery partner. Whether to ship from the store or the distribution center. Whether a bike or a truck was the right vehicle. Whether your preferred time slot was feasible. Whether an eco-delivery option made sense for your route.

That’s before the order is even dispatched.

After dispatch, the decisions multiply. If the delivery fails, do you reschedule? Change the vehicle type? Flag the address for a future correction? If the customer’s building only allows small vehicles, why did you send a 14-tonner in the first place? And pulling back further, is your entire distribution network set up to serve your current demand patterns, or is it optimized for something that stopped being true two years ago?

These are the decisions Locus makes for our clients every day. Not recommendations sitting in a dashboard. Actual decisions, executed on the ground.

Financial cost and environmental cost run together

Here is something I find myself saying a lot: You cannot keep running trucks with more air than packages. That’s bad for the bank balance and bad for our cities.

These two outcomes, lower cost and lower emissions, are not at odds. When you improve vehicle fill rates from 50-60% to 80-90% on a sustained basis, your average miles per package drops by 15 to 20% in mature operations, and up to 40 to 45% in newer ones. For every delivery you make, you are polluting less. When you layer in intelligent dispatch that matches the right orders to the right vehicle type, including routing electric vehicles to the areas and orders where they perform best, the gains compound further.

Last year, Locus helped sequester as many emissions as 20,000 acres of forest would absorb. That number is a byproduct of making logistics smarter, not a separate sustainability program.

Where traditional TMS tools hit their limit

Traditional TMS tools were designed for linehaul. Predict movement within a day, within an hour. That level of accuracy is fine when your delivery window is measured in days.

In the last-mile, the entire delivery often takes an hour. A five-minute error is the difference between on time and late. Your zone of accuracy needs to be down to minutes, and your system needs to know things a linehaul tool was never built to know. Is the customer actually home? Does this address support a heavy vehicle? If I open both the ambient and frozen sections of this truck, how much does my turnaround time change, and how does that cascade across the 11 other deliveries on this route?

There is one difference I come back to often: if a linehaul truck misses a docket yard, it’s a financial loss for that day. If you miss a delivery slot with a customer, it’s a customer loss. All the acquisition spend, all the conversion work, gone. And customers today are vocal. One bad experience, surfaced publicly, reaches people your marketing never reached.

That is why we have built features that ensure, after a troubled delivery, the next attempt goes out with a rider who has at least 500 positive ratings, at least six months in the system, with a built-in 10-minute buffer. Your delivery is not a cost center. It is a revenue generator. That framing changes everything about what belongs in the platform.

Delivery as a revenue generator

We now regularly see the head of sales in our buying cycles, not just the head of logistics. That tells you something.

A company offering 48-hour delivery can start presenting a dynamic 24-hour option where the route already passes the customer’s door. Charge three to five dollars extra. Or nudge customers to accept tomorrow’s delivery in exchange for a small discount or a sustainability message. For premium brands, especially, the customer is often happy to wait. What they want is assurance and a certain brand experience around the delivery itself.

Proactive communication changes the economics, too. A five-minute delay gets a heads-up message. A fifteen-minute delay gets a ten percent discount coupon. These are programmable decisions now. The delivery experience is being stitched directly into demand generation, retention, and customer service, not just operations.

And it goes further. We work with a leading Japanese retailer whose delivery team’s average customer turnaround time was going up, not down. Most clients ask us to reduce it. Their leadership flagged it as a success. Their delivery staff had been trained to ask customers about the product, what they liked, and what they’d change. Those conversations fed directly back to the product team.

The delivery team had become a customer interaction team. That is what the technology enables when you stop treating last-mile as a cost line.

Why we gave dispatch to the agent

We have called Locus a Digital Supply Chain Officer for a long time. Officer is another word for an agent. The distinction matters: is the output information, or is it an action?

Most TMS tools surface information and leave the human to decide. The next layer is suggestion. Beyond that, a suggestion with enough confidence to be acted on without intervention. That is the agentic layer.

At a high-traffic last-mile warehouse doing 9,000 dispatches a day, even a one percent exception rate produces ninety exceptions. That’s almost one every minute. No human dispatch team resolves that at speed without errors. The algorithms handle it. The human capacity that was spent on operational triage moves to customer interaction and problem resolution.

We have been on this journey for ten years. It is not an overnight change. It is a gradual shift from information to suggestion to accepted action. Dispatch is the clearest example. A decade ago, experienced dispatchers allocated orders manually based on geography they had memorized. Today, agentic systems assign the right route to the right driver automatically, accounting for vehicle type, load composition, time slot preferences, and the customer’s delivery history.

The operators are not being replaced

I want to address this directly because I think it is the most important thing that is not being said loudly enough.

When we reduced the dependence on experienced workers for operational decisions, the first note of appreciation we consistently receive from leadership is during peak season. December is the festive season globally. It is also when experienced staff take leave. Before, that was a double problem: peak volume and knowledge walking out the door simultaneously.

A learning system does not degrade with staff turnover. It absorbs the tribal knowledge of experienced operators and converts it into structured enterprise knowledge. The more you use it, the more valuable it becomes. The operators who used to manage dispatch manually now manage exceptions, lead teams, and interact with customers. Their expertise is worth more when they are not buried in a dropdown menu.

The teams that have matured furthest are the ones that treated this as a growth opportunity instead of a replacement threat. Their people are doing more skilled, more human work. The system handles the volume.

What the next five years look like

Last-mile will become invisible. That is the goal. When you flip a light switch, you do not think about the grid. Last-mile delivery is heading toward the same expectation. If you say 5:30 PM, it means 5:30 PM. Not a window or an estimate.

Consumer demand will drive that. Infrastructure will enable it. Autonomous vehicles, ground robots operating on urban curbs, and drone delivery will all shift the cost equation, particularly the human-cost component of final delivery. As that changes, replenishment windows compress, and entirely new retail models become viable.

In the warehouse, augmented reality is one of the most underappreciated near-term changes. Vision-based systems that identify, measure, and track unlabeled packages in real time change the cost of reconciliation at pickup. These are not decade-away technologies.

Arthur C. Clarke said any sufficiently advanced technology is indistinguishable from magic. That is the job of logistics technology: make it invisible to the shipper and the customer. The package arrives. No one thinks about how. That is what we are building toward.