The last mile is the most expensive part of any delivery — often more than half the total cost of moving a parcel from warehouse to doorstep. It is also, paradoxically, where the biggest and fastest savings hide. Most operations obsess over line-haul rates and warehouse throughput while the real money leaks out on the final few kilometres. This piece breaks down why that happens and what to do about it.
Why the last mile is so costly
Long-haul freight benefits from economies of scale: one driver, one vehicle, one long trip carrying a full load. The last mile is the opposite. Trips are short, dense, and unpredictable. A single driver might make dozens of stops in a shift, and every one of them is an opportunity for something to go wrong.
A driver idling in traffic, doubling back for a missed address, or waiting at a closed gate burns fuel and hours with nothing to show for it. Failed first-delivery attempts are especially expensive, because they don't just cost the failed trip — they cost the entire second trip too, plus the customer-service overhead of rescheduling.
What dynamic routing actually changes
Traditional routing plans the day the night before. A dispatcher — or a static algorithm — sequences tomorrow's stops based on yesterday's assumptions. By the time a driver is on the road, those assumptions are already stale.
Dynamic routing recalculates the optimal sequence continuously. It accounts for live traffic, delivery time windows, vehicle capacity, driver breaks, and new orders that arrive mid-shift. Instead of a plan you make once, the route becomes a decision you keep making, all day long.
The differences compound quickly:
- Fewer kilometres driven per delivery, which directly cuts fuel and vehicle wear
- Higher drops-per-hour, so the same fleet completes more work in the same shift
- Less idle time, because the system routes around congestion before it happens
- Fewer failed first attempts, thanks to tighter, more realistic time windows
The 30% number, in practice
Numbers only mean something in context, so here is ours. Across our pilot lanes in the Kathmandu valley — a genuinely hard environment of dense traffic, informal addresses, and mixed vehicle types — re-sequencing routes in real time cut distance per delivery by roughly a quarter and lifted daily drops per driver by about a third.
Put those together and the net effect on last-mile cost per parcel came out to roughly 30% down. Crucially, service quality improved at the same time: on-time performance rose because the ETAs the system promised were the ETAs it could actually keep.
Getting the inputs right
Routing is only as good as the data feeding it. Three inputs matter more than any clever algorithm:
- Accurate addresses, ideally geocoded, so the system isn't guessing where a stop actually is
- Realistic service times per stop — a bulky furniture drop is not the same as a letter
- Honest vehicle constraints, from capacity to access restrictions on narrow streets
Teams that invest a week in cleaning these inputs almost always get more from routing than teams that chase a more sophisticated engine on top of messy data.
The human side
It's tempting to treat routing as a purely mathematical problem, but drivers are not interchangeable units. A good routing system earns their trust by producing routes that feel sensible on the ground — not sending them back and forth across a neighbourhood to shave a theoretical kilometre.
The best implementations keep a human in the loop for exceptions, let drivers flag problem addresses, and learn from those corrections over time. Adoption is where routing projects succeed or fail, and adoption is a people problem as much as a technical one.
The second-order gains
The fuel and distance savings are the headline, but the second-order gains often matter more over time. Tighter, more realistic routes mean drivers finish their shifts on schedule instead of chronically running late, which reduces burnout and turnover in a role where both are painfully expensive. More predictable completion times let you promise customers narrower delivery windows and actually keep them, which lifts first-attempt success and cuts the support load.
There's a capacity dividend, too. When each vehicle completes more drops per shift, you can absorb growth without immediately buying more trucks or hiring more drivers — the single largest fixed costs in the operation. In effect, dynamic routing raises the ceiling on how much volume your existing fleet can handle, deferring capital spend that would otherwise be forced on you by growth.
Finally, the data that routing generates becomes an asset in its own right. Every completed route teaches the system which addresses are troublesome, which time windows are unrealistic, and which lanes consistently run long. Fed back in, that knowledge compounds: the routes get smarter, the estimates get more honest, and the whole operation gets a little better every week without anyone doing anything heroic. The savings you can measure on day one are real, but the gains that accrue quietly over months are what turn routing from a cost-cutting tool into a genuine competitive advantage.
The takeaway
The route is not a plan you make once and hand to a driver. It's a living decision that should update as the world does. Operations that internalise that — and give their teams the tools to act on it — find that the last mile stops being their biggest cost problem and starts becoming their sharpest competitive edge.