How Delivery Routing Software Cuts Last-Mile Costs

Cost per drop is not mostly fuel. It is driver time, and driver time is consumed by three things: distance between stops, time spent stationary at each stop, and time lost to failure — waiting at the store for an order that was not ready, returning to an address where nobody answered, or re-running a drop that was refused.

This matters because the three respond to completely different interventions. Distance is a routing problem. Stationary time is a store and access problem. Failure is usually an upstream problem — a slot that should not have been sold, an order that finished picking late, or an address the customer never corrected. Operators who attack all three as "routing" tend to buy an optimisation engine and find the savings smaller than promised.

The most consequential delivery decision happens days before the van moves, when the storefront offers a window. A slot is a promise against two scarce resources at once: picking labour in the store and vehicle capacity in that geography during that hour. Offering slots that are not backed by both is how an operation books its own failures in advance.

Most oversold-slot problems trace back to capacity that was estimated rather than calculated — a fixed number of orders per window, set once, that takes no account of basket size, the store's staffing that day, or how far apart the drops in that zone actually are. Twelve orders in a dense urban window and twelve orders spread across a rural zone are not the same commitment.

Customer demand for delivery windows is not evenly distributed. Everyone wants weekday evenings and Friday afternoons; almost nobody wants mid-morning Tuesday. Left alone, that concentration forces the operation into its most expensive hours and leaves capacity idle the rest of the week.

The lever is slot pricing and incentives — making quiet windows cheaper or free, and premium windows carry a fee — which moves demand toward capacity instead of turning customers away at the point of checkout. This is a commercial decision with operational consequences, and it works only if the pricing is set against real capacity data rather than intuition about which slots feel busy.

Grocery delivery is constrained by time windows, perishables, traffic patterns, store readiness, and customer expectations. Basic routing logic misses too many of those dependencies — it optimises distance while ignoring that a chilled load has a maximum time on the road, that a window is a hard commitment rather than a preference, and that a van has a capacity in crates rather than in stops.

When route building is detached from store operations, the fleet absorbs the mistakes. Drivers wait, windows slip, and support teams spend the day resolving preventable exceptions that were created hours earlier by decisions nobody connected to the delivery plan.

The best dispatch decisions happen upstream. Operators need visibility into order readiness, slot demand, store workload, and the shape of the delivery day before routes are finalised. A route built at six in the morning against orders that will not finish picking until ten is a plan that will be rebuilt under pressure.

Routing software adds value when it connects those inputs into decisions about batching, sequencing, ETA quality, and driver utilisation — and when it re-plans as the day moves, so a late order or a cancellation reshapes the remaining run instead of invalidating the whole schedule.

Lower costs do not come from a generic promise of optimisation. They come from fewer failed windows, fewer empty miles, better batch quality, and less manual replanning by dispatchers. Each of those is a specific, measurable change rather than a percentage improvement claimed by a vendor.

That also improves customer communication, because ETA logic is tied to the same operating reality that store and fleet teams are managing — which means the tracking message a customer receives reflects where the van actually is rather than where the morning plan expected it to be.

Slot behaviour is one of the most under-used datasets in grocery. Which windows sell out first, which are abandoned at the point of selection, how basket size varies by window, and which zones consistently run over their planned time all describe the operation more honestly than a delivery cost average does.

Sessions that end at slot selection are the most valuable signal of the set: a customer who built a basket and then left when they saw the available windows is telling you the operation, not the storefront, lost that order. That number rarely appears in ecommerce reporting because it looks like a website problem and is not one.

Delivery economics are decided by the worst weeks, not the average ones. Holiday trading and large promotions can double volume in a period that also has fewer staff available, and the controls that matter are the ones set beforehand: capacity limits that actually hold, contingency capacity released deliberately rather than by exception, and a defined rule for what gets rebooked when demand exceeds what can be served.

The failure mode is releasing extra slots under commercial pressure in the days before a peak, then discovering on the day that the picking capacity behind them never existed. That converts a busy week into a week of failed deliveries and refunds, which costs more than the orders were worth.

Supermarkets should review on-time performance, route adherence, cost per drop, dispatcher intervention rate, and delivery-window success by store or zone. Cost per drop needs to include failed attempts and redeliveries, otherwise the metric flatters exactly the operations with the worst failure rates.

Track slot utilisation alongside these, but do not optimise for it in isolation — a fully utilised slot book with a high failure rate is worse than a partially utilised one that delivers what it promised. Routing and slot software is worth the investment when it improves both margin discipline and day-to-day operational control, and the pairing is what makes either one work.