How to Optimize Delivery Routes and Track Shipments in Real Time

Route optimization sounds technical until you realize the customer doesn't care about your algorithm — they care that the driver is at their door at 2:14 like the SMS said. That's the whole job. Everything else is plumbing.

Last-mile delivery is the most expensive leg of the supply chain. It is also the only leg the customer sees. A driver running 90 minutes late costs you the cost of the fuel, the cost of the driver's time, the cost of the failed-delivery reattempt, and — if it happens twice — the customer. Most dispatchers running 5 to 50 vehicles are still building routes by hand in the morning, then radioing drivers to reroute around traffic, then fielding 'where is my package' calls all afternoon. That stack is held together by tribal knowledge and one veteran dispatcher who knows that the bridge on Route 9 backs up at 3:15 every day.

The modern alternative is a five-step loop: optimize the route before the driver leaves the yard, track the driver in real time, push the customer SMS updates as the ETA shifts, capture proof of delivery at the door, and run a clean reattempt or return flow when a delivery fails. Here is how each step actually works in 2026.

What you are solving is called the vehicle routing problem with time windows and capacity constraints. In English: given N stops, M vehicles, each stop with a delivery window the customer chose at checkout, each vehicle with a max weight or pallet count, what is the order and assignment that minimizes total drive time?

Three levers matter more than the algorithm itself.

Time windows. A 4-hour window (8 a.m. to noon) is roughly 10x easier to optimize around than a 1-hour window (10 to 11). If your customer-facing checkout offers 1-hour windows, you are paying for that promise in route inefficiency. Most dispatchers find a 2-hour window is the sweet spot — tight enough that customers stay home, loose enough that the optimizer has room.

Vehicle capacity. Cubic feet, weight, and pallet count are different constraints. A van full of pillows hits cube-out before it hits weight. A van full of dumbbells does the opposite. Tell the optimizer which one matters for each vehicle or you will get routes that look great on the screen and fail at the loading dock.

Re-optimization frequency. Some shops optimize once at 5 a.m. and lock the route. Better shops re-optimize every 30 minutes against live traffic and any new orders that came in. The cost is compute and driver disruption — drivers hate getting a new manifest at 11 a.m. — so most settle on a morning optimization plus mid-route re-routing only when traffic is more than 20 minutes out of plan.

GPS on the truck is half the answer. The other half is geofence checkpoints — invisible circles around each stop, plus the depot, plus high-friction zones (the warehouse loading dock, the apartment complex with the bad gate). When a driver crosses a fence, the system fires an event: arrived-at-stop, departed-stop, entered-no-service-zone.

Those events are what drive the ETA recalculation. A naive ETA is just 'distance over average speed.' A live ETA is 'current GPS position, plus current traffic on the remaining segments, plus the average dwell time this driver has at this stop type.' A driver who averages 12 minutes per residential delivery and 4 minutes per business delivery should not be modeled with the same dwell-time constant. Most dispatch software gets this wrong out of the box; the ones that learn the dwell time per driver and per stop type are the ones whose ETAs are within 5 minutes 80% of the time.

The other thing live tracking gets you: an honest answer when an attorney calls about an accident at 4:47 p.m. on Tuesday. GPS breadcrumbs at 30-second intervals plus geofence timestamps will save you a deposition.

The single highest-ROI thing in last-mile delivery is the 'driver is 10 minutes out' SMS. Customers who get that text are home for the delivery. Customers who don't get it are in the shower or on a Zoom call, which is how a successful delivery becomes a failed one.

The minimum SMS pattern is three messages: morning-of confirmation with the time window, on-the-way at depot departure, and arriving-soon when the geofence around the customer's address is 10 minutes away on current traffic. For high-value or signature-required deliveries, add a fourth: an on-the-way photo of the package on the truck so the customer knows the right item is coming.

If the driver is going to be more than 30 minutes late on a window, send a late notification proactively. Customers forgive late deliveries. They do not forgive being told a different story than what is actually happening. The dispatcher who sends the 'we are running 45 minutes behind, new ETA 3:30' text saves four phone calls and one bad review per week per driver.

Proof of delivery is the artifact that ends a 'I never got it' dispute. Three pieces, in order of cost.

Photo at the door. Standard for residential. The driver snaps the package on the porch, with enough of the door or address visible to identify the location. Stored against the stop record with timestamp. This alone resolves most porch-pirate disputes — if the photo shows the package on the correct porch at 2:14, the carrier liability is clear.

Signature for high-value. Anything over a threshold (set yours — $200 is common for retail, much higher for medical or industrial) requires a captured signature on the driver's mobile device. The signature is the legal handoff.

Geo-coordinate stamp. Every photo and signature should embed the GPS coordinates where it was captured. This is the single thing that resolves the 'driver dropped it at the wrong house' case. If the geo-stamp is at 123 Maple and the customer is at 132 Maple, you know in five seconds — without dispatching a second truck to investigate.

The combined record (photo + optional signature + geo-coordinate + timestamp) is what gets attached to the customer-facing tracking page and to the carrier's chargeback dispute file.

Failed deliveries are not a fringe case — for most carriers they run 3-8% of stops. If you do not have a clean flow for them, every failure becomes a dispatcher phone call and a driver decision made in a moving truck.

The rules to codify before launch:

Reattempt rules. Most carriers reattempt twice within a 3-business-day window. After the second failure, the package goes to a customer pickup hold or returns to sender. The system should fire the reattempt event automatically — driver marks 'no answer,' system schedules tomorrow's route to include the address, customer gets an SMS with a calendar link to confirm or change the next-day window.

Customer pickup window. If the customer chooses to pick up at your depot or at a partner location (a locker, a corner store), the package needs a pickup-by date and an SMS reminder at day-1 and day-3. Anything past day-7 in most contracts goes back to sender at the shipper's expense.

Return-to-sender flow. When the package goes back, the dispatcher needs a clean record of why (refused, address bad, no answer x2, damaged in transit) because that reason is what the shipper bills against. Vague return reasons are how carriers eat the cost of returns they shouldn't.

The difference between a fleet that loses 1.5% of revenue to failed deliveries and one that loses 0.4% is entirely in this step. Not the algorithm. Not the SMS template. The rules-as-code that turn 'no answer at the door' into the right next action without a human in the loop.