How AI Route Optimization Cuts Last-Mile Costs by 30%
We benchmarked 2,400 real delivery runs across three US metros. Here is what manual dispatch leaves on the table — and how AI clusters recover it.
On-time rate is the most-quoted KPI in last-mile — and the most quietly fudged. A short field guide to the definitions that matter.
Three vendors will quote you three different on-time rates for the same fleet on the same day. They are not lying — they are choosing different denominators. Here is the field guide.
Definitions in the wild
- Window-based: delivery within a customer-promised window. Strictest, most honest.
- Slot-based: delivery within a system-assigned slot. Easier to hit because the slot is internal.
- Tolerance-based: delivery within ±N minutes of ETA. Compounds drift across the day.
- Attempted-based: counted as 'on-time' if the driver attempted the address in window. Often masks failures.
Fleetune reports window-based by default and shows the others in the dashboard so dispatchers can reconcile against their customer SLAs.