Last mile delivery software: managing the promise to the recipient

The promise and the plan must be the same object

A window offered at checkout constrains everything afterwards, and the common architectural mistake is promising in one system and planning in another with different assumptions about capacity and travel time. Slot availability should be derived from real routeing capability for that day and area, which means the promise engine consults the same constraints the planner will use. Where they diverge, the operation spends its day apologising for windows it was never able to serve.

In-day adjustment and recipient communication

Once vehicles are moving, arrival estimates change with traffic, dwell and additional stops. Communication is the mitigation: a narrowing time estimate as the vehicle approaches, a notification when it is next, and a channel for the recipient to redirect, defer or authorise a safe place. Each of those options has an operational cost, and the ones worth offering are those that reduce failed attempts by more than they disrupt the route. Offering unlimited changes late in the day is a common way to make a plan unworkable.

Failed attempts are a process, not an exception

A meaningful share of attempts end with nobody present, access refused, an unsafe location or a rejected item. The system needs coded outcomes, a decision about what happens to the goods, retry, neighbour, collection point, return to depot, and a rule about how many attempts are worthwhile before the economics turn negative. Recording the outcome accurately matters commercially, since attempt data drives redelivery cost, customer contact volume and, where the delivery was on behalf of a client, the charge that may be raised.

Mixed capacity: employed, subcontracted and self-employed

Final-mile capacity is rarely homogeneous. Systems must dispatch to employed drivers, contracted courier companies and independent drivers simultaneously, with different pay or charge structures, different data visibility and different obligations. Modelling them as one pool with an attribute keeps routeing simple while allowing rules about which work may go where. The commercial and employment status of delivery workers is a live legal question in many jurisdictions, and system design should not assume a classification that the relevant national authority may treat differently.

Collection points and out-of-home delivery

Delivering to a locker or shop counter changes the model: capacity is a physical box or shelf, the handover is to an intermediary rather than the buyer, and a second handover occurs when the recipient collects. Systems must track occupancy, expiry and the return flow for uncollected items, and they must express the customer promise differently, since the goods arrive at a place rather than at a time. Operators mixing home and out-of-home delivery need both models in one plan, which is more than a routeing setting.

Frequently asked questions

How narrow should delivery windows be?

As narrow as routeing capacity supports without inflating vehicle requirements. Tighter windows constrain the plan sharply, so the choice belongs to commercial and operations jointly, tested against what the fleet can actually hold on a busy day.

Does live tracking reduce failed deliveries?

It helps when it prompts an action, such as confirming presence or nominating a safe place. Tracking that only displays a moving icon informs the recipient without changing the outcome, so the value sits in the interaction it enables.

Should final-mile software also plan the routes?

Many products do both, which suits operations with a stable service model. Complex constraints, multi-day territory design or large mixed fleets often justify a dedicated planning tool feeding the execution layer.