The article compares three models of the "last mile" - address courier delivery, pick-up points, and drone delivery. Based on simulation calculations for three types of urban morphology (center, suburbs, barrier zones), the cost of one order, the time from registration to receipt, the share of first successful attempts, the risks of returns, and infrastructure requirements are estimated. It is shown that pick-up points form the "skeleton" of low cost and stable terms, couriers are needed for valuable segments, complex addresses, and heavy shipments, drones are effective on short congested "bridges" and in hard-to-reach areas with favorable weather and clear tolerances. The best results are given by a hybrid with dynamic channel selection based on address density, traffic jams, and weather data. A practical switching framework is proposed that reduces total costs without degrading the service. The quality of address data, accounting for repeated attempts, and pre-agreed switching rules are critical.