Last-mile delivery is the final stage of a goods-distribution process, in which a shipment moves from a local depot, shop, fulfillment facility, or other dispatch point to its intended recipient or collection location. The term describes a position in a distribution network rather than a fixed distance: a last-mile route may cover a few city blocks or extend across a sparsely populated rural area. It is a component of logistics that connects large-scale freight movement with individual households, businesses, and institutions.
This stage is operationally demanding because consolidated loads are divided among many destinations, each potentially imposing different access conditions, delivery windows, and handling requirements. A parcel can travel efficiently between major sorting centers yet require substantial labor to move from a parked vehicle to an apartment door. Last-mile performance therefore depends as much on building access, recipient availability, and street conditions as on vehicle speed.
Although closely associated with electronic commerce, last-mile delivery also includes postal rounds, grocery distribution, restaurant deliveries, medical supplies, and bulky goods. Its organization involves interacting decisions about facility locations, inventory, routes, workforce arrangements, handover methods, and technology. Its wider effects include competition for urban space, transport emissions, accessibility, and the distribution of work and costs among carriers, retailers, recipients, and public authorities.
Origins, scope, and changing service expectations
Last-mile delivery predates its modern name. Postal services, merchants, milk rounds, and local tradespeople long operated routes that combined multiple deliveries within a neighborhood. These systems developed recurring solutions to problems that remain important: grouping nearby customers, sequencing stops, securing access to premises, and collecting payments or returned containers. Regular rounds could exploit stable demand and familiarity with recipients, while urgent messengers offered a separate service for shipments that could not wait.
During the nineteenth and twentieth centuries, mail-order retail connected geographically dispersed customers with centralized inventories. In the United States, Rural Free Delivery began experimentally in 1896, and domestic Parcel Post service began in 1913. Such developments illustrate the relationship between distribution networks and access to goods outside major commercial centers. Motor vehicles subsequently increased carrying capacity and route range, while standardized addresses, sorting machinery, and shipment identification improved the coordination of large networks.
The expansion of online retail from the 1990s changed the composition and visibility of delivery demand. Instead of transporting many products together to a shop, a retailer could dispatch separate orders to individual homes. This did not eliminate retail logistics; it redistributed activities among warehouses, delivery networks, stores, and consumers. Home delivery replaced some shopping journeys, while also creating additional stops and making packaging, tracking, and failed handovers more prominent operational concerns.
The expression “last mile” is also used in telecommunications for the connection between a network and an end user. In freight distribution, its boundaries vary by organization. A parcel carrier may define it as movement from the final delivery station to the consignee, whereas a grocery operator may include store picking and dispatch. Comparisons of delivery costs or emissions consequently require a clear statement of which activities are included.
Service expectations differ substantially across markets. A scheduled furniture delivery may require two workers, a large vehicle, and installation time. A restaurant order may have a short useful delivery window and little opportunity for consolidation. Business deliveries often follow predictable opening hours, whereas residential deliveries encounter variable occupancy and access. Same-day, next-day, and scheduled services are therefore not simply different speeds of an otherwise identical operation: they create different inventory, staffing, routing, and capacity requirements. Greater urgency usually reduces the time available to combine orders, although high local demand can partly offset this disadvantage.
Distribution networks and local facilities
A last-mile network normally forms the downstream end of a larger system. Goods may arrive through long-distance road transport, air freight, railways, or intermodal freight transport, then pass through regional sorting centers before reaching local delivery facilities. The upstream network benefits from consolidation into large loads. The local network must reverse that process, separating shipments into routes and individual stops without losing identification, custody information, or delivery commitments.
Facility location creates a fundamental trade-off between property costs and transport performance. Large facilities on the urban fringe can offer lower land costs and efficient access for heavy vehicles, but may require long journeys to delivery areas. Smaller facilities near recipients can shorten these journeys while increasing rent, handling costs, and competition for suitable space. The outward relocation of logistics facilities, often called logistics sprawl, can make the distance between storage locations and demand centers an important component of delivery traffic.
A microhub is a small transfer facility close to a delivery area. Parcels may arrive in a larger vehicle and leave on smaller vans, cargo cycles, or walking rounds. Unlike a conventional warehouse, a microhub may hold little or no long-term inventory. Its main purpose is to change the scale or mode of distribution. Performance depends on transfer efficiency, available space, predictable inbound arrivals, and sufficient nearby demand; an additional handling stage can outweigh transport savings when volumes are low.
An urban consolidation center combines shipments for an urban area, potentially across multiple suppliers or carriers. Consolidation can reduce separate vehicle visits to shops, construction sites, or districts. However, it introduces questions about who pays for the facility, how commercial data are shared, and whether participants surrender control over customer service. A consolidation center is therefore an organizational arrangement as well as a physical building. Its effects depend on participation and load utilization rather than on the existence of the facility alone.
Retail premises can also serve as local fulfillment nodes. Ship-from-store operations use shop inventory to fulfill remote orders, while customer collection shifts the final movement to the purchaser. Dedicated local picking facilities separate order preparation from the public shopping floor. These configurations connect inventory placement with delivery design: storing goods closer to demand can shorten transport, but splitting inventory across many locations can increase stockholding and replenishment complexity. The best-performing arrangement for frequently purchased groceries need not suit infrequently ordered furniture or a retailer carrying a very large product assortment.
Cost structure, density, and operational performance
Last-mile costs are strongly influenced by how many successful deliveries can be completed per unit of labor and vehicle time. Important components include wages, vehicle ownership or leasing, fuel or electricity, insurance, depot handling, dispatch systems, failed attempts, claims, and customer support. For some shipments, the dominant activity is driving; for others, it is walking, waiting for access, locating an address, or unloading. Consequently, distance alone is an incomplete measure of operating effort.
A simplified cost expression is
where is cost per successful delivery, represents allocated fixed costs, is distance traveled, is cost per unit distance, is labor time, is labor cost per unit time, represents other handling and exception costs, and is the number of successful deliveries. This is an accounting framework rather than a universal pricing formula. Its terms must be defined consistently to avoid counting the same expense twice.
Delivery density has several meanings. Geographic density concerns the concentration of destinations, while stop density concerns the number of stops along a route. Shipment density can also refer to parcels delivered per stop. A large apartment building may offer high geographic density but slow service if access is difficult and parcels must be carried to individual floors. Conversely, a business receiving many cartons at one loading dock can provide high shipment throughput despite substantial unloading time. Population density is therefore only an imperfect proxy for delivery productivity.
Failed delivery attempts affect both cost and service. If a recipient must be present, an unsuccessful visit may generate another journey, storage, notifications, or redirection to a collection point. Under a deliberately simplified model in which each attempt succeeds independently with probability , the expected number of attempts until success is . Real operations depart from this model because later attempts are rescheduled, customers change instructions, and carriers may stop attempting delivery after a specified limit. The model nevertheless shows why improving handover success can matter as much as shortening routes.
Performance measurement commonly includes on-time delivery, first-attempt success, cost per shipment, stops per hour, vehicle utilization, damage rates, and customer complaints. These measures can conflict. Increasing parcels per route may reduce unit cost but increase late deliveries or physical workload. A narrow delivery window may improve convenience while reducing route flexibility. Claims that last-mile delivery represents a fixed percentage of total supply-chain cost are not universally applicable: the share changes with product value, network boundaries, delivery density, service level, and the treatment of warehousing and order preparation.
Routing, scheduling, and information systems
Route planning is commonly formulated as a vehicle routing problem: a set of customers must be served by vehicles subject to capacity, travel, and operational constraints. George Dantzig and John Ramser’s 1959 paper on truck dispatching is a foundational contribution to this field. Modern applications extend the basic problem to include delivery windows, mixed vehicle fleets, driver shifts, pickup requests, restricted streets, and uncertain travel times. These are problems of mathematical optimization, not merely shortest-path calculation.
A shortest path determines how to travel between locations. Routing determines which locations each vehicle should visit and in what order. Scheduling adds the timing of those visits. The distinction is important because the geographically shortest tour may be infeasible: a loading bay may open only at a particular hour, a customer may require an appointment, or a refrigerated vehicle may have limited operating capacity. A practical objective may combine distance, labor time, lateness penalties, vehicle use, and workload balance rather than minimize a single quantity.
Many routing variants are computationally difficult, especially at large scale. Operational systems therefore combine exact optimization methods, heuristics, and repeated adjustments. Geographic clustering can divide a service area into manageable territories, after which stop sequences are optimized within each territory. Real-time dispatch can accommodate urgent orders or disruptions, but changing a route has costs: parcels may already be loaded in stop order, drivers may have local knowledge, and customers may have received arrival estimates.
Reliable data are as important as optimization methods. Address standardization and geocoding connect customer records with map locations. The Global Positioning System supports vehicle tracking, while digital maps describe roads, restrictions, and travel times. A correct street coordinate may still fail to identify the relevant entrance, loading dock, or apartment access point. Delivery systems therefore often maintain operational details beyond ordinary navigation data, including gate instructions, preferred access routes, and historical service durations.
Machine learning can estimate travel time, stop duration, demand, or the probability of failed delivery. These estimates remain sensitive to changing conditions, incomplete records, and biased historical data. An apparent delay may reflect legitimate access difficulties rather than poor driver performance. Estimated arrival times must also accommodate uncertainty: a single precise timestamp can imply more confidence than the underlying traffic and service-time predictions justify. Routing quality is consequently evaluated through field outcomes, including reliability and feasibility, rather than solely by the numerical improvement achieved in a planning model.
Handover, product requirements, and reverse flows
The handover is the point at which transport becomes a completed delivery. It may involve direct receipt, a signature, an authorized unattended location, a building reception desk, or a collection facility. Different methods distribute responsibility and effort differently. Doorstep delivery minimizes recipient travel but can involve waiting, access problems, or theft exposure. Collection may improve carrier productivity while requiring the recipient to make a separate journey. A completed vehicle stop and a successful customer handover are therefore distinct events.
A parcel locker provides individually secured compartments accessed through a code, application, card, or other credential. Lockers aggregate many shipments at one location and allow collection outside staffed opening hours. Their capacity depends not only on compartment count but also on compartment sizes and how quickly recipients collect parcels. A locker with many uncollected shipments can become unavailable even when daily delivery volume is moderate. Overflow, oversized parcels, and inaccessible locations require alternative arrangements.
Staffed pickup and drop-off points, often hosted by shops, provide another form of aggregation. They can handle exceptions and assist recipients, but depend on opening hours, storage space, staff training, and commercial compensation. Accessibility varies across both staffed and automated systems. Screen interfaces, compartment heights, walking distance, payment requirements, and the availability of non-smartphone access can determine who can use a service. Out-of-home delivery is thus not an interchangeable substitute for home delivery in every setting.
Product characteristics create additional constraints. Food and certain medical products require a controlled cold chain, with appropriate temperature conditions during storage, loading, transport, and handover. Insulation, refrigeration, monitoring, packaging, and time limits work together; a vehicle’s nominal temperature setting alone does not establish product conditions throughout a shipment. Bulky goods may require two-person handling, lifting equipment, or installation appointments. Dangerous goods, high-value items, and age-restricted products introduce separate documentation, security, or verification requirements that vary by jurisdiction and commodity.
Reverse logistics covers flows back from customers, including returns, reusable packaging, repairs, and product recovery. Combining collections with delivery rounds can use spare capacity, but adds uncertain quantities, inspection tasks, and loading constraints. A returned product may need identification, condition assessment, repacking, and routing to a different facility from the original dispatch location. Customer returns are therefore not simply ordinary deliveries run backward. Their economic significance depends on product category, return policy, resale value, and the time required to restore goods to saleable inventory.
Labor, access, and public governance
Last-mile delivery relies on substantial human work even where sorting and dispatch are highly automated. Drivers and couriers interpret ambiguous addresses, judge safe stopping locations, negotiate building access, carry goods, and manage exceptions. Their local knowledge can compensate for incomplete digital records. Work intensity is affected by parcel size, stairs, weather, parking distance, recipient interactions, and the time allowed for each stop, none of which is fully captured by a count of completed deliveries.
Employment arrangements include direct employment, subcontracting, franchise structures, and platform-mediated work. These models allocate vehicle costs, scheduling authority, commercial risk, and responsibility for insurance or equipment in different ways. Legal classification depends on jurisdiction and the actual working relationship, not only on contract terminology. Comparisons of delivery models must therefore distinguish the price paid by a retailer from the full resources used and from costs borne by individual couriers.
Digital monitoring raises questions about data privacy and the use of performance metrics. Location records, proof-of-delivery photographs, customer messages, and timestamps can support shipment tracing and dispute resolution, but also reveal information about workers and recipients. Photographs may capture entrances, household details, or people who are not parties to the transaction. Access controls, retention periods, and clear operational purposes are relevant to the governance of these records. Cybersecurity also matters because dispatch platforms and access credentials can affect physical delivery operations.
The curb is a particularly important interface between freight and public space. Delivery vehicles compete with buses, taxis, cyclists, private vehicles, pedestrians, and other commercial users for limited stopping space. When legal loading areas are unavailable or poorly located, walking distances and delivery times increase, and vehicles may obstruct traffic or pedestrian routes. Urban planning measures such as loading bays, building service areas, and managed delivery windows can change these conditions, but their effects depend on enforcement, local activity, and the needs of other street users.
Public governance also addresses geographic coverage and service access. Rural routes may involve long distances and low shipment density, while informal settlements or newly developed areas may lack standardized addresses. Postal universal-service obligations can influence coverage, although their precise requirements differ among countries. Private carriers may use service exclusions, surcharges, local partners, or collection points in difficult-to-serve areas. These arrangements show why network efficiency and equitable access are different analytical questions: a commercially efficient service configuration does not necessarily provide the same level of convenience to every location or recipient.
Environmental effects and transport choices
The environmental effects of last-mile delivery depend on the entire delivery arrangement rather than the propulsion system alone. Relevant variables include distance, vehicle load, failed attempts, customer collection journeys, packaging, refrigeration, facility energy, and upstream replenishment. A home delivery route can replace multiple shopping trips, but it may also serve orders that would otherwise have been combined into an existing journey. The appropriate comparison is between complete alternatives for obtaining the same goods.
Electric vehicles eliminate exhaust emissions at the vehicle during operation, but their overall environmental profile also includes electricity generation, vehicle production, and battery manufacture. Stop-start routes can benefit from regenerative braking, while depot-based fleets may have relatively predictable charging opportunities. Practical constraints include payload, range, winter energy use, charging access, and the capacity of the local electrical grid. Auxiliary equipment, including refrigeration and cabin heating, can materially affect energy demand.
Cargo cycles and electrically assisted cargo cycles can be effective where stops are close together and shipments fit their capacity. They occupy less space than vans and may reach destinations through networks unavailable to larger vehicles, subject to local rules. Their limitations include payload, parcel dimensions, weather exposure, terrain, and rider workload. They are often paired with a nearby transfer point because repeatedly returning to a distant depot would erode their advantage. Walking rounds similarly depend on compact territories and suitable methods for moving or replenishing loads.
Environmental assessment commonly distinguishes total emissions from emissions per parcel. A service can reduce emissions per delivery while increasing total emissions if shipment volume grows sufficiently. Likewise, shorter delivery times can reduce consolidation and increase vehicle requirements. An illustrative measure of transport emissions intensity is
where is energy use for an included activity or energy source, is its corresponding emissions factor, and is the number of shipments. The result depends on system boundaries and does not automatically include vehicle manufacture, packaging, or customer travel.
Life-cycle assessment provides a framework for examining impacts across production, operation, and end-of-life stages. Results remain sensitive to assumptions about vehicle lifetime, utilization, electricity supply, and the alternative being displaced. Lockers may reduce van stops yet generate extra car journeys if poorly located. Consolidation may reduce local traffic but require additional handling and feeder transport. Policy evaluation therefore distinguishes greenhouse-gas emissions from local air pollution, noise, road danger, and space use; an intervention can improve one outcome without improving all of them.
Automation, delivery robots, and aerial systems
Automation already supports sorting, labeling, shipment scanning, dispatch, and customer notifications. Automating the final physical movement is more difficult because public streets and building entrances are variable environments. A machine must handle obstacles, unusual surfaces, temporary works, animals, weather, and unpredictable human behavior. Reaching a mapped destination is only part of the task: the system must also provide a safe and usable handover and recover when normal operation is impossible.
Ground delivery robots range from small sidewalk vehicles to road-going automated vehicles. Their sensing systems may combine cameras, lidar, radar, and other sensors, while sensor fusion integrates measurements for localization and obstacle interpretation. Motion planning converts a destination and an environmental model into feasible movement. Small robots typically carry limited payloads and travel slowly, making them more suited to some compact service areas than to dispersed, high-volume routes.
Autonomy is not necessarily the absence of human involvement. Remote staff may monitor fleets, authorize maneuvers, assist with unusual situations, or retrieve immobilized vehicles. The ratio of operators to active robots is therefore an important operational variable, alongside battery charging, maintenance, loading, and cleaning. A demonstration that a robot can complete a delivery does not establish that the service is economical at scale. Reliability across many trips and the frequency of interventions determine how much labor has actually been displaced or reorganized.
Delivery drones can bypass road congestion and physical barriers, making them relevant to selected lightweight, time-sensitive shipments. Their constraints include payload, range, wind, precipitation, noise, takeoff and landing areas, and safe separation from people and other aircraft. Some systems land, while others lower packages or release them at designated locations. Operations beyond the pilot’s visual line of sight generally require an appropriate regulatory basis, and aviation rules differ among jurisdictions. Medical transport to remote locations and routine residential parcel delivery consequently present different operating and regulatory conditions.
Automation also changes the handover problem rather than removing it. A recipient may need to open a robot compartment, reach a designated landing point, or retrieve a package from a secured area. Stairs, locked buildings, and complex indoor routes remain significant barriers. Assessment therefore includes successful deliveries per operating hour, intervention frequency, accessibility, vehicle utilization, and safety performance—not merely maximum speed or nominal range. The deployment of automated systems depends on a defined operating environment in which technical capabilities, service requirements, and legal permissions align.