The Last Metre

Let us grant the optimists everything.

Suppose the warehouse is fully automated, and picks any item from ten million in ninety seconds without a human hand. Suppose demand is predicted well enough that the item was already in a building four miles from you before you ordered it. Suppose the trunk network runs on autonomous vehicles overnight at costs that make today’s look extravagant. Suppose routing is solved to within a percentage point of the theoretical optimum, that the sortation is perfect, that nothing is ever mislaid, that the vehicle arrives at your door at the exact minute predicted three days earlier.

You still have a box on the pavement and a person on the fourth floor.

That gap — the doorstep, the stairwell, the hallway, the hand — is the residue that no amount of upstream perfection touches. It is where every fantasy in this industry eventually arrives and stops, and it is smaller and more stubborn than the mile that precedes it. Call it the last metre.

───

The proposed solutions cluster into three approaches, and it is worth seeing them as a set, because each removes one friction and creates a question in its place.

Take the person out of the transaction: unattended delivery. The safe place, the locker, the parcel box bolted to the wall, the drop box in the porch. These work by lowering the security requirement until presence is unnecessary. The question they create is: who bears the loss, and how much loss is acceptable? We have already established that the answer is the customer, quietly, via a dropdown.

Take the door out of the transaction: access technology. The smart lock, the one-time code, the app-controlled garage, the building access system that admits a verified courier for ninety seconds. These work by making the barrier permeable to authorised strangers. The question they create is about trust and about the security of an access-granting system that is now a target.

Take the destination out of the transaction: aggregation. The locker bank, the parcel room, the collection point, the concierge. These work by moving the handover to a place designed for it. The question they create is who pays for the place and who is excluded by it.

Every serious proposal is a variant of one of these three, and every one of them has been tried.

from The Last Mile: How the Most Expensive, Difficult and Human Part of Delivery Became the Infrastructure of Everyday Life (2026)


GPT-5.6 Sol:

The passage’s central move is strong: it changes the unit of analysis from distance to handover. “Last mile” suggests a routing problem; “last metre” suggests an interface problem. Once the parcel and the recipient are physically close, the remaining costs arise less from transport than from the awkwardness of transferring custody across the boundary between a logistics system and private life.

That distinction matters because upstream logistics is unusually susceptible to industrial optimisation. Warehouses can be standardised; trunk routes can be consolidated; vehicles can be highly utilised; sortation can be mechanised. The final handover has almost the opposite characteristics: destinations are heterogeneous, recipients behave unpredictably, buildings have different access rules, parcels have different security requirements, and each successful transaction often has to be completed individually. The fourth-floor flat is therefore not merely another 20 metres of travel. Economically, it may be more troublesome than the preceding four miles.

There is a useful general principle underneath the passage: optimisation tends to push cost toward the least standardisable boundary of a system. Automating everything before the doorstep does not abolish the doorstep. It can actually make the doorstep loom larger, because once other costs have fallen it becomes a greater proportion of the total. This is analogous to Amdahl’s law in computing: once 99% of a process has been accelerated, the stubborn 1% determines how much further improvement is possible. Logistics eventually encounters a social and architectural boundary rather than a computational one.

The three-part taxonomy is also illuminating because all three strategies alter one component of the handover.

  • Unattended delivery removes synchronisation: courier and customer no longer need to be present simultaneously.

  • Access technology removes the physical boundary: the courier is temporarily permitted to cross it.

  • Aggregation removes individualisation: many separate destinations are replaced by one logistics-friendly destination.

Put differently, the industry has three variables available to manipulate: time, access and place. That is probably the deepest conceptual point in the extract. If the customer need not be there, change time; if the courier cannot reach the customer, change access; if reaching every customer is inefficient, change place.

But the taxonomy is a little too neat. Its categories are not mutually exclusive. A parcel locker is both unattended delivery and aggregation. A staffed concierge is aggregation but may also solve access. A smart parcel room combines all three. In practice the interesting systems are often hybrids precisely because no single axis solves the whole problem. So “every serious proposal is a variant of one of these three” works rhetorically, but as an analytical claim I would weaken it to “almost every proposal manipulates one or more of these three variables.”

There is another category hiding in the argument: change the agent performing the final handover. Sidewalk robots, drones, autonomous delivery carts, building robots and even customer collection from an autonomous vehicle do not obviously fit the taxonomy unless they are forced into “unattended delivery.” Yet their distinctive proposition is substitution of capital for the courier. The passage may deliberately exclude this because the robot eventually faces the same locked door, stairs, lift and recipient. That would actually strengthen the argument: autonomous delivery does not eliminate the last metre so much as move the last-metre problem from a human courier to a machine that is usually worse at irregular environments.

That is where the fourth-floor example earns its keep. A wheeled robot is impressive on a pavement and useless before many staircases. A drone can cross kilometres of difficult terrain and then encounter the rather embarrassing question of where, exactly, it is supposed to put the parcel. The closer delivery gets to the human recipient, the more the environment is organised around human capabilities: opening doors, reading improvised signs, negotiating entryphones, deciding whether something is safe, noticing that “leave behind bins” is inappropriate because it is raining. Humans are expensive partly because they are extraordinarily general-purpose actuators.

The discussion of risk could also be sharpened. “Who bears the loss?” is correct, but unattended delivery does not simply lower the security requirement. More precisely, it substitutes probabilistic security for verified transfer of custody. Traditional handover provides evidence that somebody accepted the parcel. Leaving it behind a flowerpot removes that evidence and replaces it with an expected-loss calculation. If failed or stolen deliveries cost less than arranging attended delivery, the system may rationally tolerate some theft. The startling thing is that the “solution” to a logistics problem may simply be to price failure into the system.

Likewise, access technology transforms rather than solves risk. A one-time entry credential is attractive because it narrows authorisation: this courier, this building, this interval. But once access becomes programmable, physical security becomes partly an information-security problem. Credential theft, impersonation, compromised courier accounts, faulty revocation and platform outages become logistics risks. The lock ceases to be merely a lock and becomes an endpoint in a distributed identity system. That can be much more convenient, but its failures can scale in ways a lost brass key cannot.

Aggregation is probably the economically strongest of the three because it attacks the fundamental diseconomy of residential delivery: the requirement to make an enormous number of low-volume stops. A locker bank converts many one-parcel destinations into a single multi-parcel stop. The courier’s productivity rises enormously. But the passage correctly notices that this does not make the last metre disappear. It assigns it to the customer. The consumer now walks from flat to locker. That may be perfectly efficient socially, but it reveals something important about claims that a system has “eliminated” a cost: often the cost has merely moved outside the firm's accounting boundary.

This is the recurring theme connecting all three approaches. None destroys the final friction. Each reallocates something:

  • Unattended delivery — gains: no need to synchronise with recipient; transfers: theft/damage risk

  • Access technology — gains: ability to cross the boundary; transfers: security and trust risk

  • Aggregation — gains: fewer stops and greater drop density; transfers: travel/time burden to customer

That framing is stronger than saying merely that every solution “creates a question.” The questions are not incidental side effects. They are the economic mechanism by which the efficiency is obtained. Someone accepts more risk, grants more access, travels farther, provides floor space, operates the concierge desk, installs the locker, or tolerates reduced convenience. There is no free removal of the handover problem.

One claim in the passage is weaker: “We have already established that the answer is the customer, quietly, via a dropdown.” That may work in the larger book because earlier chapters presumably establish it, but standing alone it is overly categorical. Retailers, carriers, insurers, landlords and customers can all absorb losses depending on consumer law, commercial policy and contract. Even when the customer clicks “leave in safe place,” the legal allocation of risk does not necessarily follow the interface design. The more defensible broader claim is that digital interfaces are increasingly used to solicit consent for degraded custody guarantees, sometimes without making the economic significance of that consent particularly salient.

There is also a social dimension that the final sentence about exclusion only gestures toward. Optimising the last metre tends to privilege people whose lives and buildings are compatible with the optimisation. A locker is excellent if it is nearby and you are mobile; much less so if you are disabled, caring for children, elderly, or ordering something heavy. Smart access works better in digitally managed apartment buildings than in fragmented housing stock. Safe-place delivery favours detached houses with private space over flats opening directly onto streets or communal corridors. Hence logistics infrastructure can encode inequalities in urban form: the nominally universal service may impose very different burdens depending on where and how someone lives.

The most interesting conclusion, then, is slightly more radical than “the last metre is difficult.” It is that the last metre is where logistics becomes politics, property and social organisation. Before that point, the parcel moves through infrastructure largely controlled by logistics firms. At the end it encounters domains they do not control: a person's home, a landlord's building, a public pavement, a recipient's time, someone's willingness to accept risk. Optimisation suddenly requires rights rather than merely algorithms: the right to enter, the right to leave property, the right to occupy space with lockers, the right to transfer liability.

That also explains why the last metre is likely to remain human for longer than futurist accounts imply. The difficult thing is not locomotion. It is authorised, context-sensitive completion. A delivery driver does not merely transport an object. At the end of the route, the driver interprets ambiguous instructions, negotiates barriers, judges security, communicates with recipients and improvises around an environment that nobody designing the logistics system fully controls.

So the passage is persuasive in its main thesis. Its strongest insight is not that the final few metres cost money, but that they belong to a fundamentally different class of problem from everything upstream. The industry can make the network extraordinarily efficient and still encounter a boundary that is architectural, contractual and human. The better formulation might therefore be:

The last mile is a transportation problem. The last metre is a custody problem.

Once stated that way, it becomes clearer why better routing alone cannot solve it.

Previous
Previous

Reels

Next
Next

On Testimony