The Milk Round
A word of praise for something that has almost disappeared, because it was better at all this than we are.
Consider the milk round in its heyday. In Britain in the mid-1970s, the overwhelming majority of household milk — the figure usually quoted is around nine-tenths — arrived on a doorstep, in a glass bottle, from an electric float driven at walking pace by a man who knew everybody. By the 2010s that share had collapsed to low single figures. Milk moved to the supermarket, and the milkman became a nostalgia object, and everyone agreed this was progress.
But look at the milk round as a logistics system, and it is an astonishment. Almost every parameter is at its ideal value.
Demand was predictable: the same households, the same quantities, week after week, adjusted by a note in the empty bottle. No forecasting problem, no long tail of exotic items, no seasonal spike. The route was fixed: no daily reoptimisation, because the stops did not change, and the driver had internalised the sequence so thoroughly that it required no cognition at all. Density was superb: every third or fourth house on a street, not one house in forty. Service time was minimal: no signature, no scan, no doorbell, no interaction — the transaction was completed without either party being present, which is the holy grail of delivery and which we have spent twenty years and a great deal of venture capital trying to reinvent under the name of "unattended delivery". Packaging was reusable and collected on the same trip, so the vehicle ran full in both directions, solving the empty-backhaul problem that plagues every freight network. Payment was periodic, settled weekly at the door or by standing order, so no transaction cost attached to each delivery. The product was standardised: three variants, maybe five with the orange juice and the eggs. And the timing was habitual on both sides — he came at the same hour, you left the bottles out the night before, and neither of you had to coordinate.
I have read a fair amount of last-mile innovation literature, and I want to report that essentially every idea in it is an attempt to recreate one of those properties. Subscription boxes: predictable demand. Fixed delivery days: stable routes. Parcel lockers and safe places: unattended handover. Reusable packaging pilots: closing the loop. Route density initiatives: proximity. The milkman had all of them at once, in 1974, with no computers.
He also had something no algorithm has, which is that he knew when Mrs Attwood at number 12 had not taken her bottles in, and he knocked. There is a genre of newspaper story, evergreen since about 1955, in which a milkman notices three days of untouched milk and saves an elderly person's life. The stories are real. They are also a description of a delivery worker performing a welfare function that no organisation asked for, priced or noticed — the first of many appearances here of unpaid social infrastructure holding the machine together.
So why did it die?
Because the shape of demand changed underneath it. Refrigeration meant you did not need daily milk. Supermarkets sold it as a loss leader at a price the doorstep could not touch. Cars meant a weekly shop was possible. Women's entry into full-time paid work at scale meant the household was less often occupied during the day, and the milk sat warming. And crucially, choice arrived: semi-skimmed, oat, almond, lactose-free, organic, filtered, the two-litre and the six-pint. A round works because everyone wants nearly the same thing. Product proliferation is the enemy of the round.
Which brings me to the central and rather uncomfortable observation of this chapter. E-commerce did not just change how we buy. It destroyed the conditions under which delivery is cheap.
An online retailer offers not three products but nine million. Orders arrive at random times, in random combinations, to random subsets of households, at unpredictable intervals. There is no round. There is a new and different set of a hundred and forty addresses every single day, drawn from a pool of forty thousand, and the software must construct a fresh route each morning from a demand pattern that is essentially noise.
We took a system with predictable demand, stable routes, standard products, reusable containers and habitual timing, and replaced it with one that has none of those things. Then we spent billions on optimisation software to claw back some of what we had thrown away.
I am not making a reactionary point. The new system delivers nine million products instead of five, which is a real and enormous gain, and I am not going to pretend that a world of oat milk options is worse than a world without them. But it is worth being clear-eyed about the trade. What we bought with variety and immediacy was paid for in density. And density is the only thing that makes the last mile affordable.
Sometimes technological modernity destroys an optimisation that an older system had achieved socially, through habit and repetition and everybody wanting the same thing at the same time. The milk round was efficient because it was boring. Boredom, in logistics, is worth a fortune.
