Inside Ocado's Luton warehouse a robot arm picks groceries at a rate no human matches, because a tote is not a hedgerow
On the edge of Luton sits the most productive grocery warehouse Ocado has ever run. In the half year to 31 May 2026 it picked an average of 315 units for every hour a person worked in it, against 267 across the company's six British robotic warehouses, up from 240 a year earlier. Those figures come from Ocado Group's interim results, published on 16 July, and they are worth a Saturday of your attention, because they are a working measurement of the thing this paper exists to watch: a machine absorbing human work, in Britain, on a published schedule.
Units per hour, UPH, is the ratio this industry runs on: items picked per hour a person works, measured wall to wall across the building. Read it upside down and it becomes the amount of human time inside your shopping. At 267, a fifty-item weekly shop contains about eleven minutes of human warehouse work. Every point the number rises, the human minutes in your basket fall. That is the substitution, not forecast but measured, and Ocado publishes it every half year because a listed company must account for its costs.
Here is the machine behind the number, simplifying slightly. An Ocado warehouse is a grid: a vast lattice of stacked plastic totes, each holding one kind of item, with wheeled robots running on rails across the top. When your order lands, the system computes which totes must surface and sends robots to dig them out and carry them to pick stations. For twenty years a person stood at every station, lifting your shopping from tote to bag. The change is that the arm has arrived on the grid itself. On-Grid Robotic Pick, the company's name for it, is a camera-guided arm that Ocado says can pick more than 70 per cent of an extensive grocery range, and its sibling, Auto Frame Load, hoists the finished orders onto delivery frames, which the company has described as the most physically demanding job in the building. Both descriptions are Ocado's own, made while selling the machinery to other grocers.
Now set this beside the piece we published on 28 July about strawberries. The hard part of picking fruit is not seeing the ripe one; it is the two seconds after, the grasp of something soft, irregular and half-hidden, on uneven ground, in weather. Field robots are still losing to those two seconds. The warehouse solved the same problem by refusing to have it. Nothing in an Ocado shed is occluded, wet or growing. Every item arrives in a known tote, in controlled light, on a flat floor, with a barcode. And where an item still defeats the arm, the software simply routes that tote to a human station instead: the system decides, pick by pick, which species does the work. The field robot must cope with the world as it is. The warehouse was rebuilt so the machine never has to.
Hold both truths about what this is. It is efficient: the same interim results show the six British sites running at 103 per cent of their original design capacity, which means growth is coming out of buildings that already exist. And it is labour leaving a cost line: warehouse costs rose 1.7 per cent in the half against 12.8 per cent more orders, and the document says those costs primarily comprise labour, property and consumables. A cost line is not a headcount, these filings publish no audited headcount for the warehouse workforce, and we will not borrow one from anywhere less careful. What the record does show is which site closed: Hatfield, the oldest and least automated, shut after a 2023 network review that carried £7.5m of closure costs including redundancy, per the audited annual report signed last September.
What the grid cannot do is leave the building. Every tote's journey ends in a van with a person at the wheel, and the same set of results prices that leg too; James Vahid takes it from here. The forward question for this desk is narrow and mechanical: each additional slice of the range the arm learns is a set of totes that stops visiting a human station, and the published UPH series is where you will see it land, twice a year, on schedule.
The clearest live instance of machines absorbing human work in Britain: the mechanism explained, the measurement named, and the reason the warehouse fell to automation before the field did.