A milking robot costs £120,000. Milk is currently selling for less than it costs many farms to produce it.

Start with the price of the thing, because it decides who can do this at all. A milking robot runs at about £120,000, and that is before the shed. Robots need large, open buildings that let a cow walk in when she chooses, so a farm that buys them is usually rebuilding as well as purchasing.

Now the price of the milk. The levy board warned on 7 August that average farmgate prices have fallen to around 34 pence a litre, which leaves many farmers on market related contracts earning well below what it costs them to produce. Autumn calving is about to start and the choice in front of a lot of dairy businesses is whether to dig in for another bad year.

So the capital spending is happening in the middle of a squeeze rather than on the back of a good year. That is unusual, and it tells you these machines are not being bought out of optimism.

What one farm got for the money is on the record. Ben Hembrow in Somerset put twelve robots into an 850 cow herd in June 2023. On his own account after twelve months, milk output per person rose from about one million litres to about 1.6 million, cows averaged 36 litres a day against 32 before, and his full time staff went from twelve and a half people to nine. He is clear about why: he expected to struggle to find labour, and moved before the problem arrived.

Three and a half posts on one farm. That is the number, and it belongs to that farm.

The national version does not exist, and the reason matters more than the gap. British dairy has been consolidating hard on its own. The levy board's April survey of major milk buyers puts Great Britain at 6,850 dairy producers, down 2.7 per cent in a year and down roughly 15 per cent in five. Average milk volume per farm has climbed to about 1.90 million litres, up 7.6 per cent in a year. The national milking herd stands at 1.59 million head, down 2 per cent.

Fewer farms, bigger farms, more milk from each. That trend was running before robots arrived at any scale and it would be running without them. Any figure showing dairy employment falling nationally would be measuring both things at once, and nothing published separates them.

Which is why this desk is stopping at one farm. The claim is that a named British business rebuilt itself around a machine and its headcount fell by 28 per cent. The claim is not that British dairy is shedding jobs to automation, because the evidence for that does not exist, and asserting it would be the thing this paper was set up to argue against.

The number to watch is the second farm. One employer with a before and after is an anecdote with arithmetic attached. Three or four would be a pattern, and nobody is collecting them.

What would change my reading is a levy body or trade association publishing labour data split by milking system. That is one extra question on a survey somebody already runs. It has not been asked.