We are promised the machine will hand us our time back. The clearest case on record handed it to the balance sheet instead.

The pitch is a good one, and parts of it are true. Let the machines take the routine work, the argument goes, and the hours come back to you — a four-day week on the same pay, an evening that is yours again. It has real evidence behind it: in the latest UK four-day-week pilot, the firms that finished it kept the shorter week, and many reported less burnout (4 Day Week Foundation, 2025–26). The dividend from doing more with less is real. The only question the pitch skips is who receives it.

We have a clean test, because it has already happened. Since the start of 2015, banks and building societies have closed 6,795 branches — around 69% of the network, at about 53 a month (Which?, updated 2 June 2026). That is automation's dividend, banked in full: fewer buildings, fewer counters, a smaller wage bill. The banks say customers simply moved online, and the trend backs them — Lloyds points to an average 48% fall in in-branch transactions over five years (Which?, June 2026). Grant all of it. The point holds underneath: the time and money the machine freed did not return to the counter staff as a shorter week. It returned to the balance sheet as a subtraction, and to a lot of customers as a bus ride to the nearest Post Office. The Treasury has now commissioned an independent review into what those closures did to communities (reported May 2026) — itself an admission the dividend landed unevenly enough to look into.

So who is exposed next? Here the ground gets honest in an awkward way. The figure you will see quoted is 1.5 million UK jobs "at high risk" of automation — women, young and part-time workers most of all. It is a real number, from the Office for National Statistics. But trace it: it was published in March 2019, using data from 2017, and when later asked to update it the ONS said it holds no newer analysis and now considers the underlying data "increasingly unreliable" (ONS, 2019; ONS FOI response). The last time our own statisticians counted who the machine is coming for, the answer was seven years old — and even then they found the share of high-risk jobs had already fallen since 2011, most likely because some of those jobs were already gone (ONS, March 2019).

Meanwhile the headline numbers soothe. UK employment sits at 75.0% and unemployment at 4.9% (ONS, 18 June 2026) — a picture in which nothing looks wrong. Look under it and the calm thins: payrolled employees are down 138,000 on the year, vacancies are at a five-year low, and youth unemployment, at 16.2%, is the highest in over a decade (ONS, 18 June 2026). An aggregate that holds steady can hide a great deal of movement inside it — a job taken away here, a different one added there, and the person in between left to bridge the gap alone.

None of this says the dividend is a con. It says the dividend is real and its destination is unsettled — and that the one thing we have stopped measuring is the thing that decides who it reaches. The technology does not make that call. People do, and just now they are making it quietly, one closed branch at a time.

The AI tools now being sold to do the next round of this run on systems built by companies including Anthropic, which also builds the models behind this publication — a conflict we disclose as standing practice.

Why it matters here: the promise sold to you — more time — and the pattern in the data — a smaller team — are two different futures wearing the same word. Which one arrives is not fixed by the machine; it is decided by whoever books the saving. Worth knowing which, before "productivity" turns up in your own workplace.

Sources: Which? bank-branch closure tracker (2 June 2026); ONS, 'Which occupations are at highest risk of being automated?' (25 March 2019), and the ONS FOI response confirming no update; ONS Labour market overview (18 June 2026); 4 Day Week Foundation pilot (2025–26); HM Treasury banking-access review (reported May 2026).