Nobody Voted for the Algorithm Making the Cuts

Liverpool City Council has worked out that deploying artificial intelligence across its children's social care service will save the authority roughly £150,000 by the end of the 2027-28 financial year. The council's overall budget gap for the same period is upwards of £29m. Put the two numbers next to each other and you have, in miniature, the entire story of how English local government proposes to survive the next three years: with a rounding error, dressed as a strategy.

The scale of the problem the rounding error is meant to address is not in dispute. Councils across the UK are cutting £3bn this financial year alone. Despite a round of funding reforms that ministers described, with the usual confidence, as addressing the sector's structural weaknesses, a £4bn gap remains. Windsor and Maidenhead faces a shortfall of £66.9m in 2027/28 — £419.71 for every resident. Barnet needs close to £80m in government support just to keep functioning as what its spokesperson, with the particular stoicism of a council press office, called "a tightly run ship" caught in "the perfect storm." Scottish authorities are staring at a combined £562m blackhole. Into this gap, councils are pouring chatbots.

The catalogue is not small. Birmingham uses AI-based tools to identify cost savings, though the council has not said how much it has actually saved, which is its own small data point. Oxfordshire uses it to plan school transport routes for children with additional needs, to transcribe social workers' notes during care visits, and to staff a virtual assistant that can, its council leader says, detect when a caller is angry and adjust its tone accordingly. West Berkshire fits cameras to council vehicles to photograph the road surface every ten metres and flag potholes before anyone reports them, which earned it a green rating from the Department for Transport. None of this is stupid. A machine that spots a crack in the tarmac before a resident's suspension does is a machine doing something useful, and a system that stops a social worker typing up the same conversation twice frees up an hour that might otherwise go on paperwork. The case for efficiency here is not fantasy — it is the honest argument, made by people who are not lying to you, that a service starved of money should use every tool available to stretch what remains.

The trouble starts with what "efficiency" is actually being asked to do. Jonathan Carr-West of the Local Government Information Unit put it more plainly than most officials manage: "there is no silver bullet… we have seen some savings being made from AI, but they are, to be frank, pretty marginal compared to the scale of the problem." That is the head of a sector body telling you, on the record, that the technology being rolled out as the answer to a £4bn hole is not the answer to a £4bn hole. It is a way of making the hole slightly smaller while everybody agrees not to mention that it is still there. One trade publication cataloguing these projects — ten different ways councils are using AI, filed under the header of transformation — noted at the bottom, in the interests of disclosure, that the article itself had been written with the help of ChatGPT. Nobody involved seems to have found that funny, which is somehow the most damning detail in the whole file.

The deeper problem is not the size of the savings. It is what the savings are made from. Andrew Reece of the British Association of Social Workers has flagged the use of AI to draft care assessments — the documents that determine what support a vulnerable adult or child is entitled to. "This is stepping into the area of making decisions," he said, "and we don't think that's appropriate." He is right, and the distinction he is drawing is the one this whole rollout keeps blurring: a tool that transcribes a conversation is a stenographer, but a tool that drafts the assessment is deciding, in the first instance, who gets help and how much. Nobody stood for election promising that. No councillor voted for it in a chamber where a resident could watch and object. A procurement officer signed off a contract with a vendor whose product was demonstrated at a conference, and the decision about who receives care in 2027 travelled from that meeting room into a family's life without ever passing through anything a citizen could observe, let alone unseat.

This is the pattern the column keeps finding wherever it looks. A decision too politically expensive to make in public gets relocated somewhere no one can vote on it. Rate rises are unpopular, so the gap is closed instead by a software licence, and the licence's failure to close the gap becomes, conveniently, nobody's fault — not the council's, which can point to the innovation; not the vendor's, which sold a pilot rather than a cure; not the minister's, who announced a funding settlement two years ago and has moved on to the next brief. Attlee's government, with the country broke and the shops empty, built the health service and 200,000 houses a year because a minister was prepared to be the person who decided and the person who answered for it. Liverpool's council has automated £150,000 of a £29m problem and called it modernisation.

The technology is not the scandal. What it is being asked to disguise is: that somewhere between the funding settlement and the care assessment, a decision worth making in public is being made instead by a system nobody elected, procured by an officer nobody can sack, to manage a shortfall a minister has already declared solved.