The Independent Revenues and Benefits Monday Discussion Group tackled two subjects that promise change for local government: artificial intelligence and council tax. One may help councils manage growing workloads. The other shows how long-standing differences in local funding have become harder to defend. Across both discussions, speakers kept returning to the same practical question: what happens when a promising idea meets the realities of running a service?
Malcolm Gardner opened the AI discussion by asking whether it is reducing work, improving quality and protecting residents, or simply creating another layer of checking. He drew a distinction between AI that helps an officer draft or find information, AI that advises on a case, and AI that makes a decision. The risks, he suggested, change considerably as a council moves from one use to the next.
Tom Clark described a modest but useful role for AI in Liverpool. He uses Copilot to sort through emails after leave, identify what needs attention and find information. The service also uses it to help triage complaints, although he has seen a less helpful side: residents can use generative AI to produce lengthy complaints, which officers must then read and answer. Liverpool is not using it to determine benefit entitlement. Tom felt that controlled, everyday uses can improve productivity and cautioned against leaving colleagues with the impression that every use of AI is inherently alarming.
Ian Savigar reported a similarly early stage in Reading. AI can help improve the clarity of reports and responses, but that may involve writing something first and then asking the tool to improve it. For now, he sees better quality more clearly than a straightforward saving of time. He also found that the detail of the instructions matters: a better prompt usually produces a more useful result.
Rachael Walker was more enthusiastic about the time saved in her own work. She described talking through a problem at length and asking AI to make sense of it, as well as experimenting with an interface to a modelling tool. The output still needs verification, she stressed but used well it can feel like an additional capacity to think through the work. Rachael also raised a question of access. Paid and free versions can offer very different capabilities, both to councils and to the people contacting them.
Robert Fox said Swindon had made limited use of AI in revenues and benefits processing. Its experience with a telephone voice bot had prompted a rollback when callers could not get the answers they needed. Like Tom, he had seen AI make complaint handling harder, including a letter containing around 60 points that required attention even though he considered most of them irrelevant. He was, however, interested in whether AI could help organise the reports and correspondence occupying officers’ time.
For Julie Smethurst, the immediate issue in Tameside is how to introduce the technology properly. Copilot has only recently become available to her team, and she is using it cautiously, checking its output as she learns. Training matters, she argued, because staff need enough knowledge to judge an answer. AI can support expertise; it cannot supply the expertise needed to recognise when an answer is wrong. Julie also wanted software suppliers to build useful capabilities into core systems, reducing the work councils face in putting them into practice.
Penny Mitchell offered a different organisational experience from Salford. There, she said, an engaged corporate IT team is helping services explore AI within agreed controls. She described its use to record and summarise social care conversations, alongside plans for telephony supported by an internally developed AI layer. Her view was that revenues and benefits teams need that corporate involvement if they are to use the technology responsibly and make progress.
Paul Howarth widened the discussion to the economy. The Treasury may look for savings from AI, he suggested, but faster handling of complaints could also encourage more complaints. He argued that the greater opportunity is to use the same resources to do more, improving productivity rather than assuming every gain should become a cut.
Gareth Morgan pressed the group to be more sceptical about what sits behind an AI answer. A useful tool still requires people who understand its purpose, its data and its limits. He warned particularly against treating drafting, analysis and systems that take actions as though they carried the same risk. Gareth cited the possibility of hidden instructions in documents being read by AI as an example of why safeguards matter. Tom responded that council tools can operate within controlled environments, while Gareth’s wider point remained: each proposed use needs to be examined on its own terms.
From AI, the conversation turned to Wandsworth and the prospect of a sharp council tax rise. Malcolm noted the striking gap between its current Band D bill and bills elsewhere and asked whether the dispute reflected a genuine funding shock, a political confrontation, or both.
Paul’s answer was both. Keeping council tax low has been politically attractive in places such as Wandsworth and Westminster. At the same time, he saw the dispute as part of a larger question about property taxation, wealth and how councils are funded.
Rachael argued that the divergence had been building for years through the way tax bases and local funding work. She had little sympathy for the idea that Wandsworth was uniquely hard done by when households in less wealthy areas face much higher bills. Yet she accepted that a large increase would still be a shock to individual residents and said transitional measures deserved consideration. She also pointed to a consequence for council tax reduction: when bills rise substantially, the cost of supporting eligible households can rise with them.
Gareth questioned the fairness of historic funding differences that leave some places better able than others to pay for local services. Kevin Stewart recalled Wandsworth’s earlier zero community charge and observed how far council tax has moved from its original design. He also raised the political and practical questions surrounding a large increase.
Bob Wagstaff described a system whose parts have drifted apart over time, including property bands, discounts and tax bases. Robert Fox brought the discussion back to ability to pay. Drawing on Swindon’s collection experience, he argued that poorer areas can face greater demand for services while finding it harder to collect the tax needed to fund them.
Tom added an operational warning from his experience at Enfield. A higher bill does not automatically produce the income shown in a forecast. Extra correspondence, complaints and staffing can absorb part of the gain. Peter Haywood asked the group to consider London’s high housing costs alongside council tax, particularly for younger workers on modest incomes. He also recalled steep annual tax increases at a council where he had worked, illustrating why limits on rises had an appeal.
On those limits, Rachael saw both protection and constraint. A ceiling can restrain unaffordable increases, but it can also prevent councils from building their tax base over time and encourage them to treat the maximum permitted rise as the normal increase. Malcolm questioned whether a long-standing cap still works as intended when council tax carries a greater share of local funding than it once did.
Ian brought the discussion to a close by asking whether the Wandsworth dispute points towards a wider rethink of council tax and local government finance, potentially alongside local government reorganisation. No one offered an easy replacement. The group’s concern was more immediate: a system can continue to operate for years while its underlying differences grow harder for residents, officers and councils to manage.
The AI discussion reached a similarly practical conclusion. There are useful gains available now, from finding an email to improving a report or helping an officer work through a complex problem. Those gains depend on skilled staff, suitable systems and checks that match the task. Whether the subject is an AI answer or a projected council tax yield, the people responsible still need to understand what produced the number, what it leaves out and what happens when it reaches a resident.
The Recording can be found here