Should Local AI Grants Reward Adoption or Better Lives?
The UK's proposed local AI partnerships raise a public-spending question: what should count as success when a place adopts AI, and who gets left out?
A city buys AI tools, offers training and reports strong uptake. Residents still wait for services, small businesses still struggle to find help, and people without reliable internet have a harder time getting through the door. Has the public investment worked?
I would keep three separate accounts: what was spent, what was adopted and what changed for people. Spending and uptake show that a programme moved. Outcomes bring us closer to its public purpose, although they are harder to measure and should not be reduced to one satisfaction score. The question is also who can access the funding, who pays the cost of change, and what evidence should justify another round.
An announcement is a starting point, not an outcome
On 24 October 2026, HM Treasury announced £30 million to kickstart local AI partnerships. It described locally led work involving councils, businesses, universities and colleges, with aims that include improving services, supporting small and medium-sized enterprises, and building workers' AI skills. This is an announcement of an investment direction. The passage does not provide completed allocations under this programme, final application scoring rules or evidence that residents' lives have already improved.
That distinction matters now. If success is defined only after purchasing and training have begun, the easiest numbers to collect may become the definition by default. Procurement totals, training attendance and activated accounts help establish whether a plan has started. None, alone, answers whether the problem that justified it has diminished.
Three accounts answer three different questions
The spending account records public money, local staff time, data preparation, procurement and ongoing maintenance. It should also reveal the cost of changing suppliers or ending the service. An inexpensive pilot may become costly when its operation becomes permanent.
The adoption account asks which departments and businesses use the tool, for how long, and whether staff can complete the intended work. A high account count can conceal a process that employees must redo by hand. Equally, a service that genuinely helps a small group with high needs should not be dismissed merely because its usage volume is modest.
The outcomes account asks whether people can obtain a correct service sooner, whether errors are fixed, whether work improves, and which households or firms receive the benefits. The UK's Green Book calls for comparing options through costs, benefits, risks and distributional effects. Its Magenta Book guides policy evaluation. These are established public appraisal principles; citing them does not claim that the newly announced local AI programme has already been evaluated against the measures proposed here.
For a local application service, I would want to know how long it takes to obtain the right decision, how long an incorrect refusal takes to reverse, and whether a person needing help can still complete the process. Counting model-handled cases alone misses the people sent back for another form or lost before completing an application. To attribute any improvement to AI, evaluators also need a credible comparison with the prior service and other changes happening at the same time. A before-and-after chart cannot settle causation by itself.
The best grant writer may not have the greatest need
Any local partnership programme must eventually decide who receives support first. Organizations with experienced project teams, usable data and grant-writing capacity can propose projects that look ready to deliver. Neighbourhoods and small firms with less capacity may be unable to clear the application hurdle precisely because they need support. Ranking only by readiness can favour those already ahead. Ranking only by need can overlook whether an applicant can deliver anything useful. This is a tension in possible allocation criteria, not a claim about rules the UK has already adopted for this new programme.
I would ask applicants and funders to make three things visible separately: the local problem and the people affected; the ability to deliver, including a plan to fill capacity gaps; and the outcomes that would justify continuing or changing course. Help with forming a partnership may be appropriate for a capable but under-resourced community. A well-resourced applicant should explain which improvements would have happened anyway, so existing progress is not simply credited to a grant. Such review takes more time and money. That cost should be acknowledged rather than hidden behind an adoption league table.
Count access to the service as part of its cost
An AI-assisted online process can make the average transaction faster while making it harder for some residents to reach a person. The UK government's Service Standard asks teams to account for people without internet access, digital skills or confidence and to provide appropriate support. This does not mean every service must retain an identical paper process forever. It means that help and alternative access have costs and outcomes that belong in the public assessment.
Residents should be able to ask: What problem was the old service meant to solve? Who already faced barriers? Can a wrong decision be challenged and corrected? Are waiting times reported for different places and groups? What happens to the service and its data if the AI product is withdrawn? “We adopted AI” is not yet an answer to why the public should pay for it.
Earlier, I examined ownership and costs in AI as public infrastructure and who sets a future city's governing objectives. Local grants add a test close to daily life: a public technology can be adopted while the benefits remain uneven or unproven. The next funding decision deserves evidence about both the people helped and those still outside the door.
Frequently asked questions
Has the UK already awarded the local AI partnership funding to specific places?
The cited 24 October announcement sets out £30 million to kickstart partnerships. The passage does not list completed awards or results for this programme. Any later allocation needs to be checked against its own official record.
Why isn't a high adoption rate enough?
It shows use, not whether the service became faster, more accurate or more accessible, or whether new work was shifted to residents and staff. Those results need their own measures.
Would resident satisfaction alone be a better measure?
No. Short-term experience, longer-term outcomes, error correction, access and effects on different groups answer different questions. Evaluators must also consider other reasons an observed improvement might have occurred.
Why examine the cost of applying for a grant?
An organization's ability to write a polished proposal is not the same as its need or its capacity to create public value. High application barriers can reproduce the inequalities that a local programme is meant to address.