The easy savings are gone: why AI pilots can't close a structural deficit

ICS AI

11 minutes ago

6 min read

By Martin Neale, CEO and Founder, ICS.AI

Local government is past the point of innovation conversations. This is a budget conversation.

I opened our recent SMART: AI Assessment webinar with a handful of numbers that most council leaders now carry around with them. In February, the government agreed around £1.5bn of exceptional financial support for a record 35 councils for 2026-27, up from 29 the year before. Since 2020-21, the total value of capitalisation directions granted to the sector has reached £6.9bn. The LGA estimates councils face a £7bn funding gap by 2028-29 just to keep services at today's level, and half of councils' finance chiefs (51%) lack confidence they'll have enough funding to meet their statutory duties by then.

Those numbers change what "AI strategy" has to mean. A saving identified for 2028 does not help a council that runs out of road in 2027. If the savings don't land in-year, they don't save the services.

The easy savings are gone

For more than a decade, councils have pulled every lever available to them: shared services, headcount freezes, council tax rises to the maximum allowed. Each one has done its job, and each one is now largely exhausted.

The next savings won't come from cutting budgets. They will come from changing how the work gets done - the hours spent answering the same enquiries, re-keying the same data, chasing the same documents and writing the same letters.

That is why AI matters now. It is the one lever left that changes the economics of the work rather than simply squeezing the budget around it. It can absorb demand that would otherwise land on stretched teams, and it can give staff time back to spend on the residents who need them most. McKinsey estimates generative AI could raise customer care productivity by a value equal to 30-45% of current function costs, and the UK Government's own trials found staff saving an average or two hours a week.

The 56% paradox: everyone is exploring, almost nobody is banking

Here is the uncomfortable part. 95% of councils say they are exploring or using AI. Yet 56% don't know whether it has saved them any money, and only 22% report realised cost savings. Councils are far more likely to report productivity benefits (35%) or service efficiencies (32%) than savings that reach the bottom line.

The difference isn't the technology - it's the operating model. The value is real, but it stays trapped upstream as "productivity" and never turns into money a Section 151 officer can bank.

Why pilots are structurally incapable of producing organisational value

Most councils have arrived at AI the same way: a transcription pilot in one team, a few hundred Copilot licences in another, an AI Assistant on the website, a document tool in revenues. Each one made sense on its own. Together, they create five problems:

Fragmented investment - every pilot comes with a different vendor, different guardrails and a separate bill.
Artificial boundaries - each tool stops at the edge of the service that bought it, even when the work flows across several.
Compounding governance overhead - every new tool needs its own DPIA, security review and guardrails, so the assurance burden grows with every project.
Optional adoption - when a tool is a pilot, using it is optional - and optional rarely becomes a standard practice.
Disconnected systems - point solutions sit beside the council's core systems rather than reaching into them, so the work still gets re-keyed.

The council is left carrying the job of turning each pilot into savings, and the pilots never compound. Duplicated governance and architecture also push up the cost of running AI, because every supplier's usage is metered separately.

There is a deeper point here, too. Until recently, cost and procurement did some of the governance work for councils; a new system took months to buy and a business case to justify, so councils couldn’t end up with too many. AI has removed that limit. It has never been easier to approve a small, isolated AI project - and every isolated project approved today is a silo the council will pay to unpick later.

What works instead: one programme, not a dozen pilots

Multi-million-pound recurring savings need a particular shape. The value equation is simple - coverage, multiplied by adoption, multiplied by productivity lift and quality, minus the cost to run, equals banked value. Pilots can't survive that arithmetic, because each one has narrow coverage and optional adoption.

A programme that can survive it looks different. It is led from the top and owned by leadership; it runs across the whole organisation, and it is radically transparent about what is working. It operates at scale with hundreds of use cases, thousands of staff and hundreds of thousands of residents.

That also changes the technology question. Instead of a dozen platforms, a single unified platform lets a council deploy new use cases through configuration in days rather than procuring each one, apply one audited governance policy that every use case inherits, manage everything from one operating view, and route work across different AI models to keep running costs under control. Nothing has to be ripped out - AI reaches into the systems where the work already lives, from housing and revenue systems to the spreadsheets services rely on.

Governance done once is an accelerator

Governance is often seen as the brake on AI. Done once, it becomes the accelerator. Rather than repeating a DPIA, a security review and a set of guardrails for every new tool, a council can move to one platform governance framework that every use case inherits.

That also makes the provability test achievable. For every use case, can you show what the AI did, why it did it, and what information it was grounded on? With unified governance, the answer is yes.

A continuum, not a destination

New generations of AI capability now arrive every 18 months to two years. That means a transformation programme is never finished - it is, in effect, a photograph of a moving train.

The good news is that generations co-exist. The AI front door a council builds today isn't thrown away when the next model arrives; it deepens. Adoption is additive and every stage produces standalone value. In our experience, around 60% of the benefits in a typical 24-month programme land in the first 12 months.

Counting benefits honestly

Done as a joined-up programme, the value for a typical unitary council is in the region of £8m-£19m a year, on lower-bound assumptions. But the way benefits are counted matters as much as the headline. We separate cash-releasing benefits (such as licenses that can be exited), capacity (hours returned to the front line) and retained income (such as funds recovered). And two rules apply throughout: every benefit is measured against a baseline, and nothing is claimed until it can be evidenced.

Derby City Council is where we have seen this work at scale. After proving AI with residents through its AI front door and using the SMART: AI Assessment to build a council-wide plan, Derby moved into a wider AI transformation programme, governed by an AI Ethics Board with an independent chair and a DPIA for every use case. Across that wider AI transformation programme, tracked against a baseline from the start, Derby has identified £12M+ in savings.

Where to start

You cannot plan a journey without knowing where you’re starting from. Most of the councils that can't say whether AI has saved them money started without a baseline.

That is why we see the SMART: AI Assessment as the first act of transformation, not homework before it. In 8-16 weeks, it shows where your council stands across five areas of readiness, measures the opportunity in your highest-volume services using your own data, and gives your SLT an evidenced, costed decision about how far and how fast to go. Arithmetic, not argument.

Discuss an AI Assessment for your council, or watch the full webinar on demand.

Blog series:

Read the previous blog in the series – “Where should your council start with AI? Six takeaways from our SMART: AI Assessment webinar”.

Look out for the next blog: "If the tools already work, why do we need an assessment?" What councils tell us when they come to us, by Aneeq Star (launching 20 October)