Sky News just revealed something that should make everyone working in public service transformation frown. Since 2018, the UK government has spent £3.35bn on AI contracts. The Met Office got a billion pounds for weather forecasting. Transport for London got £259m. Meanwhile, the Treasury and Department for Work and Pensions, the departments that actually deal with millions of vulnerable people every day, are in the bottom three for AI spending.

Let that sink in. We’re throwing money at predicting rain whilst people on Universal Credit still can’t navigate the benefits system without having a breakdown. We’re investing hundreds of millions in transport optimisation whilst HMRC’s systems remain so hostile that people would rather pay penalties than try to comply.

I don’t have a problem with billion-pound contracts. I have worked on them myself. And I understand that various AI tools in everyday use throughout the civil service are not included in these figures. But my impression is not that the government’s big announcements on “AI revolutionising public services” are made in parallel to ongoing investment in transforming other services through human skill and expertise from dedicated digital transformation teams like the ones I have worked in for many years.

Are we even trying to solve the problems that matter? 

I’ve been banging on about vulnerability and AI all year, but these numbers reveal something darker than just poor implementation. They show we’re not even trying to solve the problems that matter. The spending patterns expose our actual priorities, and they’re nothing like the rhetoric about digital transformation serving citizens better.

The Met Office’s billion-pound contract with Microsoft is for building “the world’s most powerful weather and climate forecasting supercomputer”. Impressive. Genuinely. Weather forecasting matters. But here’s what bothers me: getting weather predictions wrong disappoints people. Getting benefits assessments wrong destroys them. Yet guess which one gets the billion-pound investment?

Transport for London’s £259m went to Init, a German public transport technology company. Again, useful stuff. Better transport systems matter. But TfL users are generally people who can afford to travel. The people navigating DWP systems are often choosing between heating and eating. Where’s their quarter-billion investment in making that journey less traumatic? Universal Credit is still a bureaucratic nightmare that is definitely not trauma-informed after more than a decade of hard work.

The data, analysed by consultancy Tussell for Sky News, shows we’ve been steadily increasing AI contracts year on year. The trajectory is clear: we’re committed to AI transformation. But transformation of what, exactly? Because it’s clearly not the services that vulnerable people depend on for survival.

We’re still treating human complexity as an inconvenience

I worked on the Cabinet Office’s Spotlight counter-fraud tool back in December 2020. Even then, we knew that fraud detection was just one piece of the puzzle. The bigger challenge was understanding why good people give bad information, factors such as trauma, cognitive overload, shame or fear. Four years later, with billions spent, we’re still building systems that treat human complexity as an inconvenience to be automated away rather than the core challenge to be solved.

The bottom three departments for AI spending tell their own story. These are the departments dealing with the most complex human needs, the highest levels of vulnerability, the greatest potential for digital systems to either transform or traumatise lives. And they’re getting the least investment. It’s not accidental. It’s revelatory.

Think about who interacts with HMRC and DWP. People in debt. People with disabilities. People who’ve lost jobs, lost partners, lost hope. People whose lives don’t fit in dropdown menus. People who need help, not algorithms. These are the departments where trauma-informed design isn’t a nice-to-have; it’s essential. Yet they’re at the bottom of the spending league table.

The pattern becomes even more stark when you consider what these departments actually need. DWP doesn’t need a supercomputer. It needs systems that can recognise when someone’s inconsistent answers indicate a crisis rather than fraud. HMRC doesn’t need machine learning models. It needs interfaces that don’t require an accountancy degree to report basic income changes. These aren’t technically complex problems. They’re human problems that require investment in understanding, not just processing.

We’re automating the easy stuff while leaving the hard

But human problems are messy. They don’t have clear ROI calculations. You can’t easily measure the value of preventing someone’s mental health crisis triggered by your benefits form. You can’t put a number on the cost of shame when your system assumes everyone has a permanent address. So we invest in the clean problems instead. Weather has patterns. Transport has routes. These are problems that look good in procurement documents.

I spent years in journalism before moving into government digital services. One thing journalism taught me: follow the money, and you’ll find the truth. The truth here is that we’re building AI for problems that are intellectually interesting rather than socially important. We’re automating the easy stuff whilst leaving the hard, human stuff to rot.

The procurement process itself is part of the problem. I’ve sat in enough government meeting rooms to know how these decisions get made. Big contracts need big business cases. You need metrics, KPIs and clear deliverables. “Reduce weather prediction error by 15%” is a metric. “Reduce human suffering in the benefits system” isn’t. So, guess which one gets funded?

There’s also the vendor problem. Microsoft can deliver a weather supercomputer. They’ve got the infrastructure, the expertise, the shiny demonstrations that make ministers feel like they’re leading innovation. But who’s the Microsoft of vulnerability? Who’s the billion-pound vendor for trauma-informed design? They don’t exist, because we haven’t created a market for them.

Government procurement favours the familiar, the safe

The Sky News investigation also revealed something telling: Alphabet, the company behind Google, has just two contracts with government, worth £2.5m. One of the world’s most sophisticated AI companies, and we’re barely engaging with them. Why? Because government procurement favours the familiar, the safe, the companies who know how to navigate Whitehall. Innovation takes a back seat to institutional inertia.

This isn’t just about money, though the money matters. It’s about what the spending patterns reveal about our values. Every pound spent on weather forecasting that could have gone to fixing Universal Credit’s hostile architecture is a choice. Every million allocated to transport optimisation instead of making HMRC navigable for people with cognitive disabilities is a priority revealed.

The timing of this data is particularly galling. We’re in the middle of a cost-of-living crisis. Benefit claimants are being sanctioned at record rates. The tax system is becoming increasingly complex as more people join the gig economy. These are the burning platforms that need AI investment. Instead, we’re building weather supercomputers.

I’m not naive about government spending. I know how budgets work, how departments compete, how ministers want announceable wins. A billion-pound supercomputer sounds impressive at a conference. “We made the benefits system less traumatic” doesn’t make headlines. But that’s exactly the problem. We’re optimising for announcements rather than outcomes.

Develop AI that reduces complexity

What would meaningful AI investment look like for vulnerable citizens? Start with DWP. Invest in systems that can detect distress, not just fraud. Build interfaces that adapt to cognitive load, that recognise when someone needs human help, that don’t punish people for not fitting standard patterns. Fund research into how trauma affects form completion. Create AI that serves as an advocate, not an assessor.

For HMRC, invest in simplification, not just digitisation. Build systems that can explain tax in plain English, that can guide people through changes without assuming prior knowledge, that recognise when someone’s making an honest mistake rather than evading. Develop AI that reduces complexity rather than just processing it faster.

These aren’t technically harder problems than weather forecasting. They’re just harder to specify in procurement documents. They require admitting that our current systems harm people. They need vendors who understand vulnerability, not just vectors. They demand success metrics that value dignity, not just efficiency.

The revolution in public services won’t come from tech giants

The £3.35bn spent on AI since 2018 represents choices. Choices about whose problems matter. Choices about what kind of transformation we want. Choices about whether we see vulnerable citizens as edge cases or core users. The spending data suggest we’ve made our choice and it’s not the one our rhetoric suggests.

Next time a minister announces another AI initiative, ask them where the money’s going. If it’s another supercomputer, another optimisation engine, another pattern-recognition system for a clean problem, you’ll know their priorities. The vulnerable citizens navigating hostile systems? They’ll still be waiting for their turn, which these spending patterns suggest might never come.

The revolution in public services won’t come from billion-pound contracts with tech giants. I can tell you from first-hand experience that most projects I’ve worked on that ran into trouble, did so because the tech side let them down (Universal Credit, CAP Rural Payments, etc). The real revolution is going to come from recognising that the hardest problems, the human ones, are the ones most worth solving. Until our spending reflects that truth, we’re just playing with expensive toys whilst real people suffer through systems we refuse to fix.