Big Automated Government in a Time of Austerity
This set of three blog posts started life as some of the notes and “thinking aloud on Bluesky” that I did in preparation for giving oral evidence at the Science, Innovation and Technology Select Committee’s inquiry into the digital centre of government.
The Blueprint for modern digital government sets out an abundant approach to technology in which joined-up, personalised services combine AI with private-sector design principles. But can AI be trusted to deliver this new version of Big Government against a backdrop of domestic austerity and geopolitical turmoil?
“A blueprint for modern digital government” is a strategy of centralised automation that uses technology to grow the government’s footprint; the context of its delivery is one in which AI and automation becomes a more significant force in government while other, more traditional state capabilities, are diminished. But as the Trumpish uncertainty of the next four years emerges, the bets made here on a bigger, more technocratic approach to technology feel misplaced; rather than doubling down on big systems, 2025 feels like a good time to stay responsive and open to alternative paths to innovation.
Well before this week’s Spring Statement it became clear that AI would be central to delivering government's efficiency plans, and the Blueprint comes closest to explaining how those changes will actually look.
The mood of the Blueprint suggests a shift to a bigger, more interventionist style of government that delivers “proactive … services [that will] come to users”, promising “Government will move at the same pace that people’s lives do.” Efficiencies will be gained by “harnessing AI and automation technologies” and via One Login programme — but this invest-to-save methodology is not accompanied by clear timelines or much clarity on expected results.
Overthinking Government
One of the biggest risks of the Blueprint is that it overthinks the role of government in people's lives at a time when non-digital capabilities are shrinking. It describes services that anticipate our needs without recognising that some complex anticipatory services need just-in-time logistics and it also risks overstating the role that government has in people’s day-to-day lives.
Peter Kyle has previously suggested that “the average citizen” spends “1.5 working weeks every single year” dealing with government admin, which he says “gives people less time to do the things they like and to be with the people they love”.
1.5 working weeks a year equates to perhaps 60 hours overall, or 69 minutes a week, and it is almost impossible to average this out in a meaningful way. For instance, the never-ending bureaucratic load of a parent or carer advocating for a child’s special educational needs is not comparable to the one-off annual flurry of a freelancer who hasn’t kept very good records for their tax return. The administrative burden of managing a life with multiple long-term conditions will not additionally be experienced by everyone who also needs to fill out a pension credit form.
Added to this, some people find administration soothing or want to spend extra time completing and triple-checking important tasks; others will look for any reason to get out of the house to complete an errand, perhaps because they want to get their steps in for the day or just see another friendly face. The digitally efficient norm of constant multi-tasking - getting boring things done on your phone while travelling to work or watching the TV - is a model that works for some people but not for everyone, and it is not necessarily government’s job to dictate how people complete those tasks.
“Our day-to-day life as citizens involves many modes of being”
Our day-to-day life as citizens involves many modes of being: the same person can be simultaneously a worker, a student, a caregiver, a person with a chronic health condition, and a taxpayer. While it is absolutely right that reducing the hassle of government admin will be a bonus for that busy person, the generalised burden of government admin will only sometimes be the thing that makes that person’s life a terrible hassle. Every now and then, universal needs - such as booking a Covid vaccine appointment - will arise, but the big administrative burdens people deal with tend to be specific and contextual.
For instance, Tell Us Once is a very effective digital one-stop-shop for notifying different parts of government about a person’s death. Having used it recently, I can attest to its excellence, but it is also, god willing, the kind of service that most people will only use a handful of times in their lives. The death of a family member is a crisis, not the norm; the State rising up to meet us at those times is both compassionate and efficient, but the 69-minutes a week Kyle reckons people spend dealing with the State are not all filled with such reproducible, predictable crises.
Constrained resources
Such an approach also risks ignoring the “last mile problem” that affects many complex services. “Last mile” is supply-chain terminology for delivering the difficult-but-essential final components of a service: getting the package from the warehouse into the customer’s hands, the electricity from the grid to the kettle, the 5G signal to the phone so the person can complete the call. In digital public service terms, not matching service patterns with last-mile capabilities risks building a world in which there are intuitive digital journeys across an otherwise austere, stripped-back landscape of state provision. After all, it doesn’t really matter how easy it is to book a GP appointment online if there are no GP appointments to be found.
This speaks in a way to the fact that the Blueprint also feels preoccupied with making government feel more beautiful on the inside; indeed, its biggest outcome is an internal one - the Digital Centre of Government - rather than a set of tangible external improvements. However, it’s probable that the internal operations of government will never be perfect or fully automatable; there will always be some messiness and friction within the system, and a more can-do approach to getting things done while also slightly tidying up may feel less heroic, but it will ultimately be more effective than repeatedly moving the chairs.
And I can see how this happens. People who are interested in technology are often attracted by very big systems, and few systems are bigger or more important than those in government - but we only have to look to Elon Musk’s DOGE to understand that submitting to technological triumphalism comes with very high financial and democratic costs. Extremely joined-up government is a nice idea on paper – the allure of “completing bureaucracy” by resolving all the data standards/APIs/identity-management components/final numbers in Coldharbor is significant - but it is more than just a pitch for efficiency. Big Automated Government is also a move towards centralised social, political, and economic control - a political project, as well as an administrative undertaking.
Is “Just enough” an alternative?
One of the great things about digital technologies is that they can be implemented in many different ways.
In an attempt to get a grip on government’s approach to digital, I’ve read many policy papers and speeches over the last few weeks, and looking around at the budget cuts, the defense requirements, and the uncertainty of our relations with both Europe and the US it doesn’t seem to me that 2025 is a great time to shift to towards big, generalised automation. The combined risks of many things being half completed, of new dog-leg systems that never quite get integrated, of new capabilities being needed but not budgeted for, of innovation being mistaken for delivery, and of complex problems going unsolved seems significant.
“Technology is, regrettably, not free - even when the license is gratis ”
I mean, I would say this, but 2025 feels like a good time to be both specific and frugal. Technology is, regrettably, not free - even when the license is gratis - and every system needs implementation, management, and maintenance. Every shiny new system is a future legacy project, and many claims of AI efficiencies are relatively unproven. Working efficiently within one’s means can require ingenuity and innovation of a different sort, layering maintenance into management is a form of futureproofing. As we found at Doteveryone in 2017 - better service delivery is likely to arise from a much more precise approach: rather than trusting in technologies to deliver, meeting the needs of high-touch superusers and then scaling those improvements where needed is more likely to unknot the wicked problems than investing in big universal service components. Those kind of fixes may be less fun to live demo, but government digital services are not iPods and should not be treated as such.
Aiming to improve everything a little bit for everyone is, after all, very difficult to deliver, almost impossible to measure, and not very memorable. Delivering concrete improvements for those that need them most will reduce costs, free up capabilities, and improve people’s lives in ways they are also more likely to notice and remember.