POD! On The Abundance Agenda podcast this week, I explain why phone signal on British trains is so bad – and how the French have fixed it. Then Martin digs into the shitstorm at the Office of National Statistics, where it turns out we can’t necessarily trust our numbers. And we speak to Labour MP and the government’s Growth Champion, Dan Tomlinson about the importance of growth to achieving progressive goals. Listen on Apple, Spotify, Substack or wherever you get your pods.
I’m lucky that so far in my life, I haven’t needed to interact with the welfare system.
Because I’m a childless, middle-class-ish professional, in reasonable health,1 who has never experienced significant unemployment,2 I have no first-hand experience of what it is like engaging with the state to claim benefit payments. And similarly, I’ve no first-hand experience of what it is like actually living on a low income.
As a result of this good fortune, it means that the only thing people like me really know about Universal Credit (UC), the government’s programme for distributing welfare payments, is that when it launched it was, well, a bit of a shit-show.
Why? Cast your mind back to the 2010-15 coalition years, when Iain Duncan Smith was the Work and Pensions Secretary. Citing a trip to the Easterhouse estate in Glasgow a few years earlier, where he saw some poor people and had a cry, he launched a complete rebuild of Britain’s welfare system.
At the time, the plans were unsurprisingly controversial. Universal Credit received a tonne of criticism for imposing onerous terms on some of the most vulnerable people in society, such as forcing new claimants to wait 42 days before receiving their first payments, and for more widely imposing conditions and sanctions, such as requiring people to prove they were looking for work.
And then there were problems with the actual roll-out, which was disastrous. The implementation of UC simply didn’t work – and the government at the time was forced to write off an IT system, and rebuild the system from scratch.
So this all sounds pretty terrible – so why the hell does my headline above suggest that UC is a model that other government projects should follow?
This is because of what happened next.
Earlier this year, the Department for Work and Pensions (DWP) quietly published a “Final Lessons Learned” document, and it reveals how Universal Credit was dramatically turned around – and how a team of heroic civil servants upturned the way government usually works to build something that, well, actually works.
And this is where we get to the bait and switch in the title.
Broadly speaking, when it comes to welfare policy, because I’m a squishy lefty, I’m in favour of a much more generous welfare system. On a values level, I would like people in difficult circumstances to receive more cash, and fewer punitive sanctions. (Though I do accept that governments sometimes need to make “tough decisions”, because economics and money are real.)
But we can disagree about the policy, along predictable ideological lines – because what I actually want to talk about is delivery.
So when I say that UC is a model programme, I mean it in terms of the technical delivery of the project – and how the UC system was rebuilt to both meet the government’s policy goals and better serve the people who rely on it.
And hey, now you’re here anyway – read on to find out how the programme was rescued from disaster.
Like nerdy articles about politics, policy, tech, infrastructure and media? Then you’ll like my newsletter. Subscribe now for free to get more of This Sort Of Thing direct to your inbox.
Understanding the basics
Before we get into what went wrong, it’s important to understand what Universal Credit actually is, because I bet many of my readers are people in similar fortunate circumstances to me.
On a conceptual level, the big idea was that UC would combine six of the most claimed benefits and tax credits into a single monthly payment.
Before UC was created, if you wanted to claim, say, Jobseeker’s Allowance, Child Tax Credits, and Housing Benefit, you’d need to apply and fill in paperwork for three different bits of government: DWP, HMRC, and your local council respectively.
For each, you might have to provide answers to a bunch of questions and submit reams of payslips, bank statements, childcare invoices and so on. And then each benefit would be paid on that department’s schedule into your bank account.
And because these benefits were all from the pre-Government Digital Service era, processes behind the scenes were very analogue, and very paper-heavy.
So Universal Credit, in theory, simplifies this. It’s digital-by-default, and if your circumstances change (such as getting a job or having a partner move in with you), you just need to update your information in one place online to notify all of the relevant departments.
Then even better, you don’t even have to provide payslips, as after you’re onboarded on the system, it taps into HMRC’s real-time information system to keep tabs on your earnings.
However, the new system was not just about convenience. The other major advantage was that combining the various benefits would allow them to taper down as you move into work, reducing the cliff-edge where benefits would previously have just cut off.
So this is all to say that there were actually a bunch of good ideas baked into Universal Credit, regardless of what you think of Iain Duncan Smith, or the specific policies around the level of payments, or who qualifies for what.3
What went wrong
Universal Credit didn’t launch all at once. Over the first few months, different groups of claimants were moved over to the new system at different times – with the intention being that by 2017, everyone would have switched over.
However, it quickly emerged that the system that had been built, partially by outside contractors Accenture, IBM, and HP, was dramatically not fit for purpose.
For example, the new system was essentially only capable of handling the most basic claims, and anyone with more complicated circumstances was forced to phone in and have their claim updated manually, because it was not possible using the “self-serve” digital system online.4
Then behind the scenes, many of the processes that needed to take place, like calculating claimants’ income, or checking that payments were being made accurately, were simply not automated. Instead, staff had to check all of this by hand by messing around with spreadsheets.
So there was no way the system could ever scale to support the millions of people who would need to use it. In a damning report, the National Audit Office later concluded that “Such checks will not be feasible or adequate once the system is running nationally.”
This wasn’t the only criticism. Parliament piled in too. The Public Accounts Committee did its own analysis, and concluded that “Management of the Universal Credit programme has been extraordinarily poor.”
And it appears that politics exacerbated the problem, as the same report also concludes that “It is extremely disappointing that the litany of problems in the Universal Credit Programme were often hidden by a culture prevalent in the Department which promoted only the telling of ‘good news’.”
In other words, it was a total disaster. The system didn’t work, Universal Credit was undeliverable, and the DWP itself was toxic.
So eventually, much like HS2 in more recent times, a wholesale “reset” of the programme was ordered. And this is when something incredible happened.
Fixing the mess
The good thing about hitting rock bottom is that the only way you can go is up – at least, that’s what I tell my partner when she returns home from an important business meeting, to find me two days un-showered, feeling sick and sitting next to an empty pack of biscuits.
The same, however, was also true for Universal Credit. Because it was such a high profile disaster, nobody could deny the problem – which opened up the political space for the new team in DWP to try a more radical approach. And to explain this, I’m afraid I need to lapse into the sort of turgid project management jargon that makes your eyes glaze over, and forces you to confront your life choices.
Simply put though, the reason the first version of UC was so disastrous was that it followed the same “waterfall” methodology as most other government projects.
Essentially, “waterfall” is where you work out the requirements of a system, then design it to meet those requirements, then you build it – and then you test and deploy it.
This works great for, say, building a motorway where you can survey the geography ahead of time – but as one person close to the project put it to me – not everything is a motorway.
So with nothing left to lose, the new team was able to try something new: an “agile” test-and-learn approach. This is a way of working most commonly found in software development. But the team didn’t just apply it to the literal writing of the software code that ran the system behind the scenes – they applied it to the whole programme.
Test and learn
The first step to trying this new approach was to start again – from scratch. So the team dumped £40m of code that had been written for the first version of the programme, and began building again with an in-house team – albeit one that was physically located some distance from the main DWP headquarters, to consciously create some distance with the old way of doing things.
And they immediately started doing things very differently to before, such as by starting deliberately small, and focusing on one postcode area – Sutton in South-west London.
In fact, later on they took proximity even further, actually co-locating the team building the system to the office above one of the Job Centres, so they could experience people using it in real life, and could respond to feedback on what worked and what didn’t more quickly.5
Anyway, more important than this was the underlying philosophy behind the approach.
This time around, the team knew that it had a set of outcomes it wanted to achieve, that had been set by the Minister, but instead of predetermining every aspect of the programme, the new approach was agnostic on the question of how stuff should work until it had tested different options in real life.
So rather than limit the first users of the new system to specific, particularly easy, individual claimants, the team included couples who both claim in the pilot from day one. This exposed the designers to the sort-of messy, real-life complexity that could better inform how UC’s “operating model” should work.
And it led to lots of experiments. When it came to, for example, deciding which meetings with job coaches should be face-to-face and which could be handled over the phone, the team tested both approaches at two different Job Centres, to see which worked best.6

Another example is how the team tested the design of Claimant Commitments – which are when the job coach and claimant agree the conditions the claimant has to follow, such as pledging to apply for a given number of jobs per week.7
Here, instead of just building a system to handle it from scratch, the first step was literally using Post-It notes on a piece of paper, so the team could see how real-world claimants understood the conditions they were given and so on. Then if anything didn’t make sense, or needed clarifying, the job coaches could improvise on the fly – before anything got etched into stone when the digital system was built.
Delivery matters
What you may not have noticed in the years since Universal Credit was regularly in the headlines is that the new, rebuilt service is now up and running, and that it is basically a huge success.
This isn’t to say that welfare policy is never controversial – you only have to glance at, er, the news literally this week to see that.
But in terms of the underlying infrastructure that enables the state to pay people who are entitled to payment, it is now robust, reliable, and relatively easy to use.
You can see this in the numbers. In 2017, 25% of all new claims were paid late – on average four weeks late. But since 2020, with the new system up and running, 96% of claimants were paid on time, with 85% of new claimants getting their first payment on time. That’s obviously a significant improvement, especially when you’re in a precarious financial situation.
Another indication is how the programme fared during Covid, when demand multiplied many times overnight. Despite needing to retrain staff overnight, and effectively turn the workforce into call-centre staff, the system was agile and automated enough to handle almost a million new claims in the first two weeks after lockdown was announced. This Institute for Government report describes DWP’s response to Covid as having performed “extremely well”. I’ve also heard Universal Credit described as the only major welfare system around the world that didn’t fall over when faced with the pandemic.
The weeds and state capacity
So why have I spent two thousand words telling you about in-the-weeds design details of a government programme that started out as a disaster, but turned into a huge success?
I think it goes back to state capacity.
It’s worth obsessing over the details, and learning about why things are successful because this is what will make the state more effective and responsive.
I’ve mentioned the example before, but as Robert Colvile observed, the effectiveness of the Covid response essentially came down to databases. When the government offered support, interventions were based on what systems could reach people, and were flexible enough to adapt.
And UC was one of them – because by then the system was so effective, it was a mechanism that made the pandemic less devastating, as people were able to claim the support they needed.
So we should aspire for more of government to be like Universal Credit. And this is especially important, as across government and even inside DWP, there are other enormous projects currently in the works. For example, the forthcoming Jobs and Careers Service will integrate the system for searching for jobs with training and career coaching. It’s not entirely clear what approach that project is following – but I hope it has learned from UC’s agile experience.
However, the huge irony is, of course, that UC’s eventual success was only possible because of the spectacular failure that precipitated it. The team who fixed it were only allowed to be radical because politically, it couldn’t have got any worse for the government.
So I hope that the government can follow the agile approach with similar programmes in the future – and that it can learn, rather than make the same mistake all over again.
Phew! You reached the end! If you enjoyed reading this, please subscribe (for free) to my newsletter to get more takes on policy, politics, tech and infrastructure direct to your inbox. And if you really value my work, please consider taking out a paid subscription as it is only with your support that I’m able to keep writing ultra-nerdy stuff like this.
Just don’t ask my bathroom scales.
Though I have been made redundant twice, and second time around I was unemployed for around ten weeks, which at the time drove me a bit mad. So I can’t even begin to imagine what it is like for people in more difficult circumstances.
Of course you might also want to sprinkle in some political cynicism as there were definitely Tory-coded aspects to the design, such as the switch from weekly to monthly payments was arguably to encourage the lower orders to learn how to budget, and benefits moved to being judged on a month-by-month basis instead of annual, arguably so that feckless scroungers with their flat-screen TVs can’t claim more than they deserve.
As you might imagine, people on low-income tend to have disproportionately difficult or complicated circumstances.
I’m sure that at some point I was told that in the beginning there were more people working on the programme than there were claimants using it – deliberately – so claimant behaviour could be closely studied. But I can’t find a reference to this in any of my notes, or recall who I was talking to! I’m a real journalist, I swear.
Here you can see the software origins of the agile approach - as this sort of A/B testing is common in software.
This is a footnote to remind you that I’m interested in the delivery here, not the policy specifics, like whether conditions imposed on people were too onerous.
Thanks for highlighting the report and the approach - it really is an example of how agile state capacity can fix, change and deliver things. Test and learn as an approach is quite common over a number of public service areas, including in local government - the challenge has been how to grow it over time, and the capacity needed to move to new models while still managing the old. You might find this interesting, which is the Cabinet Office led approach to public sector reform, which is explicitly about test, learn and grow and has been working with local places on experiments
https://www.local.gov.uk/sites/default/files/documents/Public%20Service%20Reform%20webinar%20slides%20-%20complete%20pack%20-%2023%20Jan%202024%20%281%29.pdf
This is now being supplemented with an initial £100m for innovation in local government - not a lot and when local government is facing massive and existential challenges such as temporary accommodation costs - but it’s the beginning of a commitment, and a way of proving to e.g. treasury that this sort of thing can work and deliver real reform. https://questions-statements.parliament.uk/written-questions/detail/2025-03-12/37802/
Good piece and not a criticism but I was waiting for the eureka moment. In fact, how UC was turned round came down to common sense, starting small, continually adjusting approach, (always with an eye on the policy goal) and basing roll out on people's ACTUAL FUCKING needs.
Aaaaaaargh!
It's what everyone else calls customer focused delivery.
As you say, let's hope lessons have been learned.