Restart: Being Paid Handsomely for the Sun Coming Up

Every morning, at roughly the same time, the sun comes up. This is reliable, well documented, and has been going on for some considerable while. It is also, as far as anyone can tell, entirely unaffected by whether or not a man in a car park waves a stick at the horizon and shouts encouragement.

Now imagine that the man with the stick has a government contract. Imagine he is paid per sunrise. Imagine he files a quarterly report showing a 100% sunrise rate across all regions, and that the department responsible nods gravely, describes the results as encouraging, and renews him. This is, I am sorry to say, not a particularly exaggerated description of how we judge employment programmes in the United Kingdom.

Which brings us to the Restart Scheme, and the figure everyone keeps quoting: roughly one in three participants placed into work. Thirty per cent. Applause, ribbon, press release.

I would like to suggest, as gently as a man can while holding a calculator, that thirty per cent is not a success. It may not even be a result.

 

What Restart Actually Is

Restart is the Department for Work and Pensions' flagship programme for people who have been unemployed for a long time, launched in July 2021 as the economy staggered out of lockdown blinking at the light. Private providers are contracted to give participants up to a year of intensive, tailored support: CV workshops, interview coaching, motivational chats, and the occasional free biro.

According to the DWP's own statistics, 840,000 people started Restart between July 2021 and April 2025. Of those, about 250,000 (30%) achieved what the department calls a "job outcome". That has a precise definition, and in fairness it is not a trivial one: earnings equivalent to 16 hours a week at the National Living Wage, sustained for six months, achieved within 18 months of starting.

The contracts set a minimum performance level of 26% and a target of 34%. So the scheme has landed comfortably above the floor and somewhat below the ceiling, in the zone that in any other walk of life we would describe as "fine, I suppose."

And here is the part I find quietly magnificent. The DWP itself says these job outcome rates should be read "very cautiously" as a guide to the scheme's effectiveness. This is the bureaucratic equivalent of a restaurant putting a sign in the window that reads "Food: possibly."

 

The Seven Who Are Still Looking

Let us start with the bit that the press releases are curiously shy about. If three in ten participants reached a job outcome, then seven in ten did not. Not all of them are still looking (some are working fewer than 16 hours, some found jobs that did not last six months, a few will still get there), but they are the majority, and they are not a margin of error.

If an airline announced that 30% of its passengers had arrived at their intended destination, it would not describe this as a strong performance on connectivity. It would be the subject of a public inquiry, several documentaries, and at least one very angry pilot's memoir.

The headline claim is "one in three placed into work", and the unavoidable mirror image of that claim is "two in three not." You cannot celebrate the glass being one third full without someone noticing what the rest of the glass is doing.

 

The Three Who Would Have Done It Anyway

But the seven are not actually the interesting problem. The interesting problem is the three.

Long-term unemployment is not a sealed room. People leave it all the time, by themselves, through the ancient and deeply unglamorous mechanisms of applying for things, knowing someone, getting lucky, or the local warehouse finally taking on staff. The economists have a word for the outcomes a programme would have got anyway, regardless of whether it existed. They call it deadweight, which is a lovely word, because it sounds exactly like what it is.

So the question is never "how many participants found work?" The question is "how many found work because of the programme, who otherwise would not have?" Those are wildly different numbers, and only one of them is worth paying for.

Here is where it gets good. The National Audit Office's December 2022 report sets out the department's own working assumptions, and they are refreshingly candid. Left entirely to their own devices, about 30% of participants were expected to find work anyway. Restart's job was to add roughly six more per hundred on top. Put those together, and the department's own assumptions point to something in the region of 36%.

Not 30. Thirty-six. And lest anyone think I have invented that number on the back of an envelope (I did do it on the back of an envelope, but the envelope was very well informed), the contracts themselves set a target of 34%. Whichever way you slice it, the people who designed this thing expected it to land in the mid-thirties.

Now, I want to be scrupulously fair here, because the careful reader is already reaching for a pen. That 30% baseline was a pre-launch modelling assumption, not a measurement of these particular people, and it was not necessarily calculated on exactly the same basis as the job outcome figure, so adding six to one and comparing it with the other is an illustration, not a verdict. I am not going to stand here and announce that Restart has achieved precisely nothing. I cannot prove that, and neither, crucially, can anyone prove the opposite. And the 30% headline is not the final score either: participants have up to 18 months to reach a job outcome, and the most recent starters still have as much as six months left on the clock. The number will creep up. It may well creep into the low thirties.

But notice what that does and does not rescue. Even if the cohort finishes a few points higher, it is still being measured against a business case that pencilled in the mid-thirties, and the honest version of the poster is still not "one in three placed into work." At best it is "about one in seventeen placed into work who would not otherwise have been," and on the current numbers the scheme has not yet demonstrably delivered even that. You do not need to believe Restart does nothing to be uncomfortable. You only need to believe the department's own spreadsheet.

 

Nobody Has Actually Checked

You might reasonably assume that, five years and well over a billion pounds into the programme, someone has gone and measured whether the six actually exists. You would be wrong, which is a sentence that does an awful lot of work in British public administration.

No counterfactual impact study of Restart has been published. The closest thing is the 2024 Learning and Work Institute and Ipsos evaluation, which surveyed 635 people who were referred but never started. The report is admirably candid that this is "not a direct control group" and "not an impact assessment". Which is to say: it is a survey of some people who did not do the thing, and it is not to be confused with evidence about what the thing does.

In June 2026, a parliamentary answer offered that 35% of participants were in work, compared with 26% of a comparable group of non-participants. That nine-point gap sounds encouraging until you remember the first rule of statistics, which is that people who sign up for things are not a random sample of people. Those who start a scheme, turn up, and stick with it may simply be the people already most likely to find work. It is correlation. It wears a suit and a lanyard, but it is correlation.

A proper propensity-score impact evaluation is apparently underway, due later in 2026. So the answer to the question "does Restart work?" is, as of this morning: we will let you know. This is a slightly unusual order of operations. Most of us check whether the parachute opens before we buy 840,000 of them.

 

The Price of a Sunrise

Let us talk about money, because the money is where the joke stops being entirely funny.

The NAO found that the expected cost per participant had risen to £2,429, with the scheme's total cost reaching around £1.68 billion. Over the same period, the expected benefit-cost ratio fell from £3.80 per pound spent to £2.44. Still positive, you might say. But remember what that ratio rests on: the assumption that the extra six job outcomes per hundred are real.

Do the sum yourself. One hundred participants at £2,429 each costs £242,900. If that buys six additional jobs, each additional job costs roughly £40,000. That lines up neatly with an outside estimate reported by PoliticsHome of around £40,000 per additional job (calculated when participant numbers were lower than they are now, so treat it as a sketch rather than a portrait).

Forty thousand pounds per additional job is not obviously absurd. One could argue it is a decent investment if that person stays in work and off benefits for years. But now run the same sum with the less charitable assumption that the deadweight has eaten the lot, and the additional outcomes are not six but zero. £242,900 divided by zero is, mathematically speaking, undefined, and fiscally speaking, a very large hole shaped exactly like £1.68 billion.

The Public Accounts Committee saw this coming. It warned that if the six-point estimate proves too high, "the benefits of the scheme will quickly fall away". Committees do not usually write sentences like that for fun. That is a committee clearing its throat very loudly in the direction of a cliff edge.

 

A Business Model I Would Very Much Like

I have spent a career building businesses, running technology organisations, and occasionally explaining to boards why the thing they bought does not do the thing they thought it did. So I hope I can be forgiven a moment of professional envy.

Consider the shape of the opportunity. You are paid to take people who are, in a meaningful proportion of cases, going to find work anyway. You are paid when they do. Your minimum performance level (26%) is set below the department's own estimate of the rate at which they would have done it without you (30%). Nobody measures the difference between what you did and what would have happened if you had stayed in bed. And when the figures come in, the department that commissioned you cautions everyone to read them "very cautiously," which is the warmest possible blanket for a contractor to be wrapped in.

It is, in short, the perfect business. It is a weather forecasting firm that only gets paid on sunny days and is permitted to define "sunny." It is an umbrella company billing per dry pedestrian. It is a cockerel invoicing for the dawn.

Where do I sign? I have a stick. I can wave it at a horizon with real conviction. I can produce quarterly reports in a range of reassuring fonts.

 

In Fairness to the Cockerel

I should be clear about what I am not saying. I am not saying the people who work on Restart are idle, or that no participant has ever been helped. I have no doubt there are advisers who have changed lives, and participants for whom a bit of structure and someone who believed in them made all the difference. Those people exist, and they deserve better than to be buried inside a number that cannot tell them apart from the ones who would have been fine regardless.

And it is possible that when the propensity-score evaluation lands, it will show a genuine, durable, cost-effective uplift. If it does, I will cheerfully write a follow-up post saying so, in a slightly smaller font.

What I am saying is that a 30% placement rate, by itself, proves precisely nothing about whether the scheme works. It is a gross figure masquerading as a net one. It tells you how many people found jobs, not how many found jobs because someone was paid to help them. And measured against the department's own expectation of somewhere around 34% to 36%, the honest reading of the headline number is not "a third of people helped." It is "a programme running below its own business case, whose one valuable ingredient nobody has yet measured, at a cost of rather a lot."

 

What Good Would Look Like

None of this is difficult, which is what makes it so maddening.

• Measure the net, not the gross. Every contract should be judged on additional outcomes against a properly constructed comparison group, not on how many people happened to find work while enrolled.
• Evaluate before you scale. Randomise a slice of referrals from day one. It is cheap, it is ethical when places are limited anyway, and it is the only way to know.
• Pay for the difference. If providers are rewarded for outcomes the market would have produced anyway, you are not buying employment. You are buying a commentary on it.
• Report the seven as well as the three. The people still looking are not a footnote. They are the point.

 

The sun will come up tomorrow. If the department's own modelling is right, roughly 30% of the long-term unemployed will, over the next year and a half, find their way into work, as they always have, through grit and luck and the mysterious machinery of local labour markets. Somewhere, someone will be standing in a car park, waving a stick at both, and sending in the invoice.

I do not begrudge him the stick. I would just quite like someone, at some point, to check whether it does anything.

All Posts