Stop Building Innovation Labs

Labs were never the point. The conditions surrounding them were.

Nearly all of my career, intentionally or not, has been spent inside organisational experiments.

Back in 2017 I worked at Bosch Power Tools in an internal UX and product innovation lab. Shortly after, I took part in IBM’s Extreme Blue programme – multidisciplinary teams in rapid delivery sprints with organisations like the NHS, designing and prototyping new products in weeks rather than months. My time at Wilson Fletcher sat somewhere adjacent: digital transformation and organisational change. Then HMRC Policy Lab, a public sector innovation lab explicitly designed to explore policy problems differently. And now CustomerFirst within Government Digital Services (GDS), which is attempting something else entirely – a NewCo-style model focused less on generating insight and more on building, testing and learning in live environments.

Looking back now, the pattern feels clearer than it did at the time.

The first wave genuinely worked

Over the last two decades, governments across the world have built hundreds of innovation labs, policy labs, behavioural insights teams and multidisciplinary transformation units. They emerged from a growing recognition that traditional institutional structures were struggling to respond to complexity, digital transformation and increasingly interconnected public problems. Labs offered a different organisational form: protected spaces within existing systems where smaller multidisciplinary teams could experiment outside normal bureaucratic logic.

And for a while, they worked remarkably well.

The early generation of public sector innovation labs genuinely changed government. Denmark’s MindLab helped bring ethnographic and participatory approaches into policymaking long before most governments were seriously talking about user-centred design. Helsinki Design Lab explored how strategic design could help states work across interconnected systems rather than departmental silos. Policy Lab UK experimented with multidisciplinary approaches inside Whitehall, while the early GDS movement in 2011 fundamentally reshaped expectations about what public digital services could be.

These were not vanity projects or post-it-note theatres. They produced rigorous work, brought new professions into government, changed the legitimacy of user-centred design inside the state and created new ways of understanding public problems. More importantly, they created conditions most institutions struggle to sustain – protected authority, proximity to decision-making, permission to experiment and the ability to generate evidence through building rather than reporting.

Then the conditions disappeared

But over time, many of these environments lost the conditions that had made them effective in the first place.

MindLab was eventually folded into a new government taskforce following shifts in political sponsorship and administrative priorities. Helsinki Design Lab closed in 2013 after SITRA concluded that strategic design had become more valuable embedded across the organisation than concentrated inside a standalone lab. Despite Policy Lab being more than 10 years old, there is very little evidence that design-led approaches have had a significant impact on high-level policy development. And even GDS, arguably the most influential public sector transformation movement of the last twenty years, gradually became more institutionalised, governance-heavy and operationally constrained as it matured, something Mike Bracken, its founder, later reflected on after leaving government.

The common explanation is political churn. Budgets disappear, ministers move on, leadership changes. This is all true. But the deeper issue is structural.

Schuurman and Tõnurist famously described innovation labs as “islands of experimentation”. That framing still feels right. But islands can become isolated. And isolation, over time, becomes impotence. What the last few decades have revealed is that the further labs moved from live delivery and meaningful authority, the more they drifted toward insight generation detached from operational consequence – describing the future rather than testing it.

The architecture never changed

At the same time as labs proliferated, governments more broadly began adopting professions and methods designed around iterative learning – service design, agile delivery, user research, multidisciplinary product development. These approaches changed how institutions thought about services, users and delivery.

But the deeper architectures of decision-making remained largely intact. Major commitments continued to be shaped through sequential processes where assumptions harden early and learning arrives late – often only after implementation has already begun failing at scale.

The result is a strange institutional contradiction where governments have spent the last two decades building highly capable professions for iterative learning, while continuing to position those professions too far downstream from the decisions where iterative learning matters most.

Why the early GDS story still matters

The early GDS story matters not because government should nostalgically recreate 2011, but because it demonstrated what becomes possible when authority, learning and delivery remain connected long enough for evidence to accumulate.

When Tom Loosemore, Mike Bracken and the early GDS team described digital transformation, they were not really describing a design function or an innovation programme. They were describing a protected delivery environment: politically sponsored, operationally empowered and capable of generating evidence through working products rather than strategy documents. GOV.UK did not become persuasive because GDS produced a compelling vision deck. It became persuasive because working software changed the institutional argument in real time.

The working thing was the evidence. Governance, funding, resources all followed the evidence. Not the other way round.

No innovation until everything works. Early GDS – Ben Terrett

That is the part organisations have consistently struggled to reproduce – partly because it requires political capital that is genuinely difficult to sustain, and partly because the governance systems that eventually surrounded GDS were designed for predictability and assurance, not for the kind of learning-through-delivery that made GDS persuasive in the first place.

And yet, I do not think the lesson from the last twenty years of public sector innovation is that governments should stop experimenting. If anything, the opposite may be true.

Rebuilding the conditions

This is why models like CustomerFirst are worth taking seriously. Not because they represent “the answer”, but because they are revisiting some of these conditions again: smaller multidisciplinary teams, protected delivery environments, evidence generated through testing and building, and governance that follows learning rather than attempting to pre-approve every pathway in advance.

Yes, CustomerFirst is early and has not yet demonstrated sustained influence over major institutional decisions – the kind of high-stakes, high-commitment choices where the gap between assumption and evidence most damagingly widens. And yes, it is also operating within GDS, an organisation that has itself become significantly more institutionalised since 2011. But the more interesting question is whether a NewCo-style model can maintain protected authority as it scales, or whether it will gradually be absorbed into the same governance logic that constrained its predecessors.

Because the risk is not that CustomerFirst fails outright. The risk is that it succeeds at the level of products and services while the deeper architecture of decision-making – where assumptions harden, where commitments calcify, where learning consistently arrives too late – remains essentially unchanged. That is precisely what happened to the first generation of labs. They produced excellent work and lost the argument about how decisions are made.

The conditions are the point

After spending nearly a decade moving through different organisational experiments, the pattern underneath all of them feels surprisingly simple. The environments capable of creating meaningful change are almost always the ones that shorten the distance between assumption and evidence – ensuring that learning happens before institutions become too committed to decisions that are difficult to reverse.

Test, learn, repeat – The Radical How, 2024

This is the logic sitting underneath approaches like the early Government Digital Service or Public Digital’s “test and learn” approach, which feels like a contemporary evolution of that earlier ethos. The point is not to move recklessly fast or abandon rigour. It is to test earlier. Build sooner. Learn before institutions become too committed to decisions that are difficult to reverse.

It is not a rejection of research so much as a shift from generating insight about doing, to learning through doing. From discovery work that describes hypothetical futures, to prototyping that creates partial versions of those futures and tests them against reality before institutions fully commit to them. Essentially, the work becomes about speeding up the feedback loop around the things that matter most.

But a faster feedback loop only changes things if it is attached to something consequential. Which brings the argument back to conditions rather than form.

The honest lesson from twenty years of public sector innovation labs is not that the model was wrong. It is that the lab itself was never really the innovation. The innovation was the institutional conditions that made the first wave effective in the first place.

What the next generation of transformation environments needs to get right is not what they are called – lab, unit, NewCo, programme – but whether those conditions can be created, sustained and kept genuinely connected to the decisions that matter most. A protected experimentation environment without political literacy, legitimacy or stewardship of the wider system does not solve the problem. It risks being wrong faster.

The question has never really been whether governments need experimentation. It is whether they are willing to create and protect the conditions that allow experimentation to shape decisions before systems fully commit themselves to the wrong trajectory.

That was always the harder problem. It still is.

---

This piece draws on CustomerFirst, alongside earlier CIVICWORKS writing on institutional learning, decision-making under uncertainty and the architecture of public transformation — particularly Unfit for Uncertainty and Beyond the Waterfall State. It builds on research into public sector innovation labs, strategic design and adaptive governance, particularly the work of Andrew Greenway, Ben Terrett, Mike Bracken and Tom Loosemore, Schuurman & Tõnurist, Jenny Lewis, Anna Whicher, Francesco Leoni, Public Digital and others exploring how institutions learn under conditions of uncertainty.