From ‘pilotitis’ to a learning system: Growing impactful change in complex public services

This need for new approaches reflects a fundamental truth about working in complex public services – from backlogs in health and justice, to AI adoption in local government, to technology-enabled care for vulnerable adults. While central government holds many of the big levers – funding, legislation, regulation – that set the parameters for services, the real impact and experience of our public services is felt at the human level. To determine ‘what works’ in Gateshead or Gloucestershire, a structured approach to explore how best to deliver in different contexts is needed. In steps ‘test, learn, grow’.

The dilemma of scale in public service delivery

In the pressure for delivery, it can be tempting for government to seek to drive change through big bang reforms and use national programmes for ‘roll-out’. Or to get stuck in ‘pilotitis’, announcing generous but short-term funding of highly specific interventions in a few areas to see if they work, only to be confounded by the replication challenge when so much of what makes change successful and helps it stick is in the place-based and human factors that vary with contexts.

This could sound like a counsel of despair for policymakers and politicians, but it doesn’t have to be. As we have found in our research on how to turn Ideas to Impact across the economy and society, there needs to be a shift of attention towards setting the ambitious ‘north star’ for systems through a mission and then creating the conditions and capabilities for collaboration, learning, and adaptation across the ecosystem – backing races, not horses. This is a key focus of the Cabinet Office, which has a remit to test, learn, and grow this way of working for government. That doesn’t mean ‘pushing out’ bright ideas into new places, but developing organisational cultures, structures, and practices that get alongside the place-based reformers and enable them to create real value for citizens, while pulling in learning and input from others.

Unleashing the energy and ideas in the system

The structural and cultural factors that can be barriers to learning and collaboration in government are well known. Budgets are siloed, accountability fragmented, and both internal programmes and external procurements can be geared too often to fixed specifications rather than adaptive delivery. There is no lack of ideas and energy in our public services, but it often sits at the ‘edges’ and we need to reimagine the role of the centre to unleash it in service of improvement.

From our breadth of work across the UK public sector, five principles stand out to create the conditions in government departments to move towards a more adaptive learning system:

1. Focus on scaling impact, not interventions

Impact is seen and felt in the outcomes that citizens experience, which often don’t fit neatly within the boundaries of agencies and departments, and which can be hard to measure. But defining success clearly and embedding evaluation from the outset, not retrofitting it at the end, helps change efforts navigate to value.

Leading indicators – such as uptake rates or early behaviour shifts – move more quickly than outcomes and can be used to help teams decide whether to double down or pivot, as well as to ask practical questions about what else would need to change to amplify the effect. In keeping with test, learn, and grow principles, teams shouldn’t be fearful of walking away or trying a new intervention when the needle hasn’t shifted on their first hypothesis. A cultural shift is required to ensure that leaders and teams are not fearful.

Look at how HM Courts & Tribunals Service began by testing responsible uses of AI in low-risk administrative tasks, underpinned by clear guardrails. Learning by doing in these areas built trust, and informed where they could take it next, expanding into more complex areas. Following the impact like this requires leaders and teams to hold their solutions lightly and keep revising their understanding, rather than stating upfront what it is they want to implement by when.

2. Create an intelligent backbone to enable learning

Test and learn cannot happen without collaboration – and collaboration takes organisation. Learning from the well-established practices of collective impact, a critical success factor is being able to build an adaptive but strong infrastructure for working together, particularly where this cuts across traditional siloed structures that can default to a predict-and-control mindset.

This backbone includes the analytical capacity to work with qualitative and quantitative evidence emerging from practice and synthesise it into actionable insights. Increasingly, some of the processes underpinning the capture, collation, and analysis within learning loops can utilise AI to build a continuously updating picture. And this frees up the humans in the loop to collaborate creatively on the practical changes required.

A profound and welcome shift would be if teams across government increasingly viewed their role as providing this support to a wider coalition of collaborators from across the public, private, and voluntary sectors – at the national, regional, and local level. During the pandemic, similar principles applied to the ventilator mobilisation effort: shared data and common standards allowed best practice to travel quickly while expert discretion stayed where it mattered.

3. Fund the learning curve and protect it from short-termism

Programmatic funding often ends just as valuable learning emerges, leaving pathfinder organisations or areas stranded. Departments need to invest in infrastructure for learning, including the backbone team, and agreeing longer-term funding that can be released iteratively to make sure it is sustaining new learning, not just becoming business-as-usual budgets.

This kind of staged investment, where resources increase as evidence accumulates and confidence increases, avoids the sunk-cost trap and rewards delivery teams for learning. This approach is gaining traction in digital and AI programmes, where small tests unlock larger tranches once risk reduces, and the same iterative delivery can apply beyond digital services. For example, with Defra, we co-designed the recyclability assessment methodology that underpins new packaging legislation. By creating shared definitions and incentives, technical consensus became a lever for market-wide change, growth by design, not accident.

4. Use evidence-led storytelling to make learning sticky

Fundamentally, public services need to be human centred and – just as design depends on user research and user stories – sharing knowledge and learning between places requires more than a data dashboard. However real-time the metrics, a good grasp of what is happening won’t on its own give you the actionable insights about why, nor the legitimacy to share what works with others. Getting those closest to the change, and even better – the users themselves, to talk about why a solution worked in one place and what would need to be the case for it to succeed elsewhere, prevents blind replication and accelerates tailoring to context.

Both internally across teams and externally with stakeholders, there needs to be a commitment to working transparently. For example, regular short ‘what we tried/learned/changed’ updates help citizens, staff, and partners embrace non-linear progress and maintain trust.

5. Make it easier to create multi-disciplinary teams with external expertise

The Cabinet Office Test, Learn and Grow team have been building a blended team and bringing in credible leaders and practitioners from the wider public sector ecosystem. In doing this, they have reflected previous practice in central units, such as Prime Minister’s Delivery Unit, or Government Digital Service. In the wider Civil Service, this has been recommended for over fifty years as one of the thrusts of reform to Whitehall – an ability to inject specific external expertise and experience into the machinery of government to complement officials who may have strengths in policy processes, ministerial management or legislation.

But even more fundamental than bringing in the right people is making sure the operating system is geared for collaboration, rather than reverting to control and compliance. As we have long argued, collaboration is often paid lip service, when it should be a core capability. This begins with creating the right authorising environment for teams to be set up around a clear objective and then the right blend of officials and practitioners, external experts, and technical specialists to be brought together and empowered. It is the diversity of thought and practice that sparks innovation and drives learning – having become the norm in digital and data service design over the past decade, this multi-disciplinary team with shared agile practices now needs to be the new normal across policy development and public service reform.

Moving beyond pilotitis

None of this requires heroics. Landing learning in impactful ways depends less on the perfection of the pilot and more on the resilience of the runway, but that can feel messy and hard to manage.

If test and learn is often about curiosity and humility in the face of complexity, ‘grow’ is about courage, and the willingness to think differently about how organisations are structured, funded, and held accountable so that learning can travel.

While government’s delivery priorities remain urgent and its fiscal position tight, departments don’t have to choose between the broken logic of big bang change or the endless disappointments of pilotitis. When so much is expected of the public sector, we need to lean into complexity by fostering resilient services designed with citizens, where impact grows through a state that is more comfortable with collaboration and adaptation.