Healthcare is generating an increasingly promising landscape of AI innovation. The harder question is what happens after a successful pilot.
A technology may work. Clinicians may see its potential. Evidence may be encouraging. Yet many innovations struggle to move from experimentation into routine, sustainable use.
This is often explained through familiar barriers: regulation, procurement, funding, data, workforce readiness or organisational resistance.
Our research suggests that looking at these challenges separately misses the bigger picture.
Scaling AI in healthcare is not simply a matter of removing individual barriers. It is a system problem, shaped by how innovation pathways, governance, organisations and markets interact.
When these elements fail to align, even promising technologies can become trapped in the pilot-to-scale gap.
Looking beyond individual barriers
Drawing on our evaluation of the NHS AI Lab, we examined how AI innovations moved through development, evaluation and adoption across the health system.
We initially identified 15 challenges spanning the wider policy environment, healthcare organisations and technology suppliers. But these challenges rarely operated independently. A procurement problem could also be a funding problem. Regulatory uncertainty could weaken the supplier business case. Limited organisational capability could prevent an effective technology from being embedded into practice. Unclear routes beyond pilots could discourage further investment.
Looking across these interactions revealed four deeper system dynamics.
Four Dynamics Behind the Pilot-to-Scale Gap
1. Fragmented innovation pathways
Healthcare systems create many opportunities to test AI. What is often less clear is how successful innovations move from pilot to routine adoption.
Who buys the technology? Who funds implementation? What evidence is still required? How does it become part of normal service delivery?
The gap is therefore not simply between pilot and scale. It is between proving that something works and creating a viable pathway for what happens next.
Scaling needs to be designed into innovation pathways from the beginning.
2. Fragmented governance
Healthcare AI crosses multiple institutional boundaries. Regulators, policymakers, funders, healthcare organisations, suppliers, procurement teams and evaluators can all shape whether an innovation progresses.
The problem is not necessarily a lack of governance. It is often too many disconnected parts of governance. Each actor may perform its own role effectively, while the overall pathway remains difficult to navigate.
Having all the necessary actors does not automatically create an aligned system.
Scaling therefore requires stronger coordination, shared learning and mechanisms through which decisions can be connected across the system.
3. Organisational transformation deficits
AI adoption is often approached as a question of where technology can be inserted into existing workflows.
That may generate efficiencies, but it can also constrain what the technology is able to achieve.
Our findings suggest an important distinction between adopting AI and transforming through AI. Some opportunities require organisations to rethink workflows, roles and service models rather than simply automate what already exists.
4. Misaligned markets and incentives
A technically effective AI product is not automatically a scalable one.
Suppliers need sustainable routes to revenue. Healthcare organisations need confidence that implementation costs are justified. Funders, procurement processes, evidence requirements and commercial models all need to align sufficiently for an innovation to survive beyond its pilot phase.
Scaling therefore depends not only on technological performance, but on creating a sustainable configuration around the technology.
These four dynamics reinforce one another: Fragmented governance creates uncertain adoption pathways. Uncertain pathways weaken the commercial case. Organisational capability gaps make implementation harder. Difficult implementation makes evidence of value more difficult to establish.
The result can be a cycle in which promising technologies repeatedly demonstrate potential without becoming part of routine healthcare.
This is why addressing one barrier at a time is unlikely to be enough.
The pilot-to-scale gap is produced by the system, so closing it requires action across the system.
Four shifts for moving beyond pilots
- Build pathways, not just pilots
Innovation programmes should connect development, evaluation, regulation, procurement, implementation and funding much earlier. For innovation teams, funders, healthcare leaders and policymakers, planning for scale should begin with a simple question: “If this pilot succeeds, do we know what happens next?” - Create system alignment
Scaling requires policymakers, regulators, suppliers, healthcare organisations and evaluators to do more than optimise their own part of the process. They also need mechanisms for coordination, learning and collective problem-solving. For policymakers, regulators, suppliers, healthcare organisations and other system actors, the key question is whether coordination extends beyond individual responsibilities: “Are the actors needed for scaling able to learn and act together?” - Think beyond automation
AI can automate existing tasks, augment human expertise and potentially transform how services are organised. Not every implementation requires radical redesign. But organisations should at least ask whether AI creates possibilities beyond improving the existing process. For healthcare leaders and implementation teams, the question should go beyond where AI can be inserted into existing processes: “Are we simply improving what we already do, or considering what AI allows us to do differently?” - Evaluate to learn, not only to judge
Healthcare understandably emphasises demonstrating safety, effectiveness and value. But evaluation should not only happen at the end of a pilot. Formative evaluation can help organisations learn while implementation is unfolding - identifying emerging problems, testing assumptions, adapting implementation and recognising new forms of value. For evaluators, programme teams and decision-makers, evaluation should support learning as well as judgement: “Are we evaluating AI only to judge its performance, or also to learn and adapt as implementation unfolds?”
Beyond the Pilot
The challenge facing healthcare AI is not simply generating more innovation. It is creating the conditions through which useful innovations can become sustainable parts of healthcare delivery. That requires more than another pilot, another funding call or another procurement framework.
It requires connected innovation pathways, coordinated governance, organisational transformation, sustainable incentives and continuous learning.
The question is therefore no longer simply: “Does this technology work?”
But:“Can the system around it turn what works into sustainable value at scale?”
Source
Mozaffar, H., Williams, R., Anderson, S. & Cresswell, K. (2026). A system-level analysis of challenges and strategies for scaling artificial intelligence in healthcare: A qualitative study of the NHS AI lab. DIGITAL HEALTH, Volume 12. First published online 10 August 2026.