The promise of AI in healthcare has been a beacon of hope for improving patient outcomes and alleviating the pressures on overworked medical staff.
Yet, a new UK study reveals a more complex reality: implementing AI tools in NHS hospitals is a far greater challenge than initially anticipated.
Led by researchers at University College London (UCL), the study found that the journey from an AI concept to clinical reality is riddled with hurdles, including governance issues, a lack of staff training, and difficulties integrating new technology with existing NHS IT systems.
The findings serve as a crucial learning opportunity for policymakers, including the UK Government, which has identified digital transformation and AI as key pillars of its long-term plan for the NHS.
A pioneering programme and its unexpected delays
The research, funded by the National Institute for Health and Care Research (NIHR), focused on an NHS England programme launched in 2023.
The initiative aimed to introduce AI tools to help diagnose chest conditions, including lung cancer, across 66 hospital trusts.
Backed by £21m in funding, the initiative was designed to enhance diagnostic services by prioritising critical cases for review and flagging abnormalities on scans.
While previous laboratory-based studies had suggested that AI could significantly benefit diagnostic services by improving accuracy and reducing errors, this new research provides one of the first real-world analyses of AI implementation in a healthcare setting.
The UCL-led team, which also included experts from the Nuffield Trust and the University of Cambridge, conducted a deep dive into the procurement and early deployment of these AI tools.
Through interviews with hospital staff and AI suppliers, they uncovered both the pitfalls and the practices that helped smooth the process.
The findings showed that the rollout of the AI tools took significantly longer than expected.
Contracting alone was delayed by four to ten months, and by June 2025, 18 months after the initial target for completion, a third (23 out of 66) of the hospital trusts were still not using the AI in their clinical practice.
Human and technological hurdles to AI adoption
The study identified a number of key challenges that slowed down the implementation. A major issue was the sheer difficulty of engaging clinical staff who are already grappling with incredibly high workloads.
Many staff members also lacked a fundamental understanding of the new technology and expressed a degree of scepticism about using AI in healthcare.
This was particularly true for more senior staff who had concerns about accountability and the potential for AI to make decisions without human oversight.
Technological challenges were equally daunting. The new AI tools needed to be embedded within the NHS’s ageing and diverse IT systems, a complex task that varied from one hospital to another.
The technical nature of the procurement process also proved to be a hurdle, as some staff members found themselves overwhelmed by the sheer volume of detailed information, increasing the likelihood of critical details being overlooked.
The path forward: Best practices and recommendations
Despite the challenges, the study also highlighted several factors that contributed to a smoother implementation.
Dedicated project management was a key success factor, as was a high level of commitment from the hospital staff leading the implementation.
The research also found that shared learning and resources between local imaging networks, as well as strong national programme leadership, helped facilitate the process.
In their conclusion, the researchers noted that while AI tools can provide valuable support for diagnostic services, they “may not address current healthcare service pressures as straightforwardly as policymakers may hope.”
The study’s authors made several important recommendations to improve future rollouts of AI in healthcare.
They emphasised the need for early and ongoing training for NHS staff on how to use AI effectively and safely, stressing that this training must address concerns about accountability and clinical input.
They also suggested that creating a nationally approved shortlist of potential AI suppliers could help streamline the procurement process for individual hospital trusts.
Next steps for research
The researchers are now conducting further studies to understand how the tools are used once fully embedded and to explore the perspectives of patients and carers, a crucial aspect that was not part of this initial phase.
This ongoing work will continue to build on the limited but growing body of evidence about real-world AI implementation, paving the way for a more effective and successful integration of technology into the future of the NHS.
