Reality check for NHS AI rollout: UCL study details delays, scepticism, and IT hurdles

NHS AI rollouts are harder than they look, UCL study finds

AI can triage scans and surface abnormalities. But moving from pilot to practice inside NHS hospitals is proving slow and messy. A new UCL-led study shows the gap between promise and delivery is wide-and fixable with better governance, training, and integration.

What the national programme set out to do

An NHS England initiative launched in 2023 sought to deploy AI tools for chest conditions, including lung cancer, across 66 trusts. Backed by £21m, the goal was to prioritise urgent cases and flag issues on imaging to ease pressure on diagnostic services. This study provides one of the first real-world looks at how that plays out on the ground.

NHS England and the NIHR supported the work.

What actually happened

Rollout took longer than planned. Contracting alone slipped by four to ten months. By June 2025-18 months after the original target-23 of the 66 trusts still were not using the AI tools in clinical practice.

The data confirms what many teams feel: the bottlenecks are as much human and organisational as they are technical.

Why progress stalled

• Staff capacity: Clinical teams already under heavy load struggled to engage with procurement, testing, and change management.
• Confidence and accountability: Skepticism was common, especially among senior staff concerned about decision-making and oversight.
• Training gaps: Many lacked a baseline grasp of how the tools work, where they help, and where they fail.
• Integration friction: Embedding AI into ageing, varied IT estates differed trust by trust and often slowed progress.
• Complex procurement: The volume and technical depth of documents led to confusion and risked missing key details.

What helped

• Dedicated project management with clear clinical and IT ownership.
• Committed local leads who could unblock issues quickly.
• Shared learning within imaging networks to avoid repeating mistakes.
• Strong national programme leadership to align standards and expectations.

Practical playbook for NHS teams

• Lock in ownership early: Name clinical, radiology, IT, and information governance leads. Set a weekly decision cadence.
• Start training before go-live: Role-specific sessions for radiologists, radiographers, reporting radiographers, and clinicians. Focus on oversight, escalation, and limits of the model.
• Build a clear clinical workflow: Map who reviews AI flags, how disagreements are handled, and what gets documented.
• Integrate into existing systems: Target PACS/RIS/EPR integration and single sign-on from day one. Avoid standalone viewers where possible.
• Simplify procurement: Use a nationally approved shortlist and shared templates to cut contract time and reduce variation.
• Measure what matters: Track turnaround time, backlog size, false alerts reviewed, user overrides, and safety incidents. Report weekly for the first 90 days.
• Share what works: Publish configs, SOPs, and training packs across your imaging network to speed others up.

Policy implications

The researchers warn AI tools "may not address current healthcare service pressures as straightforwardly as policymakers may hope." The path to value is execution: standardised contracts, common integration patterns, protected training time, and a tighter supplier shortlist can shave months off delivery.

What's next

The team is now studying how these tools perform once fully embedded and how patients and carers view their use. That perspective will shape safer workflows and clearer communication in clinics.

Action checklist for the next quarter

• Confirm named leads across clinical, imaging, IT, and governance with protected time.
• Approve an integration plan with milestones for PACS/RIS/EPR and sign-on.
• Run role-based training and a supervised pilot in one pathway before expanding.
• Stand up a weekly dashboard covering turnaround time, backlog, review rates, and safety signals.
• Adopt national templates and, where available, a pre-approved supplier list to compress contracting.

Need structured upskilling?

If your team needs foundational AI skills and practical workflows, explore role-based options at Complete AI Training - Courses by Job or browse Latest AI Courses. Building confidence early shortens rollout and improves safety.