Legacy IT and red tape slow NHS AI adoption, UCL study finds
UCL finds NHS AI rollouts lag due to fragmented IT, governance delays, and limited clinician time. Many trusts remain offline; strong leadership and project managers help.
AI in the NHS: UCL study reveals hidden barriers slowing healthcare innovation
A new University College London study details why AI adoption across NHS hospitals is moving slower than planned. The short answer: fragmented IT, governance delays, procurement hold-ups, and limited staff time for implementation. The full paper is published in The Lancet eClinicalMedicine.
What UCL studied
In 2023, NHS England funded £21 million to deploy AI for chest imaging (including suspected lung cancer) across 66 hospital trusts. Trusts were grouped into 12 imaging networks to scale access to specialist tools. Researchers interviewed hospital teams and AI suppliers to track procurement and setup in real settings.
Key findings
- Implementation took longer than planned. Contracting ran 4-10 months over initial timelines.
- By June 2025-18 months after contracting was due-23 of 66 trusts were not yet using the tools in clinical practice.
- Top blockers: limited clinician time to engage, integration with ageing and varied IT systems, and scepticism or low awareness of how to use AI safely and effectively.
- This was one of the first large-scale looks at real-world AI rollout in NHS diagnostic services.
What helped
- Clear national programme leadership.
- Active collaboration across local imaging networks.
- Committed clinical and managerial leads inside hospitals.
- Dedicated project managers where available-these services moved faster.
Why this matters for patient care
AI can support reporting, case prioritisation, and follow-up. But without the right governance, IT fit, and clinician time, the benefits stay on paper. Pace and consistency of rollout also affect equity of access across trusts.
What NHS leaders can do next
- Stand up a fast-track governance route with clear templates for data protection, clinical safety, and local approvals.
- Fund project management and informatics time in each trust's plan; protect clinician time for selection, validation, and go-live.
- Run an "integration readiness" check: PACS/RIS compatibility, network/security review, image routing, identity/access, and audit logging.
- Agree common interface standards across the network (e.g., DICOM worklists, HL7/FHIR where relevant) to avoid bespoke builds.
- Use shared procurement packs across trusts: requirements, evaluation criteria, and service-level expectations.
- Start in shadow mode, define acceptance criteria, and monitor real-world performance before full switch-on.
- Set simple metrics: time to report, backlog, escalation accuracy, false positives/negatives, and clinician acceptance.
Guidance for AI suppliers working with NHS sites
- Ship with out-of-the-box connectors for common PACS/RIS and clear on-prem vs cloud architectures.
- Provide a deployment playbook: network needs, security, roles, training, test scripts, and rollback steps.
- Prepare a complete information governance and clinical safety pack, including post-market surveillance and update process.
- Support shadow mode, site-level performance monitoring, and explainability where clinically relevant.
Training and culture
The study recommends training NHS staff on safe, effective use of AI. Focus on practical modules: what the tool does and does not do, local clinical workflows, data protection, interpreting outputs, and escalation rules.
- Train-the-trainer models inside imaging networks speed up adoption.
- Include short simulations and real cases to build trust and reduce scepticism.
If your team needs a structured starting point for AI upskilling, see these resources: AI courses by job.
What the researchers said
"The NHS is made up of hundreds of organisations with different clinical requirements and different IT systems, and introducing any diagnostic tools that suit multiple hospitals is highly complex."
"We found it took longer to introduce the new AI tools than programme leads had expected. A key problem was that clinical staff were already busy-time for selection, IT integration, and local approvals was limited. Services with dedicated project managers moved more smoothly."
"AI tools can offer valuable support for diagnostic services, but they may not address current service pressures as simply as policymakers may hope."
![Agentic AI in Healthcare: Autonomy, Transparency, and Patient-Centered Care [Q&A]](https://cdn.completeaitraining.com/news_images/agentic-ai-in-healthcare-autonomy-transparency-and-patient-centered-care-q-a_2025-09-20.jpg)
