IPM Take
The UK’s problem may not be inventing health AI. It may be deciding what happens after the invention works.
England’s 10 Year Health Plan identifies data, AI, genomics, wearables and robotics as transformative technologies and states an ambition to make the NHS the world’s most AI-enabled health system, with AI integrated across clinical pathways. It also proposes population-based polygenic risk scoring and broader use of genomics for personalised prevention.
Yet evidence to the House of Lords Science and Technology Committee suggests that the route from innovation to routine NHS use remains much less defined.
Brainomix CEO Michalis Papadakis told the committee that the company had deployed its stroke-imaging AI across 26 NHS hospitals and trusts through an NHSX AI award, supporting an evaluation involving more than 450,000 patients. But he described a persistent gap between successful pilots and adoption at scale, with different governance, IT and purchasing processes across NHS organisations.
That creates an increasingly important policy question: if the NHS wants AI embedded in clinical pathways by 2035, who owns the pathway from evidence to procurement today?
Executive Summary
The House of Lords Science and Technology Committee launched its inquiry into Innovation in the NHS: personalised medicine and AI in March 2026, examining why advances in genomics, AI and personalised therapies can struggle to move from scientific development into routine NHS delivery.
At its 19 May session, the committee heard from Genomics plc, Autolus Therapeutics, Brainomix and Scancell. The technologies represented ranged from polygenic risk scoring and AI-assisted stroke imaging to CAR T-cell therapies and personalised cancer vaccines.
Papadakis argued that innovators lack a clear national route linking clinical need, validation, regulation, procurement, reimbursement and scale-up. He told peers that the NHS has infrastructure for running pilots but described a “chasm” between pilots and widespread adoption.
Genomics plc CEO Sir Peter Donnelly raised a parallel issue for personalised prevention. He told the committee that his company had conducted NHS validation work on polygenic risk scores and completed regulatory steps, but that adoption remained difficult because prevention lacked a single NHS budget or clear organisational owner. Those statements represent the company’s evidence to Parliament rather than independent findings of the committee.
Why it matters
- HTA bodies: AI and personalised diagnostics increasingly challenge sequential assessment models. If regulation, evidence generation, HTA and procurement operate as separate stages with different requirements, developers may have to generate overlapping evidence before adoption is possible.
- Payers: Scaling digital technologies is not cost-free simply because the product is software. Licensing, integration, data infrastructure, cybersecurity, workflow redesign, monitoring and staff training all affect the real cost of implementation.
- Industry / innovation partners: The central commercial risk may increasingly sit between regulatory clearance and procurement. A successful NHS pilot does not automatically create a national customer, reimbursement route or scalable adoption mechanism.
The UK has no shortage of ambition for artificial intelligence in healthcare.
Its 10 Year Health Plan says England should become the world’s most AI-enabled health system and envisages AI working alongside genomics, electronic records and predictive analytics to identify disease risk earlier and personalise treatment. By 2035, the plan envisages AI being integrated into most clinical pathways.
The House of Lords inquiry is testing what stands between that ambition and routine practice.
One of the clearest examples came from Brainomix, which develops AI imaging technology used in stroke care. Papadakis told the committee that NHS support had enabled deployment across 26 hospitals and trusts and a three-year evaluation involving more than 450,000 patients. Yet the company still encountered different IT, governance and commercial processes as it moved between NHS organisations.
His evidence focused on a structural problem rather than the performance of a particular algorithm.
There is substantial support for experimentation, he argued, but no equivalent end-to-end pathway that automatically takes a technology from clinical validation through regulation, reimbursement, procurement and national scale. He called for a national adoption pathway and funding mechanisms designed for implementation rather than pilots alone.
Genomics plc presented a similar issue from another corner of personalised medicine.
Donnelly told peers that the company’s polygenic risk-score technology had undergone analytical and clinical validation, including work with NHS general practices. He said 98.5% of patients in one trial reported finding the approach helpful and that GPs changed patient management in 13% of cases. But he argued that prevention technologies face a structural obstacle because NHS prevention funding and accountability are dispersed. These figures and interpretations were presented by the company to the committee and should be treated as stakeholder evidence rather than conclusions reached by Parliament.
The inquiry subsequently widened beyond industry. The committee heard from the Health Data Research Service, clinicians, Health Innovation Network representatives, NICE, the MHRA and, finally, ministers and senior DHSC and NHS officials. The final government evidence session explicitly considered failure to scale pilots, fragmentation across trusts, regulation, workforce barriers and the NHS’s role as a customer for innovation.
That matters because the technological ambition is accelerating.
The 10 Year Health Plan proposes universal newborn genomic testing, population-based polygenic risk scoring and an NHS in which AI can interact increasingly closely with health records, diagnostics and clinical pathways.
But adoption infrastructure is part of health technology too.
An algorithm can be accurate. A genomic score can be validated. A regulator can clear a product.
None of those steps guarantees that a clinician can actually use it across the NHS.
The next phase of the UK’s AI strategy may therefore be judged less by how many new pilots it launches and more by how many proven technologies can make the much harder transition into routine care.
