Andrew Davies
Executive Director, Digital Health (ABHI)
The NHS is embracing AI, but not all AI is treated equally. Immediate operational pressures influence adoption more strongly than long-term digital ambition.
Not all AI carries the same risk level. Non‑clinical applications — workforce planning, demand forecasting, pathway optimisation and waiting‑list management — sit outside direct decision‑making, so governance is lighter, procurement faster and the consequences of failure more contained.
Clinical AI is the opposite. Tools that influence diagnosis, treatment or patient prioritisation must evidence safety, meet regulatory requirements and fit within clinical governance — stretching timelines and increasing delivery risk.
Proven clinical use cases
Diagnostic image analysis is a proven clinical use case. AI supporting radiology and pathology is already deployed across parts of the NHS, helping clinicians triage high‑risk cases and manage capacity within established workflows and specialty‑led governance.
This shows clinical AI can scale where risk is well understood, benefits are clear and assurance frameworks are mature — conditions not consistent across frontline settings.
Population health management illustrates how the clinical and non‑clinical AI boundary is blurring. Analytics can identify at‑risk cohorts, predict demand and support planning across care systems. While they shape clinical priorities, they typically sit upstream of individual decisions and attract lower governance burden. This makes it clinically meaningful, yet often quicker to deploy than point‑of‑care decision support.
Short‑term pressure versus long‑term change
Reducing elective backlogs, improving throughput and meeting recovery targets remain immediate imperatives, so AI that delivers measurable productivity gains in months rather than years is easiest to justify.
By contrast, frontline digitisation programmes and clinically embedded AI demand sustained investment, workforce engagement and cultural change. Benefits accrue more slowly, and implementation can absorb scarce capacity — from funding and programme leadership to clinical and operational time — diverting resources from short‑term priorities.
clinical AI can scale where risk is well understood, benefits are clear
and assurance frameworks are mature — conditions not consistent across frontline settings
This challenge is familiar: balancing short‑term delivery with long‑term transformation.
Non‑clinical and hybrid AI should be treated as enablers, not end goals. Used deliberately, they can build data foundations, governance confidence and delivery capability for clinical digitisation; used narrowly, they risk entrenching optimisation rather than transformation.
NHS leaders must sequence investment so operational pragmatism accelerates, not postpones, the transformation needed for sustainable improvement.
