The NHS faces a critical funding challenge, with projected GDP growth unlikely to meet the resourcing required for future healthcare demands. With pressures from rising defence spending and a decade-long squeeze on public sector budgets, productivity-enhancing technologies, particularly AI, are viewed as essential to sustaining high-quality care. For AI to deliver on this promise, it must be developed, tested, and scaled rapidly.
Learning from global leaders such as Kaiser Permanente, Mayo Clinic, Johns Hopkins, Samsung Medical Centre, and Rigshospitalet, successful AI adoption shares common features: long-term investment in data infrastructure, innovation hubs for clinicians and vendors, a mix of in-house and vendor-led AI development, governance units overseeing real-world implementation, partnerships with tech giants, and built-in staff training. These institutions focus on AI applications that improve productivity, reduce administrative burdens, streamline patient pathways, enhance clinical safety, and support personalized medicine through data integration.
Evaluating AI benefits and risks requires evidence from real-world implementation. Initial steps involve ensuring AI models are accurate, safe, and usable in patient care. Subsequent evaluation considers clinical outcomes, return on investment, wider benefits or drawbacks, and model adaptations over time. Scaling AI across diverse contexts demands attention to local management, clinician engagement, work systems, procurement practices, and staff training. Traditional evaluation processes, while thorough, are often slow, taking years due to grant applications, ethics approvals, and multi-site coordination. Rapid evaluation teams, funded by NIHR, can shorten this timeline to 6–12 months, though in-service evaluations within NHS trusts, while quicker, often lack standardization and robust evidence, limiting their credibility.
Without credible, rapid evaluation, even highly accurate AI tools may fail to be adopted or spread, while overclaims from poor-quality studies risk misallocated investment, opportunity costs, and potential patient safety concerns. International examples, such as Kaiser Permanente’s short-term formative evaluation of ambient voice technology or NYU Langone’s rapid RCTs, show that robust, fast, real-world testing is possible. The NHS could adopt similar approaches to enable early-stage, implementation-focused evaluation, generating actionable evidence for broader rollout.
Context matters: AI adoption succeeds in sites with strong leadership, data infrastructure, and skilled clinical teams. Testing in such mature sites first helps isolate genuine signals of effectiveness, avoiding misleading results from less-prepared environments. Post-market surveillance and ongoing evaluation are also essential, given that AI applications evolve over time.
Ultimately, the NHS must modernize its evaluation and assurance methods to keep pace with rapid AI innovation. By using standardized, credible approaches in favorable contexts, technologies can be tested faster, scaled effectively, and integrated safely into patient care. Without this, progress risks relying on hope and overclaims rather than evidence, potentially wasting time and resources while the healthcare system faces mounting financial pressures.
