AI & health tech
AI in healthcare is a governance problem before it is a technology problem
The models are improving faster than the institutions deploying them, and trust, with regulators, clinicians and patients, is the real barrier to adoption. The companies that build governance discipline now reach the ward sooner, and stay there. I sit where NHS adoption decisions are made, on the committees and informatics boards that decide what reaches a patient. That is where I work.
A board-level test for clinical AI
The Five Questions
Is it clinically safe here?
Not "was it validated somewhere", but validated in a population like ours, on a pathway like ours, with failure modes we have mapped and can detect.
Is it fair?
Does performance hold across age, sex, ethnicity and deprivation? A tool that works brilliantly on average can still fail the patients who need it most.
Who answers when it errs?
Clear lines of clinical and corporate accountability before deployment, not worked out for the first time in a coroner's court.
Can we explain it to a patient?
If a tool influences a decision about someone's care, they are owed an explanation a reasonable person can understand. Consent is a process, not a checkbox.
Will it survive contact with the ward?
Workflow fit, workforce impact, maintenance and a business case that still stands in year three. Most clinical AI fails here, not in the model.
I use these questions in my committee work and when advising health-tech teams; they draw on my clinical governance roles and our Faculty-endorsed Delphi work on AI in intensive care.
Why me
Royal College of Anaesthetists
Committee member, Legal and Ethics Committee, appointed for expertise in the compassionate and ethical implementation of AI in healthcare.
AI demand-signalling in intensive care
Co-lead of a Delphi study on AI demand-signalling in intensive care, endorsed by the Faculty of Intensive Care Medicine, building national consensus on where AI genuinely helps.
Electronic patient record implementation
Member of the Clinical Informatics Board implementing a new trust-wide electronic patient record: the unglamorous reality of making digital work on real wards.
Founder-operator perspective
Director of London ITU Limited since 2010, with full P&L responsibility, and an Executive MBA with all taught modules complete. I understand the pressures on a health-tech business as well as on a hospital.
The regulatory window
Updated September 2026
The rules are moving on both sides of the Channel. In the UK, the MHRA's National Commission into the Regulation of AI in Healthcare reported on 10 September 2026: staged authorisation, with new tools deployed under supervision before full approval; continuous real-world monitoring for the whole working life of a device; public safety information, including incident data; and stronger enforcement powers for the regulator. In the EU, the high-risk obligations of the AI Act have been deferred, not cancelled: to December 2027 for standalone systems and August 2028 for AI built into medical devices.
Read together, the message to boards is plain. Approval will no longer be a single moment; it will be a supervised, monitored, publicly visible process, and someone will have to answer for it. That is exactly the ground The Five Questions covers: safety here, fairness across subgroups, named accountability, transparency to patients, and a plan that still stands in year three. The organisations that build this discipline now will move faster later, because trust is the real barrier to adoption.
Building or governing clinical AI?
Whether you're a founder seeking a clinical voice you can trust, an investor doing diligence, or a board that needs AI assurance, I'd welcome a conversation.