The Ground-Up NHS

A note on method

This paper is written by a doctor, not a department. It is built on published, verifiable sources: NHS England statistics, the NHS Staff Survey, General Medical Council workforce and workplace research, National Audit Office and parliamentary material, peer-reviewed evaluations, and government policy documents. Where a figure is a modelled projection rather than an observed statistic, it is labelled as such and its assumptions are stated in full. Nothing here is invented. Where the evidence is thin, this paper says so rather than filling the gap with confidence.

That last point matters more than it sounds. A great deal of NHS reform writing fails not because its ideas are bad but because it quietly inflates its numbers, and then collapses the first time someone with a spreadsheet reads it properly. The proposals below are deliberately costed conservatively and phased slowly enough to be real.

Executive summary

The National Health Service does not primarily have a money problem, a building problem, or an ideas problem. It has a systems and dignity problem. The people inside it are treated, in a hundred small daily ways, as though their time and comfort are worthless — and the patients inside it are treated as though the information held about them belongs to someone else.

Both of those are fixable, and neither requires a decade.

This paper sets out five pillars.

Pillar One — Dignity at work. Start with the material basics: food, rest, and the physical conditions of a twenty-four-hour job. National standards for healthcare food and drink already exist and are already contractual. They are unevenly delivered and barely enforced. The Secretary of State has an explicit statutory power to set minimum food and drink standards in hospitals. Use it, measure compliance publicly, and fund the gap. Cost: modest. Timeframe: eighteen months. Symbolic value: enormous.

Pillar Two — The end of the secret record. Communication is consistently the single largest category of written complaint about NHS hospital and community services — larger than complaints about clinical treatment itself in the way the data are grouped by subject. Meanwhile, forty-one million people are already registered on the NHS App, and government has already committed to a Single Patient Record viewable by patients from 2028. The missing layer is not the data. It is a conversational interface over the patient's own record, available to the patient and to their nominated relatives, that can answer "what is happening to my mother, and why". This paper specifies exactly what needs building, what it costs, and how to pilot it inside twelve months.

Pillar Three — AI triage as standard. From the 999 and 112 call, through 111, through the front door of A&E, and — critically — across the internal handovers between hospital teams that nobody currently measures. The national ambulance triage architecture rests on decision-tree systems designed decades ago. AI-assisted triage is already being piloted and already producing measurable results. The proposal is to make it the default, with clinician override always available, and to extend it to the internal handover, which is where a large share of avoidable harm actually occurs.

Pillar Four — Retention over bricks and mortar. The buildings are, for now, adequate. The system behind them is not. One in five doctors surveyed by the GMC indicated an intention to leave UK practice. Around a third report being unable to cope with their workload at least weekly. The proposals here concern training bottlenecks, consultant availability, psychological safety for resident doctors, and a specific, measurable programme of bureaucracy removal — not a slogan about bureaucracy, but a list.

Pillar Five — Global Britain, global London. Inbound private and international patient work is a genuine and underdeveloped revenue stream. It is currently worth substantially less than commonly claimed. This paper is explicit that a £2–3bn annual figure is a growth target from a low base, not money currently being ignored, and sets out what would have to be true for it to be reached.

Pillar Six — The hospital as a place, not a building. Large hospitals have the footfall, dwell time and captive attention of a major transport interchange, and are commercially managed like a storage cupboard. Retail units sit empty or are lost to planning inertia; advertising space is essentially unmonetised; there is no

national commercial estate function comparable to the one that generates hundreds of millions annually for Transport for London. This paper proposes a national NHS commercial estate body, a mandatory ethical advertising framework, and a realistic — not inflated — revenue trajectory.

Running underneath all five is a measurement architecture: continuous, automated computation of the outcome indicators that already exist — the Summary Hospital-level Mortality Indicator, morbidity and complication rates, readmissions, appraisal and revalidation data, complaint categories — so that every intervention in this paper is judged against what actually happened to patients, not against what a press release claimed.

And a final answer to a question that will be asked immediately: no, this does not require a Palantir-type company. Section 4.6 explains precisely why, and what it does require instead.

Part One — The diagnosis

1.1 What the staff data actually say

The NHS Staff Survey is one of the largest workforce surveys in the world. The 2025 results, published in March 2026, are the most recent full picture.

They are not a catastrophe. They are something more corrosive: a plateau. Most scores remained broadly stable against 2024, but of the twenty-one sub-scores tracked, none improved, and two worsened. The headline advocacy measure — whether staff would recommend the NHS as a place to work — fell to 58.05 per cent from 60.79 per cent the previous year. Indicators on workload, burnout and pressure were worse than the preceding year. Confidence that employers would take positive action on wellbeing fell, even as staff continued to report that their immediate line managers were supportive.

That last combination is the important one. It tells you the problem is not the ward sister or the clinical director. It is the layer above them: the institution's demonstrated capacity to convert stated concern into material change. Staff believe their manager cares. They do not believe the organisation does.

There is one genuinely good number worth naming, because credibility requires acknowledging improvement as well as failure: reported experience of discrimination from managers or colleagues fell to 8.77 per cent, the lowest recorded since the survey began, continuing a year-on-year downward trend. Something in the NHS is working. It is worth understanding what, because the same mechanism might be applied elsewhere.

1.2 What the doctor-specific data say

The GMC's workplace experiences research, drawing on its 2024 Barometer survey and published in 2025, found gradual improvement for a second consecutive year — burnout risk down from 25 per cent in 2022 to 18 per cent in 2024. Again: acknowledge the improvement.

But underneath it:

• 74 per cent of respondents said they were likely to make a change to their career in the next twelve months.

• 19 per cent indicated plans to leave the UK.

• Around a third reported being unable to cope with their workload at least once a week; 63 per cent regularly worked beyond rostered hours.

• 23 per cent had taken a leave of absence due to stress in 2024 — close to double the 12 per cent reported in 2019.

• Only around two in five felt able to progress their career in the way they wanted.

The Royal College of Physicians reported that only 44 per cent of respondents to its next generation survey were satisfied with their clinical training, with many citing lack of access to senior clinical and educational supervision. A third of those respondents intended to work abroad at some point; nearly half of that group attributed it to burnout or wellbeing concerns.

And a structural fact that ought to alarm anyone planning workforce policy: doctors who qualified outside the UK make up around 42 per cent of the workforce. In 2024, 4,880 such doctors left UK practice — a 26 per cent increase on the previous year's 3,869, and the first significant year-on-year rise since the pandemic. Joining numbers levelled off in the same period.

Set against this, the pay picture is real but not sufficient as an explanation. Resident doctors have received substantial uplifts over recent years, and the Nuffield Trust's analysis nonetheless estimates average real-terms earnings for 2025/26 still trailing 2010/11 levels by somewhere between 4 and 10 per cent. Systematic review evidence on why health professionals leave the NHS finds that pay was cited as inadequate in most included studies, but was the single most common reason for leaving in only one of them — with relatively small influence on turnover intentions in most others. The finding that recurs is that staff are concerned about fairness of pay relative to job demands and expertise, rather than about absolute pay levels.

That is a crucial distinction, and it is the empirical foundation for this entire paper. If pay were the dominant driver, the only lever would be the Treasury. Because it is not, there are levers a Health Secretary can actually pull.

General Medical Council, The state of medical education and practice in the UK: workplace experiences 2025. Stress-related leave is nearly double the 12% recorded in 2019.

1.3 What the patient data say

NHS England's collection of written complaints about hospital and community health services categorises complaints by subject. Across the periods published, communication is consistently the largest single subject area — around 17.7 to 18.5 per cent of complaints by subject in the quarters examined — ahead of patient care including nutrition and hydration at around 11.6 to 11.9 per cent. Clinical treatment accounts for a larger share when aggregated as a category, but communication as a discrete failure mode leads the individual subject areas.

Read that again in plain terms. The most common thing people formally complain about is not that the wrong thing was done to them. It is that nobody told them what was happening.

The Parliamentary and Health Service Ombudsman's own thematic work reinforces the point, noting that failures in communication with patients — and equally, communication between services — cause avoidable

distress and injustice. The NHS Constitution already pledges that decisions will be made in a clear and transparent way so that patients and the public can understand how services are planned and delivered. The pledge exists. The delivery mechanism does not.

A meaningful proportion of complaints that are not resolved locally are subsequently upheld by the Ombudsman, and unresolved complaints have a well-documented tendency to escalate into litigation. Every complaint that is genuinely about information rather than about clinical error is, in principle, addressable by a system that simply tells people what is going on.

NHS England Digital, Data on Written Complaints in the NHS. Clinical treatment is aggregated across all clinical specialties; communication is the largest single subject area.

1.4 What the money data say

Two figures frame the fiscal context.

First, the reorganisation. In March 2025 the government announced the abolition of NHS England and the transfer of its functions into the Department of Health and Social Care, with headcount across the two organisations expected to fall by around 50 per cent — the two bodies combined employed over 18,000 people — and reported savings of around £500 million. The NHS Modernisation Bill was introduced to Parliament in May 2026, with abolition targeted for around October 2026 and full reorganisation intended to conclude by approximately April 2027. The Institute for Government's assessment a year into the process was blunt: a great deal of time and money committed, benefits yet to materialise, and at one point a senior civil servant describing the process as chaotic.

Second, the technology investment. NHS England is rolling out AI tools including integrated ambient voice technology, backed by a £10 billion investment over three years, alongside NHS Online, the Single Patient Record, and access to Microsoft Copilot for more than 500,000 NHS staff following a trial that reportedly found an average saving of two days per month on administrative tasks.

This is the single most important strategic fact in this paper. The money for the digital layer has largely already been committed. The proposals in Part Two of this document are not primarily a request for new capital. They are a request to spend committed capital on a specific set of things, in a specific order, with a specific measurement regime attached.

1.5 The synthesis

Put the four evidence streams together and a coherent picture emerges.

Staff are not principally leaving over pay; they are leaving over demand, progression, supervision and the felt sense that the institution does not value them. Patients are not principally complaining about clinical error; they are complaining about not being told what is happening. Both of these are information and dignity failures, and both are structurally cheaper to fix than the things politicians usually promise to fix.

Meanwhile, a very large digital investment is already flowing, a national patient-facing app already reaches forty-one million people, and legislation to give patients default access to a unified record is already in the parliamentary pipeline.

The opportunity, in one sentence: the plumbing is being laid anyway; this paper is about what to run through it.

Part Two — Pillar One: Dignity at work

2.1 The case for starting with food

There is a temptation to treat hospital food as a trivial opening for a serious reform paper. It is the opposite. It is the correct opening precisely because it is small, visible, cheap and universally understood. Every member of NHS staff, every day, has a direct sensory encounter with the question of whether the institution regards their basic comfort as worth organising. If the answer at 3am on a night shift is a vending machine and a cold kettle, no amount of wellbeing strategy will land.

And this is not a fringe view. The Independent Review of NHS Hospital Food, chaired by Philip Shelley and published in 2020, addressed staff catering directly. It recommended funding for the upgrade of existing hospital kitchens and provision of ward kitchens so that a 24/7 service could be provided — from a hot drink and a snack in the middle of the night through to meals for new mothers on maternity wards and for patients hungry after a long pre-operative fast. It further recommended that where a 24/7 food service is not available for staff, they must have access to appropriate facilities to safely store, prepare and eat their own meals at any hour, with facilities to prepare hot drinks available to all staff.

The National Standards for Healthcare Food and Drink, published by NHS England in 2022 following that review, made this explicit: all staff and visitors should have equal opportunity to access food and drink that supports their nutrition and hydration needs 24/7, including drinking water. NHS organisations are legally required to meet the standards as part of the NHS Standard Contract. Organisations must have a designated board director responsible for food, with compliance reported at board level as a standing agenda item.

Over half of all food provided in NHS hospitals is served to staff and visitors, not patients. The NHS provides in the region of 199 million meals per year across all three groups. At the time the standards were published, around 60 per cent of trusts in England were assessed as compliant or in a good position to become so.

So the policy exists. The review exists. The standards exist. The contractual hook exists. The board accountability exists.

And staff still cannot get a decent hot meal at 3am in a large number of hospitals. That gap — between a well-designed policy and its lived non-delivery — is the actual subject of this pillar, and it is the template for almost everything else wrong with NHS reform.

2.2 Why it has not landed

Four reasons, and none of them are mysterious.

No published compliance data. Compliance with the food and drink standards is reported to trust boards. It is not systematically published, benchmarked, or ranked nationally in a form that any journalist, staff member or MP can look up in thirty seconds. A standard that is not publicly measured is a suggestion.

No funding attached to the capital requirement. The Shelley review recommended funding for kitchen upgrades and ward kitchens. Standards were subsequently mandated through the contract. A trust facing a deficit, told to meet a standard requiring capital it does not have, will meet the standard on paper.

Catering is outsourced and contracted on unit cost. Where catering is delivered under contract, the contractual incentive is cost per meal, not staff experience at unsociable hours. Providing a genuine overnight service is expensive per head because the head count at 3am is low. Nothing in the current structure pays for that.

The charging question is never confronted. Staff pay for canteen food. This is defensible in principle — it is a subsidised staff benefit, not a free entitlement — but it becomes indefensible when the food charged for is poor and unavailable at the hours the job actually requires. If the NHS is going to charge its own night staff for a meal, the meal has to be good and it has to exist.

2.3 The proposal

2.3.1 Publish compliance. Establish a national, quarterly, trust-level published dataset on compliance with the National Standards for Healthcare Food and Drink, disaggregated by standard, with the 24/7 staff access standard reported separately and prominently. Publish it on the same basis as other NHS statistics, with trustlevel rankings, in machine-readable form. This is the entire intervention for the first six months, and on its own it will move behaviour, because publication changes board attention in a way that internal reporting does not.

2.3.2 Add a staff-reported measure. Compliance self-assessment is not enough. Add three questions to the NHS Staff Survey, or to a quarterly pulse survey: whether hot food was available during the respondent's last night or weekend shift; whether they were able to take their break; and whether they consider the food value for money. Cross-reference self-assessed compliance against staff-reported reality and publish the discrepancy. Trusts claiming compliance while their own staff report otherwise should be flagged.

2.3.3 Fund the capital gap. Commission a national assessment of the capital cost of bringing every acute site to genuine 24/7 hot food and drink provision, including ward-level facilities. Fund it as a dedicated, ringfenced capital line rather than leaving it to compete inside a general capital envelope, where it will always lose to a scanner.

2.3.4 Fix the charging model. Establish a national maximum price for a defined "core staff meal" — a hot main meal meeting nutritional standards — available at every acute site at every hour. Subsidise the difference centrally. The subsidy is small in NHS terms and its symbolic yield is disproportionate.

2.3.5 Contract for availability, not unit cost. Where catering is outsourced, require future contracts to specify availability windows and staff satisfaction outcomes as performance measures, not solely cost per cover.

2.4 Ministerial powers required

This is the pillar where the legal position is cleanest.

The Health and Care Act 2022 expressly provides that minimum standards for food and drink provided in hospitals can be set. That is a direct, on-point statutory power. It does not require new primary legislation, a consultation-heavy reorganisation, or a negotiation with the Treasury beyond the subsidy line.

In addition, the Secretary of State's general powers under the National Health Service Act 2006 as amended by the 2022 Act — including the power of direction over NHS England and integrated care boards, and the power to set the mandate at any time and have it remain in force until replaced — provide ample mechanism for

requiring publication of compliance data. Directions made under these powers must be published, which is itself useful: the direction becomes a public commitment.

The abolition of NHS England and the consolidation of its functions within DHSC, once complete, arguably makes this easier rather than harder, because the number of bodies between a ministerial decision and a trust's board agenda reduces.

Powers required: existing. Primary legislation required: none. Secondary legislation required: possibly, for the pricing mechanism.

2.5 Cost and timeframe

All figures below are modelled estimates built on stated assumptions, not observed costs. They are presented as orders of magnitude for planning, and would require validation through the national assessment proposed at 2.3.3.

Timeframe:

• Months 0–3: Direction issued. Compliance dataset specification agreed. Survey questions drafted.

• Months 3–6: First publication of self-assessed compliance. National capital assessment commissioned.

• Months 6–12: First staff-reported cross-reference published. Capital assessment reports. Pricing mechanism consulted on.

• Months 12–18: Capital programme begins. Core staff meal pricing live at first cohort of sites.

• Months 18–36: Capital rollout across acute estate.

2.6 What success looks like, measurably

• Proportion of acute sites with verified 24/7 hot food access for staff: baseline to be established, target 95 per cent by month 36.

• Gap between self-assessed trust compliance and staff-reported reality: target below 10 percentage points by month 24.

• NHS Staff Survey score on "my organisation takes positive action on health and wellbeing": this fell in 2025 and is the natural outcome measure. Target: reversal of the decline within two survey cycles.

• Advocacy sub-score (recommendation of the NHS as a place to work): the 2025 figure of 58.05 per cent is the baseline.

A necessary caution on attribution. No honest analysis can claim that a food programme alone moves a composite morale score. Morale is multiply determined and confounded by pay disputes, industrial action, workload and national politics. The correct evaluation design is a stepped-wedge rollout — sites receiving the intervention in randomised sequence — which allows a genuine causal estimate rather than a before-andafter correlation. This paper commits to that design rather than to a claimed effect size, because a claimed effect size here would be a fabrication.

Part Three — Pillar Two: The end of the secret record

3.1 The principle

Healthcare should not be a secret.

Not from the patient, and not from the family sitting in the corridor at midnight who have been told that someone will come and speak to them and nobody has. The information exists. It is written down. It is stored in an electronic patient record that the patient has a legal right of access to under Article 15 of the UK GDPR and the Data Protection Act 2018. And yet the practical experience of being a patient or a relative in a British hospital is one of chronic informational deprivation.

The reason is not malice or secrecy. It is that the record is written in a language and format designed for clinicians, updated asynchronously, distributed across systems, and mediated by a workforce that does not have the minutes available to explain it. The information is theoretically accessible and practically opaque.

That is precisely and specifically the problem class that modern language models solve well.

3.2 What already exists

This is the good news, and it substantially reduces both cost and risk.

The NHS App. Around 41 million people are registered. More than 15 million users logged in during March 2026 alone — up by almost a third year on year. It already provides access to test results, prescriptions, patient records, patient messaging and appointment information. Every acute NHS trust in England is now linked, with around 64 per cent of hospital appointments visible in the app. Patients can view health documents including discharge summaries. Proxy access — accessing services for a child or someone you care for — already exists as a feature.

The Single Patient Record. The 10 Year Health Plan committed to unifying medical records into a single authoritative record. Legislation places a duty on providers, including GP practices, to make recorded information available to patients. From 2028 patients are expected to view SPR data through the NHS App. Clinicians will begin to get access to joined-up records as early as 2027 in some specialties, with piloting in maternity and frailty care. NHS England's stated design principles include role-based access control and audit trails on data access.

The AI investment. A £10 billion technology, digital and data investment over three years, explicitly including AI tooling and NHS Online.

Ambient voice technology. Already being deployed at scale, with a supplier registry launched in January 2026 listing 23 self-certified suppliers, and regional procurements underway — the Midlands procurement alone covering 70,000 clinicians.

Existing patient-authored content. In London, patients registered with London GP practices can now start their own Universal Care Plans through the NHS App, recording their wishes, preferences and support needs in their own words, and view clinical sections completed by professionals. This is a live, working precedent for exactly the transparency principle this pillar argues for.

The infrastructure question is therefore not "how do we build a national health data platform". It is: what specific, bounded software needs to be added on top of infrastructure that is already being built anyway?

3.3 The proposal: three agents

Agent One — the Patient Companion.

A conversational interface, inside the NHS App, operating strictly over the individual patient's own record. It answers questions in the patient's own language and at their chosen reading level: what does this result mean; what is the plan; why was this medication changed; what happens next; who is the consultant responsible for my care; when is the next scheduled review.

Critical design constraints:

• It is retrieval-grounded. Every substantive statement it makes cites the specific record entry it came from, with date and author role. It does not answer from general medical knowledge when asked about the patient's specific care.

• It refuses cleanly. Where the record does not contain the answer, it says so and offers to route the question to the clinical team as a logged query, rather than speculating.

• It does not give new clinical advice. It explains what is recorded. The boundary between explaining a recorded plan and issuing new clinical advice is the central safety line, and it must be enforced in the system design, not merely in a disclaimer.

• It escalates on risk signals. Distress, deterioration described by the patient, or safeguarding indicators trigger a defined human route.

Agent Two — the Family Access layer.

Extending the existing proxy access model, a patient may nominate named relatives or carers with granular, patient-controlled scope: a relative may be given access to "admission status, location, planned procedures and discharge planning" without access to, for example, sexual health or mental health history. Access is timeboxed to an episode of care by default. Every access is logged and visible to the patient.

This addresses a specific and enormous source of distress: the relative who cannot get through on the ward phone and does not know whether their mother has had her scan.

Where the patient lacks capacity, the layer must operate under the Mental Capacity Act 2005 framework, with access granted on a best-interests basis through defined governance rather than automatically. This is a genuinely difficult area and the design must be built with, not merely reviewed by, capacity law specialists.

Agent Three — the Handover Agent.

This is the one nobody is building, and clinically it is probably the most valuable.

When care transfers — ward to ward, team to team, day team to night team, hospital to community, ambulance to emergency department — a structured summary is generated from the record: active problems, outstanding investigations, pending results, unresolved decisions, escalation plans, and explicitly flagged open loops. Not a replacement for handover. A scaffold for it, and a persistent, auditable record of what was and was not communicated.

The clinical rationale is direct. The Ombudsman's casework includes cases where a patient came to serious harm because a critical element of their treatment was not carried across a ward transfer. Transitions are where information dies. An agent whose entire job is to notice what did not get carried across is addressing a welldocumented, well-understood failure mode.

3.4 What specifically must be built

This is deliberately granular, because "add AI to the NHS" is not an implementable instruction.

Component 1 — Record retrieval and grounding service. A retrieval-augmented generation layer sitting over the Single Patient Record and, in the interim, over trust electronic patient records via existing interoperability standards (FHIR UK Core). Its function is to fetch the specific record fragments relevant to a query and constrain generation to those fragments. This is the technical heart of the system and the main source of safety assurance: hallucination risk is managed primarily by architecture, not by prompt instruction.

Component 2 — Clinical language translation model. Fine-tuned for translating clinical documentation into plain language at controlled reading levels, in the languages actually spoken by NHS patients. It must be evaluated against clinician-adjudicated ground truth for accuracy of meaning, not merely readability scores.

Component 3 — Consent and scope engine. The permissions layer governing who sees what, for how long, under what legal basis, with full audit. This is the component most likely to be underestimated and most likely to cause the programme to fail politically if it is done badly. It should be built first.

Component 4 — Escalation and query routing. When the agent cannot answer, the question must go somewhere and be answered by someone, with a service standard. An agent that generates unanswerable queries and dumps them on ward staff will be sabotaged by ward staff within a fortnight, entirely rationally.

Component 5 — Handover summarisation service. Structured extraction of open loops and pending items at transition points, integrated with EPR workflow rather than bolted on beside it.

Component 6 — Evaluation harness. Continuous automated evaluation against held-out clinician-adjudicated cases, with published accuracy and safety metrics. This should be published, not held internally.

Component 7 — Clinical safety case. Compliance with DCB0129 and DCB0160 clinical risk management standards for health IT, the Digital Technology Assessment Criteria, and — where the tool crosses into a medical device function — MHRA regulation of software and AI as a medical device. The MHRA's Software and AI as a Medical Device Change Programme is the relevant framework, and its classification questions must be settled before, not after, procurement.

3.5 Cost and timeframe

Modelled costs. These are engineering estimates based on comparable public-sector digital programmes and stated assumptions. They exclude the underlying SPR programme cost, which is already committed and funded separately.

For context, this sits inside the announced £10bn three-year technology investment at a level of roughly half of one per cent of it. It is comparable in order of magnitude to a single mid-sized national IT programme, and an order of magnitude below the historic National Programme for IT. It is also, deliberately, not a single monolithic contract.

Timeframe:

• Months 0–6: Consent and scope engine specification and build begins. Clinical safety case initiated. Regulatory classification settled with MHRA. Public engagement launched — this cannot be an afterthought.

• Months 6–12: Patient Companion pilot in two to three trusts, read-only, single specialty, adult patients with capacity, opt-in. Handover Agent pilot in one trust, one directorate.

• Months 12–18: Evaluation published. Family Access layer piloted with capacity safeguards. Expansion to five to eight trusts.

• Months 18–30: Regional rollout aligned to SPR deployment in maternity and frailty pilots.

• Months 30–48: National availability, tracking the 2028 SPR patient-access milestone.

The programme deliberately rides the SPR timetable rather than racing it. Attempting to deliver patientfacing conversational access to a unified record before the unified record exists is how NHS IT programmes historically fail.

3.6 Do we need a Palantir-type company?

No. And the reasoning matters, because this question will dominate any public discussion.

What the current arrangement is. NHS England contracted for the Federated Data Platform, underpinned by Palantir technology, at a value reported at £330 million, with a maximum term to a potential end date of March 2031 and a break clause review scheduled for spring 2027. The programme sits within the Government Major Projects Portfolio and has been assessed by the National Infrastructure and Service Transformation Authority. NHS England's contract terms prohibit the supplier from commercialising or marketing NHS data, including on an anonymised basis, and from using NHS data to develop new supplier products, including training AI models on it.

What the criticism is. A parliamentary debate led by Liberal Democrat MP Martin Wrigley raised the argument that the arrangement is a subscription service leaving no software, no improvements and no intellectual property with the NHS after expenditure exceeding £330 million, and questioned usability and the proportion of user organisations genuinely benefiting. A campaign backed by health charity Medact and supported by a coalition including the Good Law Project, Privacy International and Amnesty International has urged trusts not to implement the platform, citing risks to patient trust. Over 47,000 patients have signed opposition. Government has indicated it would reconsider the contract if competitors offered better terms.

Why none of this is the deciding question for the proposals in this paper. The three agents described above do not require a proprietary federated data platform. They require:

1. A record to read from — being built as the SPR, under public ownership.

2. Standards-based interoperability — FHIR UK Core, already a national standard.

3. A retrieval and generation layer — buildable on commodity infrastructure, with model inference procurable competitively and swappable.

4. A permissions and audit engine — which should be built and owned in-house, because it is the component that encodes public trust and must never be a black box in someone else's subscription.

The strategic principle this paper adopts is: own the record, own the permissions, own the audit; buy the compute; rent the models; and never sign a contract that leaves no intellectual property behind.

That is not an ideological position about any particular vendor. It is a straightforward lesson from the criticism above, and it happens to be cheaper.

3.7 Predicted impact — and honest limits

What can be predicted with reasonable confidence:

Communication-related complaints are around 17.7–18.5 per cent of complaints by subject. If a functioning patient and family information layer reduced the communication-attributable share by a third — an assumption, clearly labelled — that would represent a meaningful reduction in total complaint volume, with associated reductions in local resolution cost, Ombudsman referral, and the well-documented escalation pathway from unresolved complaint to litigation.

What cannot honestly be predicted:

Any claim that this intervention will reduce mortality by a specific percentage would be fabrication. The mechanism by which better information changes mortality runs through improved handover completeness, earlier escalation and fewer missed results — all plausible, none quantified in advance.

Therefore the paper commits to measurement, not to a promised number. The evaluation design is:

• Primary outcome: communication-category complaint rate per 1,000 admissions, compared across stepped-wedge implementation sites.

• Secondary outcomes: proportion of handovers with completed open-loop documentation; time from result availability to result acknowledgement; unplanned readmission within 30 days; patient-reported understanding of care plan at discharge.

• Safety outcomes: rate of agent statements adjudicated as clinically inaccurate by blinded clinical review; rate of failed escalations; complaints about the agent itself.

• Mortality: tracked via SHMI as a contextual signal, not an attributed outcome, for reasons set out in Part Six.

Part Four — Pillar Three: AI triage as standard

4.1 The current architecture and why it is ageing

Urgent care decisions via NHS 111 and 999 are largely made using NHS Pathways, a decision-tree model designed in 2005 to help call handlers and clinicians establish a caller's condition and appropriate destination. For 999 calls, the decision on whether something is an emergency may be made using the Advanced Medical Priority Dispatch System alongside NHS Pathways — AMPDS being a structured question set originally designed in the 1970s.

The Tony Blair Institute's analysis of this landscape identified the core structural problem clearly: GPs, 111 and 999 approach the same presenting problem differently, using various tools across various entry points, producing incompatible and inconsistent assessments — causing duplication, delays, increased staff workload and poor patient experience. Public health campaigns urging Pharmacy First run alongside risk-averse 111 algorithms that direct people to GPs and A&E. The systems rely heavily on call handlers applying decision aids that are inflexible, impersonal and difficult to change.

That last phrase — difficult to change — is the crux. A decision tree encodes clinical judgement in a form that requires a governance process to update. A well-governed model-based system can be updated, evaluated and re-deployed on a far shorter cycle, provided the evaluation infrastructure exists.

4.2 What is already moving

Government policy is already pointed this way. The Urgent and Emergency Care Plan 2025/26 committed to an evidence-based clinical review of categorisation and to increasing clinical triage of 999 calls to identify patients who can be safely assessed and managed remotely, directing them to appropriate urgent care pathways. It committed to using natural language processing technology to improve call streaming from October 2025, and to developing new measures reflecting the quality of care provided by 111 — including disposition outcomes, not merely call answering speed — with the aim of implementing these from April 2026.

On the ground: South East Coast Ambulance Service ran an AI pilot scheduled to February 2026 as part of wider testing within the ambulance setting including London Ambulance Service and South Western Ambulance Service, aimed at time savings for clinicians and a consistent clinical record shared with other professionals.

And in primary care, an AI triage tool in the NHS App is expanding to more than 200,000 patients after a Sussex GP trial reduced phone queues by 29 per cent while maintaining patient satisfaction, with rollout intended to reach all users by 2028.

The direction is set. What is missing is a coherent specification covering the whole pathway, and — critically — the internal hospital handovers that nobody is currently treating as a triage problem at all.

4.3 The proposal: triage as a continuous layer, not a set of front doors

Layer One — Emergency call intake (999 / 112). AI-assisted structured intake operating alongside, not instead of, the call handler. Its functions: real-time extraction of clinical features from free speech, prompting for the discriminating question the handler has not yet asked, flagging inconsistency between stated category and described symptoms, and generating a structured clinical record transferable to the receiving crew and hospital. Category assignment remains a human decision with AI recommendation, not the reverse. This distinction is not a technicality; it is the entire liability and safety architecture.

Layer Two — 111 and NHS App self-triage. Consistency of assessment across entry points is the goal. The same presenting complaint should reach the same disposition whether entered by phone, online, or in-app. Achieving that requires a shared underlying assessment model rather than three separate rule sets, which is the single largest structural gain available.

Layer Three — Emergency department front door. AI-assisted triage at presentation, integrating the prehospital record where one exists, the patient's own record, and the presenting complaint. The specific target is reduction of the gap between arrival and first clinically meaningful decision.

Layer Four — Inter-team and inter-console handover. The internal transitions. Medical team to surgical team. Emergency department to acute medical unit. Day team to night team. Hospital to community and to primary care. This is the Handover Agent described in Part Three, applied as a triage function: what is outstanding, what is unresolved, what is deteriorating, and who owns it now.

Layer Four is where this paper diverges most from current policy, and it is the layer that most directly addresses the communication-failure evidence in Part One.

4.4 Cost, timeframe and pilot design

Modelled costs — again, estimates against stated assumptions.

Pilot design — and this is the section that matters most.

Triage is a safety-critical function. A poorly evaluated triage change kills people. Therefore:

• Shadow mode first. For a minimum of six months at each layer, the AI recommendation is generated and logged but not shown to the operator. Its recommendations are compared retrospectively against human

decisions and against patient outcomes. Under-triage rate — the proportion of patients assigned a lower acuity than their outcome warranted — is the primary safety metric, and no system proceeds past shadow mode with an under-triage rate exceeding current human performance.

• Then assist mode. The recommendation is shown, the human decides, and disagreements are logged and reviewed. This is where the majority of the learning happens.

• Never autonomous mode, at any layer, in this programme. No disposition decision is made without a human. This is a design commitment, not a transitional caveat.

Timeframe:

• Months 0–6: Governance framework, evaluation infrastructure, regulatory classification with MHRA, and shadow-mode deployment at two ambulance trusts and three emergency departments.

• Months 6–12: Shadow mode data analysis; published interim safety report.

• Months 12–18: Assist mode at pilot sites with independent oversight; 111/App consistency layer pilot.

• Months 18–36: Staged national rollout conditional on published safety thresholds being met at each stage.

• Ongoing: Continuous published performance monitoring, with automatic rollback triggers.

4.5 Predicted impact

Where prediction is defensible. The Sussex GP trial's 29 per cent reduction in phone queues is an observed result in a specific setting. It is reasonable to model a range of 10–30 per cent reduction in call-handling time pressure at 111 and primary care access points, with wide uncertainty and heavy dependence on local implementation quality.

Where prediction is not defensible. Any claim about lives saved through improved triage, in advance of shadow-mode data. The honest position is that under-triage reduction is the mechanism by which triage improvement saves lives, that under-triage is directly measurable in shadow mode, and that the programme should therefore publish its measured under-triage differential rather than forecast a mortality effect.

Part Five — Pillar Four: Retention, hierarchy and bureaucracy

5.1 The thesis

The bricks and mortar are, for the immediate term, adequate. There is a real capital backlog and it matters, but it is not the binding constraint on the next three years of NHS performance. The binding constraint is the system running inside those buildings: how many trained people stay, how quickly decisions get made, how available senior clinical judgement is at the point it is needed, and whether a junior doctor at 2am feels able to pick up the phone.

5.2 The training bottleneck

This is now the most acute retention issue and it is under-appreciated outside the profession.

In October 2025, the BMA reported that in a ballot with a 65 per cent turnout, 97 per cent of newly qualified Foundation Year 1 resident doctors in England endorsed the option of striking over unemployment and training place shortages — not over pay. Analysis has repeatedly identified growing doctor unemployment, fewer training opportunities and added workplace stress as issues running deeper than the pay dispute. Only around two in five doctors in the GMC's research felt able to progress their career in the way they wanted.

Government has fast-tracked legislation to prioritise UK medical graduates and doctors with significant NHS experience for jobs, with the stated effect of halving competition ratios for training places. That is a real intervention. It also carries the risk identified by the GMC: policies that inadvertently demoralise or drive out internationally qualified doctors, who make up around 42 per cent of the workforce and whose departure rate rose 26 per cent in 2024.

The proposal: treat training capacity as infrastructure and publish it as such. A national, specialty-level, publicly available dataset of training places against applicants, updated annually, with a five-year forward commitment on numbers by specialty. Doctors make life decisions — mortgages, families, countries — on a five-year horizon. The absence of forward visibility is itself a driver of emigration, independent of the underlying numbers.

Second, an explicit retention impact assessment attached to any workforce policy affecting internationally qualified doctors, published before implementation. Given the 42 per cent dependency, this is elementary risk management.

5.3 Consultant availability and the culture of the phone call

Two related problems, one cultural and one structural.

The RCP's next generation survey found that lack of access to senior clinical and educational supervision was affecting learning and confidence, with only 44 per cent satisfied with clinical training. The GMC found doctors in training to be the group at highest risk of burnout.

A resident doctor who is reluctant to call a consultant at 2am is a patient safety hazard, and the profession has known this for thirty years. It is not solved by telling juniors to be braver. It is solved by three structural changes:

Named, rostered, unambiguous escalation. Every ward, every shift, one named consultant whose escalation availability is a rostered, remunerated duty rather than an informal expectation. The name is visible in the electronic record and to the patient.

No-fault escalation as an explicit standard. Escalation should never be a judgement on the escalator's competence. This should be written into the Staff Standards framework being developed nationally, and — crucially — measured, via a specific question in the staff survey on whether respondents felt able to escalate concerns to senior clinicians without fear.

Consultant job planning that counts supervision as work. Supervision that is not in a job plan is unpaid, unprotected and first to be dropped under pressure. Making it a counted, funded activity is the only mechanism that reliably works.

5.4 Bureaucracy: a list, not a slogan

The abolition of NHS England was justified in terms of duplication and bureaucracy, with the Health Secretary noting that NHS England staff and admin budget had exceeded £2 billion and describing duplicated communications and strategy functions across the arms-length body and the department. Headcount reduction of around 50 per cent across the two organisations, with reported savings around £500 million, was announced. Integrated care board management budgets were to reduce by half.

A year in, the Institute for Government's assessment was that benefits had yet to materialise and that the reorganisation itself was consuming substantial time and money.

This paper takes a deliberately different approach to bureaucracy reduction, because the central-body headcount approach has now been tried and its results are contested.

The proposal: measure clinician administrative burden directly, and target it.

• Establish a national measure of clinician administrative hours per clinical hour, sampled through timeand-motion methodology at a representative panel of sites, published annually. Until this is measured, "reducing bureaucracy" is unfalsifiable.

• Institute an annual mandatory data burden review: every national data return, mandatory training module and reporting requirement must be re-justified annually by a named owner against a stated use. Anything whose output cannot be shown to have been used in a decision in the preceding twelve months lapses automatically. Sunset by default, renewal by justification.

• Establish a frontline veto pilot: a mechanism by which a defined threshold of clinicians at a site can trigger mandatory review of a local administrative requirement, with the burden of proof on the requirement's owner.

The evidence base for the direction is strong: the Microsoft Copilot trial across NHS staff reportedly found an average saving of two days per month on administrative tasks, and ambient voice technology evaluations have found a 23.5 per cent increase in direct patient interaction time and an 8.2 per cent reduction in overall

appointment length. A pilot in the emergency department at St George's Hospital in Tooting found savings averaging 47 minutes per clinician per shift — enough for each staff member to see an additional patient per shift. Great Ormond Street-led work modelled that scaling ambient voice technology to over 11,000 A&E clinicians in England could create space for over 9,000 extra A&E consultations each day.

Those are substantial, credible, published numbers. But note the Nuffield Trust's caution, drawn from the first phase of a national NIHR-funded evaluation: adoption is accelerating faster than robust evidence, and the next evaluation phase is designed to test not merely whether time is saved but how implementation reshapes work and whether anticipated benefits are actually realised. This paper takes that caution seriously. Deploy at pace; evaluate at equal pace; publish either way.

5.5 Ministerial powers required

• Training numbers and forward commitments: exercisable through the mandate and through direction, with funding routed via existing education and training budgets. No primary legislation.

• Staff standards including escalation culture: the Staff Standards framework is already in development nationally as a mechanism for consistent, mandatory expectations on working conditions. Escalation safety can be embedded within it.

• Consultant job planning: contractual, requiring negotiation with the BMA rather than legislation. This is the slowest element in this pillar and should be started first.

• Data burden sunset review: exercisable by direction over NHS bodies. Where returns are required by other regulators, cross-government agreement is needed.

• Retention impact assessments: administrative, immediate.

5.6 Cost, timeframe and measurement

Modelled cost is low relative to the other pillars because most of it is governance rather than technology.

Item Modelled cost

National administrative burden £4–7m build, £3–5m annually measurement programme

Training place expansion (indicative, per £180–260m annually at full run-rate, based on published training cost ranges 1,000 additional places) — requires separate detailed costing

Consultant job plan supervision uplift Subject to negotiation; indicative £120–300m annually depending on scope

Data burden review secretariat Under £3m annually

Timeframe: governance elements are deliverable within 12 months. Training expansion runs on a 5–10 year pipeline by definition. Contractual changes run on a 12–24 month negotiation cycle.

Measurement: GMC Barometer intention-to-leave figures; NHS Staff Survey advocacy and morale subscores; HCHS doctor leaver counts, against the baseline of 20,286 secondary care doctors leaving NHS

organisations in the year to March 2025 compared with 15,577 in the year to March 2015; and the new administrative-hours measure.

Part Six — Pillar Five: Global Britain, global London

6.1 Getting the numbers right first

There is a widely repeated claim that the UK is leaving billions per year on the table in international private healthcare. This paper's position is that the underlying opportunity is real, that the commonly cited figures are not, and that a policy built on inflated numbers will be destroyed on contact with scrutiny.

What the evidence actually shows.

The most substantial academic assessment of UK medical tourism, published in PLOS ONE, found that the UK was a net exporter of medical travellers: in 2010, an estimated 63,000 UK residents travelled abroad for treatment while around 52,000 patients came to the UK. Total spending by incoming medical tourists — treatment plus associated tourism expenditure — was estimated in the range of £397 million to £544 million per annum.

The same body of work found inbound medical tourists treated as private patients within NHS facilities to be notably profitable relative to UK private patients, yielding close to a quarter of revenue from only 7 per cent of volume in the data examined. Great Ormond Street Hospital reported income exceeding £20 million from 656 international private patients, an average of approximately £31,600 per patient. University College London Hospitals, by contrast, saw more than double the patient numbers at 1,881 but generated only £1.53 million — an average of around £813 per patient, reflecting a very different case mix.

A government-published assessment noted that private healthcare in London is a major generator of jobs and prosperity, with the medical tourism market bringing over £500 million a year to the London economy at that time, with projected growth, and with London School of Hygiene and Tropical Medicine estimates of around £219 million in additional tourism spending by medical travellers and their families annually.

The separate and distinct issue: cost recovery. This is frequently and unhelpfully conflated with revenue generation. Policy Exchange freedom-of-information work found that between 2021 and 2024, NHS trusts in England invoiced £384,245,201 to overseas patients, with unrecovered charges over the three years exceeding £250 million and a national average recovery rate of 39 per cent. The National Audit Office has previously found trusts recovering around half of amounts charged directly to patients, with wide and poorly understood variation.

These are two different policies. Recovering money owed is a debt and administration problem. Attracting international private patients is a market development problem. Confusing them produces bad policy and bad politics.

• Observed today: £397m–£544m total inbound medical travel spending
• Modelled year 10: 8–12% compound growth with sustained investment
• The £2–3bn ambition: A 15–20 year strategic objective

Observed figures: Hanefeld et al., PLOS ONE. Projections are modelled compound-growth illustrations, not forecasts.

6.2 What a credible target looks like

Against a base in the order of £400–550 million annually in total inbound medical travel spending, a £2–3 billion annual figure represents roughly a four- to six-fold increase. That is not impossible — Dubai, Singapore and Seoul have achieved comparable growth trajectories from deliberate national programmes — but it is a fifteen-to-twenty-year strategic objective, not a spending review line.

The honest framing, and the one that is far more politically durable: the UK has world-leading clinical institutions and captures a fraction of the international patient market that its quality would justify. That is a strategic failure worth correcting, and correcting it takes a decade.

Stated that way, it is unarguable. Stated as "£3 billion available next year", it is refutable in an afternoon.

6.3 The proposal

A national international patient office. Currently, an international patient seeking UK treatment navigates individual trust private patient units with no single point of entry, inconsistent pricing, and no coordinated marketing. A single national front door — routing enquiries, standardising quality assurance and handling visa and logistics coordination — is the foundational infrastructure.

Standardised, published pricing. International patients pay a premium; that is appropriate and legal. Opaque pricing suppresses demand and invites accusations of profiteering. Published tariffs do the opposite.

Ring-fenced reinvestment with published accounts. The political vulnerability of this pillar is the accusation of privatisation by the back door and of international patients displacing NHS patients. The answer must be structural, not rhetorical: every pound of international private income published, ring-fenced, and demonstrably reinvested in NHS capacity, with published data showing that international activity used capacity additional to, not substituted for, NHS activity. Without that evidence trail, this policy is politically dead and deserves to be.

Capacity discipline. International private work must be delivered in additional capacity — additional theatre sessions, additional consultant programmed activities — not by displacing NHS lists. This should be contractually enforced and audited.

Fix cost recovery separately and unglamorously. Improved identification at point of registration, standardised invoicing, and centralised debt recovery so that individual trusts are not each running a small, ineffective collections operation.

6.4 Ministerial powers, constraints and timeframe

The private patient income cap for foundation trusts was substantially altered under the Health and Social Care Act 2012, and trusts have significant existing freedom in this area. The principal ministerial levers are therefore the mandate, direction, national tariff and guidance, plus cross-departmental work with the Home Office on visa routes and with the Department for Business and Trade on international marketing.

Powers required: largely existing. Primary legislation: not required for the core proposals.

Timeframe: national office establishment 12–18 months; cost recovery improvement 12–24 months; market development on a 5–15 year horizon with annual published progress against a stated trajectory.

Modelled financial expectation. With sustained investment in the order of £30–60 million annually in national coordination, marketing and capacity, and assuming compound growth of 8–12 per cent annually from a base of approximately £500 million, inbound medical travel value would reach roughly £1.1–1.5 billion by year ten and £2.3–3.9 billion by year twenty. These are illustrative compound-growth projections, not forecasts, and are highly sensitive to global competition, exchange rates, geopolitical patient flows and UK capacity constraints. They are presented to show what would have to be true for the ambition to be realised.

Part Six-B — Pillar Six: The hospital as a place of commerce

6B.1 The thesis

A large district general hospital has the daily footfall of a mainline railway station, the dwell time of an airport terminal, and the emotional attention of neither. People arrive early. They wait. They wait with relatives. They come back weekly for six weeks of radiotherapy. Thirteen hundred staff finish a shift and want a coffee. And the commercial expression of all that human traffic is, in most hospitals in England, one tired café that shuts at four, a vending machine, and a noticeboard.

Transport for London's draft 2026-27 budget plans for £184.2 million in advertising income and £114.9 million in property income — around £300 million annually from commercial exploitation of a transport estate. That is a serious, professionalised commercial function with a dedicated property company. The NHS estate is larger, its footfall is enormous, its dwell times are far longer, and it has no equivalent function at all.

Meanwhile, NHS trusts collected £242.8 million in car parking fees in 2023/24, spending £77.2 million running the car parks — meaning the single most politically toxic form of income generation available to hospitals is also, in practice, close to the only one being systematically pursued. That is a strategic failure. Hospitals are extracting money from patients in the least popular way possible while leaving the most acceptable forms untouched.

6B.2 The Wexham Park case — and an important correction

Wexham Park Hospital in Slough, run by Frimley Health NHS Foundation Trust, is frequently cited as an example of hospital retail failure. The current position is in fact the opposite, and it is a better example than the one usually told.

Frimley Health announced that Wexham Park would gain a new Costa Coffee, a Co-op, a Chaiiwala streetfood café with a halal menu, and a 24/7 smart shop, with the existing Crossroads Café refreshed and rebranded as a Costa Proud to Serve outlet — explicitly as part of the trust's five-year strategy to improve comfort, choice and access to affordable food throughout the day and night. This sits inside a wider £50 million-plus modernisation programme at the site, alongside more than £1 billion awarded to Frimley Health under the New Hospital Programme.

That is precisely the model this pillar advocates: a halal street-food operator alongside a national coffee chain alongside a 24/7 convenience offer, chosen for a diverse local population, open at the hours the hospital actually operates. The transition has a real cost — the long-standing Royal Voluntary Service partnership ends, and the trust has said it is considering new locations for charity bookshops and an alternative to the ward trolley service. That loss is not trivial and should be planned for, not waved away.

The lesson to draw is therefore not "planning killed the Costa". It is that where a trust treats its retail estate as a deliberate five-year strategy with board attention, it can transform the offer inside a single planning cycle — and that almost no trust does this, because no one is required to, no one is measured on it, and no one has the commercial expertise in-house.

6B.3 Why hospital retail underperforms

No national commercial function. Each trust negotiates its own leases with no benchmarking, no framework agreements, no shared covenant strength and no professional retail asset management. A single hospital negotiating with a national coffee chain has no leverage. Two hundred hospitals negotiating together have a great deal.

Planning treated as an obstacle rather than a workstream. Hospital retail changes typically require change of use, signage consent, servicing arrangements and sometimes highways input. Where a trust has no dedicated commercial development capability, a straightforward consent becomes a two-year drift. This is what people mean when they blame planning: usually the problem is not the planning authority's decision but the absence of anyone whose job it is to run the application properly.

The income generation rules are poorly understood. NHS bodies have long-standing statutory income generation powers, but with a real constraint: income generation schemes must be profitable, because NHS funds cannot be diverted from patient care to subsidise loss-making schemes. This is sensible, and it also means a trust cannot easily take a commercial risk or absorb a fit-out cost against future rent — which is exactly the transaction that would unlock most sites. A national body with a balance sheet can do what an individual trust cannot.

Charitable and voluntary occupancy is unaccounted. Space occupied by voluntary services is genuinely valuable and should not be treated purely as foregone rent, but it is currently not valued at all — which means it is neither properly protected nor properly costed.

6B.4 Advertising: the untouched asset

Hospital advertising is close to zero-monetised in England, and the reasons are partly good and partly inertia.

The good reasons are real. A hospital advertising estate carries obligations that a bus shelter does not. Restrictions on the advertising of less healthy food and drink products came into force on 5 January 2026, prohibiting HFSS products from daytime television and on-demand services and from paid online media at any time. Local authorities across England — and Transport for London through its own advertising policy — already restrict HFSS promotion on their estates. Wales has committed to restricting unhealthy food and drink advertising, sponsorship and promotion in public spaces including hospitals by 2030.

An NHS hospital that runs a public health campaign on one wall and a fried chicken advertisement on the opposite wall has destroyed its own credibility. So the ethical framework is not an optional overlay. It is the product. A guaranteed clean-content advertising environment is genuinely scarce and commands a premium, which is why this pillar argues the restrictions make the asset more valuable, not less.

The proposed exclusion list — mandatory, national, non-negotiable:

• Tobacco, vaping and nicotine products in all forms

• Alcohol

• Gambling, including free-to-play and skill-gaming variants

• Food and drink classified as HFSS under the UK Nutrient Profiling Model

• Weight-loss products, cosmetic surgery and aesthetic medical services

• Private healthcare services in direct substitution for the NHS treatment the patient is attending for

• Pharmaceuticals and prescription-only medicines, in line with existing regulation

• Legal services advertising for clinical negligence claims

• High-cost short-term credit

• Political advertising

• Anything targeting children in paediatric areas beyond a defined safe list

What is left, and it is a great deal: local businesses and services, employers recruiting, universities and colleges, banks and utilities, telecoms, automotive, home services, travel, insurance, transport operators, culture and events, technology and consumer electronics, non-HFSS food and drink, charities, and public information.

On the specific suggestion of a brand-sponsored refreshment offer — a "Samsung-sponsored coffee", or similar — the principle is sound and it is how airports, stadiums and stations already work. The constraints are that sponsorship must not confer any influence over clinical services or procurement, must be transparently declared, must comply with public sector conflict-of-interest and Nolan principles, and must not create the perception that a commercial partner has purchased standing within a clinical setting. Sponsorship of comfort — free hot drinks in waiting areas, charging points, seating, wifi — is defensible and welcome. Sponsorship of care is not. That is a bright line and it should be written into the framework, not left to case-by-case judgement.

6B.5 The proposal

6B.5.1 Establish NHS Places. A national commercial estate body for the NHS in England, modelled on the arm's-length property company structure that manages TfL's commercial estate. Its functions:

• Portfolio-level negotiation with national retail, food and beverage operators, using aggregate covenant strength.

• Framework agreements that a trust can call off in weeks rather than tendering from scratch over years.

• A balance sheet capable of funding fit-out against future rental income — solving the profitability constraint that currently blocks individual trusts.

• In-house planning and development capability, so consent applications are run by people who do this professionally.

• Advertising estate management under the national ethical framework, sold as a single national inventory rather than site by site.

• Mandatory reinvestment of net income into the originating trust, published and auditable.

6B.5.2 Digital advertising infrastructure. Digital display allows dynamic content, dayparting, and — critically — instant reallocation to public health and operational messaging. A digital estate that carries commercial content for eighteen hours and wayfinding, waiting-time information, public health campaigns and emergency messaging for the rest is a materially better hospital, independent of revenue. During a major incident, a national digital estate across every acute hospital is a public communications asset of real value.

6B.5.3 Retail as a clinical and workforce intervention. This connects directly back to Pillar One. A 24/7 convenience and hot food offer is simultaneously a commercial asset and the delivery mechanism for the staff food standard. The Wexham Park model — Costa, Co-op, a culturally appropriate street-food operator and a 24/7 smart shop — does both at once. Commercial strategy and staff dignity are the same programme, and should be run as one.

6B.5.4 Protect the voluntary sector explicitly. Any commercial estate strategy must contain a guaranteed floor of space for charitable and voluntary services at peppercorn rent, published as a national standard. The Wexham Park transition, where the RVS partnership ended, is exactly the outcome that generates justified local anger and must be designed against.

6B.5.5 Rebalance away from car parking. As commercial and advertising income grows, reduce car parking charges proportionately, publish the trade-off explicitly, and let the public see that the hospital chose to make money from advertisers rather than from visitors. This converts a revenue programme into a popular one.

Modelled projection against Transport for London draft budget 2026-27 (£184.2m advertising + £114.9m property). Not a forecast.

6B.6 The revenue question — and why £1 billion is the wrong number to publish

This is the section that will determine whether this pillar is taken seriously.

The benchmark. Transport for London — the entire commercial estate of the busiest transport network in the country, with a dedicated property company and decades of professional asset management — plans for approximately £300 million annually in combined advertising and property income.

Hospital footfall and dwell time are substantial, but hospital advertising audiences are lower-value per impression than transport audiences in advertiser terms: the audience is distressed, less receptive, more restricted in permitted categories, and concentrated in locations where high-impact formats are inappropriate. Hospital retail units are also, in general, smaller and less flexible than station retail.

A £1 billion annual figure would therefore represent more than three times the entire commercial income of Transport for London, achieved by an organisation starting from close to zero, with a more restrictive content framework and a lower-value audience. It is not a credible near-term target, and publishing it would hand critics an easy dismissal of an otherwise strong idea.

What is credible, presented as a modelled trajectory with stated assumptions, not a forecast:

So £1 billion is reachable — as a twenty-year objective, not an annual expectation. And that framing is stronger politically, not weaker: "the NHS estate should be generating what TfL generates, and within a decade it can be generating more" is a claim that survives contact with a select committee. "£1 billion from hospital shops" does not.

Note also that the year-10 figure of £300–550 million would, on its own, be comparable to or greater than total national NHS car parking income — meaning the trade-off proposed at 6B.5.5 becomes genuinely deliverable rather than rhetorical.

6B.7 Ministerial powers, cost and timeframe

Powers. NHS bodies already hold statutory income generation powers, and foundation trusts have substantial commercial freedom. Establishing a national commercial estate body would most likely require primary legislation if constituted as a separate corporate entity with its own balance sheet — this is the one element of the entire paper that plausibly does. However, a shadow function within DHSC, operating through framework agreements and directions, could be established immediately without legislation and could carry the first three years of the programme. The mandatory national advertising ethical framework requires no legislation and can be imposed through the NHS Standard Contract.

Planning remains a matter for local authorities. The proposal is not to override local planning but to give trusts the professional capability to engage with it properly, plus a national permitted-development conversation with MHCLG on minor retail and signage changes within existing hospital envelopes.

Modelled costs:

The retail fit-out fund is capital that returns; the advertising estate typically pays back within three to five years at mid-market yields. On the modelled trajectory the pillar is net positive from around year four.

Timeframe: shadow function within 6 months; ethical framework published within 9 months; first framework agreements within 12–18 months; first digital advertising sites live within 18 months; legislation for the full corporate body, if pursued, in the following parliamentary session.

6B.8 Risks

"The NHS is selling out to advertisers." The mitigations are the exclusion list, published sponsorship registers, the bright line between sponsoring comfort and sponsoring care, and the explicit car parking tradeoff. The public will accept commercial advertising in exchange for cheaper parking; they will not accept it in exchange for nothing.

"It will make hospitals feel like shopping centres." A legitimate aesthetic and clinical concern. The framework must exclude advertising from clinical areas entirely — no advertising in wards, treatment areas, resuscitation, paediatric clinical space, mortuary or bereavement areas. Commercial content belongs in concourses, main entrances, and outpatient waiting areas, and nowhere else.

"Charities will be pushed out." Directly addressed by the guaranteed voluntary sector floor, and this must be in the framework from day one rather than added after the first controversy.

"It is a distraction from clinical priorities." The counter is that it is delivered by a specialist commercial function, not by clinicians or trust executives, and that its first-order clinical benefit — 24/7 hot food for staff, better wayfinding, faster information display — is real and immediate.

Part Seven — The measurement architecture

7.1 The principle: no reform without a published counterfactual

Every proposal in this paper is attached to a measurable outcome and a stated evaluation design. That is not a gesture toward rigour; it is the mechanism by which the programme survives its first hostile select committee appearance.

7.2 What the existing mortality infrastructure can and cannot do

The Summary Hospital-level Mortality Indicator is published monthly by NHS England Digital. It is the ratio between the actual number of patients who die following hospitalisation at a trust and the number expected on the basis of average England figures, given the characteristics of the patients treated. It covers patients who died in hospital or within 30 days of discharge. In the reporting period March 2025 to February 2026, across 118 trusts, there were approximately 9.3 million discharges from which 283,000 deaths were recorded; 13 trusts had a higher than expected number of deaths, 97 were within the expected range, and 8 were lower than expected.

Now the critical caveat, in NHS England's own framing, which anyone proposing to build an AI monitoring layer on top of this data must internalise: the expected number of deaths is a statistical construct and not a count of patients. The difference between observed and expected deaths cannot be interpreted as the number of avoidable or excess deaths. The SHMI is explicitly not a measure of quality of care. A higher than expected figure should be treated as a smoke alarm requiring investigation, not as evidence of poor performance.

This matters enormously for the proposal to have AI continuously computing mortality and success statistics. An AI system that continuously recalculates SHMI-like figures and presents them as performance rankings would be actively harmful. It would generate false accusation, defensive behaviour, gaming of coding practices, and the specific pathology that has repeatedly damaged NHS quality improvement: managing to the indicator rather than to the patient.

There is also a live technical caveat worth knowing: a number of trusts are now submitting Same Day Emergency Care data to the Emergency Care Data Set rather than the Admitted Patient Care dataset from which SHMI is calculated, and removal of that activity from APC data may affect a trust's SHMI value. Denominators shift for administrative reasons. Any automated monitoring system must be built by people who understand this, or it will produce confident nonsense.

7.3 What automated outcome monitoring should actually do

Redefine the ambition, and it becomes both achievable and genuinely valuable:

Not: an AI that ranks hospitals by mortality.

Instead: an AI-assisted signal detection layer that:

• Monitors existing published indicators continuously rather than on publication cycles, and flags statistically anomalous shifts for human investigation, with explicit uncertainty quantification.

• Distinguishes coding and data artefacts from clinical signals — the SDEC/ECDS migration above is exactly the class of artefact that must be automatically identified and adjusted for.

• Links intervention rollout data to outcome data with correct stepped-wedge and difference-indifferences analysis, so that when a pillar of this programme is implemented at a site, its effect is estimated against a genuine counterfactual rather than a before-and-after comparison.

• Produces structured morbidity and complication tracking: readmission within 30 days, return to theatre, unplanned ICU admission, hospital-acquired infection, pressure ulcer incidence, medication error reports — indicators that are more directly actionable than mortality and less confounded.

• Feeds morbidity and mortality meeting preparation, assembling case documentation automatically so that clinical time in M&M is spent on analysis rather than on notes retrieval.

• Supports appraisal and revalidation by assembling a doctor's activity, outcome and feedback data automatically — with the explicit safeguard that individual-level outcome data must never be presented as a performance ranking without case-mix adjustment and statistical power assessment. Individual surgeon mortality data has a well-documented history of driving risk-averse case selection, which harms exactly the patients who most need surgery.

7.4 The published dashboard

Every measure in this paper published quarterly, at trust level, machine-readable, open:

Pillar One: food and drink standards compliance by standard; 24/7 staff access verified; staff-reported availability; compliance-perception gap.

Pillar Two: communication-category complaint rate per 1,000 admissions; agent accuracy under blinded clinical adjudication; escalation response times; handover open-loop completion rate; patient-reported understanding at discharge.

Pillar Three: under-triage rate, human versus AI-assisted, in shadow and assist mode; time to first clinical decision; disposition consistency across entry points; 111 call handling times.

Pillar Four: administrative hours per clinical hour; intention-to-leave; training places against applicants by specialty; escalation-safety survey score; leaver counts.

Pillar Five: international patient income by trust; recovery rate on chargeable overseas visitor invoicing; evidence of capacity additionality.

Pillar Six: commercial and advertising income by trust; car parking charge levels against commercial income (the trade-off measure); void rate across the retail estate; volume of space at peppercorn rent for voluntary services; published sponsorship register; advertising framework breach reports.

Cross-cutting: SHMI as contextual signal with mandatory interpretive caveats attached to every publication; morbidity indicator set; readmissions; never events.

7.5 Independence

The evaluation must not be conducted by the programme delivering the reform. Evaluation should be commissioned independently — through NIHR, as the ambient voice technology evaluation already is — with a published protocol and pre-registered outcomes, and a binding commitment to publish negative results.

That last commitment is the one that will be hardest to keep and the one most worth making.

Part Eight — Consolidated ministerial powers

Summary: the overwhelming majority of this programme is deliverable with existing ministerial powers. The elements requiring primary legislation — chiefly the Single Patient Record and the provider duty to make records available — are already in the parliamentary pipeline.

This is the single most important political fact in the paper. A programme that requires a new Act is a programme for the next parliament. This one is not.

Part Nine — Consolidated cost and timeline

All figures modelled, orders of magnitude, requiring validation.

Note that Pillar Six is the only pillar that is net revenue-generating, and on the modelled trajectory its year-10 income substantially exceeds the annual running cost of every other pillar combined.

Total annual running cost: £297m – £657m

Against an NHS England budget in the order of £190bn — roughly 0.16% to 0.35%.

All figures modelled estimates with stated assumptions, requiring validation. Pillar Six is the only pillar that is net revenue-generating — by year 10 its income exceeds every other pillar combined.

Consolidated timeline:

• Days 0–100: Directions issued on food standards publication and data burden review. MHRA regulatory classification initiated. Consent engine specification. Public engagement launched. Independent evaluation commissioned. Consultant contract discussions opened.

• Months 3–12: First compliance publication. Shadow-mode triage at pilot sites. Patient Companion pilot in 2–3 trusts. Administrative burden baseline established. International office established.

• Months 12–24: Interim safety reports published. Assist-mode triage. Family Access pilot. Capital programme begins. Training forward commitments published.

• Months 24–36: Staged national rollout conditional on published thresholds. Capital works across acute estate. Core staff meal pricing national.

• Months 36–60: SPR-aligned national record transparency. Full triage rollout. First five-year evaluation published.

Day 0Yr 1Yr 2Yr 3Yr 4Yr 5

One — Dignity at work

Assess

Capital build

Two — Record transparency

Pilot

SPR-aligned rollout

Three — AI triage

Shadow mode

Staged national

Four — Retention

Governance

Contract

Training pipeline

Five — International

Set up

Market development

Six — Commercial estate

Framework

Agreements

Estate rollout

Every pillar begins within the first hundred days. None requires a new Act except the NHS Places corporate body.

Part Ten — Risks, objections and honest weaknesses

"AI will hallucinate and someone will die." The most serious objection and the one requiring the most rigorous answer. The mitigations are architectural rather than aspirational: retrieval grounding with mandatory citation to source record entries; explicit refusal behaviour rather than speculation; no autonomous clinical decisions at any layer; six months minimum shadow mode before any triage tool influences a human decision; published under-triage metrics with automatic rollback triggers; blinded clinical adjudication of agent outputs as a continuous safety metric. If the shadow-mode data are bad, the programme stops. That commitment has to be genuine or the whole thing is theatre.

"This is privatisation." Pillar Five is the exposed flank. The structural answer is capacity additionality, published accounts, ring-fenced reinvestment, and auditable evidence that no NHS patient waited longer. Rhetorical reassurance is worthless here.

"The public will not accept AI reading their records." NHS England's own engagement found strong public support for a single patient record conditional on proper safeguards — access levels by job role and audit trails. Consent is not a compliance box; it is the product. This is why the consent and scope engine is specified to be built first and owned in-house.

"Another reorganisation the NHS cannot absorb." Fair, and the abolition of NHS England is already consuming enormous organisational bandwidth with benefits yet to materialise. This programme is deliberately designed to require no structural reorganisation. It changes what is measured, what is published, what is built and what is funded. Not who reports to whom.

"The savings will not materialise." Probably partially true, as with most such programmes. Which is why every figure here is a labelled range with stated assumptions, why the evaluation is independent and preregistered, and why negative results are committed to publication.

"The food thing is trivial." It is the least expensive and most visible element, and the one on which the credibility of everything else rests. If a Health Secretary cannot deliver a hot meal to a night shift, no one will believe them about a national data architecture.

The weakness this paper acknowledges most directly: the causal chain from information transparency to mortality benefit is plausible and unproven. This paper does not claim it. It proposes to test it properly and publish the result either way.

Part Eleven — The first hundred days

Ten actions, all within existing powers, all deliverable.

1. Issue a direction requiring quarterly published trust-level compliance with the National Standards for Healthcare Food and Drink, with the 24/7 staff access standard reported separately.

2. Commission the national capital assessment of hospital catering facilities.

3. Add three staff-experience questions on food, breaks and value for money to the national survey instrument.

4. Issue a direction establishing the annual data burden sunset review, with named owners and automatic lapse.

5. Commission the national measurement of clinician administrative hours per clinical hour.

6. Open MHRA regulatory classification discussions for all three proposed agents and all four triage layers.

7. Commission the consent and scope engine as an in-house build, with a published architecture and a commitment to NHS ownership of intellectual property.

8. Launch genuine public engagement on patient and family record access — including on what people fear, not only on what they want.

9. Commission independent, pre-registered evaluation through NIHR, with a binding commitment to publish negative findings.

10. Publish the forward five-year training place commitment by specialty, and the retention impact assessment methodology for workforce policy.

11. Establish the NHS Places shadow commercial estate function within DHSC, and publish the national advertising ethical framework including the mandatory exclusion list and the guaranteed voluntary sector space floor.

None of these requires a Bill. All of them are measurable within a year.

Sources

Workforce and morale

• NHS Staff Survey 2025 results, published 12 March 2026; NHS England Staff Survey Coordination Centre; commentary from NHS Employers and NHS Confederation.

• General Medical Council, The state of medical education and practice in the UK: workplace experiences 2025.

• General Medical Council, The state of medical education and practice in the UK: workforce report 2025.

• Royal College of Physicians, next generation survey 2025 and response to GMC workplace experiences report.

• British Medical Association, medical staffing in the NHS data analysis.

• Onyejesi, James and Kharicha, Understanding why health professionals are leaving the UK National Health Service: a systematic review and narrative synthesis, 2026.

• Nuffield Trust, Exploring the earnings of NHS doctors in England (2025 update).

• House of Commons Library, NHS pay and pensions research briefing.

• UCL, Pay satisfaction and intentions to leave the NHS: a UK-based cohort study.

Food and dignity

• Report of the Independent Review of NHS Hospital Food, chaired by Philip Shelley, 2020.

• NHS England, National standards for healthcare food and drink, 2022.

• Hospital Food Standards Panel report, 2014.

Complaints and communication

• NHS England Digital, Data on Written Complaints in the NHS.

• Parliamentary and Health Service Ombudsman, casework reports and thematic publications on communication.

Digital, data and AI

• NHS England, Single Patient Record programme documentation.

• 10 Year Health Plan for England, July 2025.

• NHS Modernisation Bill, introduced 14 May 2026.

• NHS England Digital, NHS App management information and features documentation.

• NHS England, Federated Data Platform contract explainer; parliamentary debate material on the FDP.

• Nuffield Trust / NIHR Rapid Service Evaluation Team, Phase 1 findings on ambient voice technology.

• Great Ormond Street Hospital, ambient voice technology evaluation and AI strategy.

• NHS England, AI rollout announcements, July 2026.

• MHRA, Software and AI as a Medical Device Change Programme.

Urgent and emergency care

• NHS England, Urgent and Emergency Care Plan 2025/26.

• Tony Blair Institute for Global Change, Preparing the NHS for the AI Era.

• South East Coast Ambulance Service AI pilot documentation.

Mortality and outcomes

• NHS England Digital, Summary Hospital-level Mortality Indicator, monthly publications including March 2025 – February 2026.

Structural reform

• Health and Care Act 2022 and explanatory notes.

• National Health Service Act 2006 as amended.

• NHS Confederation, Abolishing NHS England: what you need to know.

• Institute for Government, One year on: where has the government's decision to abolish NHS England left the NHS?

Commercial estate and advertising

• Transport for London, Draft Budget 2026-27, and Advertising Annual Report.

• House of Commons Library, NHS hospital car parking policies in the UK; NHS England ERIC estates returns.

• Frimley Health NHS Foundation Trust retail and estates strategy announcements, Wexham Park Hospital, 2025–26.

• HFSS advertising restrictions in force from 5 January 2026; UK Nutrient Profiling Model.

• Local authority and TfL HFSS advertising policies.

International patients

• Hanefeld et al., Medical Tourism: A Cost or Benefit to the NHS?, PLOS ONE, 2013.

• Policy Exchange freedom-of-information analysis on overseas patient invoicing, 2021–2024.

• National Audit Office, reports on recovering the cost of NHS treatment for overseas visitors.

• Competition Commission private healthcare market investigation materials.

This paper was written with AI assistance in drafting and research. All statistics cited are drawn from the published sources listed above; all projections are explicitly labelled as modelled estimates with stated assumptions.

Annex — Continuity of the Programme Against Political and Bureaucratic Interruption

A.1 The problem, stated precisely

This is not a proposal for a minister who cannot be removed. Ministers should remain removable — that is basic accountability, and any paper arguing otherwise would rightly be dismissed.

The problem this annex addresses is different: NHS reform programmes in England have a documented pattern of being restarted, rebranded, or quietly abandoned every time the minister or governing party changes, regardless of whether the underlying plan was working. The person changes; the plan should not have to change with them. A programme that took three years to design and twelve months to begin delivering should not be reset to zero by a reshuffle.

The mechanism needed, therefore, is not personal protection for a minister. It is statutory protection for the plan itself — so that a direction, once made and published, continues to bind the system regardless of who holds the office that made it, until it is explicitly and transparently revoked.

A.2 The mechanism: a standing direction with statutory continuity

Proposal. Use the Secretary of State's existing power of direction over NHS England functions and integrated care boards (NHS Act 2006, as amended by the Health and Care Act 2022) to issue a standing direction implementing the six-pillar programme. Unlike an ordinary ministerial direction, which conventionally lapses or is easily reversed at the discretion of a successor, this direction would be drafted with an explicit continuity clause:

• The direction remains in force automatically across any change of Secretary of State or change of government, unless it is explicitly revoked.

• Revocation requires the same procedural step as making it: a published, reasoned direction reversing it, laid before Parliament — not silent abandonment through inaction, budget reallocation, or simply not renewing it.

• Progress against the direction's milestones is reported publicly on a fixed schedule (proposed: every six months), regardless of who is in post.

This does not require new primary legislation. Directions under s.13ZC and related provisions must already be published; what changes is the drafting convention — building in an explicit continuity and revocation clause rather than leaving continuation to informal custom.

A.3 Why this is different from ministerial immunity

The distinction matters because it is the first objection any select committee or opposition frontbencher will raise. The answer is that this protects continuity of policy, not tenure of person — a principle already accepted elsewhere in the system (the mandate to NHS England, the NHS Constitution, multi-year capital settlements).

A.4 Supporting mechanisms

A.4.1 Fixed-term implementation milestones laid before Parliament. The direction should specify dated milestones (e.g., "AI triage deployment standard published within 12 months") so that a successor's failure to progress is measurable and publicly visible, not just a matter of political framing.

A.4.2 Ring-fenced multi-year funding settlement. Where possible, secure the programme's costed elements (Part Nine) as a multi-year settlement with the Treasury rather than subjecting them to annual re-negotiation. This is a spending review mechanism, not new legislation, and several NHS capital programmes already operate this way.

A.4.3 Statutory bar on structural reorganisation within the delivery window. A specific, narrow provision preventing top-down reorganisation of the bodies responsible for delivering this programme within its delivery window (proposed: 5 years), unless Parliament explicitly legislates to do so. This directly targets the "reformed to death" pattern identified in Part One, and would likely require a short clause in the next available health bill rather than a standalone Act.

A.4.4 Independent published progress reporting. Commission the biannual progress report (A.2) from a body at arm's length from the department — e.g., the National Audit Office or a named clinical/academic panel — so continuity reporting cannot itself be quietly deprioritised by an unsympathetic successor.

A.5 Powers required

Summary: as with the rest of the programme, the great majority of this continuity mechanism is deliverable through existing powers and drafting convention. The single element requiring legislation — the reorganisation bar — is narrow, precedented in spirit (similar quiet periods have been proposed after past reorganisations), and could ride on the next health bill rather than requiring a standalone Act.

A.6 The honest risk

No mechanism can fully insulate a plan from a government that is willing to spend political capital reversing it outright and defending that reversal in public. This annex does not claim otherwise. What it does is raise the cost of quiet abandonment — forcing reversal to be a visible, reasoned, accountable political act rather than the default outcome of inertia, reshuffle, or a new minister simply wanting their own initiative to point to. That is the realistic ceiling of what any continuity mechanism can achieve inside the UK's constitutional settlement, and the paper should say so plainly rather than overclaim.