The NHS Bet the Next Decade on AI. The Bottleneck Is Still Admin.

England has published its 10-year plan for the NHS, and it reads like a manifesto for techno-optimism. AI triage. Algorithmic scribing. Genomics and wearables feeding a souped-up NHS app that turns into a personal health assistant. There's even talk of fully AI-enabled hospitals inside the decade.

I run a company that automates healthcare admin. You'd expect me to love a document like this. I don't, and it's worth being clear about why. The ambition isn't wrong. The sequencing is.

A recent analysis from Petauri Evidence makes the point that needs making: the plan is bold, but fragile. In their words, it's a hymn to efficiency that quietly sidesteps structural reality. That critique is exactly right. And it points straight at where automation actually belongs.

Detection is not treatment

The whole plan runs on early detection. Find disease sooner with AI and genomics, intervene earlier, spend less later. Fine in theory.

The problem is that faster diagnosis without the capacity to treat doesn't relieve pressure. It compounds it. You surface more patients into a system that already can't see the ones it has.

A diagnosis you can't act on isn't a head start. It's a longer queue with better paperwork.

Healthcare is labour intensive. No app replaces a clinician, a hospital bed, or the hour a consultant spends with a patient. England's own workforce numbers make the tension obvious. A pledge to cut international recruitment below 10% by 2035 sits awkwardly next to the admission that domestic training needs to grow by at least half just to cover today's gaps.

Technology doesn't close that gap. At best, it makes the gap more visible.

So where does AI actually earn its place?

Not in the consulting room. In the corridor outside it.

The boring truth of healthcare is that a vast amount of skilled people's time goes on coordination. Reading referrals. Chasing availability. Phoning patients. Re-phoning the ones who didn't pick up. Rescheduling. Confirming. This is the work that quietly eats clinical and administrative capacity. It's also the work AI is genuinely good at, because it's structured, repetitive, and high volume.

The Petauri article praises Denmark for pairing its Sundhed.dk portal with sustained investment in workforce and infrastructure. That's the lesson. Technology that protects human time, rather than promising to replace it, is the version that holds up.

That's the entire premise of what we build at Neutropy. So let me be specific about it, because the difference between automation that adds load and automation that removes it is the whole argument.

Referral-to-scheduling is the product, not a feature

We didn't bolt scheduling onto a CRM. The whole platform is the referral journey.

A GP referral comes in. The patient's identifiers are stripped before anything touches a large language model. Per-consultant rules route it to the right place. It gets pushed to the practice management system, and a voice agent calls the patient to schedule them. One job, end to end. The people who used to spend their mornings on that work get their mornings back.

And the architecture is the moat, not the model. There are well-funded companies abroad with impressive document models trained on hundreds of millions of US healthcare records. We're not competing on that, and we shouldn't pretend to. A model trained on US documents and US payer logic doesn't travel to Europe.

What travels is the data architecture. Patient identifiers never reach an LLM. Clinical data stays inside the clinic's own systems. The most sensitive category of health data never leaves the hospital. In the EU, that architecture is the product.

PMS-agnostic by design, and the voice already talks to patients

Our FHIR adapter swaps between backends on a single setting. That matters because MEDITECH Expanse sits under roughly 80% of Irish private hospitals. The adapter is what lets us walk into most of the market without a rebuild. Cliniko is live, Meditech is demo-ready, and there are 300+ tests behind it.

And the voice already talks to patients. Plenty of automation in this space still just moves paper between systems and files patient engagement under "future unlock." Ours schedules the patient directly, in a warm Irish receptionist voice, and transfers to a human the moment anything clinical or urgent comes up.

We automate the queue, not the medicine. The patient-facing step others are calling a future move is the step we already ship.

And the founder-market fit is real, not a deck slide. Brian did this job by hand as a hospital secretary. The Scheduler automates the scheduling work he did manually, and the Insurance Validator on the roadmap automates the insurance validation he did manually. We're not guessing at the workflow. We lived it.

That's the difference between automation that adds load and automation that removes it. Early detection adds patients to a queue. Better scheduling shortens the queue.

Why this matters on this side of the Irish Sea

The NHS plan is an English document, but the pattern isn't. Ireland is in the middle of its own digital health push, with eReferral modernisation underway and a €263 million commitment to digital health. Irish private hospitals run on the same scarce resource as every other system: skilled people's time.

The risk is the same too. It's tempting to read "digital transformation" as "buy more AI" and aim it at the most visible, most clinical-sounding problems. The systems that actually move the needle are the unglamorous ones. The ones that hand a scheduler back three hours a day, or let a consultant's list fill itself without a single missed call.

The NHS plan got the ambition right and the sequencing wrong. Detection before capacity. Apps before workforce. The fix isn't less technology. It's technology pointed at the right target.

Start with the admin. The clinicians will thank you for it.

Luke
Founder, Neutropy

Responding to "Digital Transformation in the NHS's 10-Year Plan: Techno-Optimism and the Limits of Efficiency" by Evelyne Priestman and Mary Baker, Petauri Evidence.