NHS AI Rollout Hits 29% Queue Reduction - But Homecare Governance Gap Widens
As NHS England accelerates a £10bn AI push backed by real pilot data, new research reveals the care sector is deploying AI at a pace its own governance frameworks cannot match - raising questions regulators have yet to answer.
The numbers are promising. The rulebook is still being written.
On 4 July 2026, NHS England set out how its £10bn, three-year technology budget will be deployed, and for once the announcement carried specific evidence behind it. A Sussex GP trial of the new AI triage tool built into the NHS App cut phone queues by 29%, while patient satisfaction held steady. The tool is now being expanded to 200,000 patients, with a target of reaching all NHS App users by April 2028.
The mechanics are deceptively simple. A patient opens the app, describes their symptoms, and the system adapts its follow-up questions in real time, requesting further detail or photographs before generating a clinical summary. A clinician reviews it before any contact is made. As Ragu Rajan, GP senior partner at Wealden Ridge where the trial ran, put it: the tool is about sequencing rather than substitution. The AI does not replace clinical judgement; it structures demand before it hits the consultation.
Advertisement
Ambient voice: the productivity case is mounting
The second priority in NHS England's programme is ambient voice technology (AVT), AI-driven notetaking that transcribes and summarises clinical conversations in real time. A major NHS study led by Great Ormond Street Hospital found that scaling AVT nationally across more than 11,000 A&E clinicians could create capacity for over 9,000 additional consultations per day. A pilot at St George's saved clinicians 47 minutes per shift. NHS England will also give approximately 505,000 staff access to Microsoft 365 Copilot after a trial found it saved around 43 minutes of admin per day.
The total expected benefit, according to NHS England's own estimates, is £41bn over the next decade, roughly half the commitments in the government's 10 Year Health Plan. These are large numbers. But as Digital Health reported on 14 July, sector figures broadly welcomed the investment while warning that delivery, data quality and staff training will determine whether it works. The recurring message was blunt: funding is the easy part.
Homecare: running well ahead of the rules
Beyond the NHS estate, a parallel and less scrutinised story is unfolding in homecare. Research published by homecare technology company Birdie on 1 July, drawing on a survey of 122 UK homecare providers, found that 70% are already using AI in some form, with adoption expected to reach 85% within a year. Around half are using AI not just for back-office administration but to shape care plans and risk assessments directly.
The results are, on one reading, encouraging: 76% of providers said AI had improved the quality of care they deliver, and among those re-inspected by the CQC since adoption, 59% saw their rating improve, with none reporting a decline. But the governance picture is stark. Only 66% of AI-using providers have any formal policy governing how the technology is used. Just 43% have a written policy in place.
Perhaps most striking is what tools providers are actually using. Adoption is dominated by general-purpose consumer AI: ChatGPT, Microsoft Copilot and Google Gemini collectively account for the majority of deployments. These are not products built for regulated care environments, and none carry the assurance that a medical device pathway would require.
This governance gap maps directly onto a structural market anomaly captured in AI Business Dispatch analysis of Companies House and IPO data. Just 64 new SIC 86.10 companies were incorporated in 2026 Q3, a fall of 79.5% on the prior period, suggesting acute contraction in formal hospital activity formation. Meanwhile, UK Class 5 trademark filings (pharmaceuticals and medical preparations) reached just 657 in Q3, down 74.3% on the prior period. Most striking of all: 98.8% of active SIC 86.10 companies hold no Class 5 trademark whatsoever. Source: AIBD analysis of Companies House and IPO data, as of July 2026.
That last figure is a proxy for a broader problem. Organisations operating in clinical-adjacent AI are not building the formal IP or governance scaffolding the regulatory environment will eventually demand.
The regulator is also mid-sprint
The MHRA is moving, but not yet at the pace deployment requires. On 11 May 2026, it published draft Medical Devices (Amendment) Regulations 2026, which formalise Predetermined Change Control Plans (PCCPs) for AI and software devices and introduce an International Reliance Pathway, enabling UK approvals to draw on decisions from trusted overseas regulators. Its AI Airlock sandbox has completed a second phase, covering large language models in clinical decision support and synthetic data for radiology validation, with Phase 3 now in development.
The gap between what is being deployed in homecare, general-purpose LLMs writing care plans, and what the MHRA's AI-as-a-medical-device framework currently governs remains wide. The call for evidence and the new medicines sandbox are promising; they are not yet policy.
Healthcare has always had to reconcile clinical urgency with regulatory caution. What has changed is the speed. AI tools that would once have required years of procurement cycles are being adopted from app stores in weeks. The question for the coming months is whether the MHRA's forthcoming dedicated AI framework, the CQC's inspection methodology, and the NHS's own governance standards can converge before the evidence base for what is actually being used in care settings becomes impossible to reconstruct.