Executive summary
A recent, widely-cited HSJ exclusive has put a headline figure on the potential of artificial intelligence to deliver savings in the NHS. The largest UK trial of ambient voice technology (AVT) in urgent and emergency care settings concluded that, if scaled nationally, the technology could deliver staff-time savings approaching £1 billion (£834m formally modelled), alongside significant throughput benefits, such as seeing 13% more patients in A&E. Tortus AI is a prominent supplier in this space, with its technology featuring in numerous NHS pilots (hsj.co.uk).
This striking claim comes as the NHS faces a well-documented productivity gap, making the need to improve output per pound a central pillar of national policy (House of Commons Library, Nuffield Trust). But beyond the headlines, how can trusts and primary care networks build a credible and realistic return on investment (ROI) case for AI? Evidence signals for savings are also emerging from patient engagement platforms that reduce DNAs and from diagnostic pathways like the NICE-approved HeartFlow FFRCT. However, it is crucial to remember that any potential savings are entirely dependent on compliant and safe deployment, as underscored by NHS England's official guidance on AI scribes.
Why the NHS needs credible AI economics now
Productivity has become the defining challenge for the NHS in 2025. Parliamentary briefings and detailed analyses from think tanks like the Nuffield Trust consistently highlight the need to do more with the available resources. For any clinical leader or manager, this means that any business case for a new AI tool cannot simply be about innovation; it must be explicitly linked to measurable gains in throughput, cost-avoidance, or efficiency at a service line level (House of Commons Library, Nuffield Trust).
The Tortus story and the AVT thesis
The recent HSJ report has provided the most significant data point to date on the economic potential of ambient voice technology NHS-wide. The multi-site study, conducted in emergency care, estimated that national scaling could lead to £834 million in annual staff-time savings, with some models pushing this figure towards £1 billion. It also reported a direct capacity uplift, with departments seeing 13% more patients (hsj.co.uk).
These findings are supported by a growing number of trust-level pilots and announcements from providers using Tortus AI, which consistently report gains in "time with patients" and reductions in administrative burden at sites including Great Ormond Street, Kent Community Health, and the Royal Devon (Yahoo News, GOSH Hospital site, Digital Health, royaldevon.nhs.uk). As the full peer-reviewed paper from the main trial is released, commissioners will be able to scrutinise the methodology and generalisability assumptions, but the direction of travel is clear.
Where AI creates value (three proven buckets)
1. Documentation & admin time (ambient voice / AI scribes)
The primary mechanism here is converting clinician minutes spent on administrative tasks—writing notes, coding, drafting letters—into time available for direct patient care. The Tortus NHS study reported by HSJ provides the headline evidence, and the national guidance from NHS England on AVT enables safe adoption, creating a clear path to realising these benefits.
2. Patient engagement & flow (fewer DNAs, digital outreach)
A significant and often overlooked source of inefficiency is missed appointments. A case study from University Hospitals Birmingham, using the DrDoctor patient engagement platform, reported an 18% reduction in "Did Not Attend" (DNA) rates, alongside measurable administrative time savings. A credible ROI model can directly link these avoided DNAs to reclaimed clinical capacity and tariff income.
3. Diagnostics & decision support (right test, first time)
Getting the diagnostic pathway right saves significant downstream costs. The classic example is HeartFlow FFRCT, a tool that uses AI to analyse coronary CT scans. In its official guidance (MTG32), NICE cost modelling found that using the tool could lead to a £391 per-patient saving compared to traditional pathways by avoiding the need for more invasive tests (NICE).
Build a credible ROI model
To build a business case that will stand up to scrutiny, you need to move from headline figures to local, verifiable data.
- Inputs: Start with your own baseline data: average minutes spent per clinical note, the fully-loaded cost per minute of your clinical staff (using Agenda for Change bands), your current DNA rates, and the per-patient costs for specific diagnostic pathways. Against this, set the full cost of the AI tool: licences, integration, training, and ongoing clinical safety assurance.
- Outputs: Be clear about what type of saving you are claiming. Staff-time savings are typically non-cashable (you don't bank the money) but can be converted into additional capacity (more appointments, hitting RTT or 4-hour targets). Cost-avoidance from a better diagnostic test is a "hard" saving.
Worked examples:
- AVT ROI: (Minutes saved per note) x (Notes per day) x (Staff cost per minute) ± (Quality effects like improved coding accuracy). Use the HSJ parameters as the optimistic "sensitivity bound" for your model.
- DNA ROI: (DNA rate reduction %) x (Number of appointments) x (Tariff impact per appointment) + (Savings from digital letters vs. post).
- Diagnostic ROI: Use the per-patient savings figures from official NICE models as the basis for your pathway-level business case.
Assurance & risk (non-negotiables that protect your savings)
A compelling ROI model is worthless if the tool is not compliant.
- Follow the rulebook: All deployments must adhere to the NHSE AVT guidance, meet the DTAC baseline, have a full clinical safety case (DCB0129/0160), and respect the MHRA classification for summarisation features (at least Class I).
- Learn from missteps: NHS England has previously ordered the pausing of non-compliant AI projects. Governance is a gating factor for ROI, not an optional extra (hsj.co.uk).
- Implementation risk: The headline savings can easily be eroded by friction in the implementation process, such as poor EHR integration or weak user adoption. Budget for this and have a clear change management plan.
Mini-case library
- Emergency & urgent care (AVT): The HSJ study provides a national model of £834m–~£1bn staff-time savings and a 13% capacity uplift. Replicate this with your local data before scaling.
- Elective & outpatients (engagement): The UHB and DrDoctor case study shows an 18% DNA reduction. Translate this into sessional capacity gains and administrative savings.
- Cardiology (diagnostic optimisation): The HeartFlow FFRCT NICE modelling provides a clear £391 saving per patient, forming the basis for a robust pathway-level business case.
Playbook: from pilot to scale in 90 days
- Define a single, primary metric that matters for your pilot (e.g., minutes per note, DNAs per clinic).
- Baseline this metric for 2–4 weeks to get clean pre-implementation data.
- Run a controlled pilot (A/B or stepped-wedge) and collect user effort metrics and safety incidents.
- Complete your assurance documentation (DTAC, DCBs, DPIA) aligned to the NHS England AVT guidance.
- Make your scale-up decision based on the measured productivity gain per pound and your governance readiness, not on the marketing hype.
FAQs
- Is the “£1bn” saving from AI scribes real?
- The HSJ has reported a national-scale extrapolation from the largest NHS AVT trial to date, which modelled £834m in staff-time savings (describing it as "nearly £1bn"). Trusts should always use their own local baselines and adoption curves to create a realistic forecast.
- What evidence for AI savings exists beyond scribes?
- Robust case studies on DNA reduction from patient engagement platforms and formal NICE cost-saving models for specific diagnostic tools provide quantifiable frameworks.
- What is the biggest risk to achieving ROI with AI?
- Non-compliance and weak integration. The official NHS guidance and past warnings show that robust governance is the decisive factor for a successful and sustainable deployment.