Dr Kola Tytler (MBBS MBA MRCGP) | 15 July 2026 | 9 min read
A 15-trust framework is not one homogeneous rollout. It is, in practical terms, at least fifteen separate implementations, potentially spanning acute trusts, mental health trusts, community providers, specialist hospitals and ambulance services, each with different documentation patterns, different clinical risk profiles and different existing electronic patient record estates. This piece is aimed at the people who actually have to make that work: trust CCIOs, CIOs, clinical safety officers and implementation leads.
Choosing the right clinical settings first
Not every setting should necessarily launch simultaneously, and a sensible rollout sequence matters more than headline speed. Candidate settings include outpatient clinics, emergency departments, same-day emergency care, inpatient ward rounds, mental health assessments, community visits, maternity, allied health professional consultations and multidisciplinary team meetings. Each carries a different documentation style, a different tolerance for latency, and a different consequence if a draft note is wrong. Sequencing deployment by clinical risk and workflow readiness, rather than by enthusiasm, tends to produce a safer and more sustainable rollout.
Establishing clinical safety governance properly
Every participating trust needs a named clinical safety officer with actual authority over the deployment, a documented review of the manufacturer's DCB0129 clinical safety documentation, a local DCB0160 clinical safety case specific to that trust's use, formal hazard identification for the settings in scope, an agreed human-verification process that clinicians are actually trained on rather than told about once, a clear escalation route for errors and near misses, a defined response plan for service outages, and ongoing monitoring after every software update, since a new release can change behaviour in ways that were not present at initial sign-off.
Integrating with more than one EPR
Fifteen trusts will not share one electronic patient record system. Realistic candidates in scope include Epic, Cerner/Oracle Health, System C, Nervecentre, Dedalus, and TPP or EMIS in community settings, alongside specialty-specific systems and local document management platforms. For each, the practical questions are the same: does the tool know which patient record is currently active, can the note be written directly into the correct encounter, are letters generated in the correct local format, is the audit trail retained, and what happens to a session if the EPR connection fails mid-consultation. These are not abstract questions; they are the difference between adoption sticking and clinicians quietly reverting to typing.
Configuring outputs by specialty
A single universal note template is unlikely to serve all fifteen trusts well. Emergency medicine notes, surgical outpatient letters, psychiatric assessments, paediatric consultations, oncology reviews, physiotherapy notes, discharge summaries and community nursing documentation all have different structural and regulatory expectations. Configuration work here is genuinely substantial, and trusts that treat it as a one-off setup task rather than an ongoing specialty-by-specialty process tend to see lower sustained adoption.
Training clinicians properly, not just informing them
Effective training covers when AVT use is clinically appropriate, how to explain it to a patient in plain language, when explicit consent is required rather than implied, how to actually review a generated output rather than rubber-stamp it, how to recognise hallucinations and omissions specifically, what should not be said aloud during a recorded consultation, how to report a problem when one occurs, and, importantly, how to guard against automation bias, the tendency to trust a fluent, well-formatted draft more than it deserves.
Measuring real adoption, not just licences issued
The metrics that actually matter are licensed clinicians versus activated clinicians versus weekly active clinicians, consultations recorded, notes accepted without significant edit, average editing time per note, abandoned sessions, specialty-level adoption variation, variation between sites doing ostensibly the same work, and, critically, the stated reasons for non-use where clinicians choose not to engage. A framework that reports only licences issued is not reporting adoption.
Preserving patient choice throughout
Every deployment needs a clear, plain-language explanation of the technology for patients, a genuine ability to decline without any effect on the care they receive, sensible alternatives where a patient is uncomfortable, and additional safeguards in higher-sensitivity consultations: safeguarding discussions, domestic abuse disclosures, sexual health, mental health, paediatrics, and interpreter-assisted consultations, where the presence of a recording device changes the dynamic of the conversation.
Conclusion
The Midlands framework creates a genuine opportunity for shared learning across fifteen trusts that would otherwise each solve these problems in isolation. But the decisive factor in whether it succeeds will be the quality of implementation at each site, not simply the availability of licences under a central contract.