System C CareFlow Ambient AI

The module exists to do one learned task: turn a spoken consultation into documentation. It transcribes the conversation, drafts letters on the trust's templates, extracts structured clinical data and outcome codes, and in emergency care fills triage, clerking and discharge content. Without speech recognition and language generation there is nothing left of the product. The surrounding CareFlow EPR is conventional record software, and the AI lives in this module.

Nothing the module produces reaches the record unreviewed. System C states that every AI output is reviewed and signed off by the clinician before it commits, that the AI makes no autonomous decisions, and that in emergency care the Emergency Care Data Set fields are verified by the clinician before submission. Within those limits the module does a lot: it drafts letters and discharge summaries, assigns outcome codes, and can start follow up workflows, orders and pathways inside CareFlow once approved.

How the draft shows its sources, such as linking a sentence to the moment in the conversation it came from, is not described.

The module transcribes in real time, launches in patient context with no separate login, drafts letters on existing trust templates including letters on hold while results are awaited, writes structured clinical data and outcome codes back into CareFlow, and enriches drafts with the patient's history and recent results. Each clinician can keep a personal dictionary of words and phrases to improve the speech to text, and the emergency version builds structured assessment content at the bedside.

The speech recognition and language models behind it are not named, and no version, training data or update practice is given.

Microsoft Azure hosts the module, and System C says it writes directly into CareFlow with no middleware and no parallel data stores, under one clinical safety case. System C has also named Microsoft as its partner for a separate social care AI product.

Which company provides the speech recognition and the language model, whether any processing passes to an outside AI service, and whether a subprocessor list exists are not published, and hosting on Azure does not say which models run there.

Benefits so far are stated as expectations, because the module is new. System C and Buckinghamshire Healthcare NHS Trust list reduced clinician time and cognitive load, faster clinic letters, less administration, better coding accuracy and fewer missed or delayed outcomes as what they expect from the outpatient go live planned for summer 2026.

No time saving, letter turnaround, coding accuracy or clinician satisfaction figure from Buckinghamshire or any other site has been published, and no evaluation of the transcription or drafting quality is available.

The design keeps data in one place. System C says the module writes directly into CareFlow with no parallel data stores and no middleware, under a single clinical safety case, and it confirms patient identity through the national demographics service before documenting. Audio recordings, transcripts and generated documents follow the trust's record retention policies rather than a separate schedule, and if the connection drops, audio is held in the browser cache encrypted with AES-GCM until it can be sent. The module also draws on the patient's history and recent test results to enrich what it drafts, so the model sees record data as well as the conversation.

How long audio is held before deletion in practice, and whether recordings or notes are used to train or tune the models, are not stated.

NHS trusts are the customers, so UK data protection law and NHS information governance apply rather than HIPAA. Patients are identified through the NHS Personal Demographics Service, and the output is written straight into the trust's CareFlow record. The data processing terms that cover recorded consultations are not on the product pages, so they sit in each trust's CareFlow contract.

System C holds ISO 27001 certification for information security and ISO 27018 for protecting personal data in the cloud, alongside ISO 9001 and ISO 20000-1, each shown on its company page with its certification body. Its G-Cloud listing for the AI module names NQA as the ISO 27001 certifying body, with a certificate dated October 2024, and gives a Cyber Essentials Plus certificate number.

There is no trust center, and no certificate scope or audit report is posted, so whether the ambient module and its speech processing sit inside the certified scope is not shown.

CareFlow Ambient AI is registered with the MHRA as a Class I medical device, according to System C, with technical documentation prepared under UK medical device regulations, and it carries a DCB0129 clinical safety case with structured post market surveillance. The registration covers the UK, which is where the product is sold; no position elsewhere is stated.

Governance runs through the device and clinical safety process: a DCB0129 safety case with a Clinical Safety Officer overseeing development and deployment, structured post market surveillance, and System C's stated principles of building AI only for real problems, ethical by design and human first. Its service documentation describes systematic measurement of accuracy, completeness and clinical appropriateness, with testing across clinical contexts.

No results from that testing are published, including by accent, specialty or patient group. Speech recognition errors fall unevenly on speakers with strong accents and on interpreted consultations, which are common in NHS clinics, and how the module performs for them is not reported.

Review before commit is the safeguard. Every draft, code and form waits for the clinician to sign it off before it reaches the record, and the module leaves a contemporaneous record of the interaction that strengthens the audit trail if a document is later questioned.

Nothing stands behind the output itself. No accuracy, error or correction rate is published for transcription, letter drafting or coding, and no warranty or remedy for AI output appears in public material, so responsibility for each document sits with the clinician who approves it. The coded outcomes that drive activity income carry the same arrangement.

Because the module is part of a record system, what counts is how its output moves beyond CareFlow. Letters and discharge summaries go out through MESH, the NHS messaging service, or through patient portals and print, with tracking and an audit trail; patients are matched against the NHS Personal Demographics Service; and the company cites support for GS1 and Transfer of Care standards. Inside CareFlow it writes structured data rather than free text, which feeds the trust's reporting. It does not work with other record systems.

CareFlow Ambient AI runs inside CareFlow EPR on Microsoft Azure infrastructure, and System C states that data is stored and processed in the United Kingdom, with CareFlow typically deployed in Azure's UK regions. Its G-Cloud listing guarantees 99.5 percent availability and a four hour resolution target for a system down or clinical risk incident.

Where the speech and language processing itself runs, whether inside the trust's CareFlow environment or in a separate AI service, and how one trust's data is isolated from another's are not stated.

Trusts buy CareFlow Ambient AI as an addition to CareFlow EPR, through a demo and a quote. On the UK government's G-Cloud 15 framework, where System C lists the module as CareFlow AI, monthly usage is charged per appointment, with overage volumes agreed in the contract. The rate per appointment, the size of the contracted volumes and any implementation cost are not public.

Because the module only runs inside CareFlow, the larger commitment is the record system itself, which System C sells on long contracts.

Two settings are covered: outpatient clinics and the emergency department, each with its own documentation and coding outputs. The outpatient version handles letters, outcome forms and referral to treatment fields; the emergency version handles triage, clerking, discharge summaries and Emergency Care Data Set fields. System C describes the design as extensible to other hospital workflows, and has said maternity AI is in development.

Transcription is in English only, which matters in NHS clinics where interpreted consultations are routine. Buckinghamshire Healthcare NHS Trust is the only named customer so far, part of an unnamed group of early adopters, and specialties within outpatients are not listed.