From Meeting Audio to Structured Minutes in Health Settings

Most health services we talk to are drowning in meetings before they get anywhere near a patient. Clinical governance committees, quality and safety, credentialling, infection control, drug and therapeutics, plus multidisciplinary case conferences can run through twenty patients in ninety minutes. Someone in the room is always taking notes on a laptop while trying to participate, and the minutes often land three weeks later—if they land at all—while the unrecorded action items quietly disappear.

This is one of the better places to put AI in a health organisation because failure modes are containable. However, it is also where teams quickly run into trouble if they treat it as a software purchase rather than a records and consent problem. As part of our Practical AI in Health series, this post covers the version we would actually build: transcription plus structured summaries for committees and case conferences, leaving the clinical record well alone.

Drawing the line early and writing it into the design is crucial. The system produces meeting records: attendance, agenda items, decisions, actions, and owners. It does not write clinical notes, does not go near a patient's medical record, and does not generate anything that reads as advice.

That line carries regulatory meaning, not just a philosophical one. The TGA's position states that a digital scribe intended solely to transcribe and translate conversations into written records is not a medical device. Conversely, one that analyses or interprets a clinical conversation—such as generating a diagnosis or treatment recommendation the clinician did not state—is a medical device and must be in the ARTG before supply. The Safety and Quality Commission highlights the same point in its ambient scribe safety scenario, noting that a tool suggesting a diagnosis can trigger ARTG requirements even when unintended by the developer.

Stock photo for illustration only, not from the actual event

For minuting, the practical consequence is a hard prompt and schema constraint. The model records what the chair declared, using the chair's words. If the meeting decides to refer a patient to radiation oncology, the minute states precisely that, without adding unvoiced reasoning or filling blank fields with plausible guesses.

Privacy sits on top of that framework. The OAIC's guidance on commercially available AI products specifically names note-taking and transcription assistants, confirming the Privacy Act applies to both inputs and outputs, and instructing deploying organisations to conduct privacy impact assessments and verify vendor terms against actual consents.