Making the most of Bio6's AI in your clinic

Bio6's artificial intelligence presents itself as an assistant that knows your clinic. It answers your questions about CNESST, drafts notes, consults Quebec practice guidelines, and creates your records, appointments and invoices on request.

This guide sets out its functions and their effect on a clinic week, before turning to deployment and the precautions to take.

Cortex, the assistant that knows your clinic

Cortex is the ✨ panel in the header. It answers in French and takes your context into account: the open patient, the current week of the programme, and your templates.

It answers your questions about Bio6 and about CNESST. For example: "How do I create a CNESST AT-138 invoice?" or "Which fields are required on the AT-505?" The question is asked where you work, without consulting a manual or interrupting a colleague.

It creates your records. Requests such as "Create a note for today's assessment", "Schedule a follow-up Tuesday at 2 pm" or "Bill the session" are prepared and then presented on a confirmation card. The record is only created when you click Create. Cortex can create a patient, a note, an appointment, a task, an invoice and an exercise programme.

It summarises a patient's previous notes. Before a session, picking the record back up goes from several minutes of reading to about thirty seconds. This is the function clinicians adopt first, since the gain is immediate and the risk nil.

It structures a care pathway. A request such as "Create an automated care pathway for post-operative ACL" provides a structure to adjust rather than a blank page.

Consulting the literature without leaving the record

This function is often underestimated before being tried, and it is the one a generic assistant cannot offer.

Cortex queries named sources and presents the answer with its citation: INESSS, OPPQ and Choosing Wisely for Quebec sources, PEDro, PubMed and Cochrane for evidence, and ClinicalTrials.gov for studies under way.

The usual case is a patient with rotator cuff tendinopathy who is not responding to treatment. You consult what the guidelines indicate, you obtain a sourced answer and you decide. The delay between the question and the answer moves from the end of the day to the session itself.

Scribe: the note written during treatment

Scribe records or imports the consultation, transcribes it, distinguishes speaker turns between the clinician and the patient, then produces a draft structured note.

Its contribution is most marked when the day overruns. After six consecutive follow-ups, notes pile up for the end of the day, and Scribe then turns writing into proofreading. The same applies to the 60-minute initial assessment: the draft captures what was said, and you correct and structure it.

Clinical relevance: decision support grounded in the diagnosis

By clicking a diagnosis in the record, or the Clinical relevance button while drafting the treatment plan, Bio6 displays a card adapted to the condition and to the week of the programme the patient has reached.

Recommendations are grouped by clinical stage, namely physical examination, imaging, intervention and referral, with the indications to favour and to avoid, the citations, a level of acuity and a confidence index.

The most useful application is not discovering new information, but quickly confirming that nothing has been omitted in a condition you encounter a few times a year.

Cortex by text message

This is the same agent, reachable by SMS, with the same permissions and the same audit trail as in the application.

The functions that are not yet delivered

The clinical agent that would propose actions to confirm in the application on its own initiative remains less developed today than the sourced answers. Nor does Cortex have a tool loop: it gathers its context before answering rather than exploring your database during the conversation.

If a demonstration left you with a different impression, ask to see the function on your own clinic.

How to integrate it

Enabling the AI for the whole clinic on a Monday morning is the best way to obtain a collective rejection the next day.

Start with two volunteers, for two weeks. Choose two curious clinicians, avoiding both the most sceptical and the most enthusiastic. Let them use Scribe on simple sessions, meaning follow-ups rather than initial assessments.

Track a single measure, namely the time between the end of the session and the signature of the note. It is the only one that genuinely speaks to a clinician.

Then extend use on the basis of a written rule. Specify what the AI may draft, what must be proofread, and what never passes through it. A written rule is more effective than several training sessions.

A note on deployment order: begin with Cortex and documentary search rather than with Scribe. These are the functions where an error carries no consequence and where the team gains confidence before letting the AI intervene in a note.

Precautions

Where your patients' data is processed

Processing involving health information runs on Bio6's Canadian infrastructure, under a signed agreement with the model provider. A call that does not meet this condition is refused by the platform. This is a technical barrier on the path of the request rather than an internal policy.

Three items are worth requesting before any signature: the information processing agreement, paying attention to the retention section; the region of execution, where the expected answer is a specific region rather than a general mention; and the nature of what is logged, since every call leaves a verifiable trace.

In its conversation replies, Cortex does not disclose identifiers from existing records, whether names, dates of birth or claim numbers. Two functions are exceptions, namely patient creation and the note summary, and they take distinct paths that are kept separate.

Who has access to the assistant

The assistant is reserved for clinic staff. Patients, caregivers and audit accounts have no access, this restriction being applied server-side rather than by simply hiding it in the interface.

Within the clinic, access follows your roles. If you have not reviewed them for some time, do so before enabling the AI.

What you can turn off

Under Settings → Clinic → AI assistant, each function is listed separately with its current state. The page reads as an inventory: what is active in this clinic right now, and what governs it.

Three states are distinguished, and the nuance matters when a function does not behave as expected.

Active covers the functions your clinic controls and can turn off itself: the conversation assistant, note generation, the CNESST report draft, and the reading of the recovery trajectory.

Unavailable on this deployment means activation happens at platform level rather than in your settings. This is the case for clinical relevance. No action on your side will make it appear, and Bio6 is the party to contact.

Consent required applies to the scribe: the function is available, but each recording requires the patient's recorded consent. This is not a setting to enable once, it is a condition checked at every session.

Turning a function off takes effect immediately and erases nothing that has already been produced: a signed note remains a signed note.

What surprises clinicians

A microphone in a silent room produces text with no basis. Transcription models generate content even in the absence of usable speech. Avoid recording an empty room or leaving the recording running between two patients.

The language of the session has an effect. A consultation alternating between French and English produces a lower-quality draft than one conducted in a single language. Warn your team, otherwise it will conclude that the tool is deficient.

The draft is not the note. The main risk is not that the model is wrong, but that it appears to be right. A well-written text is proofread less attentively than a clumsy one. That is where the real safety issue lies, more so than in the technology itself.

Frequently asked questions

Is our data used to train the model?

This is precisely the question to put to Bio6 in writing and to have confirmed in the processing agreement. A verbal answer during a demonstration is not enough: ask for the clause.

Do patients have to sign a consent?

The position adopted by most clinics is to announce that a drafting aid is used and that the clinician remains responsible for the note. Confirm yours with your professional order, since expectations vary from one profession to another.

Can the AI modify a record on its own?

No. An action is prepared and then presented on a confirmation card, and nothing is created until you click. By SMS, the agent remains read-only and can write nothing.

How long before a gain is visible?

Two weeks with two volunteers is generally enough to decide. If the time between the session and the signature of the note has not changed after two weeks of real use, it is better not to widen the deployment immediately.

Can only part of it be enabled?

Yes. The functions are listed and governed separately, so you can keep assisted drafting and turn off others, or the reverse.

Further reading

• Clinic settings, where roles and settings are found
• Getting started with Bio6, how a day unfolds
• Moving to Bio6 in 30 days, to read first if the cutover has not yet happened