Clinic Policy Concierge
Staff ask about an SOP, a payer rule, a consent requirement or a care protocol and get a short answer that quotes the exact section it came from — and says "not in your documents" instead of guessing.
A retrieval assistant that answers only from the practice's own approved policies, cites the section, and refuses when the answer isn't there.
Where this works well
The slow, invisible cost this makes visible is interruption. In a busy general practice the answer to "how long do we keep a minor's health records?" or "what's the cold-chain breach procedure?" usually lives in a senior person's head or buried in a SOP nobody can find fast — so a nurse interrupts the practice manager, or guesses. This pattern earns its keep when you have a genuine body of approved, written policy (privacy and records SOPs, infection control, cold-chain management under RACGP Criterion GP6.1, a staff handbook) and a team that asks the same operational questions repeatedly.
It benefits the front line most: reception, practice nurses and new hires in their first weeks, who need a grounded answer with the source section attached rather than a confident chatbot guess. The practice manager and clinical governance lead get the second-order benefit — fewer "quick questions," and a log of what people are actually unsure about.
Where it works badly
It is confidently useless when your "policies" aren't really written down. If the real procedure lives in custom and corridor knowledge, the tool will correctly refuse most of the time, and a wall of refusals is just an expensive way to discover your documentation gap.
The sharper failure mode is conflicting versions. If both a superseded and a current infection-control policy sit in the corpus, retrieval can cite the wrong one — and because the answer comes with a confident-looking section reference, it reads as authoritative. It also has no awareness of regulatory change: if RACGP or your state health authority updates a requirement and you haven't re-indexed, it will quote your old policy verbatim and call it grounded.
Honest test: open your shared drive and try to point to the single, current, approved version of your records-retention and cold-chain policies. If you can't do that in under a minute, the document work comes before the AI layer.
What it doesn't do — and shouldn't
It surfaces what your approved documents say and cites where; a person decides what to do with it. It does not interpret a grey area, weigh a clinical judgement, or approve its own answer. Each response is a draft a clinician approves, edits or flags, and every approval is logged to an audit trail. When the corpus has no answer, it refuses and escalates to the Practice Principal — it never fills the gap with a plausible guess.
That human-in-the-loop boundary is deliberate and it matters here for two reasons. First, this is a policy and SOP assistant, not a clinical decision aid — patient-specific clinical decisions are out of scope by design. Second, health information is regulated as sensitive information under the Privacy Act 1988 (Cth), so the accountable person staying on the decision is not optional. We accelerate the thinking; we don't replace the thinker.
What your data has to look like for this to work
You need a defined set of approved documents, each with a version and a date, owned by someone — the demo's corpus is four: a Patient Records & Privacy SOP, an Infection Control Policy, a Cold Chain Management Policy and a Staff Handbook. The documents need internal structure the tool can cite back to: numbered sections (a "§4.2") rather than a 40-page PDF of unbroken prose. There must be exactly one current version of each, with superseded copies removed from the index, and a re-index step that fires whenever a policy is approved or retired.
Most practices have some of this and not all of it. The policies exist but live in three different drives at two different version numbers; the section numbering is inconsistent; nobody owns the "is this still current" question. Fixing that is usually the real first job, and it's rarely about buying a tool — it's about how policy is captured, versioned and owned. That document work is what RMAI helps with, and it's typically larger and more valuable than the retrieval layer sitting on top of it.
FAQ
(Rendered from front-matter.)
Could it tell a nurse the wrong thing and have them act on it?
Every answer is a draft, not an instruction, and it only quotes from your approved documents — the section reference (e.g. "Patient Records & Privacy SOP v2 §4.2") sits next to the answer so the person can open it and check. When nothing in the corpus covers the question, it refuses and routes to the Practice Principal rather than guessing. A clinician approves, edits or flags each answer before it's relied on; the tool never acts.
Our policies are a mess of old Word docs and a few things only the practice manager knows. Will it work?
It only answers from what you've actually indexed, so undocumented knowledge in someone's head is simply a gap it will refuse on — which is the honest result, not a failure. Conflicting or superseded versions are the real risk: if both the 2024 and 2026 infection-control policies are in the corpus it may cite the wrong one. Getting to a single current, owned set of approved documents is usually the first piece of work, and it's the part we help with.
Does this replace our practice manager or clinical governance lead?
It replaces none of them. It removes the interruptions — the "where's the cold-chain breach procedure" and "how long do we keep a minor's records" questions that pull senior staff off higher-value work. The judgement calls, the policy ownership and the decision to act stay with the people accountable for them. Capacity gets recaptured, not headcount cut.
How current does the answer have to be, and what stops it quoting a withdrawn policy?
It's only as current as your indexed corpus — it answers from the version you've loaded, with the document version and update date shown (e.g. "v4 · updated Jan 2026"). It has no live feed of regulatory change. The discipline that makes it safe is a defined re-index step whenever a policy is approved or retired, with one named owner; without that, stale documents are the main way it goes wrong.
Where does our patient and policy data go — does it leave the practice?
Health information is "sensitive information" under the Privacy Act 1988 (Cth) and the Australian Privacy Principles, so this runs inside your own tenancy against your own document library — it isn't a public chatbot and your policies aren't used to train a shared model. The demo answers operational policy questions, not individual patient records. Exactly where the corpus and the model calls sit, and who can query them, is part of the scoping we do with you up front.
What happens when someone asks a clinical question it shouldn't answer?
It's a policy and SOP assistant, not a clinical decision aid. A question like medicinal-cannabis prescribing isn't in the approved operational corpus, so it refuses and escalates to a person — exactly the behaviour you want. It should never be wired to patient-specific clinical decisions; that boundary is deliberate.
Estimated build: 3 weeks. Most of it is template work we've already done.
Fixed scope, fixed price, fixed dates.
The cost band reflects the engagement shape, not a per-feature line item. We work on fixed scope, fixed price, fixed dates — see the services catalogue for what falls inside each band.
Considering this for your org?
The honest place to start is a bite-sized first piece — one contained change, low risk. Tell us where it hurts; we’ll play it back, scope it, and show you what’s possible.