Shirlaw AI Consulting · Discovery Brief
We looked at what AI clinical scribe tools already exist in the Australian market before scoping a custom build, and sized a lean v1 of a purpose-built scribe for IntegratePHYSIO.
This brief supports scoping the AI scribe — the highest-priority item in the wider engagement — for a development-hours estimate, ahead of a formal quote. It covers what's already on the market, so we're not building something that already exists, and a recommended v1 build approach with its estimated development time.
| Tool | AU data residency | Coreplus integration | Physio-specific | Pricing /clinician/mo |
|---|---|---|---|---|
| Lyrebird HealthMelbourne | AU-hosted, but discloses overseas processing of de-identified data via Azure OpenAI | NoBp Premier, Genie, Gentu | No | Free – A$160–240 |
| Heidi HealthMelbourne | AU-founded | NoBest Practice, Zedmed, Genie | No | A$85–150sources disagree |
| Medow | AU servers; PII de-identified before AI; 7-day audio deletion | NoGenie, Gentu, Elixir, Meditech | No | Custom quote |
| i-scribe | Claimed AU | NoGenie, Gentu, Shexie, Best Practice | Unclearan “Allied Health” tier exists | A$150–200 |
| MedicalDirector Smart Scribe | AU | NoMedicalDirector Clinical only | No | Not disclosed |
| MedTalk AI | Claims “100% AU residency”; privacy policy permits US disclosure | NoBest Practice, MD, Genie, Epic | No | A$39cheapest |
| HealOS | Not stated — HIPAA / SOC2 / BAA language (US framework); no AU-residency claim found | YesChrome-extension bridge, human confirms each write | No40+ specialties listed; physio not among them | Not disclosed |
| Perci Health | AU confirmed; APP / PPIP / HRIP compliant | Unclear“uploads to your practice management system,” unspecified | Yesbuilt for physiotherapists | Not disclosed |
| NirvaScribe | Claims APP + HIPAA + SOC2 | Unclearnot mentioned | Yesphysio + 8 other allied-health disciplines, NDIS-aligned | Not disclosed |
| Colinic AI | AU-based (.com.au); APP-compliant | Uncleargeneric “EHR integration” only | Yesphysio SOAP notes, NDIS/WorkCover aware | Not disclosed |
The Coreplus + physio combination doesn't exist off the shelf
HealOS is the only vendor confirmed to write into Coreplus, and it's a generalist tool with no confirmed AU data residency. The physio-specific tools look like better clinical fits but none confirm Coreplus support in public materials — that needs a direct conversation to resolve.
“Australian data residency” marketing often has fine print
Lyrebird's own privacy language admits de-identified data goes to Azure OpenAI overseas. MedTalk advertises “100% Australian data residency” in one place and permits US disclosure in its privacy policy. This is exactly the trap the APP 8 constraint in the original engagement brief warns about — any vendor's actual privacy policy needs reading, not their homepage.
There's an active regulatory warning on this exact category
The Department of Health and Aged Care and the TGA reportedly warned in 2026 that AI scribes widely operate without valid patient consent, carry data-sovereignty risk from offshore processing, and sit in a regulatory gap that escapes TGA device oversight entirely. This reinforces why consent capture and human sign-off are being treated as required, not optional, for our build.
v1 scope: real (not simulated) voice capture — tested specifically for accuracy against accented speech and treatment-room background noise — feeding an Australia-resident transcription service, into an LLM step that produces a structured note in physiotherapy terminology, output as clean copy-paste text for manual entry into Coreplus.
No Coreplus API integration yet; no consent-capture or audit-logging infrastructure yet, on the assumption this is exercised on staff/synthetic test consultations rather than real patients (flagged in Open Questions).
Why phased, rather than building native Coreplus integration first: Coreplus requires its product team to review any integration before production use — a lead time outside our control. Shipping a copy-paste v1 first means the note-quality and terminology work can be validated independently of that approval process, with native API write-through layered in once the notes are proven good enough to be worth automating.
| Phase | Scope | Hours |
|---|---|---|
| 0 | Discovery — confirm AU-region STT/LLM provider; gather sample physio notes from Sam per service line (musculoskeletal / pelvic health / NDIS / WorkCover) to calibrate terminology | 6–10h |
| 1 | Recording capture + real-world testing — browser-based recorder, plus a deliberate accuracy pass across accents and treatment-room-like background noise | 15–25h |
| 2 | AU-resident transcription — integrate an Australia-region (or self-hosted) speech-to-text service | 10–18h |
| 3 | Structured note generation — transcript → structured draft note in physio terminology, with guardrails against adding any diagnosis or interpretation the clinician didn't state | 20–30h |
| 4 | Copy-paste output — clean formatted text, easy to copy into Coreplus | 6–10h |
| Total — v1 | ~57–93h | |
If v1 needs to handle real patients rather than staff/synthetic test data, add consent capture (+8–12h) and a basic audit/event log (+15–25h): ~80–130h total.
Beyond v1 — native Coreplus API integration for both tenants (~25–40h), a formal logged sign-off workflow if not already covered above (~15–25h), and UAT/rollout across all ~15 clinicians (~15–25h) — roughly 55–90h more, scoped in detail once v1 is validated.
These are order-of-magnitude ranges for a quote conversation, not a fixed-price quote. They'll tighten after Phase 0 discovery resolves the open items below.
Buy — commercial scribe
A$15,300–28,800/ yr
≈15 clinicians × A$85–160/mo realistic paid tier, recurring and scaling with headcount.
Build — custom v1
57–130hrs, one-time
Plus modest ongoing hosting/API costs — materially lower than per-seat SaaS at this scale.
Not a clean like-for-like: none of the commercial options are Coreplus-native except HealOS, which isn't physio-tuned or confirmed AU-resident. It's a reference point for the conversation with Sam, not a verdict.