title: "Copilot adoption for a national allied-health network" dek: "150 practitioners, 12 clinics, ~12% daily usage of a fully-paid M365 Copilot licence. Eight weeks later: 71%." sector: "Healthcare & Allied Health" client: "National allied-health network" engagement: "Practice Retainer — Adoption Programme" duration: "8 weeks" year: "2026" outcome: "Daily Copilot usage: 12% → 71% · ~4.5 hrs/practitioner/week reclaimed" solution: "8-week role-specific adoption programme — playbooks, training, governance brief, sustaining behaviour rituals." timeSaved: "4.5 hours per practitioner per week" visual: "adoption" cardFigure: "adoption" timeMetric: "4.5 hrs" timeMetricLabel: "saved / week / person" costMetric: "—" costMetricLabel: "no per-run cost" speedMetric: "5.9×" speedMetricLabel: "lift in daily usage" publishedAt: "2026-03-22" keywords:
- Microsoft Copilot adoption Australia
- healthcare AI adoption
- allied health technology
- AHPRA AI governance
The problem
A national allied-health network — twelve clinics across four states, ~150 practitioners covering physiotherapy, exercise physiology, podiatry and dietetics — had purchased Microsoft 365 Copilot in the previous financial year. The decision had been made at COO level, on the basis of a Microsoft licensing recommendation and what the COO later described as "a lot of optimism".
Twelve months in:
- Daily active usage: ~12%
- Most usage concentrated in two clinics, both of whom had a self-motivated early-adopter on staff
- Common feedback in clinic-manager check-ins: "I tried it. It’s fine. I don’t see how it saves me time."
- The licence renewal date was four months away
- The COO was being asked, by the CEO, why the firm was paying for capability the firm wasn’t using
The COO’s instinct — and to her credit, this is rare — was to assume the problem wasn’t the technology. The technology was fine. The problem was the firm hadn’t taught its people what AI was actually for in their job. So she came to us with a brief that was unusually narrow for a Copilot engagement: don’t build us anything; teach us to use what we’ve paid for.
What we did
This wasn’t an AI engineering project. It was an adoption project — which is why most consulting firms aren’t set up to deliver it well. AI engineering teams don’t do adult education. Adult education consultancies don’t understand AI well enough to teach it usefully. We hold both, and that was the fit.
The Programme ran eight weeks. Four phases:
Phase 1 — Diagnostic (Week 1). We shadowed three practitioners (one per discipline, one front-desk manager) through a normal day. Where did the cognitive load sit? Where did they reach for shortcuts the firm wasn’t supporting? What were they already doing manually that Copilot would have eaten in seconds — but didn’t know to apply Copilot to?
The answer, repeated three times: SOAP-note formatting, NDIS report drafting, patient-communication templates. None of which they were using Copilot for, because no-one had shown them that Copilot was a draft-writing tool, not a search box.
Phase 2 — Role-specific playbooks (Weeks 2–3). We built four short playbooks, each four pages long, each rooted in a real workflow the practitioners did every day:
- Physiotherapy: SOAP-note dictation → Copilot structuring → clinician sign-off
- Exercise physiology: NDIS exercise plan templates → Copilot personalisation → clinician review
- Podiatry: Patient communication letters → Copilot drafting in clinic tone
- Front-desk: Email triage, appointment-confirmation language, complaint-response drafts
Each playbook had a one-page version, a screenshot walkthrough, and a five-minute video. The five-minute videos were the most-used artefact — by a wide margin.
Phase 3 — Live training and governance (Weeks 4–6). Six 90-minute sessions, one per clinic group, mostly in-person, mostly during the clinic’s already-scheduled team meetings. We weren’t adding to anyone’s calendar — we were replacing fifteen minutes of "any other business" with practical AI training.
Alongside, we delivered an AHPRA-aligned governance brief to the clinical director: where Copilot could be used in patient-facing work, where it couldn’t, where human review was non-negotiable, and how the firm would document its position if asked. This was the artefact the clinical director said she’d been "trying to write for six months and never had the time".
Phase 4 — Sustaining behaviour (Weeks 7–8). A named adoption owner at each clinic. Weekly five-minute check-ins (we ran these for the first month after handover). A monthly "what changed" newsletter the firm could keep running internally. And usage-metric dashboards the COO could read at a glance.
The outcome — at programme end (week 8)
| Before | At week 8 | |
|---|---|---|
| Daily active Copilot users | 12% (~18 staff) | 71% (~107 staff) |
| Average time reclaimed per practitioner per week (self-reported) | n/a | 4.5 hours |
| SOAP notes drafted with Copilot assistance | 0% | 84% |
| NDIS reports drafted with Copilot assistance | 0% | 67% |
| Clinical-director-signed AI use policy | No | Yes |
| Practitioner-reported confidence ("I know when to use this and when not to") | 23% | 89% |
The COO renewed the Copilot licences. The clinical director presented the governance brief at the national clinical-managers’ conference — uncredited, which is fine; the brief is now informally circulating across the allied-health sector.
We’d already paid for Copilot. We needed someone to make us use it well. The adoption programme is now a model we use for everything we roll out, not just AI.
— Chief Operating Officer, National allied-health network
What we’d do differently
Map clinical software integration upfront. The biggest week-six question was "can Copilot read from our practice-management system?" The answer was not natively, but here’s how. If we’d mapped that in week one, we could have shipped an integration pattern alongside the playbooks.
Train the trainers, formally. We named adoption owners in each clinic in week 7 — too late. They should have been named in week 1 and trained alongside us through the programme, so they were ready to own it the day we left.
What we didn’t do
We didn’t build any AI. We didn’t deploy any agent. We didn’t train a model. We didn’t replace any clinical software.
We taught 150 people in twelve clinics what their existing AI licence was actually for in their job. The technology was the easy part. Capability transfer was the work.
This is what we mean when we say training is the moat.