Australia's aged care sector struggles with a double bind: a rapidly ageing population and a chronic shortage of skilled nursing staff. In residential homes, the clinical handover – the moment when one shift of carers briefs the next – often becomes a bottleneck. Mounting paperwork, inconsistent note-taking and the risk that a critical observation about a resident slips through the cracks all contribute to a dangerous information gap. Regis, one of the country's largest providers with 72 homes, has answered that challenge with a custom AI agent built on Microsoft Copilot Studio. Since September 2025, RegiCare Assist has been silently reshaping how frontline teams manage the flood of overnight reports and shift-change briefings.

The tool performs three core tasks every morning. It digests all handover notes logged by the night shift, distils them into concise summaries that highlight changes in a resident's condition, flags urgent concerns – a missed medication, a new cough, a fall risk – and sorts the aggregated reports so that the incoming team can immediately see which residents need the most attention. What used to take a clinical coordinator up to three hours of trawling through typed and handwritten notes now takes minutes.

The clinical handover: a perennial pain point

Handovers in aged care are a high-stakes communication ritual. A 2023 study by the Australian Commission on Safety and Quality in Health Care found that 42% of adverse events in residential facilities were linked to poor shift-to-shift information transfer. Nurses frequently rely on paper logs, verbal cues and ad-hoc sticky notes. Even homes that have adopted electronic health records often face an avalanche of unstructured text – free-form notes that are rich in context but cumbersome to search.

Regis knew the problem intimately. With 72 sites scattered across metropolitan and regional Australia, standardising the handover process was an operational headache. Each home had its own flavour of documentation, from bespoke templated forms to clinical software that produced free-text summaries. Attempts to enforce rigid templates had historically failed; carers needed the expressive freedom to capture nuance. The AI solution had to work with that messy reality, not against it.

From chatbot to clinical companion

RegiCare Assist did not begin as a grand digital transformation project. According to Regis's internal innovation blog, the genesis lay in a hackathon where a small team prototyped a Q&A bot that could answer simple questions from the organisation's policy library. The prototype, built with the low-code capabilities of Copilot Studio, demonstrated that a conversational AI could parse dense policy documents and return accurate, sourced answers. That pilot so impressed the executive that the team received a green light to tackle the handover challenge.

The jump from policy Q&A to clinical summarisation required a fundamentally different architecture. Instead of a simple bot that retrieved documents, the team needed a retrieval-augmented generation (RAG) pipeline. Every morning, the system pulls worker notes from the electronic care planning system – a mix of structured fields like medication times and unstructured remarks like "Stan seemed confused at 3am". These notes are chunked and indexed in a vector store. When the day-shift coordinator opens the RegiCare Assist dashboard, a Copilot Studio–orchestrated flow fires queries against that index: generate a summary grouped by resident, list potential red flags, and rank the most critical tasks.

Inside the architecture

The solution leans heavily on native Azure services, all mediated through Copilot Studio's newly expanded agent capabilities. The Copilot agent orchestrates the pipeline by calling:

  • Azure AI Document Intelligence to extract text from scanned PDFs of handwritten notes that some homes still submit.
  • Azure OpenAI Service (GPT-4o model) to perform the summarisation, flagging and triaging.
  • Azure AI Search as the vector database that stores embeddings of the previous night's notes.
  • Power Automate flows to schedule the 6am batch job and distribute the output to facility managers' Teams channels.

Crucially, the sensitive resident data never leaves Regis's tenancy. The entire pipeline runs inside the provider's own Azure landing zone, governed by the same role-based access controls that protect the electronic health records. Compliance with the Aged Care Act and privacy regulations was a non-negotiable design principle.

Measurable impact

Regis shared preliminary metrics from the first four months of operation. Across the 72 homes, the AI now processes an average of 2,400 handover entries per day. The time saved per facility equates to roughly 15 nursing hours a week – time that has been redirected into direct resident care. More telling, the number of incident reports attributed to missed handover information dropped by 28% over the same period.

The tool has also uncovered blind spots. Because the AI applies a consistent set of severity rules to every note, it has begun flagging subtle deteriorations that a tired human reader might overlook. In one instance described by Regis's chief clinical information officer, the system alerted a facility to a resident who had recorded a slightly elevated temperature for three consecutive nights. The pattern was confirmed as an early urinary tract infection, treated before it became acute. "Previously, that trend would have been buried in three separate shift reports," the officer noted.

Staff adoption and cultural change

Introducing AI into a care setting is as much a people challenge as a technology one. Regis ran a deliberate change management program alongside the rollout. Instead of imposing the tool from head office, the innovation team spent six weeks in a representative sample of homes, refining the user experience with direct feedback. Nurses and carers played a significant role in shaping the summary formats and the severity classifiers.

One early finding: the AI needed to explain its reasoning. So the final interface includes a "show evidence" button that highlights the specific sentences in the original notes that informed each summary point. This transparency built trust. It also helped clinical leads dispute a classification when the AI got it wrong – which happened in about 8% of cases during the first month, typically because of colloquial language or sarcasm in the notes. The team uses that feedback to continuously fine-tune the prompt templates and classification logic.

Beyond the handover

With the morning handover summarisation now stabilised, Regis is looking at extending the Copilot agent's responsibilities. A pilot is already underway to have the same AI analyse weekly trends across a facility and generate a pre-read for the fortnightly clinical governance meeting. Another trial is investigating whether the triage engine can be deployed in real-time during the afternoon shift when the bulk of doctors' calls and family updates occur.

The provider is also exploring how RegiCare Assist can become a two-way knowledge base. Currently, it reads notes and produces outputs; the next phase would allow carers to ask natural language questions about a resident's recent history, such as "Has anyone reported eating difficulties for Margaret over the last two weeks?" The RAG backbone makes this technically feasible, though the team is cautious about adding a query layer until the summarisation feature has been hardened in more complex scenarios.

Broader implications for aged care technology

Regis's experience offers a template for a sector that has been slow to embrace advanced data tools. Most aged care software in Australia remains firmly rooted in the Electronic National Residential Medication Chart (eNRMC) or basic care planning modules. Integrating generative AI into these workflows while respecting privacy constraints demonstrates that the technology can be made safe, governable and valuable.

The implementation also underscores the maturation of Microsoft's Copilot Studio. Its evolution from a chatbot builder to a full-blown AI agent orchestrator means that organisations without deep machine learning teams can stitch together powerful RAG workflows. Regis's digital team consisted of just four developers and a product manager – a lean crew by any measure.

Challenges ahead

The project is not without its risks. Maintaining the accuracy of the AI as local documentation styles drift is a continuous governance task. Hallucinations – where the model invents a plausible but non-existent symptom – remain a theoretical concern, although the RAG design and the evidence-linking feature appear to have kept them to negligible levels so far. Regis acknowledges that the tool is a decision-support system, not a decision-maker; clinical judgement always has the final say.

Cost is another factor. While Azure consumption for the batch summarisation is modest, the vector storage and the growing corpus of historical notes will push up the monthly bill. Regis is amortising the expense against the nursing hours reclaimed, and the early return-on-investment calculations look promising. Still, smaller providers with tighter budgets may struggle to justify the initial investment.

Lessons for the community

The RegiCare Assist story surfaces several principles that apply beyond aged care. First, the hackathon-to-production path shows that modest, curiosity-driven prototypes can scale when they solve a concrete, daily pain. Second, designing for messy, real-world data – messy handwriting, colloquial notes, incomplete entries – is essential; a system that demands clean, structured inputs will inevitably fail in a care environment. Third, the people side is paramount: the AI's usability hinged on its ability to show its work and invite corrections.

For the broader Windows and Microsoft ecosystem, Regis's journey is a testament to the versatility of the Copilot stack. While Copilot for Microsoft 365 often grabs headlines, niche agents built in Copilot Studio are quietly transforming frontline industries. Aged care, which sits at the intersection of healthcare, hospitality and operational logistics, is an ideal sandbox for these purpose-built agents.

What's next for Regis

Looking ahead, Regis plans to open-source some of the prompt engineering patterns and severity classification heuristics they have developed, contributing to an emerging community of practice around responsible AI in aged care. They are also engaging with the Digital Transformation Tech Talk series hosted by the Department of Health and Aged Care to share the blueprint with the sector.

As the sun rises over another Regis home, RegiCare Assist will be parsing the night's collection of notes, turning a chaotic set of observations into a clear, prioritised picture of who needs what. It is a quiet revolution – one that promises to give carers back time, give residents better safety, and give families peace of mind. The handover will never be just a clipboard and a coffee chat again.