At a glance
Industry
Healthcare
Location
Australia
Challenge
Regis Aged Care’s new generative AI assistant cuts clinical handover review from two to three hours to under an hour, allowing staff to spend more time with residents.
Success Highlights
- From two to three hours to under an hour to review 200+ pages of handover notes
- Zero hallucinations across trials with accuracy at or above a registered nurse baseline
- Across more than 70 homes, senior clinicians rated the assistant "very easy" to use
The challenge
Regis Aged Care is one of Australia’s largest residential aged care providers, supporting thousands of residents across more than 70 homes. Every shift begins with an essential handover process for staff to understand the needs of every resident across the 24-hour cycle. With handover notes at any given residence covering 200-plus pages of clinical notes, reviewing can take up to two to three hours per day. The operational load was becoming substantial for team members and an impediment to maximizing the time spent directly working with residents.
Our approach
Cognizant worked closely with Regis to understand the lived experiences of residents and the daily challenges faced by staff, establishing a deep engagement from the start to genuinely understand both the opportunity and the risks. Through a series of workshops and technical explorations, the teams worked together to identify the specific, complex business challenge that would deliver the most value and serve as a launchpad for future AI initiatives.
The collaborative process was essential for designing and delivering a clinically trustworthy generative AI assistant that ingests clinical progress notes, produces structured summaries for each resident and answers both standard and ad hoc queries through a familiar chat interface.
Beginning with a proof of concept in two Regis homes, a trial was conducted with a solution built on Microsoft Copilot Studio with the authentication and security controls required for a healthcare environment. Following the successful trial, the solution was moved into full production across the Regis network.
Engineering a clinically trustworthy assistant
With such lengthy documents to handle, designing the AI to know how to process over 200 pages of notes required careful testing. The techniques used ensured the model could process the material in segments while never losing the thread of information, which was then reassembled to be analyzed by an extensive custom prompt. The prompting runs to several pages of logic, structured like a natural-language code where each small linguistic choice holds significant importance in achieving the required outcome.
Tuning was treated as an ongoing scientific process, not simply a one-time configuration. We worked iteratively to achieve the required precision expected by the Regis clinical team.
Benchmarking a gold standard
We used a benchmarking process designed with a senior panel of clinicians who decided on a set of control answers across a range of health categories. The AI assistant’s outputs were compared to those of a registered nurse as well as to a default Microsoft Copilot baseline, covering clinical concerns such as agitation, medication, infection and non-regular medication. While the model matched or exceeded a registered nurse baseline for a majority of categories, the benchmarking process is repeated on a regular basis to monitor ongoing performance. Maintaining high confidence in the AI assistant’s abilities helps Regis and its teams know how best to feel supported by the tool so they can do the very human work of supporting residents in care.
Preparing clinicians for an AI assistant
The rollout went beyond technical excellence, requiring a holistic change management program that was critical to adoption. Regis understood that the introduction of AI created uncertainty for some staff and that building trust was essential.
The program focused on addressing these concerns early, building an understanding of the AI assistant not as a replacement, but as a supportive tool. The goal was to give staff the ability to spend more time with residents, focusing on human outcomes over operational tasks. This approach has proven successful, with uptake considered very strong compared to previous system rollouts. Nearly 60% of staff are now frequent users, utilizing the tool daily or several times a week. Regis continues to work with hesitant staff to build their confidence and familiarity with the tool.
Business outcomes
Having met stringent requirements, including zero hallucinations, the Regis AI assistant was deemed a success and has expanded from its two-site trial into full production across all Regis Aged Care Homes. Around two hours per clinician per day are saved; across more than 70 homes, this translates into thousands of hours annually redirected toward safer, more timely care for the residents.
- From two to three hours to under an hour: Reviewing 200-plus pages of handover notes is now a task that takes under an hour for staff to review.
- Zero hallucinations in trial: Through rigorous technical design, prompt engineering and benchmarking, no hallucinations occurred across the trials and accuracy of note review was the same or better than the registered nurse baseline on the majority of clinical note categories
- Ease-of-use rating: Most senior clinicians rated the AI assistant “very easy” to use, with daily or several-times-weekly usage by close to 60% of users, with hundreds of thousands of interactions logged since launch
