Early trial shows human + AI model reaching 7x more priority patients
Early results from the first rollout of an AI Health Coach within New Zealand primary care suggest a hybrid model combining AI with clinical staff could help reach significantly more high-need patients, while increasing consistency of workforce practices and extending support between appointments without adding equivalent pressure to the workforce.
Tāmaki Health launched the AI Health Coach across selected clinics in July with New Zealand health technology company Groov, following its development and co-design with patients, workforce and clinical leaders. The rollout was the first time an AI Health Coach had been integrated into routine primary care in New Zealand, designed to work alongside Tāmaki Health’s existing workforce.
Early data from the first six weeks, across six Tāmaki Health clinics working with six Health Coaches, is now providing an indication of what that model could deliver at scale, including:
Proactive outreach using AI alongside human coaching engaged 7 times more patients than comparable previous human-only text campaigns.
Overall collection of patient-reported health outcome information increased by 8% with the AI Health Coach.
The AI Health Coach provided 105 additional patient touchpoints between appointments without additional workforce time.
Importantly, the outreach was not aimed at people already highly engaged with their healthcare. Tāmaki Health deliberately targeted people living with diabetes whose average blood glucose levels indicated poorly controlled diabetes and who had not attended a clinic for six months or more.
Tāmaki Health CEO Dr Lloyd McCann says the early results provide encouraging evidence for combining technology with the existing health workforce. “Early results are promising and indicate that further roll-out and investment where we blend the best of our human teams with technology is a sustainable path forward.”
Reaching people primary care needs to reconnect with
One of the clearest early results has come from Tāmaki Health’s proactive outreach to people living with diabetes who had not attended a clinic for at least six months. Rather than waiting for those patients to return to a clinic or be referred by a GP, they were proactively contacted and offered access to both the AI Health Coach and a human Health Coach. The outreach achieved a 2.7% enrolment rate compared with a 0.4% seasonal benchmark for comparable human-only outreach, representing a 7-fold increase. This result is based on a relatively small, higher-need cohort and is indicative rather than a controlled comparison, but provides an encouraging early signal about the ability of the model to engage people who may otherwise remain disconnected from support.
Groov CEO Matt Krogstad says the ability to proactively reach priority populations is where the potential health system impact becomes particularly important. “This is a group the health system really wants to reach: people living with diabetes whose blood glucose levels indicate their condition is not well controlled and who have been out of care for six months or more. Instead of waiting for them to come back through the door, we can bring support to them proactively.”
Across the pilot more broadly, enrolment in the AI Health Coach as a proportion of patients seen increased from 41% in week one to 75% by week six, 4.4 times the enrolment rate of the existing digital portal used as the closest available benchmark for this population.
“The early results suggest that combining the reach of AI with human healthcare can help us engage significantly more people without having to increase the workforce at the same rate. If we can reach people earlier and help them manage their health before problems escalate, there is a real opportunity over time to reduce avoidable pressure on both primary and secondary care,” he adds.
The early results are also encouraging from an equity perspective. Pacific Peoples and Māori together account for 56% of patients enrolled in the AI Health Coach, a 1.4x higher proportion than their representation across the clinics where the pilot is operating. Pacific Peoples make up 38% of AI Health Coach enrolments compared with 27% of the clinic population, while Māori account for 18% of AI Health Coach enrolments compared with 13% of the clinic population.
Tāmaki Health Chief Digital Officer Sam Ranchhod says increasing access cannot rely simply on increasing the number of appointments available. “Primary healthcare cannot simply rely on adding more appointments to meet growing demand. We need new ways to support people earlier, help them stay engaged between appointments and make it easier to access care when they need it.”
Collecting more health information while reducing administration A second important finding is the potential for AI to improve the amount and consistency of health information collected without creating another administrative task for frontline healthcare workers. Patient-reported outcome measures capture information directly from patients about their health and wellbeing and help healthcare providers understand whether support is making a difference. Collecting this information consistently is important for individual care, but also gives the wider health system better information about programme performance and where services may need to change.
In the first six weeks following the rollout, Tāmaki Health’s pilot Health Coaches recorded an 8% increase in the overall collection of patient-reported outcome information for patients using the AI Health Coach compared with their performance before the AI Health Coach was introduced. The biggest improvement was in collecting this information at the beginning of a patient’s care, where completion increased by 15%. Importantly, 55% of all collection was performed by the AI Health Coach and this has been accelerating, reaching 78% collection rates in week six and demonstrating the opportunity to relieve the workforce of significant administrative burden.
Among patients using the platform, 84% completed their patient-reported health forms through the AI Health Coach, meaning information that would otherwise require manual collection by the workforce could instead be provided directly by patients. Fifty-five per cent of patients also had an active health action plan on the platform.
Krogstad says this could become one of the most important benefits if the model is expanded. “What we are seeing here is the potential to collect more consistent information directly from patients while taking some of that administrative load away from healthcare teams. That helps the workforce, but it also gives the system better information to understand the return it is getting from healthcare investment and where improvements need to be made.”
Support that continues when life happens
The pilot also showcases how AI can extend the reach of primary care beyond the clinic itself. The AI Health Coach is available around the clock, giving patients somewhere to return to when they have a question, encounter a problem or need to adjust a health goal between appointments. It works alongside human care, with patients directed back to Tāmaki Health’s clinical teams when additional support is needed.
During the first six weeks, 27% of enrolled patients had a conversation with the AI Health Coach between clinic visits and 12% returned to check in on their action plan. Overall, the AI Health Coach created 105 additional patient touchpoints between appointments without additional workforce time. Patients rated its helpfulness an average 8.9 out of 10.
Dr David Codyre, Clinical Director, Wellness Support Team at Tāmaki Health, says extending support beyond the consultation is central to the model. “Most health improvement happens outside the consultation room. The AI Health Coach gives people practical, evidence-based support to turn advice into action, build confidence and sustain healthier behaviours between appointments, while ensuring they can be directed back to human care whenever needed.”
In one case, a patient raised exercise-related pain with the AI Health Coach that they had not mentioned to their healthcare professional. Their action plan could be changed immediately rather than waiting until their next appointment. Unprompted, patients also began holding conversations with the AI Health Coach in their native tongue. The multilingual capability had not been promoted and remains in beta pending further safety review, but the early behaviour points to another potential way of making support more accessible between appointments. Human oversight remains built into the model. Every AI Health Coach conversation generated during the pilot is anonymised and then clinically reviewed, with zero unsafe responses identified to date.
What could this mean at scale?
The next phase will extend the AI Health Coach to Tāmaki Health’s Health Improvement Practitioners. The same model also has the potential to be used for other priority populations where earlier engagement could change the trajectory of care, including people affected by gout or obesity, young people and pre- and post-natal populations.
Already, several large Primary Health Organisations have also expressed interest in implementing the model across primary care settings. Krogstad says the early results point to a broader opportunity to increase the capacity of New Zealand’s health system without assuming that every additional interaction with a patient requires another appointment. “New Zealand does not have an unlimited healthcare workforce, yet we need to reach more people, collect better information about whether our programmes are working and provide support when people actually need it.
“The opportunity with a hybrid model is to make the workforce we already have go further while removing burden from their heavy workdays. The early data suggests we can reach significantly more priority patients, collect more of the information the system needs and continue supporting people between appointments, while escalating them back to a human whenever that is the right thing to do.
“It is still early, but if those results continue as the model expands, this gives us a potentially scalable way to increase the reach and capacity of primary care while keeping people at the centre of it. This has significant positive impacts for the community, primary care and secondary care costs where high costs and long wait times are a major issue.”