Heidi: the AI scribe shaking up NZ healthcare

WEBVTT - Heidi: the AI scribe shaking up NZ healthcare This week on the Business of Tech, we're heading into the emergency department and the GP's office to look at one of the most concrete examples of artificial intelligence actually changing how healthcare works. Yes, Heidi Health, the Australian startup that wants to double the world's healthcare capacity without dehumanizing it. You're listening to the Business of Tech. I'm Peter Griffin and my guest this week is you Liu, co founder of Heidi Health, which has become one of the most widely deployed AI tools in New Zealand's public health system.

If you've turned up to the emergency department in the last few months, there's a good chance your clinician has had Heidi running in the background, listening to the consultation and drafting the notes, referrals and follow ups that used to soak up their time at they're hours. In Hawks Bay Hospital's emergency department, a pilot saw documentation time drop from around seventeen minutes to just over four minutes per patient, freeing clinicians to see roughly one extra patient per shift and slashing after hours paperwork by over eighty percent.

That's the promise of AI and healthcare pretty much in a single statistic, more patient scene, less burnout, and a better human interaction in the consultation room. You and his co founders actually started with something different called Oscar, a medical chatbot to help students practice their exam skills, essentially a simulated patient to train bedside manner and clinical questioning. It was clever, useful, very technical, but there was one big problem. No one was willing to pay for it at scale, and it didn't touch the most painful bottleneck in healthcare, clinicians drowning in admin instead of spending time with patients.

So You and the team of it at hard into the clinical world. They built Heidi Scribe, an ambient AI scribe that sits in on consultations, captures the dialogue, and produces high quality clinical notes tailored to each hospital's templates and workflows. It turned out to be a hit almost from day one. You says they saw doctors using it in every consultation, calling the founders directly when it went down because they literally couldn't imagine going back to typing up everything themselves.

That kind of dependence is rare for a young startup, and it underpinned Heidi's rapid spread around the world investors have noticed. Heidi first raised around six point three million US in a Series A funding round in late twenty twenty three to expand its aiscribe and GP footprint in Australia. Last year, it followed up with a sixteen point six million US round led by Headline, with Anthropic Linked Anthology, Local Globe, Blackbird and others owning in to accelerate the mission of doubling global healthcare capacity.

Then came a US sixty five million Series B, valuing the company at around four hundred and sixty five million US and lifting total funding so far too close to the one hundred million dollar US mart. So what does doubling healthcare capacity actually mean? The way you Liu puts it, if you give every clinician an AI partner that handles the documentation, form filling, evidence search, and routine follow up communication, you effectively add another pair of hands to every consult without hiring another doctor.

Today, Heidi says it supports more than two point five million consults every week and one hundred and ten languages, and a company estimates it has already returned tens of millions of hours to frontline care in New Zealand Health and Z is rolling Heidi's aiscribe out across all emergency departments with around one thing clinician licenses and additional licenses from mental health crisis teams, and Heidi is now moving beyond being just described. The company has launched Heidi Evidence, which surfaces relevant clinical research and guidelines at the point of care, and it's expanding into pre chart summaries, automated referrals, and spoken commands that trigger real actions in health systems and email.

The long term vision is an AI care partner that takes over everything non clinical, the admin, the follow up logistics, the task management, so clinicians can focus on diagnosis, judgment and human connection. So in this episode, we'll unpack how Heidi went from a student training tool to a globally scaled AI care partner, what it took to get New Zealand's public health system comfortable with ambient AI in the consultation room, and whether doubling healthcare capacity is a realistic goal or a seductive pitch line.

Here's my interview with Heidi Health co found Yuliu. You welcome to the business of tech. Great to have you, Heidi. I think is a company that is sort of getting on the radar of a lot of kiwis, both in the venture capital world.

Very impressive capital raising story that Heidi has had to date in the venture capital community, people in the startup community very interested in the company, but also in the health sector right across now licensed software across every emergency department in the country. Clinicians all over the country are now using Heidi. I think it's probably one of the most popular AI tools deployed across the New Zealand health sector. So really it's a drill into how you manage to get that uptake so quickly, but really want to start at the beginnings for people who don't know much about It's a really interesting story from.

What I can tell. I think you started with a slightly different startup in mind. It was Oscar was really around clinical training and it was a pivot there towards AI. Documentation tell us about that.

Not many companies or startups have the prime or have their luck to build one thing and keep building that thing for many, many years towards the success, and we certainly are not one of them. So we're started with the Osborne, which is a medical chap up for medical students to tlean up their oscy skills. It's really need for them to learn how to talk to a patients when patients comes in to see them presenting these issues. And we've spent a lot of time on that product for sure, and then I myself built entire like back end systems and also all the machine learning models training out from scratch and figuring out how we can give better training to students when this situation happens or when that situation happens.

It takes a lot of the considerations on tech and a product to make to a level of the students can verny use it to train their skills. But eventually we found out this product is great, it's a nice tool, very useful for students to learn and practice their skills, but eventually it doesn't solve that impact for problems, which results in no one is learning willing to pay for it, especially for students. They are not at their prime time as a doctor yet later they will be, but not at their student period.

So when pivoted the businesses model to focus more on the clinical industry where we think we can make big their impacts, and then we see a lot of gaps in the industry operations. All doctors are spending a lot of time in documentations, in different admin tasks that they have to do well as they could rather spend more time with the patients doing eye contacts or doing more like human stuff that only humans can do. And that's the original idea of hiding and of course, like we have been on a roller coaster going around from different products and finally converted into this idea I described.

And since the very beginning we've seen great uptick on the usage data the doctors, the early adopters who want to use it, they use it every day, use they use it for every session they do with their patients. And whenever we have any issues with our models, with our systems, they will called split up to our phone numbers asking what's going on because they can't they can't do the consults without our stacks anymore. And that's the moment we figure out this is something different that is really really impactful.

And two and a half years later now we've seen great results. We've saved more than forty seven million hours across the globe and that time can be really invested into patient care. That means more patients get seen and this is like very simple data. But beyond data, there are a lot of behavior changes we are observing in how doctors are really treating the patients.

In the past, they have to make sure they have a really complete notes capturing everything. So while they are doing consults with patients, they have to spare some brain powers spring on the time to put all the fundings and the learnings into the computer, so they have to face the computer typing and all things seen. Well as now they don't have to do that anymore. They just focus on the patients because the haiti as our AIK partner will set aside will help them to document everything even in a better quality, in a more complete way.

So that's that's the impact we are making, like giving learning, giving the better experience to patients as well, and not only to save more doctors time. It's an incredible story. And when we talk about the promise of artificial intelligence that the supposed productivity gains, I think this is one use case that it is emerging very early on as a huge productivity booster. You know, a lot of us go into our general practitioner, our local doctor, and it frustrates me that my doctor who is fantastic, knows me for many, many years, but doesn't use these tools, so there's no record of our conversations or the nuance off it, what was discussed.

None of that goes on to my patient portal, So it's it's a very old fashioned way of doing things. She probably spends twenty minutes or something writing up the notes about our meetings, so it's automating all of that. Just before we get further into what Heidi actually is interested in. The founding team here, so you've got doctor Thomas Kelly who was a vascular trauma surgeon, and Willie Musa.

Where did you meet these two? We met at a previous business called Bruiser's Cancer. Tom and Walid we're building a tutor and business for the students. That's how we came up with the idea of having Oscar to clean up the medical students on the joined that company as a technical like a leader, and worked on this Oscar project as a side project.

That's very beginning, but literally when we brought this, when we brought this product into a more complete form, we figured, oh, there are more impacts we can do, and that's how we met each other. And how we get to work with each other. For the last six years, it hasn't been always in there like very smooth period. We've been a.

Lot of the downs as well, and I can remember there are a couple of times we just figured, oh, this product direction just doesn't work. We have to do something about it. On that pay is runly tough because every day we cut, you have to think about what am what am I going to do today? Do we have we find the right direction or it's the right direction?

Do we need to do more? Like you always feel you can do more as a founder, that's for sure, but when you don't have the direction, that's the that's the worst bit and good thing is not for long. We because we are a group of people full of ideas. We have new ideas every every single day until today.

So we have lots of ideas, which is don't know which one is the is the. Is d one? The words us investing the next years, next ten years, and later on we figured, oh, we want to do more for clinicians, for health organizations to bump up more productivity than double up the capacity for the doctor's care. And that's why we started from scribe.

We think that's the biggest gaps at the moment in the system. But now we've seen actionally more so beyond documentations, doctors are doing a lot are doing a lot more at main tasks which consume a lot of time, I AM and energy. We want to take over more admitting tasks for them. That's why we launched highly Evidence where we can enable doctors to do more AI enabled searches and we bring those cominical facts, relevant communical facts in front of doctors so they can make better judgments and they can do more like with more confidence.

And also we are looking to do more like computer use side of things for doctors and what it means is in the past, like when you want to do any action on HR or any other softwares, you have to click through a lot of screens and feel in this form, feel in that form cnnick the button and it pop up a new screen and you need to select a bunch of fields and then you can click into the next screen. And that time think is really huge and doctors spend a lot of time on that wellas now with our Heidi product, you can just automated your entire workflow with that.

So that's what we are looking to do next to deep thought more productivity for key issues. Yeah, you integrated into clinical workflows and the platforms that health systems are using. You started, I think in twenty nineteen, so that was a tough time I guess to start a startup. You had the whole COVID era, so even collaborating was challenging.

But that was also before the debut of chat, GPT and large language models like that. There was clinical transcription stuff around in that era, but to what extent really did the debut of llm's going mainstream really changed the game and inspire you that, wow, there's something addition know here that we can turn into something really powerful. I came from a traditional machine learning background, so we've played around in the different transformer models and lots to train them from scratch on different tasks, and it is not as as weak as we would think now in l M era.

It's still super powerful. If you have the data and you train your model to words of one certain task, it really works super super well. These are on how we find How we find the GVT models much more powerful is the fact that you don't need to train them much anymore, and then you can get to a good state, and that's when we realize, Okay, this is going to be a game changer. And it already has a lot of capabilities of dealing hundred or south zions of different tasks and super super powerful.

But on that note, is super super good for general tasks, for many many tasks, but we in the in regards to one specific task, for example, describing documenting the clinical documents for doctors, it can do eighty percent ninety five percents, but you can never do ninety nine percents or even close to one hundred percents. Why because the general models they optimize for the general tasks. They not assais that they have are about hundreds of thousands of different tasks, so they optimize for general capabilities, whereas in healthcare we do need more specialized models to make it super super powerful.

For clinical tasks, general models can proprietary models can give ninety five percent, whereas in Heidi Heidi, specific clinical models can do ninety nine percent. People be saying these four percent difference is not huge. Well as in reality, if doctors their notes always have four percents, just can't get it right. This the only need to spend like two millions five minutes on the notes to make sure right, they need to scan everything.

Whereas at Heidi we can give more confidence, we can give higher accuracy, so doctors can actually spend less time to add it the clinical dotomatation we present to them. So that's our unique advantage. So you're sort of using a blend of you know, the big models that the large language model frontier model creators are building, the Geminis and the Claudes, but you've developed your own models as well to get that higher accuracy and that what's. Needed for the clinical setting.

That's correct and also anymore so for us. Rule number one is the model needs to sit in that region to serve the region sales organizations or clinicians are working there. It has to be localized the data sovereignty is not a not something we can compromise whatever. And secondly is the model we used, we will train it up towards the clinic.

Greed, what I mean is a lot of models. Yeah, they are great gut models, great German and models agree, but they often get medicine names around, for example, or they don't get the right context. To to to to to. Figure out what doctors really want to put into the notes.

That's where Highly our own models can you know, uplift the the the accuracy or laws. So take us into the that the life of a clinician uses Heidi. Maybe in an emergency department, you've got someone coming in, maybe they're not in great shape, they've been in an accident or something like that. Doctor wants to have a conversation trying to establish what conditioner and I guess that's when you start heading record or is it ambiently recording all the time.

It depends on different scenarios and for ED doctors. Now we have released a product called highly Remote, which is a tiny AI weariable where you can clip to your closest or you can put it on your linear and you can just wear it, go into the room and when you are about to go to an operation or go into a consult with the patients, click the button on what we are. If you want to leave it, that's fine, but we would suggest that doctor scale do multiple presses to separate different sessions.

So imagine a real ED departments. I am a doctor, I'm going to see these patients before I see it, parents button on the variables and I talked to the patients completely, no need to worry about the documentations. Feelish the consult known for us, and then I'll have the notes showing up on my computer already. So that's the workflow we want to bring into the to the ED department.

And I believe I think you and you can still do it. But you're doing a lot of sort of live transcription, but you sort of now you're batching it. It's more efficient to save those recordings and then do all the transcription and everything in one go. That's right.

So one thing is we don't ever store any recordings, so those voice recording we don't store it. So the way that works is we will have those al variables capturing transcribing all the conversations, and we do it in chance so we can rarely support the doctor's royal workflow because ied departments doctors have to work around work around means sometimes your any remote will disconnect with the mobile or disconnected with the backstop. But we don't want that disconnection disrupts the transcribing.

So HAIDI remote has dedicatingly designed for that scenario. You can go on connected or you can go on disconnected. That's all fine, All these conversations will be transcribed on device on HIGHI remote and next time when you finished the consults, you can connect your Heidi remote to your mobile or to your desktop and the all the loots will be there will be synced up to your application right away. Has that been received by doctors?

I know, you know they've had pages for a long time and you know mobile phones and nets and tablets in many cases in terms of actually moving into the wearable space space where they have something clipped on maybe when they're even doing surgery. How has that been received by them? Can we have more than thousands, like thousands of doctors are currently using we We do have. Do you see a lot of the minds in the markets and we are working super super hard to make sure we have efficient stocks for everyone.

And at the moment we've seen very large cohorts using in New Zealand, the Canada, in Australia, US UK and a lot of doctors they can't do their consults without it. Now. In the past they will bring their phones, They have to bring their phones if they want to use Hidi. Now they can just bring this highly remote to carry with them and just do the consults for food day, especially for those home care each care doctors who have to go to different places to do the consults.

And I just thought to a founder from Fly to Hells like they want to use remote for rural remote areas where the doctors have literally have to take their aircrafts to fly to rural areas and then do the consults there. Can you imagine like the doctors if they have to bring a laptop with them to do the notes, it would be very very troublesome for them to do so, whereas now they can just clip on the highly remote, take their aircraft, see see the patient for whole day, and come back to the hotel and then they can have all the notes synced up and they have all the notes ready for them to be That's incredible.

You know, you talked about the accuracy of the large language models isn't sufficient for your requirements. And when you start getting into medicine, just the names of the drugs are so complicated. So that alone, so I guess you know, when you when one one or two percent error rate is the most you can tolerate, sort of what thresholds do you have to consider? You know that are acceptable for different use cases in a clinical setting, if you're in surgery or if you're having a discussion with the GP, and then the notes are going to outline what drugs you're going to be prescribed.

Are there different tiers of sort of accuracy and reliability that are needed, So. Generally across the border we want to offer the best. So especially for those cinical safety side of the sense, we just do our best. Like even zero one person is not at all in r OKS.

In reality, the everything has error rates right even like you are getting ont of a plan. So we do a lot of evaluations in hopes to make sure the stuff we present and give to the doctors are close to bullet proof. And for example, we have a team of doctors in the house. They are the first cohort of users to check and use review all the model capabilities whenever we have a newer version comes up.

And second is that we will have a lot of larger language models. As a reviewer, we use the models to review as well. And also when we really when we feel ready that the model is good enough, we want to push this new ver. Now, we don't just like roll it off everyone.

We gave it to small cohort of highly users that who are willing to help us to test and evaluate the effectivities of the improvements we give to them asking for feedback. We do like rounds of different feedbacks until we are finally comfortable to give you to more people. Yeah, it seems like part of the successive Heidi and the quick uptake around the world, including here in New Zealand, is it's a sort of a it's been a bottom up approach. You get clinicians using it, trialing it.

I know there were some very successful trials in the Hawk's Bay here in New Zealand. It showed that, you know, the number of minutes that a clinician was having to devote to writing up notes reduced dramatically. So, yeah, that's been an interesting approach. Maybe if you talk a little bit about that, how you manage to really get traction, for instance, in the New Zealand market so quickly.

We always believe we have to build the product that makes sense for the end the users, and we invested a lot on that. That's why we still have a lot of the direct lines with the users, with myself and the coup founders, like we want to hear first hand feedback from clinicians and make sure the product are at the level of quality that can do well. And second is we lead by we grow our customers by word moss or a lot and a lot of good results. We've seen with Hawk's Bay Hospital, ed shows that exactly.

For example, in the past, they have to spend seventeen minutes per patient on documentation. Now they spend like only like four minutes around four minutes. That's a huge time singing And because of that, they can see one extra patients per shift on average, and that means more patients get seen, more lives get saved, and they feel more like reliefs because they don't have to bear the burden that they have to memorize everything put it in a documents. After effects and.

Also like they don't have to do a lot of after our st petitions. So eighty one percent reduction in documentations for after hours and last year is more like how we bring these kind of results to another, to more and more organizations and make sure more organizations can impact these into their workflows and for different hospitals, we uually customize different templates depending on what they want in their templates in their communic documentations, we can customize it for them.

So the customization is really really important for continuitions. And you're dealing with probably around the world health systems that are under stress financially, there's more demand on their resources. So anything that frees up doctors, it's very We have a real crisis in our health system here, particularly in the regions getting doctors to be gps and small communities, So anything we can do to free up times great. Have you also mentioned the other side, uh, the human side?

Is there is there any evidence to show that the human side of consultations is really improving as a result of using Heidi. I think that's a good and useful. Positive impact for a product means that you don't like you don't just have really nice usage data, but you are really changing how the users, the end users are operating in day to day. And now we've seen a lot of the patients higher engagement scales from patients.

Do you feel more much happier when their doctors are really listening to them? And those stuff will ship the long term patients Klinian relationship. We believe that's that's definitely one thing we can uh, one thing we can help more in the normal for the entire healthcare system. I mean you talk about Heidi as like an AI care partner.

It's not just describe. You talked about Heidi evidence there, so you know, I've seen my GP. When I'm talking to her, she'll often go on Google and just start googling things, and she'll go to good sources, you know, official source of information. But it sort of amuses me that she's doing doctor Google in the doctor's office.

But what sort of scribe adjacent workflows do you see beyond evidence and transcription and that sort of thing. Are we talking about things like, you know, follow up, patient communications, decision support, those sorts of functions. Anything that is non technic along clinical and we want to automotive for them. For example, like after the consults, we want to send a referral letter to specialists, and in the past they have to open their email box, copy pass the under they finish the referral letter and then put it uploaded into the email box and send you this a specialists, whereas now they can just say it out loud to HEII what they want to do, and Heidi, we'll send.

All the emails. Well, do all the actions, real actions for them, and our idea is just that doctors focus on care and we take care of the rest. But we don't really want. To do a lot of the effect how doctors are making judgments, Like we want to leave the human part and the clinical decision part two doctors and we believe they are the best one to do it.

Yeah, do you see a time in future and as on on your radar sort of heily having patient facing tools where patients can bring their own AI records or summaries into a consultation. We've got wearables, We've got all sorts of stuff, rich data coming to us in the home environment. And you know, the health sector has been telling us for a long time we need to have interventions earlier in the process so people, you know, aren't just turning up once every three years to a GP and and finding out, oh, you've got high blood pressure or you've got diabetes.

Is the scope you think. For AI to play a role much earlier on with patients having their own tools. A lot of patients like they are already doing it. They're already doing a lot of say the moment cold or open AI.

But we, like I said earlier, we believe the health start, I should never leave the country. So we are very well positioned to help lift thought on a lot of experience on this part and Bernie make sure the UH, the patient data, especially on a population level won't leave the country. And that's for one. For two is right, there's a lot of things we can we can help, especially for the patients.

To some of the patients. There are some cases the patients they don't speak English as their first language. We've seen some cases like that, so recently we just partniar with the with the Bona University to build these translation apps for them so they can UH. The patients can speak a different language, but they can they can get it translated so they can still communicate it super well with the Kinians.

And yeah, there are some other. Like a patient app to help the patients to manage their health records and to for Kininians to easily share stuff for information with the patients over those over those platforms will be something some areas we were looking to. One thing I think you've done really well is a lot of partnering with health accelerators in New Zealand, the likes of Streamliners, Hendrix's Health and others. So there's a local ecosystem in terms of you know how Heidi and interacts with our health tech platforms here.

You know, the perception of New Zealand health system is it's very fragmented, there's a lot of legacy tech. There. Are we any different really to the UK or Australia or the US And in terms of how robust systems are. By understanding like New Zealand is a country runal value the data, sovereignty and the security and Ronnie want to give credits to the government and the whole society for that because in my point of view, like health starter is the most important how you can have as a human being, as a society, It's something we need to be a big good use of and also we need to put a lot of efforts to protect the data.

That's why, Heidi, we have done a lot of efforts to bring the models, to bring the compute even closer to where the data is. That's why we have these a lot of optionalities for enterprises to bring the compute on site and that's what. Katie and a lot of departments are working on. At the well.

You've had a really successful capital raising journey. You did a really healthy. Series A about sixteen million, Australian Series B sixty five million. I think you've raised over one hundred million US dollars in capital so far.

Some great names near Blackbird which is a big investor in New Zealand, possible ventures, Local Globe, some good, really good names there. What's that been like. As a as a founder going on that journey, the venture capital journey where it can be exhilarating, But boy, they have really strong targets that they. Want you to reach.

They want you to grow and exit really quickly. Yeah. Of course, like every venture capital back, the companies have their urg and they are to go big, more revenue, more revenue, and bigger, bigger. But for us, it's more like revenue is just like a slightly effect of the impacts and the and the outcomes we bring to the world.

In the end, we want to do something good to to to the entire world, to the patients, to the to the comdition, to the health organizations. That's our intention, and revenue is just a slight effect. So no matter how importantly it is to hit our milestones, we have to do it in the right way, in a campin, in a way that we should respect us to the local people. We are optimizing for a lot of the girls, but at at the same time we're understanding whole critical EPA is to build our foundation running running, well, that's something candal tern.

Sustain our success. Yeah, it's really important, particularly at the moment as we talk. You know, we're hours away from the SpaceX I PO. You know that essentially now is an AI company with a rocket launch division bolted onto it.

We've got Anthropic has filed to IPO documents. Open AI is going to do the same. So we are entering the era of AI like never before, trillions of dollars potentially in market value here. I guess from your perspective about the future trajectory of Heidi, what does that mean for you.

It's obviously it's obviously good that the AI is mainstream here, but I guess there is also the that Claude or chet GPT starts to do more of these sorts of functions themselves. Yes, yes, yes, Like I said, a lot of people are using Claude for their patient data management, which is great. But what highly uniqueness is we are deep in the health care systems. Our goal is to be the default ai AI layer for the entire health systems.

Like when you think about ALI, any health organizations believing I can make their productivity much higher, they go to Heidi, that's something Claude and something Claude and the opening I or space X, I say, I like they never what you do is they will go down these medical device paths and they optimize their models for only health care. And but that's something we are investing in a lot of time and energy. Like as a. General models, like they can do a lot of the great stuff, but when it comes to clinical tasks, they just don't have a lot of signals like we do to make the models as good as it should be.

And also the product of the abuty is more for the general purposes, but for us than we are keeping these systems. We embed ourselves into the workforce to become the default ear layer for the entire industry. And just finally, you I think you know your company mission. You framed it as sort of doubling the world's health care capacity, which would save lives and save money as well, because you know, healthcare is such a hungry part of the government's budget.

It's just like a bottomless pit. We have an aging population, we have more need than ever before. What's it going to take to get to that point where we literally when by any metric, we have doubled the world's capacity and healthcare. Currently we are saving average problem one ority two hours every day for clinicians, but there are more room we can do.

I think doubling the capacity generally just means we in the world in different In a lot of regions, especially in rural areas, we still have a lot of the demands of patients that who can't really see a doctor in time, and how we can solve those problems from population level is a big topic for us and a big task for us. We started from documentation because we see that as the biggest gaps, but there are a lot of other gaps in the in the society or in these systems that we are working on right now as we speak to get those clues.

That's why we brought We have brought the wearable into reality and because we see we saw a lot of doctors the only use highly when they sit on their desk with a conference's mic, with a stable wildfile. That's totally fine and that works super super well. But when they go to ed department when they go to World rounds, when they go to rural area, whether the internet has been a problem or they can't running like take their phones, they won't have good mics that they can't utilize highly enough.

That's where remote comes to play and comes to effect that we cover more scenarios, cover more situations. The doctors do the concerts, and that's how we think, like, what we can do more for kinicians? What do we can do more for health organizations to make this more efficient, to run it, double the capacity. Well, it's a great mission you and you've made so much progress in a relatively short period of time.

Millions of sort of notes been taken around the world all the time, thousands of clinicians using so congratulations on your success and thanks so much for telling your story on the business of tech. That's it for this episode of the Business of Tech. My thanks to you Liu from Heidi for lifting the lid on how ambient AI is reshaping frontline medicine from Hawk's Bay to hospitals around the world. Do a lot of interviews on the Business of Tech and in Business Desk about the efficiency gains from AI, but really when it boils down to it.

The most dramatic efficiencies and the ones that actually impact people the most, are happening in healthcare. GPS clinicians using tools to free up time so they can see more patients, which is exactly what we need in a country where our health sector is under real pressure. So that's great to see, and if they can extend that, through Heidi and others to other parts of the health sector, even better. If you enjoyed this conversation, follow or subscribe wherever you get your podcasts and share the episode.

But someone who's curious or maybe even. A bit skeptical about AI and healthcare, I'm Peter Griffin. Thanks for listening, and I'll be back next week with more stories from the people building the future of tech in New Zealand and beyond.