638 - Beyond the Demo: How Facere AI Is Building AI Receptionists That Actually Work in Clinics

0:00Peter Birch: Have you ever heard a term in digital health and just quietly had to Google it later? The Talking Health Tech Glossary has got you covered. We've got hundreds of terms in plain English explanations, written for our industry, and all free to use. You don't even have to sign up for anything. Just go to talkinghealthtech.com slash glossary. Welcome to the Talking Health Tech podcast. My name is Peter Birch. Today on the show, I'm joined by Bin Rong, CEO and Bowen Zheng, Head of Engineering at Fuchero. AI. G'day gents, how are we going? Good, good. How's your day? Really good. Better now I'm talking to you.

0:36Peter Birch: You know the drill. It's great to have you on the podcast. You're really keen to dive into this a little bit more as I've learned more about the company and the work you're doing and this kind of how far things have progressed and really the stage you're at now. But let's unpack it a little bit more because I think a lot of people are going to get some new information, kind of confirm some understanding and kind of even potentially open up some eyes about the role that AI can play and automation in all parts of healthcare. But firstly, maybe if I go to you guys to introduce yourself, tell us a bit more about you and what you do.

1:15Peter Birch: And also tell me if I got the name right, Fichara AI? Fichara. It's a unique name, Ben. I'll start with you. Tell us a bit more. Yeah, thanks for having us here. And yes, a lot of customers, their first question is how to pronounce our company name. Actually, when we're trying to get the company to digest, we find all the good name has been used. But luckily, we found this Latin word, supposed to be good meaning, which created something out of nothing. But the pronunciation is a little bit strange, called facara. So we had a customer, she has a bit of italian heritage and she suggests we have closing what kind of faceri uh which sounds very good so we pronounce us from there got it yeah no yeah i wasn't too far off no but a bit more about um uh actually bowen as well firstly like you and your background and the work you do at uh faceri yeah so i'm the bowen i'm the head of engineering here at faceri um so been working with ben since actually we met from before future um being in tech our whole lives um really saw this uh basically since you know three just over three years ago now when we first saw a gp 3.5 really hit the world and everyone everyone saw what the current you know generation of ai could really do we saw a lot of the opportunity um in this and basically we would provide you know automation across you know the world but we really saw that you know healthcare is such a you know underserved industry and we felt we could definitely do a lot of good in here and really help with healthcare both in Australia and also across the broader world as well.

2:53Yeah, 100%.

2:54Peter Birch: What about you, Ben? Yeah, and I did my PhD in computer science in Australia and cut my teeth in IBM, TJ Wilson, New York Centre, research on the first generation of big data. Then I went back to Australia to build Australia's first mobile taxi hailing product for Silver and mobile payment as well. So I built a lot of working AI and computer science for about 20 years. So before we come here, I worked closely with Bowen, our previous company. Yeah, got it. And then what brought you to then, you know, the opportunities that present in healthcare? Tell me a little bit more about that why and why you're on this journey.

3:38Yeah, that's actually we first started this company because we have some tragedy story. And in my previous company, I always plays tennis with one of the former colleagues who very unfortunately lost both his mom and son during COVID for suicide mental health. And we've been talking about in such a wealthy country with all the fundamental social welfare, why this still could happen. and we realized is that, you know, the government has spent a lot of money on these NGOs and to help people when they need it. However, when people are trying to make the call for help and the cost success rate, cost through rate is surprisingly low.

4:28Could be something very, very low, extremely low. I cannot share a number, but it's surprisingly low. And that's why. But the problem is we find out because this social worker, they probably spent 10, 15 minutes talking to the people, but they still spent maybe 45 minutes to fulfill the paperwork because that's the real reason they cannot help as many people as they want. And that was late 2022, just when ChatGPT came out. We thought we could do something to relieve the admin burden from these people. That's why we, after building preparation and building prototypes, we finally launched this company in August 2023.

5:12And at that time, there's no such product on the market. So we've been talking to a lot of specialists, the GP, even hospital mental health. And we realized, okay, because these kind of people are spending more time on the main task manually executing everything. So we say we build the current, our product offering like a document routing, document allocation. So people, rather than doing manual work, our AI does everything automatically end to end. And also, we all had the experience when we go into any clinics, half the staff are on the phone, the other half are on the computers. As you can see, a lot of clinics are bombarded with 200, 300, even 1000 phone calls every day.

5:58That's why I offer AI receptionists to really book the patient in, check appointments, cancel everything. So we built this company with the dream to help people, to make people's life much better. So the people who needed a healthcare service can make it more accessible.

6:16Peter Birch: That's why we're here, right? And that problem that you outlined and then, you know, what you've built to address it being, from what I heard, you know, document routing, automation, general admin, and then, you know, this AI reception piece as well too. So, you know, a few moving parts and, you know, from, I think you said, you know, 2023 up until now, there's been, you know, varying interests in primary care and other parts of healthcare too in terms of some of these tools. What stage would you describe which you guys are at at the moment? Because I guess from 2023 until now, being like, you know, mid to the second part of 2026, what's been keeping you busy to now and like where you are?

7:05So I give a high level overview and Bowen can jump into more details. So basically when we started and we're building two products, one is document routing, document allocation. And I would say that's about 5 % to 10 % into the product building. We have product in a pipeline will come out in a month or two, will be sometimes better and can automate almost 95 % of work. And the other piece I want to talk about is AI phone, AI receptionist. And even including ourselves, we greatly underestimated the difficulty of building it. Because AI phone is such a piece of technology, it's so trivial to build a demo, a shining demo.

7:50However, extremely, extremely difficult when working in production. And from what we can see is without fundamental technology breakthrough and all the AI receptionists can probably only complete 20 % of the customer request at the most. However, we've been, in the past three years, we invest heavily in IND now we can complete roughly 70, 75, pushing 80%. Because I can tell you why. Maybe you think it's easy, but I can tell you why. Because the AI, for half of a time, barely understand what you're talking about. That's for one conversation turn. Let's say you have 50 % of understand one turn, five turns later, you probably AI understand nothing.

8:36The other thing is latency, because the current technology goes through telephoning, everything. and when the latency go above 300 three seconds most of the customer hung up yeah but we already resolved this so probably bowen can give you even more details yeah i mean it's an interesting

8:54Peter Birch: like if we're talking about ai reception specifically the it feels like the ultimate thing to one of the big things to crack because you've got that engagement the difference between like scribe or automation like any of the ai automation tools it impacts the healthcare worker um be it the clinician or the admin and look if there's a delay or whatever like it's all part of the workflow but if you've got like real-time engagement with someone who they're not they have no sop on what they're going to say in a phone call and you know like they've got any kind of requests under the sun that's um whilst it does and that's why it does take a lot of time for someone to try and work out.

9:33Peter Birch: So it seems like the right kind of issue for technology to address. But I guess, yeah, a lot of challenges. Love to hear a little bit more, Bowen, about what's being built. Yeah, definitely. And I guess when it comes to, especially the AI receptionist, it's always a balance between basically, you want an AI that is quick to respond to someone. For example, us talking over the line right now, I probably respond back to you, within a couple hundred milliseconds. And if you feel that I'm responding much slower than that, you feel like something's gone wrong. Is it lagging? Is it Wi-Fi bad? I'm not sure if the other party is there.

10:12The other challenge to that is we want to be able to be intelligent. So I can respond to you, but I should be able to respond to you currently. I shouldn't start driving the conversation down the wrong path. If you're asking a request for me, I need to go give you back. Something coherent is an answer. Honestly, the final thing is the cost. These models themselves are big. They cost a lot of energy to run. So we're basically balancing what we call here, at least at Futura, the impossible triangle between cost, latency, and intelligence. So we've been working tremendously hard over this problem for the last two years now.

10:52A lot of different technology breakthroughs in that process, basically pushing the state of art of not even just in Australia, but even compared to Silicon value as well. So in terms of our agents, we're basically able to perform at very high competence intelligence. So, you know, closely ranking, basically passing our internal metrics at, you know, 98%, be able to do core completion. That's just the intelligence piece. That's just, you know, if you say something, I say something back to correctly. The next piece of that was latency. You know, you're going to say the same thing back, but within a short amount of time.

11:24You know, this is still something that's being pushed right now in the state of technology. most of the top Silicon Valley companies really pushing around two and a half seconds of latency. That's probably like the best you see on the market right now. Really happy to announce that over here at Futura we've been able to push below two seconds. So we're basically bringing you the most intelligent voice agent and the fastest that the world has ever seen.

11:50Peter Birch: So yeah, digital health has got a pretty serious acronym problem and we've all got our favorites, but it's a lot, right? And you don't really want to be the person in a meeting that has to ask what something stands for after you've been talking about it for the past couple of weeks. That's why we've built the THT Glossary. It's a free, ever-growing library of digital health terms that we explain in simple language written for people who actually work in the industry. So that's useful if you're brand new in the space or you're one of the veterans that just want a refresher before a meeting. Or maybe you want to check to make sure we've got everything in there.

12:26Peter Birch: It's all there for you. Have a look at talkinghealthtech.com slash glossary. Bookmark it. Send it to your team. It's the kind of thing that should exist. So we created it. Hope you enjoy. I'd love to understand what that interaction looks like a little bit more. And maybe like from the, you know, we've got the ability here on the video podcast episode to share some examples with people too. I guess a testament to your technology as well. I guess if I was to, um, reveal to the audience, this in fact, actually isn't even Bowen the human. This is his AI avatar. No, I'm joking. But, uh, the, I guess that's the stage that we want.

13:04Peter Birch: I thought that was hilarious. But, um, the, uh, um, the stage we want to get to, oh, I don't, I'm interested to maybe later in the conversation. The, um, is the stage that we want to get to a point where it's indistinguishable eventually on engaging with a human versus, um, AI from a patient's point of view? Or do we want to get to a stage where it's, you know, known to a patient, it is AI, but it's just as helpful, if not more helpful. And there's not that element of that feeling of deception. Sometimes clinicians might be like, I don't want to deceive my patients with, you know, engaging with an AI receptionist.

13:38Peter Birch: So a lot will, you know, notify up front and all of that. But look, there's so many questions and things that I imagine clinic owners and stuff would think about. But before we dive into that, maybe there's some things you can show us, Bob. Yeah, absolutely. I guess with maybe he's prefaced a little bit more we've basically been working closely with a lot of different specialists and hospitals so kind of starting from that kind of document routing allocation piece especially across more complex workflows this is kind of like basically what we built in across like hospital and specialists across the kind of workflows that we've then kind of trickled down into the primary care this will kind of eventually tie into that kind of voice agent.

14:21But maybe I'll kind of start from the document routing and then show you how it ties into how the voice agent works. So you can imagine if you're a specialist, let's take the example of mental health since that's kind of where we kind of started. So as a mental health kind of practice, I'm sure those out there that are listening, they're absolutely flooded with referrals. So you can probably get 100 of these per day. Your team is manually going through your emails checking each email that comes in. Is this a referral? Is this a new referral? Is this an existing patient? Then saying after the doctor, hey, when do you want to see this patient?

14:58The doctor comes back maybe, you know, after a couple of days, we have to helm them and chase them up. Eventually come back to you and say, yeah, let's book this in for this timeframe. And then you go and call. So what we have is basically AI that automates this whole process end to end. Now, from here, what I've got here is I've basically, this is our little kind of demo environment here. what I've done is basically I've sent myself an email. So this contains a little referral inside it and our AI is reading the inbox and automatically picks it up. So no longer needing to manually read your inbox, our AI kind of picks it up, can tell there's a referral, can tell what needs to happen with it.

15:34So I've got that here. So here we've basically got that referral that's come through in the email. Basically on the right-hand side is where the actions are being taken. So throughout the process, when we know healthcare is a very environment where getting things done correctly is very important. We basically built it so that the AI is always prompting the end, basically admin staff, and they're in the driving seat to get things done. So the very first step here is matching that with the patient. So we've got this rule come in for Ava. We've actually looked inside the PMS and found that, oh, there isn't actually an exactly matching patient record for Ava.

16:13So So let me go locate that patient. In this case, we've actually got a few different Ava's here. So I'm just going to pretend this is a brand new patient. So let's just say there's a brand new referral. I'll go create a brand new patient. So our AI has automatically detected all the different demographic details from the referral. Me as admin staff, I just have to confirm that I believe that's all correct. I can pop that in. And that also creates a patient inside the PMS as well. Now, the next step here is basically letting me to decide if I need to triage this. So I see that Ava doesn't have any upcoming appointments and I know that she's a new patient.

16:45So let me send it off to the doctor triage. Send it off. I'm just going to send it to myself just so we can show you what that all looks like. But I can send it off to the doctor from here. Now, the doctor gets this. They get all sorts of notifications in various ways. One of them might be over email. The doctors can immediately just jump straight into the triage. So from there, as a doctor, I can see the referral. I can see the patient is for. I can see any of those previous or upcoming appointments. Then let's just say, I'm happy to see this patient and it's a routine appointment that I need for them.

17:18I can submit that. And that just goes back to my admin team. Traditionally, this is a process of printing off referral or forwarding over to the doctor, maybe walking over to the doctor's rooms and saying, hey, when do you want to see this patient? And you can't always get a hold of the doctor. They're in consults most of the time. And so that becomes a three-day job just to hear back from the doctor. Now, with our workflow, we've automated that process. the doctor gets that, they can action that, and comes back to the admin staff without needing to go through this manual, kind of remembering where everything is and chasing out people.

17:49So when it's back to the admin staff from here, I can see that the doctor has basically requested for an appointment to be arranged. I go ahead and arrange the appointment. I can go ahead and first kind of just file away the referral. So let's quickly just set doctor, include that for filing, and the next step here is to book that appointment in. So at this point in time, that referral is filed against the PMS. So all those kind of billing records, any claim records to be created, those are all set up. So it takes away all the kind of manual effort of popping those into the PMS as well. Now, leading to the next part, so let's go and call the patient.

18:27Of course, we won't have the human call the patient. We'll have the AI call the patient. So as a staff member, I can trigger that off, call the patient with AI, and from there what you can imagine is I'm Ava here just you know going about my day eventually I get a little phone call so I get a call on my phone um and hi Ava I'm Sarah I'm autonomous agent calling on behalf of the rooms of Dr Johnson we've received your referral and I'm calling to book your appointment does next Wednesday at 2 p.m work for you oh that's not so great for me. But is next Thursday at 10am available? Let me check the schedule for you.

19:08Yes, Dr. Michael Johnson is available next Thursday, September 24th at 10am. Would you like to book that in as a standard appointment or a telehealth appointment? Just standard is fine. Thank you. That has been booked in for you. Your standard appointment with Dr. Michael Johnson is confirmed for Thursday, September 24th at 10am. We look forward to seeing you then. And that's one of my appointment booked. And you can hear from that demo that's very responsive, intelligent, understands what I'm asking for. It's not just, does this time work for you? No, okay, I'll hand you to a human. Can I actually work with you, get the right time, and I'll respond quickly as well.

19:47So it's taken us tremendous effort to get here, but this is the kind of level of automation that we can provide to our practices and really help with the load and the admin side.

19:58Peter Birch: So, of course, great to see that from start to finish. you know we should be looking at workflow more on the show because it's great to visualize that and um people if they're listening out and about on things on audio encourage you to jump in and see some of those screens on the video feed of this episode which you can get in the show notes um and i'm thinking as you're going through this process it's you know that particular workflow from a referral in and triage and doctor look and then out to an appointment a great use case of that but i could also think of a number of different examples in healthcare too where there's just a lot of stuff that needs to happen that potentially AI can assist with and automate as well.

20:36Peter Birch: Directionally, what's the vision? Where does this go? What are you hoping to achieve over a longer period of time with this kind of capability perhaps? I think as we've spoken before, what do we have shown and what do we have shipped to the customer, not even 10 % done. So we have a new fully automated workflow coming in, which is 10 times more powerful than this. As you mentioned, this covers a particular workflow. And the new product is going to cover majority of the workflow, let's say 90, 95 %-ish, your daily work. But rest assured, we got this question asked a lot by some of the customer.

21:14Are they coming to take our jobs? Now, far from that, we make you far more efficient, make less errors and you're still in the driving seat and you you decide when what's the next action to take ai just help you automate the execution yeah well that's the other piece too

21:33Peter Birch: like the um you know automate the execution i'm interested in both your takes perhaps you know as we we round out the conversation when you think about primary care and the the specialist world and even the hospital landscape, it's too, like what's ripe for automation and what is just like we're not touching, like that's not our role in this kind of journey of streamlining things. Maybe we'll start with Barlan if there's anything else you want to add. Yeah, definitely. I think with the automation, what we basically see, there's basically two parts of it we're looking to help practices with. One is your orchestration.

22:10So right now I'm sure a lot of practices, they have a manual spreadsheet that they update day to day on who's doing what. And having a practice manager as the chase up, hey, Sarah, can you make sure that this is done? Or is this done? Where is this at? So that's where AI is best to basically know where everything is happening inside your practice and automatically make sure that things are getting done in the practice. So that's one part of it. So that's kind of like what you see, the referral coming in, getting the triage done, getting it put into the PMS, and then obviously calling the patient.

22:41That's the kind of like the general workflow that you saw automated. So tackling that side, but also tackling those kind of individual tasks. So the actual filing of that document, the AI gets all the manual data entry ready. The human's just in that driving seat, having a final check, happy with everything that's been done with the AI suggestions and authorizing that action. What about you, Ben? I think on top of both of them, I think first is from the first day of founding of this company, we are very clear. We stay away from clinical side. anything involved doctors clinical judgment and we don't get involved at all and that's one part the other part is anything requires high level a professional human judgment like a nurse after years of training we know this case is urgent or not we're not getting involved so we're only getting involved non-clinic manual at the main part that's clean yeah no look i think um you know

23:40Peter Birch: through the demonstration that you've provided and just, you know, hearing about the journey and the work that's been done. I'm really looking forward to seeing how, you know, putting this kind of capability in the hands of more clinics and the healthcare system, it's going to continue to feed on itself in terms of hopefully more efficiency, but also just, you know, the, the, as soon as you start to put these tools into people's hands, you then opens up even more opportunities and it becomes, you know, the roadmap becomes remarkably long and that's a good thing. I think there's so much, as you say, being an, and as Bowen has also kind of highlighted that, uh, if there's more of that kind of busy work that can be not even busy work, it's like the important kind of administration work that needs to be done to then, um, allow those humans to focus on the clinical, then, then that is wonderful.

24:24Peter Birch: And that's the healthcare system we need. So, uh, Ben Bowen, I appreciate you making the time for this conversation. All the details for Fuchero will be in the show notes of this episode on our website and you're in the community too. So I really appreciate you making the time for us today. Thank you so much. Thank you. Thanks, Pete. We hope you enjoyed this episode of Talking Health Tech. Make sure you like and subscribe and share this episode with someone who might find it valuable. For more information and resources about healthcare innovation, visit TalkingHealthTech.com.

24:59Peter Birch: That's it for this episode. I hope you enjoyed it. Make sure you subscribe to our newsletter on TalkingHealthTech.com slash newsletter so you can be across all the things we do at Talking Health Tech above and beyond this humble little podcast. Thanks so much.