From Intake to Outcome with Agentic AI - Insight Health

欢迎回到《医疗创新播客》。

Welcome back to the Medical Innovation Podcast.

每年，美国的医生大约有三分之一的工作时间都花在了填表 paperwork 上，而不是诊断病人或治疗病人，而是处理事先授权、转诊传真、入院表格等等。

Every year, physicians in The US spend an estimated one third of their working hours on paperwork, not diagnosing patients, not treating patients, but on prior authorizations, referral faxes, intake forms, etcetera.

今天我们请来的嘉宾，就是要解决这个问题。

Our guest today set out to fix that.

他是 Insight Health 的首席执行官兼联合创始人，这家公司打造了一个智能平台，能为独立诊所自动化整个临床运营流程。

The CEO and cofounder of Insight Health, an agentic platform that automates a full clinical operations stack for independent medical practices.

他最初是加拿大一名电气工程师，参与发明了 OLED 显示屏，后来参加了 Y Combinator，最终在 2023 年创办了 Insight Health。

He actually started off as an electrical engineer in Canada, co invented the OLED display, went through Y Combinator, and eventually started Insight Health in 2023.

今天我们非常高兴能与杰米尔·索尼对话。

We're excited to be speaking with Jaimal Soni today.

杰米尔，欢迎再次来到我们的播客。

Jaimal, welcome back welcome to the podcast.

谢谢你的邀请。

I appreciate it.

谢谢邀请我来。

Thanks for having me.

是的。

Yeah.

我们非常高兴你能来。

We're really excited to have you.

首先，我们很想听听你的故事，你是谁，是怎么走到今天的。

To start off, we'd love to just hear a little bit about you, who you are, how you got to where you are today.

对。

Yeah.

我叫杰马尔，是Insight Health的联合创始人兼首席执行官。

So my name is Jaimal, one of the cofounders and CEO of Insight Health.

大约三年前创办Insight之前，我曾负责Segment Twilio公司的医疗团队，后来还领导了Twilio的医疗部门和客户AI业务。

Prior to starting Insight about three years ago, I had the opportunity to start and lead the health care team at a company called Segment Twilio, and went on to lead parts of the health care team at Twilio and also customer AI.

就是在那儿，我认识了我的联合创始人索伦·西瓦。

That's where I met Sauron Siva, who's my cofounder.

在那之前，索隆曾带领过大小不一的工程团队。

So Sauron has, previous to that, led, small and large engineering teams.

尤其值得一提的是，他曾在维萨公司领导支付处理团队，主要负责安全和基础设施方面的工作。

Most notably, he led the, payment processing team at Visa more on the security and infrastructure side.

之后，他又在Twilio负责Infrared Security团队。

And then after that, led in Infrared Security at Twilio.

我们的创始团队里还有两位成员，一位是神经外科医生，另一位是心脏病专家。

We also have two positions on our founding team, a neurosurgeon and a cardiologist.

我和这两位已经认识好几年了。

So I've known both of them for quite a few years.

再跟你们说说我们的背景故事。

And a little bit about the backstory.

我们刚开始做Insight的时候，就仔细研究了市场。

When we first started Insight, we we looked at the market.

那时候，AI语音助手就已经开始兴起。

And even back then, AI scribes were starting to take off.

那时候，医生们在病历记录上面临一个巨大的痛点，现在很多人依然如此。

There was a huge pain point that physicians had at that time and often still do have around documentation.

AI文书助手在缓解这个问题上确实取得了显著成效。

And AI scribes were hugely successful in in helping bridge that gap.

但当我们深入观察这个领域，尤其是在专科医疗中，真正的问题往往发生在就诊前后。

The reality is, though, that when we looked at the space, especially in the specialty world, a lot of the problem areas that we we saw were actually before and after the visit.

我们并不是忽视AI文书的作用，而是更关注就诊前的各个环节，比如患者信息采集。

So not discounting the work on the AI Scribe, but really everything that leads up to that visit, things like patient intake.

所以我们第一个产品其实是患者信息采集系统，我们为西海岸一家肿瘤诊所打造了全球首个完全由AI驱动的就诊前自主交互系统。

So our first product was actually a patient intake product, and we did the first ever fully autonomous pre visit encounter powered entirely by an AI model that was for an oncology clinic on the West Coast.

从那以后，我们开始深入思考，甚至被整个医疗流程所吸引。

And we went from that to really thinking about and and almost honestly getting pulled into that entire workflow.

刚才你提到，医生们被大量文书工作压得喘不过气，花在行政事务上的时间占了三分之一。

So in the intro, Ranav, you mentioned, you know, all the paperwork and the drowning in that one third of the time spent on administrative work.

其实这些问题从最开始的转诊环节就出现了，一直延续到信息采集、就诊本身，以及后续的所有流程。

That really starts in the very first touch point, which might be the facts of referral that a clinic might see through to the intake, through the visit itself, and everything that happens after.

所以我们从患者登记入手，后来逐渐深入到了整个临床工作流程。

So we started in the intake space and then have since gotten pulled into that entire clinical workflow.

对。

Yeah.

患者登记这个环节其实特别有意思，因为一直以来，我们都在用一些系统，比如iPad之类的，搞那种结构化的登记表，但说实话，不管是让我自己填，还是让患者填，体验都不怎么样。

And that intake space is is really interesting because I think for a while now, you know, we've had, like, systems, you know, like iPads and things where it's kind of in the structured intake structure that, honestly, whenever I have to do or I have to make a patient do, it just isn't the most ideal situation.

结果就是大家只是机械地打勾，你根本没法确定这些信息到底准不准。

And you get these situations where people are just kinda checking boxes, you're really questioning maybe the accuracy of it.

我其实挺想听听，你们是怎么想到用AI来改进登记流程的？毕竟现有的方式，比如让非医疗人员来处理这些事，效果并不理想。

I guess I I did wanna hear a little bit more about that solution and how you felt like, you know, AI could handle it in a in a superior way to what existed already for intake, that might be done outside of just, you know, an actual, health care professional.

是的。

Yeah.

没错。

Yeah.

有意思的是，我们发现差不多一半的诊所还在用那种iPad登记系统。

It's interesting because we told probably one in two clinics, people do that iPad.

其实，诊所用iPad已经算是挺先进的了。

And iPad, I think, is even, like, pretty forward thinking by clinics.

但一半的诊所连这个都没接入电子病历系统。

But one in one in two clinics, that's not even integrated directly into the chart.

病人填完表，这些信息可能根本没传到医生手里，就直接来就诊了。

So somebody might fill that out, and that may not even get to the physician prior to your visit.

更糟的是——我自己就经历过这种情况。

Or even worse, they're and I've I've done this myself.

我提前十分钟到，手忙脚乱地填一堆纸质表格。

I get there maybe ten minutes early.

填完还得有人再手动打一遍进去。

I'm, like, trying to quickly fill in these these the stack of papers, and, of course, somebody has to type all that in.

所以说，我们想的是：怎么才能真正贴合患者的实际需求？

All that to be said, I think the approach that we took is how do we make it so that we can meet patients where they're at?

顺便提一句，我们最初主要做专科领域，早期业务里大部分都是肿瘤、神经外科和消化科。

So and one thing to caveat, we we initially started predominantly in the specialty world, and I would say the lion's share of our early business is oncology, neurosurgery, and GI.

以癌症治疗为例，医生需要收集非常详尽的病史。

So looking at oncology, for example, there is quite a detailed history that, a physician needs to capture.

而这些信息通常都是在就诊过程中现场收集的。

And often that's done in the visit itself.

所以第一次就诊的前一半时间，可能都花在了解病人的过往手术史、家族史和社会背景上。

So up to half of the first visit might just be history capture, any prior surgeries, family history, social history.

我之前提到的那次尝试，是和一位来自俄勒冈州波特兰的乳腺癌外科医生合作的。

So that first encounter that I mentioned we had done with that was with a a breast cancer surgeon in Portland, Oregon.

在那第一个小时的前二十到二十五分钟里，她主要会问一些像家族遗传、患者是如何走到这一步的这类问题。

And a lot of what she explores in the first, let's say, twenty, twenty five minutes of that hour first visit is things like Ancestry, like, everything that brought that patient to that point.

这些内容很难用单一维度的表格、iPad或者类似工具完整传达出来。

That is extremely hard to get across in a very one dimensional form or an iPad or anything like that.

这本质上是一场对话。

It is very much a conversation.

如果患者提到自己在服用某种药，比如降糖药，我们就得深入追问：为什么吃这个药？

So if a person person mentions they're on a drug and that happened to be a diabetes medication, digging into that and having the intelligence understand why are they taking that medication?

他们有糖尿病吗？

Do they have diabetes?

除了表面回答问题，你还应该问些什么？

Like, what beyond just answering the question at face value should you be asking?

要明白该问什么、什么时候问，真的需要专业的临床判断。

And it takes really a clinical understanding to know what to ask and when to ask that.

另外，语音AI的进步让我们能以自然对话的方式和患者交流。

The other piece to add is that the advancements in voice AI really allow us to to converse with a patient in a conversational way.

这带来了巨大的可能性——让患者直接开口讲述病史，远比填勾选框或打字能获取更多信息。

That unlocks a huge amount of opportunity because you can capture so much by just having a patient speak and share their history verbally versus just having to answer a like a checkbox or type something into a field.

我认为，这两点结合起来特别关键。

So the combination of those two things, I think, is really important.

我们的做法是把两者融合在一起，我们称之为‘视觉语音’。

Our approach to it is is a combination of both, we call it visual voice.

它包含语音交互的部分，但同时，在合适的时候，点击勾选框也是有必要的。

It's the voice aspect, but also where and when possible, it does make sense to just click a box.

如果你只是简单地回答是或不是，那确实能快速方便地获取这些信息。

If you just, you know, saying yes or no, like, that's a very quick and easy way to capture that information.

所以我们早期开发的很多技术，都是把这两种方式结合起来——既能通过对话收集信息，又能穿插一些视觉选项来加快流程。

So really a lot of the early tech that we built is that hybrid of the two, being able to capture it verbally but also interleave these visual components to help speed up the process.

对。

Right.

我觉得有些问题确实需要让患者多说一点，不能只满足于是或否、打个勾就完事，但有些情况，比如家族病史之类的，就不一样。

I think there's definitely certain questions where you want the patient to open up more, be able to really you don't want just a yes or no or checkbook answer, but there's certain things, like maybe family history or something along those lines.

你只想知道，这位患者有没有心血管疾病的家族史？

You just wanna see, does this patient have a family history of cardiovascular disease?

是或不是。

Yes or no.

有些风险因素、某些疾病情况，都是这样。

There's certain risk factors, certain conditions, etcetera.

这样安排确实很合理。

So that makes a lot of sense.

这种灵活性真的让收集病史更高效，不用像你提到的那样，每次就诊开头花三十分钟，或者只填一张纸质表格。

Having that flexibility really makes it more useful to get patient history instead of spending thirty minutes at the beginning of a visit or just getting a paper intake form like you mentioned.

是的。

Yeah.

对。

Yeah.

另一个例子是消化科。

And another example is GI.

我们在消化科诊所做了很多工作，主要针对预防性筛查。

We do a lot of work with GI clinics for things like preventative screening.

比如结直肠癌筛查。

So take CRC screening.

他们已经把筛查年龄下调到45岁，我觉得下一步可能还会降到40岁，也就是说，每隔五到十年，大家就应该做一次常规结肠镜检查或类似筛查。

Or they just lower their age to 45, and I think they're probably gonna lower it again to 40, meaning that, you know, every five to ten years or so, you should get your routine colonoscopy screening or equivalent.

但目前这个流程对诊所和患者来说都挺繁琐的。

And the process to get that today is is somewhat cumbersome for for both the clinic and the patient.

首先得完成初步筛查，然后才能安排一次胃肠科医生的就诊，再预约结肠镜检查。

Have to, first of all, get that that initial screening done, then maybe you get a visit schedule with a GI doc, and then you get your colonoscopy scheduled.

但对于40到50岁之间的大多数人来说，风险其实很低。

Whereas for the majority of the people between, let's say, 40 and 50, they're relatively low risk.

所以这类低风险人群完全可以直接安排检查，不用走那么多流程。

So you should be able to just fly through that and where low risk, just get those patients directly scheduled.

这属于临床判断的问题。

That's a clinical conversation.

就像你提到的，有些情况确实需要问一些开放性问题，比如心血管风险之类的。

And in that case, to your point, like, there are some aspects you wanna ask a very open ended question, right, like cardiovascular risk, things like that.

但有些情况就很简单，比如你抽烟吗？

In other cases, it's, you know, do you smoke?

就是个非黑即白的问题。

Yes or no type of thing.

能同时处理这两种情况，非常重要。

And those are the the kinda having that ability to do both is very important.

完全同意。

100%.

我完全赞同。

Totally agree.

在深入聊产品之前，我想先退一步说说。

I guess before going deeper into the product, would love to just take a step back.

你提到过Insight Health的灵感来源，以及你是怎么发现这个缺口的。

You mentioned sort of the inspiration behind Insight Health, how you came up with the idea, the gap you saw.

但最开始的第一步是什么？

But what was the very first step?

当你刚决定要解决这个问题时，你和联合创始人一起，团队里还有两位医生。

When you first decide, I wanna tackle this problem, you're working with your cofounder, and and you have two doctors on a team.

那时候你们最先做了什么？

What were your first steps?

是先考虑该从哪个专科入手吗？

Was it exploring which specialty to pursue first?

是先去多找一些医生聊聊吗？

Was it talking to more doctors?

是先做个最小原型，看看大家的反馈？

Was it building an MVP and seeing what people think about it?

你们当时是怎么开始的？

How did you approach it?

对。

Yeah.

这是个好问题。

That's a great question.

其实，对这个问题的早期洞察，主要来自那两位医生。

So the early foresight into, really, this problem space actually came from the two doctors.

他们俩一起跑步的时候，很快就意识到，问诊和病史采集是他们最大的痛点之一。

And the two of them, like, shared that they were on a run together and, you know, very immediately realized that, like, intake and history capture is, like, one of the biggest pain points that they have.

你看看神经外科医生们。

Like, you look at the neurosurgeons.

潘贾医生，也就是潘贾·考尔，是位于俄勒冈州波特兰的一位神经外科医生。

So Pankaj doctor Pankaj Kaur is a neurosurgeon based in Portland, Oregon.

来看他的病人来自很广的区域。

And patients come to see him from quite a wide area.

我觉得全国可能只有大约25位医生能做他这种手术。

I think he's one of maybe 25 doctors in the country that, do his particular type of surgery.

他专攻脑肿瘤和颅底手术。

He's focused on brain tumors, skull base surgery specifically.

他每次和病人见面只有45分钟，希望能充分利用这段时间，让病人获得最大价值。

And that forty five minutes that he may have with a patient, he wants to make it as valuable to that patient as possible.

所以他想提前尽可能多地收集信息，这样病人来了之后，他只需要花几分钟确认一下，就能直接切入核心话题——制定治疗方案，帮助病人康复，以及解答病人所有的问题。

So how much can he actually capture ahead of time so that when the patient gets there, he can spend a few minutes just confirming and then jumping straight into the meat of the conversation, which is the care plan, how to get that patient better and everything and all the questions that the patient might have for for him.

这个最初的构想，其实来自潘贾和另一位医生的共同经验。

So the early idea really came from both Pankaj and Doctor.

埃里克·施泰克尔是另一位来自俄勒冈州波特兰的心脏病专家。

Eric Stecker is is an EP cardiologist also based in Portland, Oregon.

他们俩在交流一些想法时，突然意识到这里面的机会。

The two of them were were chatting through some ideas, and that's really where when they saw the opportunity.

从个人经历来说，苏兰和我之前都在 Twilio 带领他们的医疗健康团队。

On a personal level, so Suran and I came from Twilio where we led the health care team there.

我们在那里专注的都是些增长和营销方面的问题。

And all of these cases that we focused on there were really on the growth and marketing side.

我们跟很多大型消费品牌合作，那些品牌你肯定都听过，帮他们提升用户留存、患者留存和增长之类的。

We work with large consumer brands, for example, that we've probably all heard of to help them with things like, user retention or patient retention and growth and things like that.

但一直有一块领域让我们特别感兴趣，那就是更偏向临床的那一部分。

There was always this area that we were super interested in, and that's really on the the more clinical side.

进入这个高度监管的领域后，问题确实复杂得多，但这也是我们目前所处的环境。

It's certainly a harder problem to tackle when you get into, like, the super high highly regulated, arena that that we currently play in.

不过，这一直都是我们的兴趣所在。

But that's that's always been an interest of ours.

所以三年前我们重新联系上潘卡吉时，发现他们也在思考 AI 如何为临床应用带来突破，那一刻，一切就都对上了。

So when when we reconnected with Buckage about three years ago and, you know, they were also thinking on the same lines as how AI is really a huge unlock for the clinical application, that was that was really where it all clicked.

是的

Yeah.

很有道理。

Makes a lot of sense.

而且你们自己和医生一起担任联合创始人，这一点特别有意思。

And it's it's interesting to you having, you know, yourself and, also having the physicians on as, like, cofounders as well.

我们最近请了Curology的创始人大卫·洛舍尔医生做客播客，他写了一本书叫《为什么医生能赢》。

We we recently had, doctor David Lorscher on the podcast from, founder of Curology, and he wrote this wrote this book, why doctors win.

书里主要讲，医生创办初创公司或加入企业时，往往能带来巨大价值，原因有很多。

And, basically, you know, over the course of this book argues why physicians building in in in startups and companies often contribute significantly and, really contribute to the success of a company, for for various reasons.

所以我很好奇，在Insight这边，你们是怎么看到这一点的？

And so I guess I'm I'm curious, like, how have you seen that at Insight?

你们创始团队里有两位非常资深的专家，这对公司产生了什么影响？当初又是出于什么考虑，选择让他们深度参与，而不是只做顾问？

You know, having two, you know, really highly qualified specialists, you know, on the founding team, How has that how has that been for the company, and and how did you decide to make that decision in the first place to have them on, in such deep involvement versus, like, something like an advisory board?

是的

Yeah.

是的。

Yeah.

这是个很棒的问题。

That's an awesome question.

我觉得这个问题有几个方面。

I think there's a there's a few parts to that.

首先是可信度。

One, instant credibility.

对吧？

Right?

我虽然在科技行业干了这么多年，但跟医生、医生管理层或者医院打交道时，他们可能也就听听我说的。

Me as somebody who's spent a career in tech, speaking to a physician, speaking to physician leadership or a hospital or anything like that, you know, they would certainly listen to me.

但如果你身边有个真正当过医生的人，那完全是另一回事。

But if you have someone who has actually experienced this firsthand as a physician themselves, it's it's a whole other it's a whole other world.

所以，我觉得第一点就是可信度。

So, like, instant credibility is the first thing I would say.

第二，我觉得医生的问题解决能力其实和创业者要应对的挑战非常契合。

Two, I think the the problem solving skill set that physicians have is actually very it aligns very well to the stuff that you deal with as a startup founder.

对吧？

Right?

作为创业者，你得不断适应、持续向前。

Because as a startup founder, you just have to keep rolling with it.

你当然有自己的一套方法论可以参考，用来解决问题。

Like, you know, you have your your, your methodologies that you can you can look to to help solve problems.

但很多时候，你之所以选择创业，正是因为你想改变或颠覆某些东西。

But often, the reason, like, you're you're in the startup game is because you're trying to change something or trying to disrupt something.

所以你得懂得怎么应对问题，尤其是那些特别复杂的问题，拆解开来，想办法快速推进。

And, you know, I think you just need to know how to approach problems and sometimes very complex problems, break them down and and and try to figure out how to move quickly through that.

因此，我觉得医生亲身经历过的那种处理复杂问题的能力，和创业中遇到的挑战有很多相似之处。

So I would say there's a lot of similarities to what I imagine being a position is firsthand to solving relatively complex problems in the startup world.

第三，我觉得让那些有不同背景、曾经从事其他行业的人来参与科技创业，也有其独特价值。

Third, I think there's there's something about having people who are perhaps a different they had a different set of experiences building a tech.

现在做产品，门槛其实已经没了，说实话。

Because if you think about the barrier to entry now when building products, quite frankly, I don't think there is a product barrier anymore.

现在的AI模型已经这么强大了。

AI models getting as good as they are today.

谁都能随便搭个云服务，做出个挺厉害的产品。

Anybody can spin up Cloud Code and build a pretty amazing product.

真正能拉开差距的，是你在科技圈之外的行业经验。

So the real differentiation is is that experience that you have in a field that's outside of tech.

我觉得第三点特别关键，就是我们对临床工作流程的了解，帮了我们大忙。

And that's where I think the the third thing I'd share is that's been a huge accelerant for us, understanding the workflows, understanding how things work within a clinic.

像转诊、Priyat这些事，你开头提过，但其实远不止表面看起来那么简单，想真正解决这些问题，你得懂临床里的方方面面。

Simple things like referrals and Priyat, you mentioned that in the intro, in front of I I think, like, there's a lot more to it than just what meets the eye, and there's a lot more that you need to understand about the clinical workflow in order to actually solve that problem.

我过去几个月一直在说，就像你讲的，做产品已经不是难点了。

I think one thing I've said over and over again for the past few months is, you know, I keep like you said, building is not the hurdle anymore.

这根本就不是问题了。

It's not a problem anymore.

谁都能做个东西出来。

Anyone can build something.

关键是怎么推广。

It's distribution.

你怎么才能让真正用得上的人拿到它？

How do you actually get in the hands of people who can use it?

你说得对，拥有信誉、适应能力，以及对工作流程的深刻理解，这才是把产品有效推到诊所、医生和管理人员手里的关键——他们才是真正能用它帮助病人的那群人。

And to your point, having that credibility, the ability to adapt, and just being able to understand the workflows very deeply is exactly how you can distribute well and get your product in the hands of the clinics, the physicians, the clinic managers who are actually gonna be able to put it to use and help patients with it.

这说得通。

So that makes a lot of sense.

对。

Yeah.

最后一块，我觉得和前面那三点同样重要，就是信任与安全。

The the last piece, which I think is is just as important as the other the other three, is the trust and safety.

我们大部分产品都是直接面向患者的，所以界面设计上我们花了很多心思，但还有很多东西是普通用户根本看不到的。

So if you think about the interface that we set so we're patient facing in most of our products, and there's a lot that we a lot that we build that most customers would never see.

其实是背后的一整套机制。

It's the the framework behind.

我们自己开发了一套评估体系，更具体点说。

We have our own evals framework to get, know, more technical here.

我们搭建了自己的评估系统，用来监控异常情况之类的。

We've built our own evals framework to monitor outliers and things like that.

我们还设置了升级流程，必要时可以直接联系诊所或首席医疗官。

We have escalation paths to the clinic or to or to chief medical officers as and when needed.

所以我觉得，这类系统必须有临床背景的人参与设计和运营，毕竟关系到患者的健康与安全。

So I think that's something that you you you have to have someone who's clinical help design those systems and and run those systems, because of the the impact of, definitely, patient patient safety and and well-being.

说得对。

Yeah.

你只要走进诊所，或者哪怕是在Zoom上跟人交流，只要名字后面有个医生背书，就能大大增加信任感，因为对方知道你至少懂这套系统的基本原理，也了解医生的实际工作方式。

Just when you walk into the clinic or talk to someone on Zoom even, having the MD behind the name can help a lot because they know you have at least a fundamental understanding of how the system works, what, you know, your experience with the physician is practicing.

没错。

So that Yep.

这确实很有道理。

Definitely makes a lot of sense.

我觉得你提到了几个产品，比如患者登记和预先授权。

I think you've touched on a couple of the products, you know, patient intake, prior auths.

但从我浏览官网和了解贵公司的情况来看，你们的产品线还挺丰富的。

But from what I saw on the website and what I've read about your company, you have a a variety of products.

你知道的。

You know?

你们还有一个智能文书助手。

You also have a Scribe.

还有其他各种智能代理。

You have a variety of other agents.

我很想听听，你们的产品平台是怎么从单一的患者登记发展成现在这一整套智能代理的？你们接下来会优先考虑开发什么？

I'd love to hear more about, you know, how did your product your platform evolve from just one patient intake to this whole suite of agents, and how do you think about, you know, what's to build next?

是的。

Yeah.

这是个好问题。

It's a good question.

我们最初确实是从患者登记开始的，后来逐渐深入到围绕这一环节的整个体系中。

So intake was certainly where we started, and we since gotten pulled into the entire ecosystem that surrounds the position.

如今，诊所和医院面临的一个重大挑战，而且未来还会继续加剧，就是成本不断上升，而报销金额却增长缓慢甚至停滞。

So one of the big challenges that clinics and hospitals face today and will continue to to face this this probable compound is the rising costs and relatively flat or slow growing reimbursements.

最终，唯一的出路就是尽可能自动化，同时为医生提供支持，让他们能看更多的病人。

Ultimately, the only path is to automate as much as possible while providing the resources to physicians to see more patients.

你几乎必须用更少的资源做更多的事。

You almost have to do more with less.

我觉得，当我们思考问题的根源时，通常都始于某种转诊。

And I think as we think about where the problem sits, it usually starts with some sort of referral.

大多数专科诊所的入口，往往都是某种形式的转诊。

The on ramp to most specialty clinics is is usually referral of some sort.

因此，这自然就成了我们接下来要拓展的方向。

And that's the natural progression for us to go.

在接待流程之前，可能是文件处理环节。

So prior to the intake might be document processing.

我们目前真正产生影响的产品，通常都是从这个最初的接触点开始的。

So where we think about our current the products that are really moving the needle usually start from that very first touch point.

比如胃肠科诊所就是一个例子。

So take a GI clinic, for example.

我们曾合作过一家诊所，他们积压的患者案例非常多。

We had one clinic where they had a backlog come back to CRC example.

他们积压了大约四千名患者， backlog 大概有六个月之久。

They had a backlog of about 4,000 patients, and the backlog was about six months or so.

其中一个难题是，他们有四五个人专门处理这些积压，但工作量实在太大，根本忙不过来。

One of the challenges with that is they had four or five people managing this backlog of of patients, and they still had a hard time clearing it just with the sheer amount of work.

而现在，我们已经能做到帮他们分担这部分工作了。

And we're now at a point where we can offload that.

所以我们开始和他们合作。

So we started working with them.

我们只用了三周就清掉了这个积压。

We we cleared this backlog in about three weeks.

这一切都始于最开始的接触点——从转诊到临床筛查，再到入院登记，最终为这些患者安排结直肠癌筛查结肠镜检查。

And that started from the very first touch point, which is a referral into a clinical screening, into an intake, and then ultimately getting those patients scheduled for their colorectal cancer screening colonoscopy.

你提到了文书助手。

You mentioned scribe.

我们确实认为文书助手是个很有价值的工具。

So we certainly see scribe as being a valuable tool.

在很多情况下，电子健康记录公司已经开始提供文书助手服务了。

That to us, you know, in in a lot of cases, scribes are starting to be offered by, EHR companies.

对吧？

Right?

比如Athena就在自己投入资源。

So we see, like, Athena spending their own.

还有Ethic，当然也在投入自己的资源。

We see Ethic, of course, spending their own.

有几家选择合作，我们自己也和几家合作了。

There are a few that that choose to partner, and we, in fact, partner with a few.

我们当然也会通过这些合作伙伴来销售产品。

And we, you know, we, of course, like, sell through those as channels.

但总的来说，Stripe 和电子病历系统的关系比较密切。

But in general, we can see Stripe as being somewhat germane to the EHR.

不过，在那之前和之后的环节，才是真正值得我们开拓的机会。

The stuff that is before and after that, though, I think really is is for us the greenfield opportunity.

回到你刚才的问题，普拉纳夫，我们会继续完善这个端到端的流程：从最初的转诊开始，到就诊前的预检，再到长期的慢性病管理与随访，全程站在患者和医生之间，为低风险、常规性的任务提供自动化支持——比如主动联系患者、完成必要操作，再把结果反馈给医生。

So back to your question, Pranav, we'll continue to build within that orchestrated workflow that starts from the initial referral to, the pre visit intake to ongoing chronic disease management and follow ups and really sitting at that interface between the patient and the provider and providing them with this autopilot layer to where possible for low risk routine tasks, reach out to the patient, fulfill whatever needs to be done, and then come back and report to the provider.

这些任务由智能代理来执行，但医生始终掌握主导权，决定代理下一步该做什么。

This is what the agents have done, and then have the the clinician still at the in the driver's seat really telling the agent what to do next.

这就是我们看到的市场走向，我们会持续优化这个流程，从患者接触的第一刻起，贯穿整个诊疗全程。

So that that's how we see the market unfolding, and we'll continue to build in that in terms of our orchestrator workflow from from the very first touch point through the entire journey with that patient.

这种协同式的工作流程也特别有意思。

The orchestrated, workflow is is really interesting too.

对吧？

Right?

你刚才举了结肠镜检查的例子，其实已经说明了你之前开发的各个智能体是如何协同配合的。

Like, you you laid out the example of of colonoscopies, but in that example, you've laid out how each individual agent that you've already built out kind of plays together to facilitate that.

这几乎让人觉得，只要遇到预防性筛查，比如乳腺钼靶检查，也能用同样的方式来处理。

And it almost feels like you've you've basically built out a model where whenever we have a preventative screen, maybe it's, mammograms, for example.

对吧？

Right?

你可以直接复用这套组件，只调整一些临床细节，就能适配到其他场景。

Like, you can maybe apply the same exact formula of, like, the building blocks and then just, like, modify some of the clinical details to that.

你是不是也觉得，既然已经有了这些基础模块，就能轻松扩展到其他临床流程中？

Do you do you see it in that in that way of, like, you can just kind of scale since you have the building blocks and apply to other clinical, clinical workflows as well?

完全没错。

Totally.

你刚才说的这一点，对人群健康的影响其实非常大。

I mean, what you just articulated, it has one has a huge impact on population health.

其次，如今系统内靠人力来大规模开展预防性筛查的能力越来越有限。

And the second is, there is lesser and lesser capacity in the system today to actually do that at scale, the preventative screening using humans.

你说得完全对。

So you're absolutely right.

说到预防性筛查，我们最初从结直肠癌入手，但像乳腺癌筛查，我们也做了很多，后来还扩展到了一些初级诊疗诊所。

When you think about preventative screening, you know, we started with colorectal cancer, but mammography, you know, we we do a lot in you know, we've since branched out into some primary care clinics.

我们为医保和医保优势计划的患者开展了大量每两年一次的健康体检。

We do a lot in terms of biennial wellness visits for, Medicare and Medicare Advantage patients.

在扩大这类预防性筛查方面，有很大的空间可以发挥。

There's a huge opportunity to help scale that preventative, screening.

不过，这种方式必须确保不破坏医生和患者之间的关系。

That said, it it has to be done in a way that doesn't replace the provider, patient relationships.

有些情况下，患者的主治医生确实希望亲自掌握这个关系。

And in some cases, their primary care providers do wanna own that relationship.

所以，像预防性结直肠癌筛查或乳腺癌筛查这类工作，完全交由系统处理是可行的，因为没人会反对这些筛查是有益的。

So where it will impossible offloading that, like, preventative CRC screening or mammography is a great one because nobody nobody argues that that's good to do.

对吧？

Right?

证据是明摆着的。

Like, the evidence is there.

没人会说这事儿不该做。

Like, nobody says, you know, this isn't a good thing to do.

你得把握好分寸，确保还是以医生为主导，所有操作都得在记录在案的医生指导下进行。

There are there is probably a fine line you have to walk to to make sure you still maintain that relationship with the physician and and not do things that aren't still at the direction of of the physician who's who's on record.

你也不希望让医生觉得被冒犯，或者被取代。

And you don't wanna step on the physician or even know APC plus toes.

你不想让他们觉得你在试图替代他们。

You don't want them to feel threatened or, like, you're trying to replace them more or less.

没错。

Correct.

是的。

Yeah.

这说得通。

That makes sense.

你刚才提到的那一点，比如Epic、Oracle，我觉得挺关键的。

I think one point you mentioned a few minutes ago was, you know, Epic, Oracle.

他们有些公司正在EMR系统里自研助手，这样一来，护城河可能就没那么深了。

Some of them are developing their own scribes in the EMR, so that's sort of there's maybe less of a moat there.

所以，把这当成核心服务可能就没那么有吸引力了。

It's maybe less less attractive to focus on that as your primary sort of service or offering.

你有没有担心，或者觉得Epic这些主流EMR系统会切入你专注的其他流程，比如患者初诊这类环节？

Are you worried about, or do you predict that Epic or some of the major c m r c EMRs will get into the more patient intake flows, the, like, the other flows that you're focused on?

因为我觉得，迟早他们也会做这些。

Because I feel like at some point, they're gonna do it as well.

那在短期内，你觉得你们有什么独特优势能让你不用太担心这个？

So what what do you think differentiates you enough to not have to worry about that at least in the short term?

对。

Yeah.

短期内，Epic 确实已经有了一套患者登记产品，这是前提。

So in the short term, there's so Epic does have an intake product, to start.

对吧？

Right?

而且他们还在持续完善这个功能。

And, you know, they continue to develop that.

实际情况是，就像阿尔文提到的，整个流程是多步骤的。

The reality is the multistep, in Arvind, you mentioned this.

你不是在替换某个单一的环节，而是把多个智能代理串联起来，形成一个完整的流程。

Like, these agents being strung together in sort of this multistep journey, you're not really it's not a single point solution that you're you're really replacing.

你真正推动的是一个结果。

You're actually driving an outcome.

那这个结果是什么？

So what is the outcome?

结果就是帮患者完成筛查，并成功预约结直肠癌筛查。

The outcome is to basically screen a patient and get them scheduled to have their CRC screening.

听着很简单，但这一路上实际上有大约十三个不同的步骤。

Sounds simple, but there's probably about 13 different steps that happen along that journey.

而短期内，像Epic这样的公司并不会把重点放在这上面。

And that is just not where, at least in the short term, where the epics of the world are gonna be focused.

长期来看，也许情况会不一样。

Long term, I think, you know, potentially could be the case.

我们会持续跟进市场的发展趋势，不断调整方向。

And, you know, we will continue to evolve where we see the market going.

对他们来说，现实是电子病历系统本来就是数据来源，现在更是成了核心数据库。

But for them, the reality is the EHR becomes the source of rec I mean, it is already the source of record and becomes really that database layer.

而运行在上面的智能代理，就像人类使用电子病历系统那样与之交互、调用数据，但目标是推动整体结果，而不是单独的功能模块。

And then the agents sitting on top of that really act much the same as humans would that interface with the EHR and use that data store, but are actually driving outcomes rather than individual product features.

所以我们看到的那些更复杂的场景，根本原因是电子病历系统压根没设计来支持这些需求。

So that's where we see the the more complex use cases just is the EHRs are just not set up to help tackle that.

除此之外，我觉得大多数诊所的工作流程——我们现在和很多使用Epic的诊所合作，也包括中型市场——都面临一个碎片化严重的电子病历环境。

And beyond that, I think, you know, the majority of of the clinics work, and now we work with many clinics that use Epic, But the mid market space as well where we see a lot of the attraction is a pretty fragmented EHR space.

所以我们看到的还挺多的。

So we see, you know, quite a few.

事实上，今天在中端市场，我们经常接触到十三四家不同的电子病历系统。

In fact, today, we see about 13 or 14 different EHRs on a regular basis within that mid market space.

它们都没法像Epic或者Oracle那样打造强大的系统。

So they're not all equipped to build like the epics can or the oracles of the servers can.

所以有两个不同的方向，一个是专科领域，那些市场的电子病历系统很愿意合作。

So there's there's two different avenues, one on the specialty side where EHR is in in that market category are happy to partner.

我们确实和一些公司有合作。

We actually partner with some.

我们会把产品嵌入到它们的白标方案里，帮它们加速发展。

So we embed our products in a white labeled offering to some of these EHR companies to help them accelerate their own growth.

但另一方面，我也觉得，目前电子病历系统根本没把重点放在这上面。

But, also, on the other side, I just think the EHRs are are not focused on that today.

这说得通。

That makes sense.

说到电子病历系统高度分散的问题，你看看像日本这样的国家，虽然我不太记得确切数字，但大概有一半的诊所还没用上电子病历，仍然依赖纸质记录或者一些老旧的软件系统。

And to your point about EMR being very fragmented, if you look at, you know, even other countries like Japan, for example, fifth maybe if I think, I feel the exact statistic, but around 50% of all clinics are not using EMR is still relying on, you know, more preliminary fragmented systems, whether it's paper or some older software.

我很好奇。

I'm curious.

你们目前只关注美国市场，还是也考虑海外？

Are you focused only on The US market, or you also looking abroad?

是的。

Yeah.

我们内部确实讨论过这个问题。

We we've talked about that internally.

现在我们只专注于美国市场。

Today, we're only focused on The US.

我是在加拿大长大的，我觉得加拿大那边也有巨大的机会。

I I grew up in Canada, and I think there's a huge opportunity just north of the border.

但现阶段，我们的重心还是主要放在美国。

But today, for for this time being, our focus is primarily The US.

是的。

Yeah.

有道理。

Makes sense.

而且我觉得，这类流程也挺有意思的。

And, I think it's it's interesting too with these kind of flows.

每个国家都有自己非常独特的操作流程。

I think each country has its own, like, very specific, workflows.

尤其是美国，当你为美国市场开发时，几乎感觉像是要重新打一场仗，再进军另一个国家。

And especially for US, I think when you're building for The US, it it almost feels like an entire, different battle moving to to another another country.

对。

Yeah.

经济激励机制也不一样，特别是单一支付体系这种。

And the economic incentives are something different too, especially, like, a single payer.

我从小在多伦多长大，安大略省的OHIP——安大略健康保险计划，很大程度上塑造了那里的行为模式。

You know, I grew up in Toronto and on the province of Ontario, and LHIP, Ontario health insurance plan drives a lot of the behavior there.

我们当然也有私营企业和科技公司在做些了不起的工作，但那里的模式完全不同。

And, you know, we certainly have, like, private enterprise and and tech companies doing some amazing work, but it's just a different model there.

经济激励机制也不一样，这就导致了完全不同的采购行为，也影响了更广泛的行为模式。

And the the economic incentives are different, so they just drive different purchasing behavior, and they drive different behavior more generally.

欧洲市场挺有意思的，因为它有点像是双轨制。

Europe, I think, is an interesting market because it's somewhat two tier.

不过美国完全是私营体系，这让你能迅速成长起来。

But, yeah, US, I think, being entirely private has you have the ability to grow up pretty quickly.

确实如此。

For sure.

我觉得Insight Health特别有意思的一点是，我们接触的很多人做医疗AI，但主要面向医生或医疗专业人员，那边有一套明确的规范。

I think I think one of the really interesting interesting things about Insight Health is, you know, a lot of the people that we've talked to are building, you know, in the health care AI, but predominantly physician facing or health care professional facing where there's, you know, there's a specific rule book on that side.

但归根结底，总得有具备临床经验的人来核对事实，或者对系统输出的内容做出判断。而一旦面向患者，哪怕只是做初步问诊，不直接提具体建议，整个安全准则和交互设计的玩法都完全不一样了——从怎么引导对话，到怎么设定安全边界，都得重新考虑。

But at the end of the day, you always have someone that has that clinical knowledge that can fact check or like, exert a level of, you know, understanding of whatever is being said, where it feels like, you know, when you're building patient facing, and even if it's like an intake and you're not really necessarily providing specific advice, it seems like the playbook is is is a completely different one in terms of what you do on a safety level, what you do, and and how you, like, instruct the agent to interact with the user.

你对那些面向患者的助手，是怎么看待的？

How do you think about, you know, any of your agents that are patient facing?

你们有没有一个统一的准则，来区分面向患者的AI和像Scribe这样面向医生的解决方案？

Do you have, like, a central kind of, like, rule book of, like, this is how we, you know, think about, patient facing, AI versus, for example, your Scribe solution that's physician facing?

是的。

Yeah.

对。

Yeah.

这个观点非常到位。

It's it's a really good insight.

我觉得两者的操作方式完全不同。

I think the the playbook is completely different.

在医生端，你可以更注重准确性。

You you can be, more correct there.

我们一直在思考的一个问题是：怎么评估效果？

I think part of what we think about is, one, how are we running eval?

当初我们启动Insight项目时，到现在为止，市面上都没有现成的评估框架能直接用，所以我们自己开发了一套。

So when we started Insight, maybe to this day, there isn't an off the shelf eval framework that would work for our use case, so we built our own.

你可以从两个不同的阶段来考虑这个问题。

I mean, think about this at at in two different stages.

一个是实时安全检查，另一个是近实时安全检查。

There is real time safety checks, and then there are near real time safety checks.

在面向患者的应用中，我们任何时候可能都在运行大约四个不同的模型。

In a patient facing application, we have at any given moment, we probably have about four different models that are running.

首先是驱动对话的核心模型。

So we have the core model that's driving the conversation.

然后还有语音转文字、文字转语音这些模型。

We might have speech to text to text to speech models.

此外还有一个监督模型，确保所有系统协同工作顺畅。

And then we have a supervisory model that ensures that double orchestras playing well together.

一旦出现异常，它就能及时介入并终止问题。

And should anything fall off the rails, it it can help stop that.

比如阻止幻觉，或者识别安全风险——像患者需要挂电话拨打911这种情况，这些都会在实时中处理。

So things like, stop things like hallucinations, stop things like if if there is a safety risk, like, you know, if a patient needs to hang up the phone and call 911, like, that's all happening in real time.

而在近实时阶段，系统会事后处理所有数据，并将潜在问题上报给诊所。

And then in near real time, it processes everything in arrears and then potentially surfaces up any concerns to the clinic.

所以如果有安全风险或需要立即处理的紧急情况，就会在近实时阶段处理。

So if there is a safety concern or something that's more urgent that needs to be dealt with almost immediately, that happen happens in near real time.

无论是电话通话、语音界面还是聊天界面，我们都从这两个阶段来考虑。

So if it's a phone call that we're running or a visual voice or even a chat based interface, those are those are the two stages we think about.

最后一步，是具备真正的临床理解能力。

And then the final piece is having really a clinical understanding.

当我们总结通话或与患者对话的内容时，必须确保总结内容经过相关专科医生的审核。

So when we think about summarizing the output of a call or summarizing the output of a conversation we have with the patients, making sure that the summary itself is vetted by physicians in that specialty.

我们大部分的上线工作其实都是和诊所一起测试，确保他们不仅清楚整个流程的每个环节，也明白如何持续监控，识别我们系统之外的异常情况——这样他们就知道，除了我们做的，他们自己也要额外把好关，在每次互动发生时保持警惕。

And we do probably about I'd say the majority of our onboarding is really just testing with the clinic to ensure that they are, one, fully aware of the various different steps of the process, but also that they are aware of how to continue to monitor and check for, you know, outliers and anomalies in these conversations beyond what we do already so that they know, you know, this is the best thing is doing doing the diligence in in addition to what we're doing when these when these interactions are happening.

所以我觉得整个流程差不多分为三个阶段。

So so then I think there's there's almost three stages to it.

第一个是实时阶段。

There's the real time.

还有近实时的处理，再加上我们和诊所合作，确保他们完全明白自己收到的是什么，也知道怎么发现那些不太对劲的地方。

There's near real time, and then there's us working with the clinic to ensure that that they they fully understand what they're getting and and know how to detect if something doesn't quite look right.

这说得通。

That makes sense.

我觉得你提到的数据和案例都表明，这套系统运行得不错。

I think the numbers you've sort of mentioned, the case studies you have show that it's working well.

患者们很喜欢，而且它不仅对患者有益，也给诊所带来了经济收益——毕竟，诊所不会买任何不帮他们赚钱的东西。

People are liking it, and it's showing not only a patient benefit, but also a financial benefit for the clinic, which, you know, clinics aren't gonna buy unless it's helping them financially.

所以你得兼顾患者护理和财务收益这两方面。

So you need to sort of balance both patient care plus the the money aspect, the financial aspect.

但你显然不是唯一在做这事的公司。

But you're not the only player out there, obviously.

还有其他人在做类似的事情，我猜口碑传播也起了一定作用。

There's other people doing very similar things, and I'm sure there's some word-of-mouth aspect in there.

那你们的市场推广策略是什么？

But what is your go to market strategy?