Kristin Myers: Now, if you ask a doctor what they hate about their job, most of them are going to say the same thing. They hate the typing, the note after every visit, the codes, the forms, hours of it, often after the last patient has gone home. This is what physicians call pajama time. My best friend has told me a lot about this.
Now, my guest’s company puts AI in the exam room to listen and write the note instead. And they just raised $90 million. But first, let’s take a look at the company. The road from an idea to the opening bell starts right now.
Doctors go into medicine to treat patients, not to spend hours typing notes, filling out forms, and navigating software. But that administrative burden has become a massive problem, and that’s the opportunity Punit Singh Soni is pursuing. The former Google and Flipkart executive founded Suki with a big idea: to use AI and voice technology to give doctors their time back.
Suki Assistant helps clinicians handle documentation, coding, and other administrative tasks, but the ambition is bigger. Suki is also building a platform that lets other healthcare companies add its AI and voice technology into their own products. And now Suki has new fuel to scale, raising $90 million as it expands across the healthcare system.
But here’s the challenge: can Suki go from saving doctors time to becoming the AI infrastructure that healthcare runs on? From Startup to Stock Exchange, Suki founder and CEO Punit Singh Soni is next.
And joining us now on Zoom is Punit Soni, founder and CEO at Suki. Punit, thank you so much for joining us today. So I want to start a little bit with your background. You were at Google, and then you were the chief product officer at Flipkart in India. I’m curious to know, given that background, how did you end up dealing with what you called a health crisis in the American healthcare system, dealing with clinical documentation?
Punit Soni: Yeah, but first of all, thank you for having me. You know, when I left Flipkart, Flipkart was India’s largest e-commerce company, so we were dealing with a completely different set of problems. And I wanted to do something that would add value to the world, where my background probably has an advantage and where time would be a superpower. Oddly enough, Kristin, healthcare. It takes a few years to even be credible in it.
And so when I started looking around in healthcare, I noticed this thing that I was telling you earlier. About 88% of all doctors do not want to recommend their profession to their kids. People are retiring early because of the administrative burden that’s laid on top of them. Rural areas don’t have enough doctors because, you know, they’re just not practicing as much as they were earlier because of all of this administrative burden.
And so this idea came about: could we use AI to bring presence back to healthcare? Doing that would probably add value to society. It would require a lot of time. And because of my background running mobile apps at Google and working on voice applications, it would be something that I could actually do. So that was the reason I got into it.
Kristin Myers: So talk to us a little bit about how Suki works, right? If you are a doctor, you walk into the exam room, Suki is running. What essentially happens from there in the background?
Punit Soni: Yeah. I mean, think about the last time you went to your doctor. Most of the time, their backs are to you. They’re typing furiously while they’re actually trying to catch up with everything that they have to do with respect to you. The idea of Suki would be, it’s an application, you know, on your phone or on a computer. It listens to the doctor and the patient talking to each other, and it’ll create a clinically accurate note, which is actually a document that’s needed that represents that encounter.
Now, that note doesn’t just have the encounter information. It has all the problems that were discussed. It has the plan. It has the diagnosis. It has, on top of it, financial information like ICD-10 codes, which is how, you know, insurance actually then ends up reimbursing the doctor or paying the doctor. And so you end up getting all the financial information.
Now, when it has the note and it has the financial information, you could also then provide clinical reasoning. So pertinent to you and the discussion you were having, the AI could provide the latest [unclear: “Qi”] protocol or the latest information the doctor needs to practice at the top of their license. So you can see how this builds up: an assistant that can document, that can actually capture financial information, that can provide clinical reasoning, and ultimately should be able to just do the tasks that came out of that encounter.
Maybe you have a referral. Maybe there was a prior authorization required for a medication switch. Maybe there is a doctor you need to go meet. Agents running on top of the assistant can then do that too. So that’s how it works. It’s an assistant that can actually take care of all the work around the clinician, so they can focus on what they trained for, which is taking care of you.
Kristin Myers: So as you just mentioned, this AI assistant has the codes that get billed. And these notes, if anyone doesn’t know at home, do go into a permanent medical record for the patient and decide what gets billed to insurance. So let’s talk about oversight. What happens when that AI assistant gets something wrong?
Punit Soni: Yeah. I mean, first of all, the single biggest product here is not the assistant. It’s the safety around the assistant. It’s the evaluations, the quality, all the work that we do to make sure that the note’s exactly right, that the diagnoses and the problems that are extracted are exactly right, that the codes that are attached are pertinent. And for this, we have a lot of guardrails and a lot of things like hallucination checks and other things that we put into place to make sure that this product can be used safely across the ecosystem.
Think about it. In the time you and I are talking, there are going to be at least 100 patient encounters across the country that would have used Suki, across 99 specialties in 81 languages. That’s a tremendous responsibility. And so before I even talk about what happens and what the oversight is, the first thing to mention is that that is the actual product: clinical quality and making sure everything is captured accurately.
Now, at the end of the day, like all tools that we use, you know, you could be doing an interview and capturing notes. The tool is effectively an assistant to the doctor. And so the clinician has oversight. And our job is to make sure that the right information is shown in front of the clinician in a relatively easy way, so that it doesn’t create more cognitive burden in being able to provide that oversight.
So the clinician will take a look at the note. They will look at all the codes that are attached, the problems that have been extracted. And typically, they will sign the note and say submit, and it will automatically break up and go into all the different aspects of the medical record system, documented for perpetuity. So that’s how the product would work.
Kristin Myers: Talk to us a little bit, since you said the doctor is the one who would be signing off on these notes. I’m curious to know, one, if you know how carefully the doctor is checking these, but also if you could say a little bit more about what went into essentially building Suki, right, so that everything that was being done by the agent is accurate, medically accurate and clinically accurate, not just in English, but, as you mentioned, it’s obviously handling these in multiple different languages.
Punit Soni: Yeah. I mean, you know, we could go down that rabbit hole quite a bit, depending on how far you want to go. But the structure of how the product works is, effectively, under the hood, you have a speech recognition system that listens to the words and is able to convert speech to text. Then you have, on top of it, intent extractors. Intent extractors are things that can understand the intent of what you said. “Hey Suki, you know, create a new note.” “Suki, submit it.” “Suki, why don’t you insert a typical physical review of systems and adjust this.” Those are commands you can give to your assistant, and Suki needs to understand that.
Then you have, effectively, a manager of large language models. We use a variety of different large language models, including in-house models and some frontier models. Depending upon the specialty and the setup, different models are actually more effective in providing higher quality. Now, all of this is packaged into an application. That application is what the clinician actually uses on a day-to-day basis.
Our job is to make sure that the output of this application is evaluated and qualified and certified regularly across 99 specialties, 81 languages, across inpatient, ambulatory, urgent care, telehealth, and other settings. And so we have a large team of MDs who are constantly taking samples of these notes and running them through agents that do things like auto-adjudication to make sure that they can check the quality.
If we see any issues in the quality in the evals, we will actually tune our models further and not let it be released into production till it’s actually above the current production quality bar. And so those are some of the ways you actually go about building, not just, to be honest, Kristin, Suki, but also almost all AI application products in the market.
Kristin Myers: Let’s talk about time savings. How much time is actually saved by the doctors, and how are you measuring that?
Punit Soni: Yeah, well, you know, you mentioned pajama time. So the first thing to do is to be able to get a sense of that pajama time. There are two aspects to it. Obviously, there’s work that they’re doing with their clicking in the medical record system, checking boxes, typing things. You can actually calculate that because it’s on the computer, so you know what the baseline is. You can get an estimate of the amount of time the clinicians are taking when they’re sitting at home late in the night, going through their notes, preparing for the next day.
And so usually, when we go into a health system and deploy, we start by getting a baseline of how much of this pajama time is being taken. Now we have that baseline. We also get a baseline of the note quality of the encounters that we have. We get baselines of how the code captures are, so financially how things are being done. We also get baselines on how many patients a doctor is able to see if they wanted to see them. So we have all of these metrics.
Then we deploy the product, and three months in, we will do a recheck on the baseline. Ideally, what you should see is time taken to do notes decline dramatically. You should see note closure rates increase, so you’re capturing better. You should see the doctor just organically has more time, so maybe they will be able to see one or two patients more a week, or not. Maybe they go home early and spend time with their family. You should be able to see that various metrics of code capture, which means the accurate representation financially of what happened, actually become more specific and more accurate over time.
We capture all of this information. On average, what we have seen is that clinicians can save anywhere from an hour to an hour and a half a day when they use a product like this. And certain reports, like KLAS, which is the healthcare Consumer Reports, have estimated that if you take all of these savings and quantify them in terms of money, then it’s about $2,688 per user per month of incremental savings and/or incremental revenue, if the doctors use AI in this way in their practice.
Kristin Myers: Now, here is the problem that my guest is selling against. American physicians spend roughly two hours on paperwork for every hour of direct patient care, and a large share of that happens after clinic hours, so much so that doctors even have a name for it. They call it pajama time. About 45% of US physicians report at least one symptom of burnout, and documentation burden is consistently named among the top drivers.
Now, the electronic health record was supposed to fix all this. It was mandated and subsidized by federal law in 2009, at a cost of tens of billions of dollars. But just 15 years later, the pitch for the next new technology is that it’s going to undo the work created by the last one.
We’re back now with Punit Soni, founder and CEO at Suki. So, Punit, I want to talk a little bit about your competition. There’s Abridge, Ambience. There’s Microsoft DAX. There’s Commure. Several of them have raised multiples of your $90 million. What do you do, and what does Suki do that they don’t?
Punit Soni: You’ve been doing this a long time, Kristin. Since when was raising a lot of money correlated to actual success of the company? You know, if the metric of success was how much money you raised, then there would be other companies that could be successful. But the genuine metric of success usually is, you know, how is your revenue growth? Are you able to build a business that’s sustainable? Is the product that you’re building being loved?
You know, we have super-high NPS scores that rival some of the best consumer companies in the market. We have almost no churn, so people who actually use us almost always continue using Suki forever. So I think user satisfaction, happiness of the user, putting a dent in the problem that you just mentioned, making sure you can do that with good revenue growth and sustainable margins, those are the better indicators of success in the world we are in. Especially in the AI world, you can raise as much money as you want. That’s never a mark of success.
Kristin Myers: So I want to talk a little bit about the fact that Suki, and we might have made it sound like this, Suki is more than just an AI assistant that is writing a note. Suki also has the ability for other companies to essentially build on top of it. So I wanted to dive into that just a little bit. But I’m curious to know, if four companies can all transcribe a visit and write a note, is that not just a feature that ends up inside of the electronic health record, instead of inside of the company itself?
Punit Soni: Yeah, I mean, it’s a good question to ask. The question is, what are we building? You know, we are not building a clinical documentation feature. We are not building a revenue cycle management feature. We’re not building a clinical reasoning tool. We are not building agents that can do scheduling, you know, referral, prior authorization, and all sorts of other tasks. We are building all of it.
All of it together is an experience, effectively an infrastructure layer. We call it ambient clinical intelligence, which works seamlessly through the documentation and administrative tasks, financial tasks, clinical reasoning tasks, and operating tasks together on the same data layer. And so, you know, it’s possible that certain aspects of these could be stuck in different interfaces and work. But it’s very difficult to build an AI-native product, ground up, that can do all of these things seamlessly, in a way that the user is just focused on the patient and everything else is taken care of. And so the idea of building ambient clinical intelligence and a true assistant that can do all of these activities is much bigger than the idea of just building clinical documentation.
Now, the other point I’ll make to you, Kristin, is that clinical documentation, even though that was the entry point, the first feature we all started with, is not easy. A lot of people think it’s a commodity. I will contend it’s a utility. What that means is every single person needs it, but there’s only a few who know how to do it at super-high quality. Because remember, you know, if we got one zero wrong somewhere in that particular note, there’s a huge consequence to it. The clinician cognitive load, oversight, you still have a lot of work to do. So from my perspective, these are very difficult, very sensitive areas which are not commoditized. They’re actually utilities that require a lot of investment to do.
Now, that was one part. You asked about the second business line. The same AI platform that we use to build all of this ambient clinical intelligence, the Suki app, we also offer as APIs and as a platform to our partners, some of the largest healthcare tech companies in the market, who use it to build their own versions of the Suki app. So we have companies like Optum, like Zoom, 15 medical record systems, and lots of other healthcare tech companies who are building their own versions of ambient clinical intelligence using our platform. That is going to bring presence back to healthcare in equal parts as building a Suki app and giving it to doctors directly.
Kristin Myers: I want to ask you about that action layer that folks could see on the screen just a moment ago, as you were describing everything that Suki is really doing and the layers that you’re building. I’m curious to know, because it said that Suki would be able to even do some of the tasks. What does that look like? Is it going to start suggesting referrals or even follow-up care? How far does it go into taking on some of those doctors’ tasks?
Punit Soni: I mean, the sky’s the limit, to be honest, but the initial foray is going to be in building an agent layer on top of all of this. So think of it: you went to the doctor. You know, the doctor said, “Well, what’s my schedule like?” It said, “Kristin’s coming.” The doctor said, “Okay, give me a summary of what I should know about Kristin before I see her.” It provides that summary. The doctor looks at it and says, “Well, you know what? I really need to see that chest X-ray. Can you give it to me?” It brings up the chest X-ray. “What medications was Kristin taking?” It gives the doctor that information.
Then you walk in. The doctor says, “Okay, you know, let’s just listen. Pay attention, Suki.” And then the doctor and you are talking. The note is created. The note has all the problems that were discussed, the plan. Then at that point, the financial information is all attached, so all the codes are being attached. And at that point in time, maybe the doctor wants to know the [unclear: “key”] protocol in a particular diagnosis that they were discussing with you, or a new kind of plan, and the Suki Ask feature actually provides that.
As a result of all this, at the end, maybe the doctor wants to refer you to go see a cardiologist. Maybe there is a medication switch that’s happening that they actually want you to get a prior auth for. What Suki will now do in the future is actually route that referral and make sure it ends up going through the staff directly to the right cardiologist. It will actually take that medication and, based on the insurance information, etc., do the prior authorization, so that the switch actually goes through seamlessly and is then sent to your pharmacy.
This is no longer an assistant that’s capturing. It’s no longer an assistant that’s just providing intelligence. It actually now becomes an assistant that’s doing the actions that come out of it. All of these are a part of the same roadmap of building a genuine ambient clinical intelligence assistant that doctors can use, so that they focus on clinical care and you, rather than all of this other work.
Kristin Myers: So let’s talk about billing and the revenue piece, because a lot of patients say, you know, healthcare is just getting incredibly expensive. They’re getting billed for all of these things. I’m curious to know if you think that Suki is actually making the billing piece more efficient, or is there room for it to essentially say, “Hey, there’s all of these codes that we could actually attach to this one visit that we’re going to bill insurance,” right? And then if insurance starts paying out more, they tend to pass that on to the consumer in terms of those premiums and what they’re paying every single month. So on which side of the equation do you see Suki sitting?
Punit Soni: Yeah. Kristin, when I looked at this from the outside, remember, you know, I have no experience in healthcare beyond the few years I have now built Suki for. It was actually astounding to me to see all this kind of work going on, and billions of dollars of revenue cycle management companies being put together, all optimizing this stuff.
At the end of the day, the question isn’t really about up-capture, increasing codes, or down-capture, decreasing codes. The question is one of accuracy. I have the fortune, or the ability, to be in the patient encounter live when a doctor and a patient are talking. So Suki’s job should be to accurately capture everything. Accuracy would mean the doctor doesn’t have to cognitively remember all the things that they wanted to make sure were captured. Accuracy would mean you would not actually be capturing more than you should. You shouldn’t be providing a level that’s higher than it is. It’s actually just exactly what it’s supposed to be.
The truth is, if you actually go any other route here, Kristin, you end up in some kind of a weird war, you know, where some AI is being used to upcode and other AI on behalf of payers is used to, you know, decline these claims, etc. And then you have this weird AI warfare. The real answer to that is actually accuracy, and being very, very transparent about exactly what you captured, and providing an observability layer so that you can go into everything that was captured and see exactly the trace behind it, which says where this information came from. If you do that properly, I think it will actually lead to optimization of the healthcare ecosystem, potentially efficiency in how revenue is shared, so that everything else then goes into what matters, which is clinical care.
Kristin Myers: So, Punit, I want to ask you something that we ask everyone. And you mentioned how long you’ve been building this for and how long you’ve been working on this company. So I want to talk, because I am on the floor of the New York Stock Exchange. You’ve got the bell right behind me. Do you think about going public? And if so, what does that day look like to you, and when do you think going public might actually happen?
Punit Soni: Well, I think about it. And I think about it with the idea that I need to be deserving of that opportunity. You know, we have spent eight years building this company. It has now actually gotten to a place where we look at this revenue and the scope and the footprint, and we feel proud. There’s 450 health systems who use us, 30 large healthcare tech companies who have incorporated us into their products. You know, our revenue doubled last year and is on the same kind of upswing this year. Patient encounters using us have grown about 190% in the last, you know, year or so. So there’s a lot of pride in what we have done.
But also, there is an acknowledgment that, you know, we need to have the permission to get to a place where we can dream like that. To do that, there’s a little bit more work to do. So is it my aspiration that one day we will go public? Absolutely. I mean, that’s why I started this company, so that we could actually build an independent company that could do that. Are we there yet? No. We have a couple of years more work to do before we can actually start dreaming about that. So I would be very happy when that day comes. But right now, I have to make doctors happy and grow this business from where I am.
Kristin Myers: Hopefully we can bring you on again when that happens. Thank you so much. Punit Soni, founder and CEO at Suki.
