John Anthony: Swim.com, WHOOP, Google Health, and Podium

John it's good to see you again. You spent a number of years in the space. You're very experienced about what's happening with wearables. You spend time in teams like Whoop, Google, and Supersapiens. So, I wanted to have this discussion and share some insights with the listeners of this podcast. Why don't we start with a small introduction about yourself? 

Yeah, sounds good. And thanks for the opportunity to talk it's good to reconnect with you again. I’m happy to be here. I have a background in biology and computer science and engineering and spent the first half of my career in Pharma biotech and financial services. And then that second half, for the last 15 years I've been in the startup space largely focused on health and fitness. There was a catalyst that supported the transition between those two opportunities.

I was running an innovation lab for a major financial services company and we were looking at the use of sensor technologies to enable new business models. So it was really around understanding, measuring and mitigating risk on behalf of our customers. That exposed me to really kind of the raw power of sensors in general and motion sensors specifically. At the time I was training for triathlons and we of course have like power meters for bikes and GPS for running but there was nothing for swimming.

So we, myself and a co-founder built the first AI based swim watch. You know, this was the 2009 time period so it wasn't a smartwatch and it really wasn't a watch, so we called it a swim tracker. But that was my introduction into product management in my first startup. Then from there, we sold that which eventually spawned Swim.com. And I did another startup, I was racing bikes and doing crit races and bike and road races.

And apparently was crashing more than anybody was comfortable with, including me. So we built the first automatic ride tracking, a crash detection device called bike tag. And so yeah I mean that really kicked off like my focus on health and fitness and you know, getting into opportunities like Whoop, Supersapiens and eventually being the product leader for Android Health as well.

Can you speak a bit more about the AI sensor you mentioned?

Yeah. So like I said this is around 2009, 2010. The Apple iPod and the API had just become available, so we had access to sensor technologies. We didn't have to go find a manufacturer. We didn't have to build our own piece of Hardware. We could just write software in order to access the technology to prove that you could use motion sensors. Specifically, accelerometers in this case, to track the movement of a swimmer, and then come up with or derive metrics like, how far you swam, what your stroke rate was, what stroke type you were swimming–so freestyle breast stroke butterfly, etc.

And so we actually built this prototype. We would vacuum seal an iPod Touch to the outside of your arm and swim with it. And we took this prototype to company called Finis, who we ultimately executed a joint venture with built the Swim Sense product. This became our dedicated swim tracker proprietary piece of hardware we built from ground up and brought that to market.

You had a lot of signal processing that had to be done in order to like turn motion data into swimming metrics. And a part of that was using machine learning to classify things like turns and stroke types. We were the first device on the market to be fully automatic, meaning you didn't have to press any buttons when you stopped swimming. At that point in time, there weren't a lot of swim tracking solutions in general.

So if I gave you a swimming set like 10 by 100s on 145 or with a 10-second rest for example, when you stopped swimming you'd have to press a button to pause it, you'd rest, and then you'd press a button again in order to start swimming. Right? And so you're demarketing your intervals manually. Just like you would if you're like running on a track. But that's like really problematic for swimmers right, because like, you're swimming, you're coming to the wall, your arm has to like reach over, hit the button as soon as possible because you want the most accurate timing.

So, we used a combination of signal processing and machine learning to automatically detect when a swimmer was swimming, versus when they were resting. I do swim with a many devices as I possibly can at all times. I'm probably biased, but I still think the swim sense algorithms we built, has roughly speaking a plus or minus one second accuracy on interval detection. Which is it's a really hard problem, because there's just so much noise happening from a motion perspective at the end of the wrist in a swimming environment.

Can we speak a bit more about that? Because I believe it's very difficult to measure accurately, even if it's a track for a runner and so on. Is this a sensor problem or is this a software problem?

That's a good question! I mean, we tackled it as a software problem. So we leverage existing sensors and I had mentioned before that we predominantly, in fact, exclusively used an accelerometer. Although a full IMU, like with a gyroscope a magnetometer an accelerometer, would certainly provide a lot more robust signaling, which would ultimately either reduce the software problem that you had or improve accuracy or even unlock new features–which we could talk about.

I think there's a whole host of things we could be doing from a swim tracking perspective if you have full 9’ed off sensing capabilities–but for us you know we weren't looking to innovate on the hardware side. And most aren't, I mean, even if you look at like Garmin for example, or Shunto, Apple watch. You know most of them, I would assume, based off of the capabilities that we see today from a swim tracking perspective, are probably just using the accelerometer.

You know Apple today can also detect kicks so if I had to guess, that would be a combination of using like a magnetometer as well as an accelerometer. But most of it is a software problem. It starts with a signal processing problem, like a classic signal processing problem, and signal filtering in order to remove all the noise that is not important for ultimately deriving metrics like stroke count, stroke rate, distance swam, and the type of stroke that you're swimming.

I think there are very specific approaches you can take where machine learning comes into play. Stroke-type classification like I had mentioned, is a great fit for a traditional classification problem. Stroke identification gets a little bit into the weeds. But if you think about what a swim looks like in a pool, you would kick off the wall and then you see some number of strokes, and then we will see a gap like something different than strokes, and then we see more strokes.

And so you can begin to imagine how you could apply machine learning to identify patterns of swimming versus patterns of resting. And if you combine that with a set of characteristics and good signal processing like we talked about all of this is very capable on existing hardware today.

Very interesting. When did you join Whoop? How early was it for the company? 

I think it was somewhere around 2019 or 2020. I was running swim.com at the time and was also a user of Whoop. Not just personally, but some of the athletes who I was coaching, and also my son who is a professional triathlete, the team he was on was using Whoop as well. So I was introduced to Whoop through a variety of different lenses. I had actually reached out to the Chief Product Officer at Whoop just to better understand if there was a collaboration opportunity between swim.com and Whoop in the world of swimming. And that conversation just led one thing to another and I was hired as their first director of product.

I believe from very early on Whoop had a very strong interest from a lot of athletes and I've seen a podcast lately mentioning that LeBron was one of the first 100 owners of a Whoop device. What did they do well early on that so many people like?

Yeah. I don't believe Whoop ever sponsored any of those athletes at least in the early days. I think there's some important lessons to be learned from a product development point of view here, which is being product-led versus being marketing-led. And I don't mean to say that like, you know, being marketing-focused is the wrong thing by any means. But Whoop is very focused on solving a very particular problem.

In the early days that was around building a sensor that could accurately capture heart rate variability, and then deploy that metric within context so that it became useful for users. And that razor-sharp focus became the bedrock for how we understand today. Their focus on the product and building a product that was both usable and from a form factor perspective, a battery life perspective, building context around physiological metrics like heart rate variability, you could argue that was their primary focus.

Versus for example, we know there are other companies who would take more of a marketing approach, really lean heavily on building the brand. You know, in some cases that dilutes the product opportunity. And we see that play out in like retention metrics or customer lifetime value metric. So I think Whoop got it right in the sense that they were focused on the product first, and then focus on growth and marketing perhaps secondarily to that. And you saw a lot of organic usage as a result of that coming from these professional folks in sports.

And you mentioned joining as the first director of product. How many people were there when you joined? And what was your first role there or maybe product that you worked on?

I was the 569th employee. So yeah, I was the first director of product, worked for Ben Foster. Ben you know, is prolific in the product management space just really like an amazing opportunity for me to be mentored by somebody who so experienced around Vision and product development. I was excited not only about the opportunity to work for Whoop and for a product that I believed in, but you know the opportunity to work with Ben specifically was just a step-function opportunity for me in my own growth.

When I first joined Whoop, we were organized in this sort of Spotify pod fashion where product managers and their teams were really focused on like functional capabilities. At that time, we were roughly organized by Sleep, Strain, and Recovery. Most of my focus actually cut across those areas so it was more foundational.

Bluetooth connectivity was an area of focus. When users have connectivity issues like the pairing with your phone gets dropped, there's just a very measurable amount of user friction and a destructive user experience that gets invoked that has a long-term impact on things like user retention. So you know, making sure that the connection to your phone was seamless, robust, could automatically recover, like when you were not within the context of your phone.

All of those things were really important for us to get right. Also, the membership services chat component that is in the app, that entire experience my team owned, developer API and connectivity ecosystem we own. So building a robust API for developers to come in, I mean you know this space well. And then there's also some features that just didn't fit very well within the structure that we have.

So like Whoop Live for example which didn't get the same attention as like Strain, Sleep and Recovery but that also fell into into my remit.

If we transition a bit to Whoop Live, I remember early on looking at videos of

athletes who have their heart rate on the screen. I think it was very interesting to understand the potential of this technology. But can we speak about it like how does this work? And what would it help to achieve?

Yeah. Whoop Live is an interesting feature in that it's technically really hard. I mean think about if you're a golfer on a tee and Whoop is capturing heart rate right, or you're in CrossFit you know going through a course and Whoop is capturing heart rate, or you're a NASCAR driver right and you're racing around a track. The technical infrastructure and the challenges in terms of capturing heart rate getting it off the device, getting it to a local networked infrastructure, that then could be pushed to a broadcast television in near real-time, so that you know you're seeing somebody's heart rate as it when it's happening.

There's a massive amount of infrastructure and technology, and intellectual property that goes into making all that happen. I think the impact is really around engaging the audience in like a fundamentally new way. It's almost like a fourth dimension, right? Because now, not only do I get to see the output of the athlete, whether that's a shot on a golf course, or NASCAR drivers racing around in their place and their performance.

NASCAR is an interesting example. The things that I'm seeing are like really exciting in terms of what a race car race means. What I don't have any appreciation for is what the driver is going through. Or what that professional golfer is going through on a tee. Unless you've done it before, it's difficult to imagine physiologically what's the response that that individual is going through. And so Whoop Live provided that kind of insight.

And it's just a single metric. It's just heart rate, but to look at the heart rate of a NASCAR driver be over 130, 140 beats a minute is amazing. Those NASCAR drivers are athletes and I wouldn't have described it that way before without having like that kind of insight through a Whoop. And similarly like the conditions as a golfer. It feels like a very calm game but internally physiologically right like you're controlling some very important responses, some sympathetic responses which is what Whoop is about.

And then it's surfacing those responses in a very consumable way. So I think it's adding another dimension for the viewer in terms of what the athletes are going through. I look forward into other areas.

I own a high-performance coaching company called Podium Coaching Group and we are working with SuperTry, formerly Super League which is the fastest-growing segment within the triathlon Market. It's really high-end downtown racing. Like high energy short course racing in amazing venues like Toulouse, France and London. And again it's the same story right as a spectator I get to see all this high intensity.

I get to see the output of that high intensity in terms of placement, athletes going in and out of the water, transitions getting on the bike. But what if I could see that you know real-time power, realtime heart rate. What if I could see stress responses that we're not talking about today. It just adds another dimension to the performance aspects and what those athletes go through when they're performing at their highest level.

Can we understand from your point when you joined Whoop? Did you have three teams that you are working together? And I'm guessing it's hyper-gross mode in Whoop. How does this feel like getting a lot of people in, a lot of people out? Like how are you witnessing things?

At the time that I joined Whoop, we were going through rapid expansion. We would talk about as a product team that it was like being on a rocket ship and that meant a lot of different things in terms of like how you plan, how you will think about how you're going to staff up, while simultaneously designing new features for which you don't even have capacity for on the bench yet. Aligned to your strategies that you know your three, four, five-year strategy you've got a lot of things in motion there.

And there's a lot of really important soft aspects. It's not like you can take a 20-person product team, lay down a plan to grow them to 40 or 50, and then expect that they're going to join and be contributing at the same level as somebody who's been a part of the team for the past two years. There's an onboarding. There's a culture. There's a velocity. There's operational aspects that these new team members have to come up to speed with.

There's the product itself. There's multiple dimensions that are kind of pushing against that velocity, while you as a team are trying to grow to create more capacity to drive higher velocity.

So it adds a lot of stress to the system, no pun intended, right? At that point, it's not just about building product. It's actually about building a product team who are all aligned on the same vision, and who are all driving towards the same kinds of outcomes the same kinds of impact. And so getting all of those pieces in place, it takes really strong leadership, a lot of communication. To be honest, it doesn't always work either, right?

There's going to be missteps. When you're growing that fast maybe you bring on individuals who ultimately, you decide aren't a good fit. And I think organizations need to be comfortable with recognizing that an interview process can only get you so far. And that ultimately you may need to course correct. That may mean a deployment into a different part of the organization. Maybe there's just not a good fit at all, right?

But those kinds of things that are easy to take for granted in a very large or stable organization all of a sudden become under a lot of pressure within a hyper-growth organization.

So for the last few questions, I wanted to ask you a couple of questions on sensors. I think there is usually two extremes. It's either there are going to be more sensors in like your shirt, your shoes, your watch, in everything that you use. Or there's going to be one sensor that overrules them all. From your experience, what do you think is coming?

Yeah. I love that question. I mean, there's really no question in my mind that the world we live in today, and certainly the world we're to be living in going forward, is one where there's more sensors, not less sensors. I do not believe in a world in which there's going to be a single device that's going to be able to capture everything all the time. 

Here's an example, I sit here today with you with a pixel watch on which is capturing my resting heart rate. If I had a pixel 2 it'd be capturing my body response as a stress measurement as a result of doing this podcast. I'm getting all these metrics passively, and I'm getting similar metrics on my Whoop as well. So, there are areas where there's overlap between two devices and that creates a different opportunity, but there's certainly areas where there's no overlap.

I can run with my pixel watch because it has GPS and I can either use Fitbit or Strava or whatever app to track my run. I get a visual display of my run, I can see my heart rate my pace, I can see what zone I'm in. Obviously I don't get anything like that from Whoop. There's no display, there's no GPS. So from an activity tracking perspective, I have a tool here that solves a very specific problem for me.

But I could say like I'm not that interested for me, like I don't sleep with my pixel watch on. There's a variety of reasons for that. One of them today is power, but Whoop very specifically solves that problem by allowing me not only to charge for 3-5 days and I can charge this device without taking the whoop off my wrist. So those things which maybe feel a little bit nuanced are critically important.

Because like every time I don't have to take the watch off, I don't have to make the decision to put the device back on again. And like I'm sure we see this with smart ring users, if we were to look at Oura ring. There's a lot of rings coming to the market from Samsung. Zep health is launching one. We're very likely going to see users with smart rings along with a smartwatch. Sennheiser earbuds were just launched which can now track temperature and heart rate through the ear canal.

Right now, I could be on a run and I could be getting heart rate from my Whoop, heart rate from my pixel watch, and heart rate from my smart buds. This creates another opportunity for the wearable space which is something that I think about around Sensor Fusion.

I think Sensor Fusion is more than just sensor prioritization, which is kind of where things stop today. And that is looking at where is the most robust signal coming from? From which device and under which context, and then using that data as the primary source. I'd argue Sensor Fusion is barely dealt with. You know, not at the level that I think we're talking about today. I see both more sensors coming into play where it's more specialized.

There's some really interesting research out of Caltech around Eskin technology using galvanic skin sensors to emit sweat responses that then opens up a whole new biowearable space. So as the sensors become more specialized, we'll see more devices within our environment sensing us. And then the challenge gets pushed down the road to how do we deal with all this additional data, and how do we merge it or fuse it in a way that is the most reliable, most robust, and most actionable from a user perspective.

I think if we get that right within a wearables perspective within our industry, then we can drive some very important outcomes for users, outcomes that we're not even talking about today.

John, for my last question, I wanted to ask if we go 100 years into the future, and we look backwards, what are the three most impactful things that are happening right now that we would be looking saying, those three things defined the next 100 years of the evolutions of wearables?

That's a great question! I think one is the new sensors and new and new algorithms. This is a key change that's happening today. And I think over the next five years within this horizon, we're going to see an explosion of new capabilities. Just to be really clear about that, we could be talking about fundamentally new censors like Eskin, but we could also be talking about fundamentally new correlations.

For example, looking at the relationship between resting rate and HRV to predict if woman is pregnant, or when she's going to give birth. Here we have two fundamental metrics that have been around for a while, but now we've been able to drive a new insight by correlating them in fundamentally new ways. So, that's one area. I think the other is our entire environment whether it's on body or off body is going to become more sensed.

I think ambient sensing is going to be more prolific, whether it's stress response, sleep monitoring. And again, the third important component, which is around Sensor Fusion. So, as we amass more data and new kinds of data sets, how are we going to leverage those insights to drive outcomes?

Now, those three things that I talked about, those are kind of like the plumbing of health and sickness. These aren't the things that I think users are going to be worrying about and thinking about. But I think that they are fundamental and foundational to driving–and I'm gonna steal something from somebody who I admire a lot Peter Atia who talks a lot about and actually in his book Outlive, talks about a shift from medicine 2.0 to Medicine 3.0.

And you know, we don't have enough time to talk about what that means but that will be a shift from population-based medicine to personal medicine, or reactive medicine to proactive medicine, or clinical kind of responses. I think if we get these three things right, I trust those will become important contributors to this shift from medicine 2.0 to 3.0. And I think in 100 years, we will look back at this shift as being one of the most important changes in terms of how we think, about and how we manage Healthcare.