Transcript: Alex Frenkel | Kai.ai | How Do You Combine a Human Team and AI to Support Mental Health?

Intro, what Kai.ai is, and the move to Hebrew after October 7

Eyal David: [00:00] Hi friends, it's Eyal. You've reached Product Builder, where you can listen to the most interesting product conversations. My journey began in 2012, when I learned what product is at my own startup. Since then I've worked in the industry, and since 2017 my company has been advising a wide range of companies. My drive is to help all of us get better as product people. And before we start, if you enjoy listening to Product Builder, please rate us five stars on Spotify. It gives me great feedback that the podcast brings you value, and it would make me very happy. So yalla, let's start. Hi Alex, how's it going? Great. With us today is Alex from Kai AI. Alex,

Alex Frenkel: [00:41] So can you tell us a bit about Kai AI's product? Sure. Kai is building a platform that makes it possible to give emotional support to thousands of people, around emotional distress, and growth, and development, and quality of life, through a combination of a human, clinical team and conversational artificial intelligence. Today we also work in Hebrew, here in Israel, providing support and help to reservists, and we also work with universities and with employers.

Eyal David: [01:13] Super interesting and meaningful. So maybe let's actually start the story from there, with the story of, really, the Hebrew story.

Alex Frenkel: [01:21] So maybe tell us how that happened, where does it start? Sure. I think the seed starts on October 7th. On October 7th, both I and Netanel, who is our co-founder and CTO, were at Google, in Mountain View in the United States. I was on the flight to Israel on October 7th, and my plane simply turned around and went back to San Francisco. It took me a few weeks to find an El Al flight back, and very quickly it was clear that we wanted to help with what was happening here. There aren't enough therapists, there weren't enough therapists before, there weren't enough therapists before COVID, awareness has grown a lot, but Kai didn't speak Hebrew. And luckily, we were in conversations with the JDC (the Joint), and they completely took up the challenge and helped us do the localization, and actually start operating the service in Israel with partners here. That was a very, very meaningful process, one we're proud of, and we feel the company is developing and really growing and learning.

Eyal David: [02:27] So yes, that's meaningful. So let's go into localization for a moment, because localization here isn't just translating texts, it's much more than that. But if we're already starting there, maybe tell us for a second what the experience in Kai looks like,

Alex Frenkel: [02:37] And then apply the localization to that. Sure. The core of the Kai experience is simply a contact on WhatsApp. From the very beginning Kai worked directly inside messaging. There's an iOS and Android app, but almost nobody talks to Kai in the apps. In the US almost everyone is on Apple Messages, young people are on Discord, there are countries where Telegram is easier, in Israel almost everyone is on WhatsApp, and Kai is simply a WhatsApp contact. What's very special is that Kai is a proactive contact, and she'll reach out, literally at the level of the morning, right? Kai wrote to me, good morning, Alex. She sends me something for inspiration, she asks me some reflection question, she remembers me and knows me, knows my daughters' names, knows where I work, knows the main topics on my mind, and she'll accompany me. She'll accompany me during the day, during the week, over time, and help me pause and practice and do breathing and do reflection exercises that come from the therapeutic world. And she's very, very good at check-ins and follow-ups, she doesn't forget, right? Hi Alex, I haven't seen you for two days, just wanted to see how you're feeling. Eyal, it's been four weeks, let's see how you're doing now, let's do some kind of assessment. And that's the core of the experience. A lot of the magic happens behind the scenes, because we actually have a clinical team that does an intake for every new person who starts working with Kai, before they start talking to the AI, and tailors a specific personal plan for each person. Because it could be that I'm dealing with parenting young children, and someone else is currently dealing with their relationship, and there's someone who's currently in Gaza on reserve duty, and each of us has different needs, and there's a plan tailored to each person. And the team actually creates another WhatsApp channel, which lets you reach the team. The team can adjust the plan later on too, it can reach out when red flags come up, showing there's a need to step in and write to someone directly. And this also creates a journal for the user, where I can see what I'm writing about, what topics are on my mind, how consistent I am, how I'm improving, and for the team it creates a kind of mirror of how people are doing, who needs more attention,

Eyal David: [05:09] So that's how it works. So just to close the localization corner, there were probably the platforms you operate on, and probably also various kinds of training for the AI, for the model, that we could talk about. I'm curious, I'm stepping out of localization for a second, what actually is the KPI for success here? How do you know you've succeeded?

Measuring success: WHO-5, engagement and retention

Alex Frenkel: [05:32] So look, we've been operating for four years now, and we've gotten to see hundreds of thousands of people. I think the most meaningful KPI is improvement in mental well-being. Up until today, even before we started working in Israel, we published two large studies, thousands of people, thousands of adults over 18. We have a lot of experience with teenagers, ages 14-18. Kai actually works from age 14 to 84, I suppose that's the range, but a lot of Gen Z. And we see a real improvement in well-being, on a World Health Organization index, a 30% improvement after four weeks, improvement a few months later as well, among women and among men too. That's something we're seeing today in Israel among reservists as well, and today we're also running a clinical study. So the impact and the improvement in mental well-being is a critical KPI. Another KPI is really the usage itself. One of the problems of the many apps in these worlds of mental health and well-being is that usage is very low. It's like a gym with amazing equipment: people buy the membership but then it's hard to stick with it. And we see something different. Kai receives five messages a day on average, per user. 30% of users messaged Kai every day, every day, every single day during the month, 80% will do it twice a week, and they stay at double-digit retention even after several months, which is very, very unusual. And Kai, even though she's AI, and we never pretend to be a person and the user knows it's AI, receives thousands of thank-you messages for the care, for the help, for the intervention, for the exercises. For us that's a cluster of data points that shows the success. So real improvement in people's condition, the high level of engagement, retention, usage, and really direct feedback from people saying it's one of the best things that's happened to them, especially in the recent period. Stunning. And do you actually measure the improvement in quality of life? How do you close the loop? So it turns out the AI is amazing at this, because it doesn't forget, right? When I start, we run some baseline questionnaire. We use the World Health Organization's index a lot, it's called WHO-5, it's really the gold standard, it's linked to anxiety, to depression, to all kinds of other measures, and it's been researched for decades. And Kai will say, Alex, it's been a few weeks, let's see again how you're doing, and she'll carry out this tracking over time. Today, since it's a clinical study, there are control groups, and still, the measures will be taken every period. And people fill it in, especially when it's in a WhatsApp conversation, because they also want to know for themselves where they stand, how they compare to others. There was something that was amazing for us at the start: people would connect, do the initial assessment, get a very, very low score that really indicates significant distress. They'd take a screenshot of the low score, post it on social media and say, okay, I feel validated, because I knew not everything was fine and dandy,

Eyal David: [08:38] And this helped me. And this helped me see that I really can do things to improve my mental well-being. Wow, that's crazy. That actually also means, from this story, it means that people also arrive, so to speak, very ready for this product. And then you're actually confirming their concerns, that's basically what she said. So it's much more engaging in order to... and then you actually create a kind of habit, a conversation over time, and then it actually becomes

Alex Frenkel: [09:07] Second nature, really. Exactly, you're touching on two very significant things. One is that we never targeted a clinical population. Kai doesn't replace therapy, and we never said, you suffer from anxiety and depression, come use Kai. And when we worked directly with the users themselves, it was a very basic question on TikTok or Instagram: how are you, how are you feeling, would you like to know, start a conversation with Kai, Kai will help you take the initial assessment. And sadly, you know, what became very, very clear, very quickly, is that the baseline level of distress is very, very high, including, you know, places of real distress, not just anxiety and depression: self-harm, eating disorders, a lot of issues of identity, values and meaning, far beyond what we expected, because we weren't reaching out to a clinical population. So that was a part that was very, very surprising,

Eyal David: [10:04] Very, very surprising. Here I assume you're referring to Gen Z, so I'm putting a pin in that and we'll come back to it, because we'll hear the story of how you started, and then I'll probably touch on that too. And I'm also really curious how specifically these young people, whose attention span is supposedly very, very short, suddenly have the attention to talk now in a conversation with an AI,

Alex Frenkel: [10:25] That's exactly the point. You touched on it earlier through habit-building, um... the toolbox in the therapeutic world is very, very good and it's been validated for years, and the tools help. Today Kai's chief psychologist, Prof. Ronit Shoshani, is a researcher of happiness and resilience, and one of the things she says all the time is that mental well-being is something you can work on. We'd often compare it a bit to doing yoga to maintain flexibility: if you practice and make it a habit and do it for a few minutes every few days, then there's psychological flexibility, and then when things get harder you have a toolbox and you're more skilled and know how to use the tools. That's a very, very significant part, and you're right. The fact that the coach doesn't write you a plan and tell you go do it now, but knocks on your door at seven in the morning and says, hey, right now, do you have a few minutes? Let's practice, how are you feeling? The fact that a journal is created automatically builds this habit, and it becomes something people simply practice, do the work, and simply improve their mental well-being, but also their psychological flexibility,

Eyal David: [11:38] Which ultimately also builds resilience. I think that's because your product is good. You know, on the other side of the fence, what creates habits is Facebook and all kinds of gaming apps; with yours maybe it pushes back toward you. So let's maybe dive in for a second to understand, not a second, but let's dive in to understand your story,

From clinical psychologist to founder: how Kai started

Alex Frenkel: [11:57] You can start wherever you want. Yes, look, Kai is the second startup I've gotten to found. I come from a psychology background, I did a master's degree in clinical psychology, and I worked as a therapist for years, at first in clinics in the south, both public and private, both adults and adolescents. Then we moved back north, I live in Haifa today, and I worked at a large center called Telem in Oranim, dozens of patients I treated, and I was the psychologist of two schools. And already then it was very, very clear that something isn't working in how this world operates today, because there aren't enough therapists. People were waiting for appointments already back then, and that was before COVID. And in the schools where I worked, there were adolescents, kids, whom I could see once every six months, and that made no sense. The impact you make one-on-one in the room is very, very significant, but it also has a lot of limitations. And I made the move to high-tech really by accident. I founded a startup in other worlds, AI for games, something that started as a hobby and suddenly became a company, and I fell in love with this combination of technology, which lets you reach scale and create impact on hundreds of thousands, millions of people. I was very lucky to work on products that reached a lot of users, together with the psychological component and the accompaniment, together with a team that came with a psychology background. And somehow, when we started Kai, all the stars aligned: Ziv and Netanel, who are my co-founders, were available and wanted to build a meaningful company together, and that combination was the beginning,

Eyal David: [13:49] There was an initial version of Kai on WhatsApp, and that's how we started. So just to get this straight for me, so Ziv is the CTO and Netanel,

Alex Frenkel: [13:58] Netanel is the CTO. Netanel worked with me at Sears and was VP R&D at Kik; before that he was CTO at another company. He's a technology leader who has led teams for many, many years. We were together, shoulder to shoulder, I was product and he was engineering lead at Sears, we worked on social commerce products. And Ziv did the master's in clinical psychology with me, but he's a serial entrepreneur. He actually built an online ticketing company a bit after his military service, decided that's it, sold it, decided he'd had enough, went to study psychology, and that's how we met. But he's a growth person, he's a person who simply makes things happen. He built the initial version of Kai himself, showed it to us, and it took us a few weeks to understand the potential of the technology for emotional accompaniment, emotional support. Think about it, this was before ChatGPT, before OpenAI. And after a few weeks we knew this is what we were going to build, and that's how we started. What was the scope of that version? What did it contain? You'll laugh, but it was mostly me and Ziv sitting on WhatsApp. Ziv's mom was the first user, then family and friends were the next users in line, and little by little we started building the daily routines, started adding the exercises and the protocols into the system, little by little the team grew,

Eyal David: [15:16] And suddenly we found ourselves with a thousand users, and then ten thousand users. Wait, wait, wait, wait, wait, wait, Alex, one second. First of all, let's take these jumps slowly, but I have a question I can't figure out: training such a model on lots and lots of populations, segments,

Human in the loop and a collection of models instead of one

Alex Frenkel: [15:32] It's not like training it now on mom. How did you actually do that? So you need users, of course. There were two things that were really, really critical, which I think are at our core to this day, and they're a real part of our credo. So from day one it was Human in the Loop. That was a principled decision. We built an amazing team in the Philippines, super, super, super smart people, with a high level of English, and they helped a lot to improve the experience in edge cases, in tagging, in identifying situations where the AI doesn't work, and that's something very important and central. The second thing that was critical is that very quickly we understood there wouldn't be one model that's safe enough and flexible enough and good enough and would work in Hebrew and in Russian and in English, and could handle a population of elderly third-generation people and a population of young people. And from the start we worked with a supplier of models. There are models that are very, very good just at returning the sentiment score in a conversation. Right, we use a Google model, and it knows how to do that very, very well: say, happy or sad in the conversation? There was a model we trained ourselves to identify prominent emotions. What's the main emotion appearing right now? And as the models keep getting better, we kind of pick and choose, plug and play, a collection of models. Our magic is knowing where to route, to which model, what to give that model dynamically, how to create the continuity and the memory, and how to do it safely. For example, to check that Kai doesn't suggest medication, that she doesn't suggest to anyone to divorce his wife, there are models whose only job is to validate other models. In Hebrew, Gemini works very, very well for us, better than other models, but it could be that in a few weeks there'll be a different model. So that's something that differentiates us from many companies: we don't train one base model ourselves meant to be some kind of LLM psychologist, but rather use the best, most flexible, most efficient models we currently have, whether open source or commercial,

Eyal David: [17:47] To really deliver the experience that is Kai. Stunning, that's it. I think Ilon from Gong also mentioned they work in a similar way, so that connects very well. Let's think for a moment, so we're actually going back to those crazy growth jumps, those user jumps. First of all,

B2C growth, Discord, and why Gen Z

Alex Frenkel: [18:03] Let's take a breath and say that at this stage you were aiming at B2C, right? Absolutely, our whole focus was B2C. We're people who come from a B2C background, with all the understanding of what it involves and how to build it properly, and it worked really well, it worked really well. We really started with family, friends, word of mouth, people spread it very easily. The decision not to send people to the App Store, to install an app, not to go through all the permissions and all those parts, was a massive game changer, because all you need is a QR or a link: you tap, you connect, no matter which messaging platform. And it was very, very efficient both in terms of organic viral growth and in terms of CACs, working with TikTok and Instagram and other platforms and working with Gen Zers.

Eyal David: [18:59] Let's stop there for a moment, park again before we move on. Making the decision now to focus on Gen Z, it isn't something that's only Gen Z, because you spread out and probably tried to see what sticks, I assume, against where you can bring the most value at low CACs.

Alex Frenkel: [19:16] So how did Gen Z actually crystallize here? You're completely right, it came very, very much from the data. I think that's something that really characterizes us: we look closely at what actually happens in reality and not at some fantasy of what we'd like to happen. And for Gen Zers it was much, much easier to trust, to share in depth, precisely because it's AI; trust was simpler for them. Shame, in terms of fear that they'd be judged, or that something's wrong with them and they're idiots and people will laugh at them. And the usage was very significant. There were other groups: young mothers, women working in high-tech, people juggling family and career simply connected more naturally, and their usage was more significant and lasted longer.

Eyal David: [20:15] It really grew out of user behavior and the data. Which is interesting, also fishing for those Gen Zers, I'm sticking with this for a second, fishing for them, going now to actually find them on their platforms, like making that decision, you had to come with some bit of creativity and that kind of place, and probably also,

Alex Frenkel: [20:34] How did you validate it before you went all in? That completely, completely belongs to Ziv. He actually built the strategy. I think one of the pivotal moments for us was the decision to work on Discord. We actually built a community for young people that simply spread like wildfire. Because of the distress, because of the difficulty, because they're online all the time. You know, we haven't mentioned it until now, but in our world there are two more problems. One is that there aren't enough therapists, it's very expensive, it's not accessible, stigma, that's one problem. The second problem, which wasn't self-evident, is that face-to-face sessions don't suit everyone, and especially not young people. I look at my daughters, they're teenagers, at the end of high school, it doesn't suit them to go meet an older person and have an open conversation with him about life. And it also sometimes doesn't fit the pace of life, because 50 minutes once a week, no, something happened now, tomorrow morning I have something that's stressing me out, and I don't want to wait. This is a generation that doesn't want to wait. And on the other side, the way Kai works, which is asynchronous messaging, and in English Kai also works in voice, suits them very well. Because it can be at 12 at night, I can't fall asleep, it can be good morning, I have no energy to get out of bed and go to school, I feel awful, awful, right. The flexibility is very significant, and it was very natural for them. And the trust, and the trust: when we worked with adults, and especially with professionals, it's funny, but the first thing they'd do is try to break the system. Almost all of them would try to be suicidal, so to speak, to see what happens and what the AI would do in those situations. And that's great, because it let us build very robust protocols around edge cases. But with the Gen Zers we saw natural understanding, commitment, a lot of usage and real impact, it helped them,

Eyal David: [22:44] That's very important. Wow, what you did is so smart, starting broad without assumptions. I mean, every marketing campaign starts like that: either you start very focused or you start broad, and then you actually dove in. Interesting, also the approach of going to Discord, amazing. How did it actually go from there? So you said there were growth jumps, I understand Ziv is a rockstar and he actually managed to make it happen. What did it look like? Because, like, you were an app you have to pay for, I assume, at this stage. We started completely free, before the economic crisis,

Alex Frenkel: [23:14] Classic hockey stick. We reached 80 thousand monthly active users, very, very fast, hundreds of thousands of users, studies we published showing the impact and the usage, very, very high engagement and retention numbers, and we raised our pre-seed entirely on the basis of the data. And along the way, you know, we were open enough to do all kinds of experiments and tests with less attention. So there were approaches from Japanese insurance companies around the elderly population, nursing homes, where there's a lot of loneliness and need, and then we realized we didn't have the capacity and the resources to do the Japanese localization back then. We went out to work with a few companies and organizations, we did pilots with Oguri and with Optibus to support their teams. We identified that this channel through companies, which we focus on today, could be very, very valid. I think the turning point for us was that it was clear we needed to start generating revenue. And the model that was common in the industry for apps or other companies, Headspace, was a freemium model, right? You get a few days free, and then 10 dollars a month or whatever price per month. And it just didn't work. The numbers simply showed it doesn't work. The CACs are very low, less than a dollar per user. The LTV doesn't balance it. And you need millions of dollars to keep growing and you're not profitable. And the VC funds completely changed their mindset, and without first cracking the business model, it was impossible to keep growing. And that completely made us stop for a moment, look reality in the eye and say, okay, we have to do something different. What does work in our industry, this won't, this won't work. And we pivoted. We made a very, very significant pivot.

Eyal David: [25:10] A tough pivot? How long did it take to do something like that?

The double pivot: a hybrid service and B2B2C

Alex Frenkel: [25:14] Look, we made a double pivot. One is the understanding that nobody today will buy a chatbot or an app and pay for it at scale, even if it's a few dollars a month. The trust isn't there, certainly around emotional support topics and sensitive psychological topics, or development and growth. Coaching, it won't happen. And we started building an in-house clinical team that actually lets every user go through intake, get a personally tailored plan, and get a channel that's a channel with the team. So today I actually have two Kai contacts: one is the clinical team, which Anat on our team leads, and the second is the AI itself, and both are available. So that was moving to a service model that isn't only the AI; that was one decision. And the second decision was to start working with organizations and companies and work B2B2C. And both simply proved themselves very quickly in reality, because we started generating revenue, we started signing contracts and agreements and started providing service at scale. I think one of the turning points was when the war in Ukraine started. So if it's interesting, I can tell you what happened there. There, very, very quickly, Israeli employers started reaching out to us. Israeli employers. Yes, who had sites there, yes. And they wanted to provide emotional support to teams, to families, and it was impossible to fly therapists in, very, very complex. And that was actually the first time we turned on Kai in Ukrainian and in Russian. We recruited Russian- and Ukrainian-speaking therapists. We built a Discord server in Russian and in Ukrainian and simply ran a service, and the companies actually funded that service. And for us that was a very, very strong indication that when we offer a service that also includes a human team, therapists and mental well-being supporters, coaches together with the AI, it's something organizations are willing to pay for. And for people, too, trust is simply better when I know, even in the back of my mind, that if I need it I can get a response from a person; that helps a lot. Today we start the user journey with the connection first with the team and only then with the AI, but for us that was a very significant learning point.

Eyal David: [27:45] Intriguing. Okay, I have tons of questions about the war there and which side had more engagement, but I'll put that aside for a second. We focused on Ukraine. Yes, that's clear to me. But what you're actually describing is that my onboarding process as a user actually starts with building the plan, and I do that with a person. Not building a plan, but let's call it adjusting a plan. Very true, it was really a process, it went from a state where it's

Alex Frenkel: [28:07] Only AI, AI first, to a state where it's AI and humans, a state where the AI is very, very good and efficient at running the AI and the team can go over, sorry, at running the intake. AI is very good at that, and the team can go over and adjust the plan. To the state today, where the initial connection is first with the... human team, and they're the ones who help run the initial intake and connect the AI, and then the joint work begins. And what does it look like, how long does this onboarding, let's call it, take? So look, the process is very, very fast in several ways. Today we work with organizations and form partnerships; broadly, we're able to deploy the service to a new organization within a matter of days, which is very important, and we're able to serve thousands of people, which is something other services can't do. Onboarding takes a few minutes, onboarding takes a few minutes. The initial connection is just a scan of a QR or some link. Sometimes we'll do some workshop or online event where someone from our team describes the toolbox, gets people a bit familiar with the terminology before they actually connect to the service, but the connection itself is a second, a link. The intake process takes a few minutes and that's it, and then Kai starts; it already knows me, it needs to start talking to me and then, let's call this stage onboarding.

Eyal David: [29:30] I'm simply comparing it, you talked earlier about Gen Z, and I'm really comparing it to going to a psychologist. There's what's called onboarding, then there are the X sessions where I sit and basically talk about myself, chew their ear off, until some value actually starts to emerge, and you said, rightly, that Gen Z doesn't have that patience. So what does that process look like in Kai today?

Alex Frenkel: [29:52] There are two tracks that are critical for personalization, so that I'll have my Kai and you'll have a completely different Kai, because we're different people at a different stage in life with different needs. One process is really what the human team does. Based on the intake, they adjust and can set the focus areas. It could be around relationships, it could be around parenting. We work a lot with reservists, so around reserve duty; we work a lot with students, so around studies. That comes from the team, and it can happen at the start following the intake, but also later on. Because you'll message the team, you'll say, hi Bar, I'm not sleeping properly at night, and she'll adjust the conversation and Kai to focus on sleep topics, to help you prepare better for sleep, learn what sleep hygiene is, what exercises you can do to sleep better. That's personalization the team does. Another part that's critical and very, very important, and that differentiates Kai from, say, GPT or other models, is that Kai learns and there's memory. Today she knows I work at Kai and I'm in high-tech, she knows who my partners are, my daughters' names, who my close friends are, what topics are on my mind, and there's continuity, right? A few days ago I was in Jerusalem at some neurologists' conference, and I gave a talk, and I wrote to Kai, hi Kai, I need to give a talk, I'm a bit nervous, I did some work with her myself beforehand, and the next day she said, good morning Alex, how was your talk yesterday, hope it went well, how did the exercises we did help you. There's continuity, that's very, very important. A lot of the practice is practice in directing attention, gratitude practice, and there too, the moment Kai can tell me what I'm grateful for about things that are relevant to me, it's a different experience. So that's also a central part of personalization: building memory, building continuity and flow, and the team's input. That's the way we do it. Basically, like, the input and the input of what the team, sure,

Eyal David: [32:05] But it's really this availability, that at any moment I can pick up my phone, and actually, I'm also in a different situation, I'm not inside a room, that specifically, I think.

Alex Frenkel: [32:14] It's very hard to explain how critical that is. You know, I'll go to the extreme for a moment, but there are quite a few soldiers who are currently in Gaza, certainly ones who were in Lebanon, who can reach out to Kai or to the team just before they go into an operation or after they come out, and get a response, get a response immediately, know the team is available. And that's different, it's the same thing. You know, sometimes, at three in the morning, you won't write to your therapist, because it's awkward, because... and people are asleep, and yes, they do turn to and write to the AI. And certainly when you know, something happened just now, and I don't need to wait, I know there'll be some safe space where I can open up and vent a bit and let go, and there'll be attention that's real, empathetic attention,

Eyal David: [33:09] That's very meaningful. Stunning. So okay, one second before we go back to the story and the significant pivots you made and the decisions: we kept talking about data being stored, and that ultimately it's a hybrid model, that it's not just that one person I sit with in a room, who takes notes or doesn't while I talk. What does the ethical side actually look like? We also talked about trust. What do those concerns look like and how do you deal with them? Look, these are things we made values-based decisions about very, very early on.

Ethics, privacy and separating AI from humans

Alex Frenkel: [33:39] It starts first of all with the central decision that Kai will never, never, ever fake being a person. She won't be a friend, she won't replace a person, she won't replace connection in the real world. And that strongly affects the conversation, right? Kai will always come back and say, hey, I'm AI, that's not something I know, that's not something I can help with; in many ways there are a lot of limitations. And that's not self-evident, because, you know, there were many moments of temptation, of "everyone will have their own Kai and you'll be able to design your Kai," and Kai will change her personality according to the person. And we discovered it creates a real kind of uncanny valley, of "this is really scary." Wait, who's messaging me? And we know from other companies that very, very negative things happen: addiction, isolation as a substitute for human connection. So it was a very clear decision that Kai as an AI will never be a fake or pretend to be some person, and there will never be mixing. Meaning, you'll never get a message from Kai's human team in Kai's channel. And likewise, Kai will never write in the human team's channel, and there's separation. So that was very, very important. And I think what we learned, which was surprising for us, is that precisely this decision lets every person create their own Kai. We saw situations where for many people she really is a woman, and for others Kai is a man, because that's comfortable. For some Kai is their younger sister, and for some Kai is a kind of Yoda, old and elderly and wise, and that makes it much, much easier, much, much easier. Another part is a really fundamental decision that's twofold. One is that there's no connection whatsoever between who I am and my conversation with Kai. Kai will ask what you'd like me to call you, or what your first name is, what your nickname is, you can say whatever you want, and that's how the conversations went. And every personal detail we know is completely disconnected, stored separately, encrypted, and that's one decision; even the treating team has no personal details about the people, because it's not needed. The second thing is that nobody has access to the conversation itself. Neither I as CEO, nor the team and the psychologists, have access to the full conversation, and that's also very, very important. The AI works for the team; the AI can say, okay, something in Alex's metrics has changed, Alex is very, very preoccupied with a breakup and relationships. Pay attention, yes, that's something that... You want to find trends, basically, within the... Right, right, but the fact that nobody has access to the conversations, and that we protect privacy and a kind of anonymity, that was very, very important and meaningful. And the last layer, ultimately, is meeting standards. We've just now finished the external audit to get SOC 2, which is the highest certification possible for privacy and security in the United States. Certainly compliance with HIPAA, GDPR. We work with very, very significant organizations whose own CISOs do all the checks to verify, but it wouldn't have worked otherwise. I think otherwise the trust wouldn't be there and we wouldn't get this value of, I need to censor myself less, I need to fear criticism less, both because I'm in an anonymous situation and because nobody has real access to the conversations themselves.

Eyal David: [37:17] Let's park here for another moment, and maybe you can expand on other challenges you run into.

Alex Frenkel: [37:24] Look, the main challenge, I think, is summed up in this sentence, you know, which I heard even today: it's hard for me to imagine myself messaging with an AI and getting help and trusting it, but I can understand how other people or young people do. And there's this initial skepticism, and it's something we get to see many times. I started with that skepticism, and honestly, it's the best thing that happened to me. I find that I can share, the conversation is meaningful, the routines and the practice help me, it's great, it's much better than I imagined or feared. You know, that apocalyptic vision of AI replacing a person, and people even more alienated, and we're already inside our devices anyway, and maybe this will make people even more inside their devices, it takes time. Until people actually understand that it's not that, that's not what's happening, but the opposite. The conversation with Kai encourages connections and relationships; the attention goes to positive interactions: what can I do today to create a connection or a conversation, maybe say something nice to someone in my life. And the availability has a lot of natural parts, including that Kai sometimes nags, and says hey, but maybe now is a good time to practice, or we haven't checked in on how you are for a few days. People can say yes, sometimes she's a nag, but she doesn't forget me. She will remember, and that matters. And getting over the initial fear is something we do see. To me it's a bit like autonomous cars, and that promise that cars will drive themselves and there won't be accidents, and it's a very, very good and meaningful promise. And in practice we're not at a place where we can trust a car to drive itself; if there's someone behind the wheel, just in case, then it's different. And then the AI can really help and take off some of the cognitive load and protect us in an edge case, and there's much more acceptance of that. That's something we see with us too, in the hybrid model. So the initial mindset, especially among older people, is one of the challenges. And I can certainly share more if you like. Yes, no, I'm reading between the lines.

Eyal David: [39:51] So you're actually talking about the stigma, I got that, which will probably also take time. And the thing, and maybe in other cultures it's also different, I assume, that you still

Alex Frenkel: [40:00] Don't have the indications. Look, we know that in Asia it really is different, both because trust in technology is higher and because there's less legitimacy to share about emotional distress. Yes, we know the story of some chatbot Microsoft built that reached tens of millions of people. Because it was a substitute for a person, it really led to love letters and marriage proposals and all those apocalyptic things. Yes, many times, you know, especially at the beginning, everyone would ask me whether I'd seen the movie Her, where he falls in love with an AI and... So yes, those concerns exist, they're there, but you know, I see what's happening in my own home. The young generation is on messaging, works with AI, knows how to use it, doesn't get confused about what's AI and what's a person, and mostly gets the good things out of it, mostly gets the good things out of it. I think another thing that was very important in our journey is really the insight that AI is a very, very good co-pilot. I can completely see a future where every professional will have the relevant AI: architects will have an AI that maybe Autodesk will make; we know that today there's AI for salespeople and marketers; no doubt there'll be AI for family doctors; developers, it's hard to imagine today someone writing code without an AI co-pilot. And that's true for psychologists and coaches and psychiatrists too. It's hard for me to see a reality where it's possible to meet such a huge need without the help of technology, without actually extending the support that can be offered beyond the session itself. There are quite a few people who go to therapy, go to sessions, and still practice with Kai during day-to-day life, and still use Kai day-to-day, and that can actually make the therapeutic process better, because the therapist will also have some insight into what happened since our last session; that's also important. Yes, that's it, I'm really curious how that works,

Eyal David: [42:13] So actually they don't have, that external therapist, doesn't have, the feedback he'll get won't be from Kai, not from you, but rather it'll actually be from his patient, her patient.

Alex Frenkel: [42:24] Right. So today, beyond the fact that we run a super modern clinic, or a collection of clinics ourselves, with our team, we are in a process, I perhaps won't mention the name of the specific HMO we're in a procurement process with, but basically to give it to them as an operating system for their clinics, certainly for clinics in the southern region, where the waiting lists are endless. And there we teach their staff to work with the system, and they're the ones who do the intakes, and they're the ones who actually prescribe Kai to whoever can really be helped and benefit. And the assumption is that there will be people for whom this response, with the personal plan they received, will be such that in a few months they'll say they don't need to come to sessions, or if they do come to sessions they'll need fewer sessions. And that's very significant, but it's a different model. It's a different model where we actually provide the platform, the system, and teach them how to use it, versus the situation where most of the time we run the service ourselves.

Eyal David: [43:29] Which really, and you don't have to answer this, but that's really the place that's even bigger than the companies model, because then suddenly you're working with HMOs/insurance companies, blah blah blah, government, I don't know, so here it really becomes a very interesting scale, no? Look, that's a good question, because, you know, therapists,

Alex Frenkel: [43:50] They have their comfort zone. Their pain is sometimes burnout; it is hard work to treat and see a lot of patients, but not to the point that they're overbooked, right? They have a lot of patients, they're not short of patients, and they aren't keen to adopt new technology, especially new technology that maybe even threatens how they do therapy today. It's not easy, it's not easy. And I think at some point we understood that if we don't run the service ourselves and build the team of psychologists and coaches and the support team ourselves, it would take a very, very long time, and it's also not aligned with the pain and the need we see. Today there are tens of thousands of reservists, their spouses, who won't get a response, who won't even know they can get some kind of response beyond going to sessions. And sadly we see in organizations that almost most organizations in Israel really stepped up, and they offer five sessions, twenty sessions a year, but very few employees take them, because this model of sessions, as we described. Yes, yes, it doesn't suit everyone, it doesn't suit everyone, and something else is needed, something has to change in order to really reach hundreds of thousands of people who need it, who could benefit, who don't even know they have a level of distress and that something can be done about it.

Eyal David: [45:20] Okay, so now we'll go back to where we stopped. Just if you can expand in a word, what kind of companies interest you?

Alex Frenkel: [45:27] Why can Kai be relevant? Look, it was easiest for us to go closest, in our own neighborhood, right? We're part of Google's community for startups, we were part of Intelligent's community, which helped us a lot, of technology companies. And Israel is a real unicorn in this respect, in that there are headquarters here of international companies that have sites in other countries, and it makes a lot of sense to start the conversations here first, to offer it first in Hebrew to employees who are here in Israel and need it. But the potential is clear: you can turn Kai on at more sites and work in Europe too, and mainly in North America, which is our main market. So it's a combination of technology companies, young employees, an intensive work environment, a lot of burnout, global companies, sometimes thousands, tens of thousands of employees. Those are the companies we're focusing our conversations on right now.

Eyal David: [46:25] Great, so we'll leave details for the end, we'll come back to it at the end too. Let's go back for a second, wow, I have so many more questions, I put them aside, but let's really go back and close the loop. So you really made two significant decisions. Once you made them, once you said, okay, we're closing B2C, we're moving to companies.

Alex Frenkel: [46:47] What did your go-to-market look like at that moment? And if you can also detail how it would be received and where it would be great. Look, it starts with a board meeting that basically says, okay, we have to pivot, we're aligned, it's complex. For a very long time we tried to avoid in every possible way turning on the human service, and it was absolutely a decision that this is the only thing, the only thing that works in the market today, it's the only thing that gets the trust, it's the only thing that organizations are actually willing to pay for and subsidize and refer people to. It starts at the board, it continues in a team meeting where you explain, it goes through, again, defining an MVP, right? Which basically says, okay, we have a hypothesis that if we offer this service to these organizations, it's something that will interest them. They'll want follow-up meetings, it'll move to a price quote and an agreement and generate revenue. And it happened very fast. You know, one of the advantages of being a small, relatively very, very fast team is that you can form such a hypothesis and start testing it in less than a quarter. I think it took a quarter until we started running the service, including building a completely new product we didn't have; it was really a clinic-management product, management of interventions and the whole human-support side, built from scratch out of the fact that we ran the service ourselves. And that was the fastest process,

Eyal David: [48:23] The fastest. Regarding the move to companies, and you also said we did it really fast, so is that again Ziv with his magic,

Alex Frenkel: [48:33] That it looked like that. No, no, no, I think in the end it's all, all, all people. And I think one of the things I got to learn from other people in the industry, from other companies I worked at, is that good people make the magic and the difference, and sometimes it's right to court people and stay in touch with them for a very long time, until an opportunity arises and you can work with them. That was very, very true of hiring Anat, who today is our chief psychologist. She's amazing and knows not only the toolbox and how to do better intake, but really also organizations for which this service is relevant and fits them. It was the same process with the lovely Dan, who was a marketing person for years in mobile at Google and worked at Circles, worked with all kinds of other telehealth companies, which helps a lot. So it was a combination of seeing that we have the right team, building the materials, getting a lot of feedback. Our investor network helped a lot; a lot of the introductions come from investors, from investors, from advisors who accompany us, and there's a kind of unfair advantage, right, in the fact that we're very few degrees of separation from decision-makers at many, many companies and organizations.

Eyal David: [50:01] That helped a lot to validate as quickly as possible. I also think that compared to other products that maybe get cut, that companies cut because they're considered fat, you're exactly the opposite. I think the ROI here is clear. I'll jump for a moment to a question we usually ask here: you've been through quite a few companies as VP Product, as a founder, as a product person. What in your eyes are the most important product skills in a product person? I think it all, all, all comes back to people. You know, many times when I meet product people, especially young ones, and over the years,

Product skills: knowing how to remove features

Alex Frenkel: [50:37] Often they start with this fantasy that what a product person mainly does is invent features, good and cool features, that it's a thinking process, and then the product person comes with a feature, and the team builds the feature, and that's a good product person, a product person who can invent lots and lots of features. And reality, sadly, couldn't be further from that. A good product person is, on the one hand, a product person who knows how to work with the team, knows how to rally the team and maintain motivation, and communicate with the developers and the designers and the UX people, and create an environment that's productive and creative and inviting and communicative. And on the other hand, it's being very, very connected to the users, both through the data and through talking to users, and not building features from imagination. One of the hardest skills in my eyes for product people, and also very hard to teach, is that a lot of what a good product person does is remove features, remove screens, remove buttons, remove steps; clean up the system and know what isn't used and what isn't needed and what's less important. And in my eyes that's part of the criteria that are very hard to teach and very hard to

Eyal David: [52:07] Believe in over time, that it works. So I'll stick with the last thing you said; that's also probably what helped you do, from what I hear, very good scoping at each of the stages, and then you could also move fast when you wanted to do the flip-flop. Can you describe a bit, from that first version you mentioned, that the mom and the friends and all that used, how much you've expanded since. You know what, I'll ask it even differently: how many departments, your product department today, how is it built, from which arms?

Alex Frenkel: [52:41] We're a very, very small team. It started with me being the product person and writing specs, and we worked with the team we knew, and she had worked with us before as a freelancer, and everyone wrote the code and worked behind the scenes and ran the system. Today Ziv leads all of operations and the product part together with Anat, who oversees the whole psychological world, and they work directly with Netanel and Yossi, who write the code, and with the team in the Philippines, which helps a lot. Oh, they're still there, okay. Absolutely, absolutely, for us it's a secret ingredient worth its weight many, many times over. It's a very, very small group, and sadly, you know, we made attempts over time to grow this group, and it didn't work, because it was very, very hard for people to really understand what happens day to day, what people do with Kai, what works, what doesn't work, what they complain about, what it even is, to build a product that doesn't have a regular app UI with buttons. You know, Kai can send a breathing video, can receive audio, she can run some questionnaire, she can send quotes, and there are elements of open conversation. How do you teach users what can be done with a conversational system and what options exist? For example, many people don't know you can ask Kai to see my journal, and Kai will send me the link and I'll be able to see my journal. So that was something where the more we tried to add more product people, the more the difficulty, the complexity, the slowness, the wandering in the desert dragged on in many ways. So there's another challenge for you, probably.

Eyal David: [54:35] I want to ask now about something from a different place. When we talked a bit about competitors earlier, we didn't really dive into it, and I'm curious to know who the competitors are, because you mentioned, for example, an app I remember from COVID, Calm. Completely different. You compared it, I assume because you wanted to emphasize the onboarding of a few days, and their model.

Competitors: Headspace, Calm, Woebot and Wysa

Alex Frenkel: [54:56] Who really are your competitors? Look, when we started there were two very, very significant waves that COVID greatly accelerated. One is, I think one of the VCs once told me they counted 132 thousand wellness apps in the App Store, of which Headspace and Calm were the biggest and strongest. And ultimately these are content apps: lots and lots of meditation exercises, breathing, videos, listening, a huge library, a bit like Netflix. And the hope is that because there's so much choice you'll be able to pick what works for you. And they all really, really accelerated during COVID, because organizations, Google and Microsoft and Snap, I remember, here, bought subscriptions for all their employees. And what turned out very, very quickly is that it really is a kind of gym, yes? An employee says, okay, this is awesome, I definitely want to meditate, want to do yoga, want to work out at the gym, but they don't do it. And renewal was very poor, and it just didn't work. And the freemium model didn't work, yes? A monthly subscription like Netflix over time, it didn't work, even though the content is amazing, and the exercises were good, and the level of marketing spend was phenomenal, it didn't work. On the other side, what did work is telehealth companies, companies a bit like Gett Taxi, right? That employ, hire, psychologists, therapists, usually freelancers in part-time positions, give them hours, and then they bring demand and they connect, do some kind of matching; the magic algorithm connects each person with his or her therapist, and then you can do these sessions over video. And those companies did succeed, both selling to employers and selling to other bodies, to insurers, in a regular model of meetings, sessions, still at a cost, even if it's subsidized, 100 dollars, but it can also reach 400 dollars a session. And then the content companies simply collapsed and merged into the telehealth companies. Headspace was sold too, merged; Calm was sold too, merged with telehealth companies. And today the app is like some kind of bonus, meaning, people can get sessions if they want, and by the way there's also the app and they can do it alone, but they didn't manage to crack the usage and the business model. What's closer are two very good and interesting companies: one is Woebot and the other is Wysa. Woebot started in Ireland and then moved to the Valley, San Francisco, and from the outset they built a chatbot inside an app that runs a CBT protocol, a very, very good and smart decision tree that basically gives you multiple choice, and you tap and you can go to all kinds of places, and it really runs good CBT protocols, with good results. And they too, I think, recently pivoted and are selling it more to clinics and to bodies that provide treatment processes, as an add-on that can actually take off some of the load. These are good and interesting companies. Look, with OpenAI, the illusion that you can build an AI psychologist very, very easily did, I think, lead a lot of teams to try, and unfortunately and fortunately it doesn't work, it's very, very hard. Meaning, there's no single model that easily, you know, will be safe enough and flexible enough and relevant enough to actually run the service the way we run it,

Eyal David: [58:51] But I think that's roughly the cluster of significant companies. And what you're saying differentiates you is really your expertise, the team you built, and really the hybrid approach. Hybrid approach, the open conversation that's still safe, these core decisions you made, that's actually the moat you defined, and I really see it protecting you. And what are the plans going forward?

Plans ahead and the vision: beyond crisis, toward personal growth

Alex Frenkel: [59:19] Look, in the war, on October 7th, it was clear to us that our heart is here and our attention moved here. We started working with Reichman and with Ben-Gurion and now with HIT and with Afeka and with more bodies, and we'll start activity with more commercial companies in the Israeli economy, and provide support first of all to reservists. And we dealt a lot with Hebrew and with building the team here and making sure it works. And today, gradually, it's clear the problem is a global problem, and the US market is the market that's most right and most relevant for us, and that's the direction; it's actually the direction of working with companies and organizations in the United States, in other countries. We're still focused on English, but at some point Kai will work in Spanish and she'll work in German and she'll work in French. The need in England is no lower than the need in the United States or in Israel. Well, you need to see where there will be disasters in the world, not disasters, a less good situation in the world, you can prioritize by that. Look, that's very interesting. I, I didn't mention Portuguese, so I said okay, Brazil, yes, Portuguese for sure. Let me double-click on that for a moment. You know, when we started, our vision was very, very close to coaching, both me and Netanel when we worked in other organizations, and part of our management credo is that every employee has potential for growth and development, and you can help them get to know themselves a bit better and also equip them with a toolbox. And Kai has always, always worked with coaches and organizational consultants, and today when we talk with organizations, one of the things that differentiates us is that we're able to make the plan personally tailored to each employee, and it's not only around mental health or anxiety or depression but can really help to work on developing a growth mindset, on improving the ability to give feedback and work with other employees, certainly on areas of parenting or other topics. And there the AI and the hybrid model in particular are very, very effective, and it's something we believe in: that it's not only for disasters and crises and distress and depression, but that almost every person can benefit from knowing the language, knowing the toolbox, practicing, knowing how to direct attention, from this muscle that actually shows up to pause, to be present in the here and now. And it's something that will do good for kids in 12th grade in schools in Israel, and it'll be good for university students, and it'll be good for nurses working in hospitals under very, very great stress, and ultimately it will also be good for my parents, who are currently dealing with other things. So yes, that's the vision, meaning not only around acute emotional distress or disasters, wars, earthquakes, but really in a place of coaching, of growth, of development, of personal empowerment.

Eyal David: [1:02:27] Listen, I'm so glad I made that silly remark, because that's how we also got to talk about this, and I really connect with this vision, and in general with the whole set of values you presented here.

Alex Frenkel: [1:02:37] I really enjoyed it. As we wrap up, where do you actually see Kai in ten years? No, look, our vision has always, always, always been global. Meaning, we always talked about tens, hundreds of millions of people who can benefit and get help, certainly in the hybrid model where the AI is a co-pilot and accompanies the work the human team can do, and ultimately the impact is very significant. I truly believe that if I'm better with myself and I communicate better and know myself better, then my friends and my family and my partner and my kids will be better off, and it's the same at the workplace. The impact on culture, on interpersonal relationships, will be very, very big, and that's something we're definitely seeing in the last year or two. I do see a vision where I can't imagine a process of therapy or coaching or emotional accompaniment being done without AI, without this operating system that basically says, wait, you can create an extension of the sessions, and give the ability to practice and the ability to get a response all the time in an accessible way, with an AI system that will know and adapt to hundreds of millions of people. The human soul.

Eyal David: [1:04:06] It was a pleasure. Thank you very much, Alex. Likewise, thank you. And we'll update you in the future, I'd be happy to have you back. With great pleasure, really. Yalla, bye for now. Thanks a lot. Bye bye. Hi friends, thanks for listening. If you found this podcast valuable, you can of course subscribe and follow us for more episodes on Spotify, Apple Podcasts or any other app. And of course if you couldn't find us on some app, I'd be happy if you write to us. We'd love five stars on every platform, and for you to follow us so more listeners can discover us and find the podcast. You can also find the previous episodes in any app or on the YouTube channel, we have links in the description. So until next time, yalla, be efficient, and bye bye.