Episode Transcript
[00:00:06] Speaker A: Good morning, everyone.
[00:00:07] Speaker B: Today I'm publishing a conversation I had with Joel Hellermark, the CEO of Sana. Sana is a next generation learning and AI knowledge assistant platform. They happen to be the company we're working with on Galileo. But this isn't really about Galileo. It's really about the learning technology market. And I think you're going to find this fascinating.
[00:00:29] Speaker A: I'm going to be doing other podcasts.
[00:00:32] Speaker B: With vendors over the next few months, particularly on the AI technologies, because they're so new. And let me just say, in the learning industry, companies like Docebo, Uplimit, Arist, Sana are really going to revolutionize this space and many others I'm sure that I haven't talked to yet because it is so well suited for AI, almost as interestingly unique as in the recruiting market. So take a listen to this, and if you are a vendor and you really want to talk to us about what you're working on, please let us know. Send a message.
[00:01:03] Speaker A: And I'm going to be writing a.
[00:01:04] Speaker B: Big article about this, which will be entitled the autonomous learning platforms that are coming, which I wrote about in the predictions. And you'll hear from Joel the vision of something that I think I started.
[00:01:17] Speaker A: Working on in 1998.
[00:01:18] Speaker B: To be honest, that has been very hard to achieve in organizations, but I.
[00:01:23] Speaker A: Think it's actually possible now.
[00:01:24] Speaker B: So take a listen, and at the.
[00:01:26] Speaker A: End, I'll do a little wrap up.
Okay. I'm here with Joel Hellermark, the founder of Sauna Labs.
[00:01:34] Speaker C: Sauna. We just go by Sana now.
[00:01:36] Speaker A: Great. And as most of you know, Joel and I are working together on Galileo, and sauna is our partner on Galileo. But we're not going to talk too much about Galileo. We're going to talk about Sauna and Joel. So, Joel, maybe to get started, tell everybody about your background and what made you decide to start sauna, of course.
[00:01:58] Speaker C: So I was very fortunate to grow up during a time where with an Internet connection, you could effectively learn anything. So I came across this Stanford courses from Andrew Eng. And at the age 13, I was able to just take them on my computer at home. And I really got obsessed with programming. And I spent quite a few years working then as a software engineer. And at that beginning, I was really interested in, from the beginning, machine learning, but equally human learning. And I got obsessed with what is this process that enables humans to acquire new knowledge. And I founded my first AI company when I was 16. And then when I was thinking about what to do next, I started thinking.
[00:02:50] Speaker A: About, hang on a minute. How long ago was that?
[00:02:54] Speaker C: That was about eleven years now.
[00:02:57] Speaker A: So you've been doing this for a long time, since you were a high school kid, basically.
[00:03:02] Speaker C: Exactly.
[00:03:03] Speaker A: Okay. All right.
[00:03:04] Speaker C: So I thought when I was looking to what I would do next, I was really inspired by this historical swedish entrepreneurs like Ingva Camprad, who founded IkEA, for example, and they really spent an entire career, 60 plus years, pursuing a problem. And I thought if I wanted to do my life's work now, I better select a problem that I care about very deeply.
And this intersection of human learning and machine learning had long been a big obsession of mine. So that's what I set out to pursue back in, back in 2016, and that led to Sana, and since we've shipped several products at that intersection of machine learning and human learning.
[00:03:51] Speaker A: So it's funny, I mean, we kind of grew up in a parallel universe. I'm a little bit older than you, but I stumbled into online learning in 1998.
And so I would agree with you that human learning using technology is a never ending challenge, really. It's quite interesting how challenging and different it keeps becoming. So when you started this, did you know that there was such a thing as a learning management system and all this infrastructure and all this history of this stuff, how did you come to know the space?
[00:04:26] Speaker C: It's an incredibly good question, because I didn't. But then I came across this great mind called Josh Burson, who taught me all about it. So I heard about learning management systems all of the time when I was out talking to customers, and I didn't quite want to frame it that way because I preferred the sort of first principles way of reasoning of, okay, what's the problem? We're trying to have folks learn something here and then architect a solution around that problem, rather than sort of get caught into the existing categories of LMSs and LXPs and authoring tools and so on. But I remember the first time I sort of saw this table you created about what sort of fit into each category. And that was actually the first time I understood what an LMS was.
[00:05:22] Speaker A: Jewel, this is going to sound really funny to you, but in 1998, when I fell into online learning, it was before the Internet was really very big. We built a product at a company that I was at called ERista Knowledge Systems, that was basically a knowledge management system for learning. And we couldn't get anybody to buy it because nobody understood what the heck we were talking about. So in many ways, you're right where the market has been trying to go for a long long time. So given that you've now learned that you call Sana an LMS, it's really much, much more than that, right? I mean, my perspective, you guys have heard me talk about this. I think learning management systems are management systems. They're not learning systems, and that was why the m was always in the abbreviation. But now they're really learning systems. So tell everybody on the podcast what you're doing. That's different from what most people would think of as an LMS, of course.
[00:06:17] Speaker C: And we want to make sure we also deliver really well on sort of the classical LMS tasks. But we also don't want to get stuck in that category, because to your point, ultimately it is about learning. And I always thought it was such a messy ecosystem. You would bring in an LMS, then an LXP, and then an authoring tool, and you would stick this together, and then you might add enterprise search and a knowledge management tool on top of that and so on. So what we aim to do is solve this in an end to end way, meaning build the best tools for you to create the knowledge. With a multiplayer editor with interactive cards that allows you to create really delightful content. An editor that is AI assisted so you don't have to create all of the content yourself. You have this AI copilot with you, you can just upload a PDF, have it automatically generate an entire interactive course. You can tell it to generate a question or a poll or an interactive card for you. A tool that solves the analytics really well. You can build custom dashboards to understand how your learners are progressing, how the content is performing. A tool that made the learning process highly personalized in the sense that it adapted the content to each individual learner, to their knowledge level, their interests, and so on. And finally, that also enabled you to just put this entire process on autopilot so you had all of the automations you needed. And when we looked at that, we also took some inspiration from marketing automation platforms in terms of how you could set up triggers and so on to deliver the right learning to the right person at the right time. So although we play in the LMS category and we make sure we deliver on everything you need there, we also want to go far beyond that because we think that's such a legacy category that every time I hear as described as an LMS, it kind of hurts a bit.
That's ultimately the abbreviation that folks are looking for when buying.
[00:08:28] Speaker A: Yeah, well, I think you have to kind of stake out that category because it's such a big space and people know what it is about this idea of autonomous learning. So as I was writing the predictions report, I was reading about autonomous driving and the five levels of autonomous driving. And I don't know if you saw this chart I put together of the five levels of autonomous learning. I mean, I think there's an analogy. There is that where you sort of see this going, where some content is generated automatically, some content is available directly on demand from the source. What about the human live part of this? How do you see this all coming together?
[00:09:07] Speaker C: I think, yeah, there are very different types of this process where one is highly collaborative. You want to sort of be discussing and learning from folks in a room, and there's others where it's way more efficient. If you have content dynamically generated, if you have this knowledge assistant effectively that knows all of the world's knowledge, all of your company's knowledge, knows exactly what you know and can pull together the exact content in the exact format you prefer it in real time. That's incredibly powerful. And that's literally just made possible the last year or so. This is the first time where we can truly deliver on that promise. And I think it's incredibly exciting because to the point of what we built with Galileo there you basically get this tutor. You could ask anything, and it's grounded in your research. It's citing every source. So I know where it's coming from, when I want to dig deeper, but it's highly personalized and highly contextual to me. So I think that's the role where the knowledge assistants can play a big role, personalizing, pulling in the content. And then you want to combine that with the more experiential learning where you discuss it in groups and go deeper on that. And what we aim to do with the platform is to both deliver on those highly collaborative parts, but also on the more sort of self study where persuasion really assists.
[00:10:34] Speaker A: I mean, I have to know. When we started with Galileo, now that we've had it out for a while, I realized this is sort of the way people want to learn. They want to ask a question, and then they want to tell the system, give me more information, give me a course, give me a video, as opposed to, here's a big library of videos, click the one you like or think you're going to like, and then 15 minutes into it, you're like, that was a waste of time. I clicked the wrong one.
[00:10:58] Speaker C: Your analogy of the system sort of the autopilot steps, what we're seeing now is also this. Assistants can be more proactive so they can anticipate the questions we'll want to ask before we ask them and then nudge us in those directions as well. And I think that will be incredibly powerful.
[00:11:21] Speaker A: Let me ask you about that particular topic. So many of the older LMS vendors tried to do that. If I think back about Sabah, not Sana, and many of the other vendors in sort of that ilk, what they would do is they would say, you're an employee, this is your job title. This is your level. This is your job description. This is your competency model. Here's your course. Oh, you're level one. Here's your course. You're level two. Here's your course. That was about as personalized as they got. How do you see Sauna capturing enough information about people to actually recommend what the system thinks they need?
[00:11:56] Speaker C: I think the first point is, what is the atomic unit? Historically, the atomic units have been so big that no matter how good your recommender systems were, it didn't feel very personalized. It was literally giving you a course on leadership because you seemed to be a leader, which is not very useful. The atomic unit now is getting so granular, it can even be down to a word level, right, where it basically assembles the sentences and from there on up, based on exactly what you're interested in. Okay, you're looking to learn machine learning.
Your role is in biomedicine. This is your knowledge level. And then it can tailor the exact content that it puts.
[00:12:40] Speaker A: If I take a course in sauna on some topic, it may not be the same course as someone else who wants to take a course in the same topic because it would be dynamically generated for me. Is that the idea? Exactly.
[00:12:53] Speaker C: And I think increasingly so.
We're now at the beginning of this phase, but even choosing the unit. So we hear a lot of folks that we just introduce automatically generating the text to spiritual on the content, and we hear a lot of folks that like to listen to the content. We also have other folks that love to chat and have the more socratic method where they keep asking questions and so on. So I think even the sort of medium in which it's presented will start getting a lot more personalized. And in learning science, it's long been sort of debunked the idea that people have different learning styles and there's no scientific evidence for that. However, I think people have very different sort of preferences in content consumption. And I think that's a very real thing.
[00:13:48] Speaker A: I agree. In fact, I think sometimes the content consumption is where you are at the time you're consuming it. I'm in the car. I'm at the office. I'm in a meeting. Right. So let me ask you kind of a hairy question. I don't know if hairy translates into swedish. So the biggest infrastructure thing going on in L. D. Right now is the skills taxonomies. And, I mean, I just had a whole morning of conversations with companies about, we talked to Google, we talked to Nvidia this morning and some other companies about how they're building their skills taxonomies. Is that not the micro data that you want or signal or where do you see skills relative to the other signals that you want to use to develop content?
[00:14:26] Speaker C: There's the higher level sort of skills. You build a skills taxonomy. You assign the skills taxonomy to all of the content you level folks at which knowledge level they're at, according to these skills, and then you provide very sort of tailor content accordingly. And that's inevitable, and it makes a lot of sense. But where I think it gets real interesting is of this very granular personalization, where people get to develop very specific skills in their day to day together with this knowledge assistance.
And that's where I think we can see a true shift.
[00:15:14] Speaker A: Since you and I are working together on all this content that we have. So if we think about this big corpus of content that some company has, and then some project comes along and the HR department decides we need a new skill in some biomedical science, if I'm a pharma company, or some compliance thing, if an automobile company, how would the Sana lms use that skills data to dynamically or algorithmically build a solution that would be relevant to this? Because the skills data that's coming from the rest of the HR system is pretty coarse. It's a word or phrase. Exactly.
[00:15:55] Speaker C: And I think that's where it's sort of higher level directional, where you say, we want folks to develop AI ethics skills as we're starting to interact a lot with these tools. And then you would use that, but then you would need to start tailoring it, because if I'm a software engineer, it means something very different from when I work in marketing or design.
And then you get another level of tailoring, and then the subsequent level is, okay, I'm in marketing, but what's my knowledge level and what prerequisites do I have? But then you have this sort of very granular level where you're in marketing, you're facing a specific issue. So you're running a new campaign, you're using this new AI tool, and you're unsure how you could be using it and what data you can use and so on, and you're faced with this very specific challenge, which you need a set of skills for when you effectively instruct the system that this is the situation, and then it pulls together all of the resources and it literally puts you at the second in the video.
[00:17:11] Speaker A: So it's really exactly like what we're doing with Galileo, in a way, is the user or the system is asking a very specific question about a scenario and a situation, and then the system is assembling content.
[00:17:23] Speaker C: And I think you need to deliver on all steps of that journey, right? Yeah, I think that's one way. Then you also want to make sure they just get assigned the courses they need to develop that specific AI ethics skill. You want those courses to be personalized and so on. But one aspect of the skills building that I think has been lacking is that sort of real time tutoring. That's really exciting. So of course you need your skills taxonomy, you need your content to be tagged to your skills taxonomy and so on. But what I find really exciting is as you develop those skills, how can you assemble the content dynamically based on the exact sort of skill to solve the task at hand?
[00:18:10] Speaker A: I completely get it because I've also seen your system and we've worked on it so much together. So given that idea, which is obviously a pretty powerful idea, how hard is it for anybody else to do this? I mean, if I'm a good software engineer and I pick up Chat GPT and understand the APIs, I don't even know how you guys built your company and where you got your engineers from. But is this going to become a common thing across all of the learning platforms? Or do you think there's something unique and competitive here that you guys have already figured out? You don't have to tell me what it is, but give me a sense of, because the reason I ask is, I know most of the LMS vendors, and frankly, most of them are not working on this. Most of them are working on small things, skills taxonomies, a content generation tool that generates sort of static content from documents. How much of a lead do you think you guys have?
[00:19:04] Speaker C: I think the final point you make there is a really interesting one, because if you're not translating it into something that's really engaging, it doesn't matter how good that AI technology is. So you have to ask yourself, what is the atomic unit of this content? If we're just pushing out this wall of text documents with multiple choice questions, it still won't be a very exciting learning experience. So one area we invest a lot into is making sure that the units that you can create in the platform are super interactive and enjoyable to consume and multimodal and so on.
[00:19:47] Speaker A: And visually interesting. I noticed, too, you guys are very visual in a lot of your design.
[00:19:51] Speaker C: Exactly. We put a lot of effort into that. What is it that you can create using this? And then the second area is the one around how does this sort of AI assemble this? And I think in order to truly get it right, it's a big ranking problem, and it's a big search problem, and that is, in fact, very difficult to solve. You want to pick the right assets created by the right author. You want to rank those effectively. And there's a lot of work that goes into building a good search engine. And that's, in fact, what is powering these generative technologies today. You combine a search engine with a generative technology so that the LLM is fed the appropriate context to do its generation.
[00:20:47] Speaker A: So there's some serious heavy duty software engineering under the covers here. This is not a simple problem.
[00:20:54] Speaker C: Exactly. And then solving it in an m to M will as well. Right. I think when you try Sana for the first time, you're surprised by how many steps of the problem that Sana is solved from the create experience. We have a virtual classroom in the platform that enables you to host really interactive live sessions. We have the analytics, the management, the search and assistance functionality, and we sort of bring this all under one roof. And I think once that is fully integrated, it creates a really delightful experience for, as the learner, you don't have to go to five different tools. And particularly as an admin and a creator, you can really build these incredibly delightful end to end journeys.
[00:21:40] Speaker A: Well, I'd certainly agree with that. I've seen it. Two more questions, sort of, about the market, and then maybe a little bit more on. We'll wrap up. So there are two sort of use cases that seem to me to be challenging with this AI oriented approach. One is compliance training, which I did a lot of at Deloitte. I actually had to do it. Take it where in regulated industries, in regulated safety applications, there is a annual or periodic compliance process. You really do want to test people in some way. Sometimes it's a practicum, sometimes it's a knowledge test, it's a simulation, and then that data has to be used. Have you guys thought about all that and how you're going to handle those kinds of use cases, how are you going to assess somebody's actual capability in something? Of course, I think it's easy.
[00:22:29] Speaker C: When you go on a podcast, it's easy to talk about the future and so on. But these are very real challenges.
[00:22:36] Speaker A: I'm not trying to simplify the problem. I just want to get your perspective.
[00:22:40] Speaker C: But compliance is not always the first topic that comes up in podcasts, but it is a very real challenge.
[00:22:47] Speaker A: It's actually a pretty big area that the LMS market is kind of locked into.
[00:22:52] Speaker C: Exactly, but it is one where we have customers with, we have one customer with 300,000 employees where they need to make sure everyone is compliant every single year. And as part of that process, they first want to put it on autopilot. Right. They want to set up the trigger, set up the entire flows so they don't need to chase it and so on. So that's where the automation capabilities come in really handy. And then secondly, there's sort of twofold when it comes to the personalization of that as well. So for certain content, you have to go through it in a particular order and so on to get assigned. But for other content, you might not only have to practice the areas where you have knowledge gaps. So what we start with there is this placement test, which is basically a dynamic assessment that will assess what you already know, and then it will only serve you the exact knowledge gaps that you have. So for folks, like a lot of our customers in finance and so on, that have a lot of compliance, their learners are delighted.
[00:24:01] Speaker A: Okay, so you could have a compliance course, and every year there's new content added to it. So somebody could test out of the old stuff and only really have to be trained and certified on the new stuff. Exactly.
[00:24:15] Speaker C: Which sits them tens of hours in compliance training each year. And also, it's a very delightful experience. I think. Finally, on that note as well, what we like to think about is we're bringing the scandinavian design ethos to this space. So we're a product which is more of a consumer grade product. So I think that's quite rare also in the compliance space that people are.
[00:24:41] Speaker A: Delighted, that's for sure.
All right. Okay, I have one or two more questions, then we'll wrap it up. So one more on learning, and then I want to talk about the company for a minute. So a big part of the learning platform space that's been very difficult for learning management systems is the human aspect of collaboration. What is the role of a teacher? What is the role of a mentor who collaborates with who? I mean, most lms have sort of discussion rooms and things like that. But what is your strategy and architecture about the human relationship between people and the teacher in your strategy for Sana?
[00:25:22] Speaker C: It was an interesting insight that we had that a large chunk of the learning happened in the live contexts, and the tools that were used for the live context were quite static and more broadcast esque. And that's why we built out Sana live, which is this really interactive virtual classroom, if you like. And what that enables you to do is set up sessions with a whole lot of interactive cards, polls, reflection cards and so on that can engage the groups. So as a facilitator, you can just click next and you have the next interactive card. Click next and you have the next one. So it's really easy for you to facilitate that. And as a learner, you move from, you know, being broadcasted too, to really engaging. And, and it's not just a single person speaking at a time, but everyone can jot down their ideas in these cards. And I think that was a very big shift where you could combine these journeys of highly interactive live sessions and really personalized self study that made a very good combination of the very human elements and the others where you just needed sort of time and have it in your own way.
[00:26:37] Speaker A: I agree. And then, of course, all that content is available for the AI to look at later to be reused for different applications as people want to get the content that came out of those sessions. Right? Yeah. Okay, so another question for you, and then we'll kind of wrap up a little bit.
I don't want to ask you how old you are, but you're a relatively new CEO. How do you build a great company? And maybe the swedish culture is unique and part of this, but how do you stay ahead of the market and thrive as a software entrepreneur in this kind of crazy world of online learning, which is pretty crowded and pretty messy, of course.
[00:27:17] Speaker C: In fact, I founded the first company when I was 16 and then I was the CEO. So I have a food list, sort of eleven years of CEO experience by now. But I think the first thing is I've just surrounded myself with a lot of senior leaders that have done this journey before for the US market. We brought in someone exceptional from Slack, that has seen this, scaled this before. So I think the reason how I've been able to sort of survive in this world is by that team that we've built.
I think for the last sort of six years or so, I've basically been a recruiter and I think that's how I've survived. It's just building that team of exceptional folks, and I think we have this very unique purpose of changing how people learn with AI that is really exciting. We're also at a sort of breakout phase now. We've raised 80 million. We're expanding internationally with offices in London and New York, et cetera. So it's a really fun time to join. So I just try to bring in the best folks to compliment me there.
[00:28:23] Speaker A: Well, I'll tell you of my experience working with you guys. I think you're doing an amazing job. I mean, it is a competitive market, but I think you're definitely on the right track, and you're going to have a lot of success. I also just want to personally thank you for your partnership with us. We have really had just a wonderful time working with you guys, and I think we're going to do some great things together.
[00:28:42] Speaker C: It's been a true blast. And coming back to that idea that we had around Galileo, maybe adding some backstory to that. So I think this was back in 2018 or so, I trained an assistant based on everything that Steve Jobs had ever said. So every interview, every book, everything effectively that he had put out, and then I could use that and chat with it all day long. So I would give my strategy to it, and Steve would give comments, and he would say, for example, he would comment on the people strategy where I was questioning during COVID whether we should go remote. He would say, never go remote. Joel, 95% of people problems are solved by putting folks in the same room. And he would give me nudges, like, to where we started the conversation. I had learned so much from you, and now I'm fortunate to get to meet with you and discuss problems with you, but back then, I wasn't. And so when we were starting this idea, we were thinking about how could we basically create this sort of Josh copilot where anyone could chat with Josh and upload their people strategy, get feedback on it, and ask any question, and get responses instantly. So it's been such a fantastic project in democratizing the access to all of the knowledge that you built out over the last 25 years.
[00:30:04] Speaker A: Well, thank you for saying that. I'll tell you, for us publishing all this content, until we had an application or a platform like Galileo, we never knew how we would ever get it in the hands of people. So it makes it so approachable. It's so ridiculously different. All right, Joel, great. Hey, thank you so much for taking time to talk about sauna, and I know we'll be talking a lot more. And folks out there in the market, you should check it out. Obviously, it's a very big market. There's lots of applications. But I think Sana is definitely an important company in this entire industry. Thank you.
[00:30:35] Speaker C: Thank you very much.
[00:30:37] Speaker B: Okay, so as you can see, there's a lot going on here. Some really important, exciting new ideas and technologies in the learning and development space. Stay tuned. We'll do much, much more on this. And as I said, if you are a vendor or you have a really interesting story as an l and D professional you'd like to share, I think it's time for us to talk about it. Please contact us. We are preparing for irresistible 2024. It's going to be a star studded event May 20 through 22nd. It's actually taking place during my birthday. So if you can find the time to come out to California, join us on the USC campus for a spectacular experience, and we'll talk much more about what's coming at irresistible in the next couple of weeks. Thanks, everyone. Bye for now.
