Welcome back, everyone. Really pleased to have with us today Nathan Schultz from Chegg. He's been at the company 16 years and is currently the Chief Operating Officer. Nathan, welcome to Jefferies EdTech Day. We've had a number of speakers today covering all aspects of education via technology. For those that aren't as familiar with the Chegg story, maybe if you could just kick off and just talk through the evolution and kind of where you're headed over the next five years.
All right. Thank you. Really appreciate you having me. Love the opportunity to always talk about Chegg. Let me step back and obviously explain who we are and kind of why we exist or what we are and why we exist. So Chegg is a direct-to-student educational support platform. We're a subscription-based model focused on students in the global higher education market. Why we exist is students more so than ever before around the world need self-directed learning support. Whether they're in a traditional tracked program or they're doing it on their own, students require or are looking for and going online for more support than ever before. So Chegg has been and will always be designed to meet the students where they are, providing them personalized learning instruction to help them gain confidence in their studies.
We continue to see the results of our work with 89% of all students surveyed saying that Chegg helps them learn their coursework and 88% of students saying that they better understand the concepts that they're studying and they're simply more confident in their work. Over the last 16 years, it's been a wonderful journey of transitioning from a very physical-based model renting textbooks to now something that is 100% digital and driven to meet students where they're at. It's been an exciting ride. We're going to continue to do that for students around the world.
Nathan, one of the big focuses today has been AI. I know last year it was a little bit of a headwind. We went and tried all the products. We subscribed to your service and then tried other services. Clearly, the other services are more bark than bite. You came back with accurate results. It was more intuitive, easier to use, all the great things. So I think everyone is trying to understand is just getting that awareness out that not all AI is created equal, but AI done right, especially in your platform with all your content can be very powerful. So can you just bring us up to speed in terms of where you're going with this, how you're thinking about the next chapter?
Absolutely. It's important to take a step back a little bit and think about Chegg as a content creator over the last 10-12 years and what we've built there because that serves as a foundation for where we're moving with AI and really advanced machine learning because not everything is going to be solved by AI. There's lots of machine learning involved. You don't need to apply very large models. You can actually apply smaller models. But we've been a content creator for the last 10-12 years. And in doing so, we've built a very strong backend that has been able to create over 100 million pieces of learning content by our proprietary 150,000 subject matter experts, which we have taken an exorbitant amount of time vetting and training. And that's the human layer that's so important to understand in AI.
When obviously the world shifted very quickly, we were set up to be thinking about how we were going to move in this way and took the time, actually, to say, "Do we want to simply be building something off of somebody else, or do we want to kind of do what we've always done, which is figure out how can we make content creation, machine learning, and now AI a proprietary part of our technology and product stack?" That's exactly what we've done. We really wanted to, first and foremost, unlike the world of generic AI, I'll call it, or generic language models, put some primary goals out there. Our goals were quality and accuracy because in education, that's what matters. You can't answer a problem wrong, right? You can't answer a problem with the wrong instructional design.
Students are there to trust you. Chegg has built a brand around trust. That trust has become from our content as the foundation of what students are using and why they trust us. We had to step back and say, "Great. We've got the content to do real hard training of models and make them accurate. We've got the subject matter experts. How do we want to go tackle this in a way that is organic to our product experience first, but then very much allows us to expand our product experience into a further personalized instructional model for each student?" When we stepped back and we looked at that, we said, "Great. Our primary product is around solving students' doubts, so questions- and- answers." Students start on the platform by asking a question.
We wanted to move our platform from one where humans were the content creators in answering those doubts to really model tuners, if you will, using human refinement. What we've set out to do and what we've now done is we have completed the production of 26 language models that cover all of the subjects that Chegg covers. Every subject has its own language model. In fact, it has two language models. One is designed to answer the question. Think of that as a single-turn model. Brent, you asked me a question. I provide you with an answer that is built around a pedagogical structure, which understands what the question is, why you may be asking that question, and what is the best way to answer that question. Because not every question is created equal. You have conceptual questions. You have factual questions. You have procedural questions.
They all need to be. You need to apply a different rubric against those questions. So we wanted to do that. And then we wanted to advance the platform and really think about, "Okay, how do we unlock a personalized learning journey for every student?" And that requires conversational elements. And so we have a multi-turn model where a student can say, "Great. Here's the question." We single-turn the answer back to you in our step-by-step proprietary format. And in that, a student says, "I'm stuck on step three. Great. Let's have a conversation about step three. Let's break down step three for you. Let's break down step four for you.
Hey, let's take that content and maybe turn it into another learning element, like a flashcard or a practice problem." Now we're in a world where our single-turn language models are fully out there, are getting out there by the end of Q1. We're now unlocking the conversational element. Where we see this going is just Chegg continuing to build AI into our current product experience and then using it as an expansion of our product experience to provide these personalized learning journeys for students.
Many ask kind of what inning do you think we're in as it relates to AI?
I don't think this is a question that's specific to Chegg. I think we're all in early innings as you think about where this could potentially go, especially in education where not every question is text-based. Many questions have computational elements. Chegg has actually been a leader in building out computational engines, if you will, calculators, if you think about them that way. We've got to do more of that in AI. You've got a fun and exciting challenge with multimodal problems where a problem actually has a diagram that is complex. You're required to understand elements of the diagram to answer the question. That's something that we're actually cracking right now. So I think we're in an early inning.
We're looking at it. We're also looking at how do we use AI in advanced machine learning, not just for the product experience, but so we can do things faster, cheaper, and more aligned with where technology is going. We're in early innings. We're super excited, but we're very, very excited where we're going with it.
That's great. You recently launched automated answers. Can you talk about the user experience and how the responses you're getting from your users?
Absolutely. We'll step back and make sure the room understands what the experience was prior to automated answering and why we call it automated answering. Prior, when you came to Chegg, you would ask a question, a doubt, if you will. Many of the times, those questions had already been answered. So you would pull that directly from our archive of 100+ million learning objects, learning solutions. But for new ones, they would go out to our subject matter experts. And that would take, I know, between 2-4 hours. We guaranteed it within 4 hours in order to come back as a customized learning solution just for that question. That was the bedrock of the Q&A experience. Either you pulled from the archive, or you got it from custom-created for you.
But we wanted to take that same experience, but now make it instantaneous. And so when we thought about our quality standards, we said to ourselves, "Well, our language models have to be as good as, if not better than human answering," right? You can't go below that because then we're chipping away at the brand and the trust that we've created. So now when a question comes in, if that question is brand new, actually, what's great is the experience between the archive and instantaneous feel like one. Before, if you got it from the archive, you got it automatically. If you had to wait for it, it felt like ordering DoorDash. And you were watching, is your Dasher picking up your question? Is your Dasher answering your question? Is your Dasher delivering your question? Now it's either it's coming from the archive. It's coming instantaneously.
It actually is a unified experience. That was actually a big program that we worked on in 2023 to get ready for the launch of these language models. And so now it comes automatically. And now you start to feel like you're having a conversation with the system versus what before was much more close. So the user experience is more unified. The user experience is absolutely instantaneous. And now we can start to think about, "Okay, how do we interweave?" We think about the next best action for the user. Is it to ask a question about that question that they just asked? Is it to transform that question into another learning object? Or is it to progress them to a new question?
And you kind of add all these things up. I guess when you think about the awareness build in your base in terms of how, again, like I said, when you test drive this product, it's so different than any of the other products. But I think students seem to be confused or they're curious because they go and try other tools. And then they're like, "Well, maybe I can use this free tool for a while." What does that conversion, how do you get them off of that and get back on? Is there some type of campaign? Is it a university tour? How do you break that? Because it's like, I think we're dealing with something now that students may not even really know to the true extent of what you have behind the scenes.
Yeah. Chegg's always kind of faced this free alternative, whether it was in our old Q&A product - questions- and- answers, apologies - questions- and- answers products. I mean, there was always Quora. There was always Yahoo Answers. There was always companies like Brainly. We really worked hard and continue to work hard in making quality and accuracy the gauge in which we measure all of our content and now language model kind of standards too. So we thankfully have a brand awareness that is extremely high. We are known for being accurate. When it matters, you're going to go to the service. Education certainly matters.
And when you're studying for that exam, you're not going to give yourself over to a platform allow yourself to use a platform where you've got real question marks around whether it's teaching you the right stuff or not. And so we're going to continue to kind of beat that drum of how we involve one of the competitive moats we have is really around those 150,000 subject matter experts. Keeping a human in the loop in AI is actually really, really important. You can't. The models don't know themselves when they are drifting or hallucinating. That's not what they can do. Even with synthetic data production, you still don't know whether synthetic data isn't designed to be honest and accurate. We are designed to be that level of accuracy.
And so we are explaining to our students actively right now what Chegg's AI position is, where AI is being used, how humans are remaining in that loop to continuously train and modify. And we don't just drop a language model and it answers, take chemistry and answers the beginning concept to the most advanced concept. There's a continuum in there where the language models can answer and then where language models can no longer answer because either the capabilities don't exist, the computational models don't exist. And we actually have built some really fascinating and exciting advanced machine learning models that are designed to analyze the question before it goes to the language model for answering, to ask some basic questions. Again, why would the student be asking this? What's the context of the question? What are the elements necessary to answer this question?
Is it all just text-based and therefore it's really easy for a language model to do? And what's the best methodology to do it? Or is there a computational element here that we have to think about and use technology to go in and out of our language model to a computational solver and back again to create a solution? So we are really educating students that what Chegg is building is a truly verticalized AI experience for education, not just AI for AI's sake, not just language model for language model's sake. We're going to continue to see, I believe, the separation as we want these as we want this technology to become really aligned to the jobs to be done, and in our case, for the students studying, understanding concepts, mastering concepts, putting faith in something that's going to allow them to progress their knowledge.
That's going to require true verticalization. I think that's where Chegg has been extremely successful in the past, and we're going to continue to do so. We're seeing great results. I mean, as the launch of our models, I know we talked about it in our Q4 call. In January alone, we saw a 3x number of questions asked by students than the past. And that was just one month with some of our not even all of our models, just some of our models out there. And so in January alone, it was almost 2.2 million new questions asked by students. It's 3x the number of last year. It's unbelievable. So when you explain why this AI exists, how it was created, you take away you build that faith in it. You build that trust in it. And that's so important to our story.
There's obviously a lot of incorrect outputs from these AI engines. Many ask how you're ensuring the quality of your output.
Yeah. And so it goes back to kind of stuff I've talked about. So first off, we spent a long time as we were building the actual language model itself; we were building this machine learning capability to study the. I think of it as the anatomy of a question. And that, again, goes to what is that question? Why would a student be asking it? Best way to teach that question? And what are the what are the what are the what's the parts of the question that need to be answered and what technology can be applied to answer it? Because there's instances where we're actually you're fully automating the answer. There's instances where you're partially automating the answer because actually, this piece has to go to a human right now.
So the way we're thinking about it is, obviously, there's constant refinement, human-reinforced learning on the models in order to get them better. That's the 150,000 subject matter experts we use. That's a wonderful competitive advantage of Chegg. The more questions we throw at the models, the more feedback we get. And that's happening outside of production. So their student experience isn't interrupted. We're constantly pushing the models and tweaking the parameters and the weights of the models in order to get them better and better and better. And then we're constantly studying what questions are capable, what questions aren't capable of being answered. So again, our standards are these models have to be as good as, if not better, than what humans could be able to answer. And we're starting to see that really.
I mean, we are 100% seeing that, which is why we have some of our models in production today. And we're seeing such a great reception to this by the number of questions students are asking.
I mean, the question we always get from investors is, "It sounds like all these initiatives are going the right direction. When does all this amount to getting the company back to growth and turning that corner?
That's a question. So I think we're going to continue to, I think, as we've said, we're going to continue to roll out our language models throughout this quarter. And then we got our big kind of next advance, which we've talked about coming up by the end of this half, is really that conversational element. And I think that's when you start to see those personalized pathways being created for students in a very organic method. Even students, there's nothing more organic than simply having a conversation, frankly. And that's when that value prop, I think, really starts to take root in students. And that's when I think Chegg can start to establish itself as taking Q&A to a new level, if you will, questions- and- answers to a new level and turning ourselves into that, which we've shown investors.
We've created some really compelling product videos around. Our goal is to be that personalized learning assistant. I'm very confident that that's going to continue to progress throughout the remainder of this year. That's going to continue to provide growth for us in 2025 and beyond.
I mean, Chegg's historically gone after the college student. But you've talked about expanding the offering so you can address a broader student base. Can you discuss kind of what that looks like and where you think this goes over the next three years?
Absolutely. I think about it as kind of going up and going down at the same time. Obviously, we've got a wonderful skills initiative that we've been working on for a number of years. Dan's talked about how AI has really improved that initiative's speed to market. So the ability to create courses takes much, much shorter time than it ever has in the past. It's cheaper to create courses around skills. So we're faster to adapt to what skills employers are looking for to make sure their employees are well trained in.
So I can continue to see that kind of push that market up and allow us to be a more competitive offering in the skills market as well as kind of well, it was really exciting to create the bridge between college and the first-time career, which is something I think is truly missing in the market, particularly around the area of durable skills, which you're starting to hear a lot about. Everyone can learn Excel. Everyone can learn SQL. Everyone can learn the more technical side of things. Not everyone can become a corporate citizen as quickly as I think corporations want. And I think that's a viable opportunity. And AI is really helping us push that boundary because we're able to create more content. We're able to create more personalized learning around it. And so that's going to continue to happen.
Alternatively, going down into that high school market, not necessarily with paid products to start, but certainly with products, as you think about because we own our models, we actually have unlocked something that's unique and very defensible for Chegg, which is we're not running like many companies who are using AI that is powered by, say, a GPT or someone else. We're not running a model that is based on an API call. So every time you call it, the cash register rings. We've built our own models. And we're really just relying on GPU power, which, frankly, is playing into every anyone who's relying on that is playing into our advantage these days with the cost of that coming hosting cost coming down and down. And it's only going to continue to as more GPU power comes online.
So when you think about that and then what we can build for students, especially in the high school market, you can start to really increase our high school awareness, our high school attach rates, our high school membership, if you will, to Chegg, that brand affinity. And so we think about really activating that student and getting them ready. So when they exit high school, whether they're in the United States or Canada or around the world, that next step they take is a very logical step of, "Great. I'm going to move my membership from free to paid," right, because now things are getting really serious. And then finally, in the college market itself, the higher education market itself globally, I think about how AI is allowing us to move faster nationally, which we're obviously in a lot of countries.
But we're taking a domestic product right now, doing some localization work on top of it, but not truly making it a native product. But now we can, right? We can move to AI-based translation. And we can just go faster in building really native products for that market. So we tighten up the product-market fit over there. And we're not just taking the students who are learning in Turkey and English, but we're taking that market overall. So we see a number of different vectors through which we can kind of expand our TAM, strengthen our service addressable market as well.
I mean, the skills business you guys have been talking about for a while. I mean, you've made an acquisition there. I mean, where does that business stand? And where do you think, what are you most excited there?
I'm most excited about the connection between our college business and the skills market. I mean, I think that is an unlock that Chegg owns, certainly, the keys to, given our brand and the affinity for our brand with the college market. These students are leaving college not with a. They don't have a trusted Sherpa. They kind of go into that next phase of their life. And Chegg has that phenomenal opportunity to be that. And that, again, goes back to the durable skills programs that we continue to look into. On the skill side itself, it's a great B2B business that continues to grow and continue to have great relationships with companies to continue to help them empower their employees.
At one point, there was a talk about the connection between college and workers. Jefferies, we hire a lot of interns and a lot of graduates. When you think about that bridge of me being able to look into, "Here's the Chegg students that are finance, accounting, marketing," and then being able to surface those relationships and give them the bridge then to these different employers, there was discussion about that at some point. Does that initiative still stand? Or how does that work?
Yeah. I mean, one thing at Chegg is sitting on top of it is a data stream on our students about where their strengths are, where their weaknesses are, where their likes are, where their dislikes are. And that is something that's an opportunity I think we can begin to unlock as more students actually get into wanting to engage with skills beyond May of their graduation year, right?
So as we pull students in in their sophomore year and the students are starting to organically get to that point where they're realizing that their colleges are not teaching them those durable skills that are going to make them successful in an internship or make them successful in their first job, I think as that becomes more and more the norm and we are set up to have those programs, I think we then unlock that next phase, which is, "Okay, how can we now better align you with the likely employer that wants your set of skills?" So I think that's the first step, certainly, let's get students into these durable skill programs. And then we can think about how do we align students with the right careers, the right jobs, the right companies.
That's certainly the pathway we're going to go.
Nathan, I mean, as you sit inside the company for 16 years, you get to see and feel and bring in a lot of other things that we can't see on the outside. So for Wall Street, what is the most underappreciated thing that we can't see you get to see?
The talent at Chegg, the mission at Chegg is something that is not to be kind of be missed. I think we own. I believe this for many years. We really do own the success of our future. It's not going to come at the hands of a new technology necessarily. It's going to come at the execution of the roadmap that's in front of us. And the employee base is something that is, frankly, unique in the Valley if you look at the tenure of our employees. You look at kind of our track record of growing, I think, intelligently versus just increasing our employee base. But we really grow intelligently and think about what roles we need and just exactly how many we need of them.
That's an underappreciated asset of Chegg that, in Chegg, we think a lot about and make sure. I've never been one to think about the fire and hire paradigm. I think of how do we actually train our employees to not just for the job they're in today, but for the job that we need them to do tomorrow. And we continue to do that and then continue to invest in them. The result is we've got a tenured employees that's kind of unparalleled in the Valley, which is phenomenal. And that is an underappreciated asset, I think, externally because getting into you guys don't get into those details. So we own the success that we're going to build into in 2025 for sure. We've got the material, the raw materials to make it happen. So I'm super excited to watch that all unfold over the next 12 months.
Yeah. Well, Nathan, really appreciate you being here. Thanks so much for sharing your story. Looking forward to staying in touch. Thanks, everyone, for joining.
All right. Thank you, Brent. Have a great day, everyone.
You too.