All right. Good morning. Thank you everyone for joining for the Coursera presentation. I'm Stephen Sheldon. I'm the EdTech analyst at William Blair, covering Coursera. I'm required to inform you the complete list of research disclosures and potential conflicts of interest is available at our website at williamblair.com. So it's great to have the Coursera team here at our conference once again, including President and CEO, Jeff Maggioncalda. We also have Cam Carey in the audience, who runs IR, along with CFO Ken Hahn. You know, as most probably know, Coursera is one of the larger EdTech companies and sits right between learners, companies, and universities, which, in our view, provides a lot of strategic value.
It's been a choppier road so far this year, but the company remains well-positioned to capitalize and support the significant upskilling and reskilling that we think, will be needed both near term and longer term. So with that quick, quick intro, I'll turn it over to Jeff.
Great, thanks a lot. Good morning. I guess I'll just do it from here. I don't have a lot of walking space. I'm Jeff Maggioncalda, the CEO of Coursera. I always appreciate when people come and hear about our story. These are pretty exciting times. We're doing a ton of stuff, and I think the opportunities are amazing. The way I often talk about this is that generative AI is like a tidal wave, and we're and others in EdTech are sort of on surfboards. And the question is: where's the wave gonna break, and when's it gonna break? And I'm not a great surfer, but the basic concept is you've got to time it, paddle out there. You don't wanna be too early, you don't wanna be too late, it'll crash over you.
But if you can catch the wave, and if you can ride it with something that's valuable and distinctive, there's a huge opportunity out there. I have been traveling quite a bit. We serve learners all around the world, and we serve institutions all around the world. It is striking what is happening with businesses, governments, and campuses from Kazakhstan to Colombia, from Seoul to India. And in the last 18 months, you know, I've been to each of these countries. We have institutional and individual clients all around the world, and everybody is fascinated by the future of learning and work, and nobody has the answers. But at a really broad level, what we see is a world where generative AI is going to disrupt jobs, it'll disrupt industries, companies, and especially institutions that do education.
The way that people learn will change, the way that people teach will change, the way that institutions will need to transform themselves, including the institutions that comprise the $2 trillion higher education market globally. There has been... You know, I was a bit wrong. I thought that generative AI was gonna sweep through institutions much more quickly than it has, but you'll see in the charts, some of the reasons why it's been a little bit slow. I do not believe it's because this is not gonna be a really big deal. I think it's because it's a lot harder than people thought.
And then finally, when we think about what's gonna be required, a combination of not just content, but content delivered in a learning experience that's powered by technology, that develops real competencies and skills that will be in demand as the need for skills shift. These skills need to be assessed and credentialized so that employers can figure out who's really got the skills to do the jobs that are needed to survive and thrive as a company. Ultimately, what the learner cares about is their career. What the company cares about is having a talented, effective, and low-cost workforce, wherever it is in the world, that can help them stay competitive and deliver value to customers. If you look at the kinds of occupations that are most likely to be disrupted, I'll just quickly show.
These are the percentage of job tasks, according to McKinsey, that could be automated using AI. The darker bars for each job category are pre-generative AI, and the lighter blue bars are the ones that are the percentage of jobs post-generative AI. You can see at the top there is educator and workforce training. This is an industry, and by the way, just higher education, where you know, people paying tuition is about $2 trillion. If you look at the corporate training market, it's probably $300 billion-$400 billion. Before generative AI, about 15% of the tasks of those types of occupations, educators, and workforce trainers, could be automated using technology. Today, the estimate by McKinsey, as of GPT-4, is that 54% of the tasks can be automated.
So it might not be the most at-risk occupation, but marginally speaking, the impact of generative AI will probably be more significant to educators than anybody else, any other occupation in the world. Like, 3x more tasks because of generative AI are likely gonna be automated. And then the number of other job roles that are gonna be done differently because this technology are gonna be vastly impacted by even the current generation of generative AI. There was a study that was done that was striking to me. It was done by Princeton, UPenn, and NYU, and they literally looked at over 800 job classifications. They looked at all the tasks, and they said, "Which jobs have most exposure to generative AI?" Literally, 11 out of the 13 top jobs most impacted are college professors.
So there's a really big global industry called higher education, with institutions that have not really been built for agility and people who have not really had to keep pace with how fast things are changing, and this tidal wave is gonna come, and it's gonna wash over this entire industry and everybody in it. If you look on the business side and you say, "Well, what kinds of jobs are potentially gonna be upgraded, if you will, with more productivity from generative AI?" Again, according to McKinsey, they expect over $4 trillion of productivity gain. When you look at the types of jobs that are gonna be most impacted and productivity stands the most to gain. $3 trillion of the $4.4 trillion comes from customer operations, software engineering, product R&D, marketing, and sales. Five basic functions.
Even though companies have not yet completely revolutionized the way they write code, and I've been talking to a lot of companies about this, they know it's going to change. Maybe it's not going to be 75% of all job tasks to software engineers immediately, and maybe it's not going to reduce the need for software engineers, but it will definitely change the way people write code. And it is absolutely changing the way people answer the phones and support customers, and it's totally changing the way marketing gets done, it's totally changing the way sales gets done. A fairly few number of functions account for a huge amount, multi-trillion dollars of expected productivity gains.
Here's the issue: You're not going to get the productivity gains unless people understand and can use the technology to perform their jobs in different ways. A recent survey. It's not often that you see 100% of anything. 100% of the people in the survey by Accenture said that Gen AI is going to transform their businesses. These are leaders in businesses. When asked about their personal understanding of the technology and whether they feel like they understand it well enough to lead change in their organization, only one in three said, "Yeah, I think, I think I got my handle on this technology well enough to lead my organization through it." When you look at employees, 90% say they're starting to use gen Gen AI, but only 5% have received any training so far. It's like, what's the disconnect here?
It is things are changing so quickly in terms of the technology capabilities that the leaders are really having trouble trying to figure out exactly what we do to make this more of an opportunity than a threat. And the workforce is saying, "Well, we're starting to use this already, but we have not yet seen a lot of systematic skilling programs." And this is where Coursera sort of steps into the picture. We were started in 2012 by a couple of Stanford professors in data science, Andrew Ng and Daphne Koller. We make top content from universities and industry partners available directly to consumers of which we have about 148 million registered, and then we sell this to businesses, governments, and campuses.
We sell our platform, which includes content and credentials and the learning experience, to institutions, including businesses, governments, and campuses. We bring to these institutions and individuals the top brands in the world in terms of university and in terms of industry. We believe that even if content is dramatically commoditized, people will want to know that what they're learning is credible, is worth their time, and will lead to a credential that they can show to somebody, "Look, I really have these skills." We believe, and I think it's pretty true, the credentials, the learning credentials that come from these institutions are the most valuable learning credentials in the world. So it's not just about content, it's about skill development and credentials that show that you have those skills.
The faster the rate of change in skill requirements, the faster the rate of learning that needs to be delivered to individuals. What this is really about is having a distribution system, which we do, that brings the learners from around the world, directly through consumers, employees at work, employees for governments and the citizens as well, and students in campuses. Bring them into branded, high-quality content and credentials on Coursera that can lead to either a college degree with what we call Pathway Degrees, where the open content counts as credit towards a college degree, and career pathways, the ability to advance your career, not just because of the content, and not just because of the learning experience, and not just because of the skills, but because you can earn a credential that you can show to your boss, "I know these things.
I can get this done." Whether that's someone that you're trying to get a job with or your current employer. And the faster things go, the harder it is to figure out who has skills, which is why credentials are so important. And an easy way to think about this is look at the cloud companies and the number of credentials that have been created, starting with, like, Microsoft and Cisco, way back in the day with networking, .NET, and look at the cloud companies. We're going to see the same thing with generative AI. There's going to be generative AI credentials all over the place to try to figure out who, in which parts of the world, knows how to use this technology in which ways. And those credentials are going to come from partners, like, like ours.
In terms of generative AI and how we're trying to take advantage of this and turn it more into an opportunity than a threat, we have basically six major things that we're doing. The fundamental predicate is that this technology is going to change the way business operates, and that's going to impact the kinds of jobs that need to get done and the skills to do those jobs. So there's going to be a wave of upskilling and reskilling that is going to be formidable. And in upskilling, a lot of people are going to have to learn new skills to do their current job, just to maintain their job. And by the way, the risk to somebody's job, especially in North America, is not just that some machine automates every single task.
It's that some machine or some combination of machines automates enough of the task that you don't need as many of those folks, or creates enough capability with learners someplace else in the world who cost less, that that job doesn't need to be done where it's being done today. I believe that generative AI is going to drive a lot of globalization of talent, where emerging labor markets will be far more able to deliver the kinds of productivity and capability to employers than without generative AI. The combination of online learning, which allows anyone to learn skills, remote work, which allows almost anyone to do a digital job from anywhere, and generative AI, which allows people to be far more effective, especially entry-level, lower-paid jobs, will be able to be done far more effectively because of these tools.
I think this is going to create a lot of demand for talent in emerging labor markets. So on that note, Generative AI Academy is a whole set of content. I'll show you a slide on this that basically trains people in their jobs how to do their job differently using generative AI. We just launched this in Q1, and we're continuing to launch content in Generative AI Academy, but this is basically for the upskilling. On the reskilling side, for people who are displaced, I think there's, like, 1.3 million call center people in the Philippines. They're not gonna need 1.3 million people in doing call centers in the Philippines. They're gonna have to learn how to do something different.
We have something called the Career Academy, which is professional certificates that allow people to skill into a whole new job. These are certificates from Google, Microsoft, IBM, Meta, Salesforce, Intuit, many different others, that are basically job training programs for someone who wants to get into a career or switch a career. So on the content side is Generative AI Academy, to teach new generative AI skills, Career Academy to help someone switch careers. On the teaching and learning side, generative AI is already fundamentally changing the way people teach and the way people learn. We are in market with a huge array of capabilities, and use them on Coursera. They're really effective. Coach is personalized tutor. It's like kinda copilot for Coursera. It's a tutor, and is grounded in the content of the courses on Coursera.
So we use something called RAG, which is Retrieval-Augmented Generation. The quality of the answer that comes back from Coach is based on the quality of the content of the course. We don't just go to some LLM and say, "Oh, what do you, what do you think is the answer to this student's question?" The student asks a question, we check for relevance to the course, we then find all the elements of the course that are relevant to answering that question, send that to the LLM along with the request. The answer that comes back is, they call it Context Grounding, is grounded in the content in the course. You put the exact same Coach on someone else's content, it will give you different answers.
And the key thing when people talk about hallucination, bias, accuracy, et cetera, is how well have you grounded the model in the content? People talk about the data advantage. We believe our data, and much of the data, certainly we have learning behavior data, but a lot of the data is the actual content in the courses that have come from the smartest experts in the world. So when you use Coach on Coursera, you're getting answers that effectively are pulling from the expertise, the authority, the credibility of the people that authored the course. Coach today does tutoring. If you have ever used the GPT-4o, which is an amazing experience, where you can talk. It's a lot better than Siri, and probably will turn out to be Siri quite soon, is my guess.
The whole pedagogy of personalized interactive learning is gonna be vastly different than it's been in the past. We are doing things, and I'll show you what we're doing with Coach, that allow you to do summaries, quizzes, you can do practice, it does grading, it does career guidance. It's striking what Coach can do. Course Builder is more on the teaching side than the learning side. This allows people to create courses extremely quickly. What we did a few years ago was we took the 7,000 courses on Coursera. These courses are usually 10 hours, about 10 hours. They're built on modules, which are three hours, lessons, which are 45 minutes, and clips, video clips, individual videos that are 5-10 minutes.
In order to make our content more bite-size, we offered the ability for our clients to take clips, and that leads you into the course, so we have about 200,000 clips. So the first stage before generative AI was unbundling courses so that people could take a little piece and then go on to do the whole course if they want to. With generative AI, we can now rebundle courses, but the courses don't have to come from the same provider. So now, in any institution, in any businesses, we just launched this a few weeks ago, in any business, in any government, in any campus, an administrator or professor can literally, in natural language, they can say, "I would like a course that teach these things. These are my learning objectives," and they can just freestyle it.
I wanna teach discounted cash flow to new, business analysts, and I wanna make sure that we could talk about discount rates, and I wanna talk about how to use a spreadsheet, and I wanna talk about errors in models, and I wanna talk about how it shows up in financial statements." I mean, literally, you just say, "I want it to be this long. I want it to come from this plurality of content. I wanna take these courses from Rice, these courses from Michigan, a couple over here from, from Intuit, and now build me a course by recombining components from all these other courses. Oh, and by the way, make sure you put in these elements that are unique to my institution." So now you have a custom course. You could literally build these things in about 15 minutes.
It is not generated from scratch. There's a big misunderstanding. Go out there and try to generate content, try to generate images, and really try to use these images. Try to draw an architectural plan using DALL-E. It looks kinda cool until you look at it, you're like, "This actually doesn't make any sense." The generative AI today cannot really effectively generate content from scratch.
It's probably going to get to a very good capability at some point, but I believe that humans using these tools and putting their human eye on it, and when the course is built, putting their institutional brand on that thing that says, "Even though I built this quite quickly, I built it, I built it using a plurality of experts, and I'm putting my brand on this thing to say this is a course worth taking." I believe that those types of courses, coupled with a credential that will say to the learner when they finish, "I learned this stuff from this institution," will be far more valuable than just content being delivered by AI. So, Course Builder allows for the custom creation of private courses using a mixture of experts on Coursera and experts internal to a company....
And it will include the ability to also generate content from scratch as a component of a given course. Machine translation, we translated 4,500 courses into 22 languages. Every course that gets created now on Coursera gets translated into 22 languages. Every private course, if you're a multinational and you wanna do your private course on name the topic, you build that course on Course Builder, and you put in 45 minutes of training that your HR team did, and that training now is translated into every language that you want. It's mixed in with experts on Coursera. Coach works on both the private and the public catalog clips, and your employees around the world can use this course immediately, from a centralized provider of the content. And then the final thing that's coming quite shortly is academic integrity.
So it's one thing to have content, it's another thing to have a learning experience that delivers it in a personalized way so that people can develop skills. I believe with the globalization of talent, people will be hunting for folks who have the talent that they're looking for. When I say people, I mean businesses will be looking to see who has the talent that I need at low cost. The ability to assess whether someone has certain skills is gonna profoundly change because of generative AI. The ability to generate assessments, whether those are knowledge assessments with, you know, do you understand this concept? Or whether they're practical assessments that says: Can you use a spreadsheet and build the discounted cash flow model or some other application of that knowledge?
The ability to assess whether someone really has that knowledge and those skills and abilities is gonna change very rapidly. And so we are introducing a whole suite of capabilities that not only generate assessments that's grounded in the content, but also can grade those assessments, and also can verify that people aren't cheating using ChatGPT. You might feel like, "Whoa, whoa, that's a big statement." We can't completely verify it, but I'll give you a preview of something we're about to ship in a couple of weeks. It's called AI-driven Viva Exams. These are like oral exams. If you've ever done... Anyone who has a PhD, you submit your dissertation, and then you have to defend it. You have to defend your work by explaining what was your thought process in that submission. We now have Coach. We're gonna shortly launch Coach.
When you do a submission, Coach is gonna interview you about your submission and ask: How did you come up with this? How did you come up with this thesis? When you said this point, what's your citation? How do you know that source was credible? So it's an interview process where the student who does the submission has to actually defend the thought process behind it. And if you just have ChatGPT do it, it's gonna be pretty tricky to explain how you came up with that. So there could be whole new ways of assessing skills, generating assessments, and ultimately, the assessment and the integrity that goes behind it is gonna be required to issue the credential. Because if you give someone a credential because they have the skill, but they just cheated, it's not really clear who has the skills.
So understanding people's real skills and abilities is gonna be changing because of Generative AI. It's easier to cheat, and so that's gonna make the assessment piece a lot more important. Generative AI Academy has three basic pillars: Generative AI for Everyone, Generative AI for Executives, we've launched both of these, and Generative AI for Teams, which we're gonna be pushing. I'll get to this in just a moment. A lot more content coming in 2024 to help functions, especially the functions that have the biggest productivity gain potential, to help those functions learn these new skills. There's also a lot of broad topics in terms of what we're teaching with respect to Generative AI. We're still somewhat early days on this.
I thought by now, Generative AI Academy would be rolling out far more quickly, but companies are really trying to figure out: what is the technology, what does it mean for my business model, and what do I need to train people on? I can say, though, that in 2023, someone enrolled in Generative AI content every minute, and so far, right now, it's four enrollments every minute, and the more content that we get out there, and the more important the Generative AI becomes to the world, we're seeing demand increase, increase, increase. By the way, Generative AI not only creates a need for new skills and content to teach those skills, so there'll be new titles to do that, it also creates the ability to upgrade content that's already on platform.
So if you have a professional certificate in UX design, and it was created before the world of generative AI, not just people who want to become a UX designer, but every UX designer is gonna need to figure out: how do I do my job differently with generative AI? They might go to YouTube and just watch a bunch of things, but increasingly, and especially as jobs become more globally competitive, the ability to take content that gives you the skills and a credential to say, "I actually know how to use generative AI to do UX design, to do project management, to do IT support, to do front-end software development, et cetera," we think is gonna be a big opportunity. I mentioned Career Academy. These are the certificates that are now available.
Each one is a digital job role, does not require a college degree, created by, you know, the leading institutions in the world. We launched a number of new certificates very recently, and there's a lot more to come, which I'll show you momentarily. We're also upgrading these certificates using Generative AI to reflect what is needed in that job in terms of knowledge, skills, and abilities with Generative AI to do the job efficiently. So, industry microcredentials, these are the professional certificates on Coursera, the ones I just showed you, have a lot of appeal to students. So 90% of students said: "I would like to take these as part of my college degree to increase the chance that I get a job when I retire...
when I graduate." And 88% of employers said they're more likely to hire someone who's got a professional certificate and a college degree, rather than someone who just has a college degree. So universities are gonna have to change, not just because the world is generally changing, but employers are looking for skills and abilities from the graduates that the universities just do not have the capability to teach right now. There's not a lot of data science professors out there. There's not a lot of computer science professors out there compared to the amount of demand. There's almost zero generative AI electives in any of the 26,000 degree-granting institutions in the world who are teaching 220 million people.... Who's gonna teach all this stuff?
They're gonna have to rent teachers from places like Coursera in order to make sure that graduates have these skills and that students want to come to the universities to learn these skills. We believe that generative AI is gonna fundamentally change not only how people teach, but what needs to be taught, because there's a lot of demand for this. So looking at professional certificates, you can see so far in 2024, a number of titles have launched. We have a lot more coming. So we're expecting to launch more than 30 new professional certificates in 2024. Here are some of the different job roles, and this includes upgrading some of the most popular certificates to include, major components on how to do the job with generative AI. On the product side, I mentioned Coach. This is basically personalized interactive learning.
It's key to know that it's the answers are grounded in the content on Coursera that comes from trusted authorities on those topics. Coach can play multiple roles. It can certainly do tutoring. It can help you find which content you need. You can literally say, "I'm in this job title right now. What jobs will be available to me if I learn certain skills? And then tell me what courses will teach these skills, and what credential will be the credential that I will need in order to get that next job. Will I need a degree? Will I need some kind of a certification? Will I need a CFA?" Once you figure out what credential you need to get into that job, then the learning happens. You get assessed, you earn the credential.
The ability to adapt is far more personalized than it's been before, and of course, everything that we do on Coursera is guided by responsible AI principles. I'll say another thing, too. Currently, Coach does not utilize any fine-tuned proprietary models. You might say, "Well, that's really bad. Where's your advantage?" The advantage is the content. The answers from Coach don't come from fine-tuning. The answers from Coach come from this RAG. It comes from looking through the course to find the right answers, and the more content that the instructor puts into the course, the smarter that coach becomes. This is just a little sample, but Coach basically sits as a sidebar, and when this is a course I did on navigating, see, generative AI for CEOs. But Coach is available all the time.
You can ask it questions, and it basically helps you go through. We found that learners who use Coach are much more likely to persist in the courses, to finish the courses, to not get stuck as they go through. Course Builder, I mentioned. I'll just give a quick, bit by bit here. You could build custom courses and generate assessments that are completely private to your institution. This allows you to bring experts from Coursera into your course and combine it with experts on, in, on your team. If you're a professor, they can basically have guest lecturers come into their course. If you're in learning development, you can have guest lecturers essentially on Coursera come in and help teach your managers new skills. It helps automatically find the content and mix and match it.
It does the translations, and this, too, is built on our responsible AI principles. We do not use fine-tuning for Course Builder either. It's all based on a combination of the content that's on Coursera, our pedagogical principles, and we have a lot of reference examples of what a good course looks like. And so we can build courses that adhere to these principles and also have the expert content in them. So this is just a little. It just kind of shows. There's on the left side, you sort of just specify what you want the course to look like. On the right side, it generates first an outline, human-in-the-loop, you can then edit the outline.
Then you can say, "Tell me what courses you want to use as source material for this course." We give you the list, and you can say, "I don't want that course. I want this course." That's gonna be your source material. Then you click a button, and it builds the course. And then you can drag and drop, you can move the modules, you can delete things, you can generate assessments, you essentially tailor the course, but it dramatically reduces the amount of time that it takes, even though a human is still responsible for building the course and putting their institution's brand on it. So you can see the course will have, you know, components from Columbia, components from Illinois, components from Google, all in the same course, tailored to the specifications of the administrator. I talked a little bit about language translations.
We are cranking on these. All the data pipelines have been in place since mid-last year, and we're just generating translations. Today, we're translating text, the transcripts, the assessments, the readings, the UI, everything is translated. But the ability to do text-to-voice, the ability to do text-to-voice with voice cloning, so it sounds like that professor's voice when she is speaking French, although the course is natively in English. There's something we call lip sync dubbing, where the professor can actually look like they're speaking Arabic or whatever other language. I, my course is actually lip sync dubbed into Arabic. It looks like I'm doing the whole course in Arabic, and it sounds like me talking in Arabic. And I think this is really just the beginning.
So our view is that we're in the early innings of Generative AI. Yeah, it will have a huge impact on the education industry and businesses in general, but it will be a transformation of institutions, and we believe that with the content and technology that we offer at Coursera, we'll be one of the players who really benefits from supporting the transformation of these industries, especially the $2 trillion higher education market. Thanks.
All right. Thank you so much, Jeff. The breakout's gonna be up in Richardson upstairs. Thanks, Jeff.
Great. Thanks.