And we’ve done the first two bricks, which is conversion and take rate. Next one is retention. And then what can you raise it to when you want to come at our promotional pricing for the customers that took it? And how long do you want to keep promotional pricing for new customers. But right now, better to fight for market share and make sure people don’t come away with the conclusion that ChatGPT is a better solution than us because it’s not.
Josh Baer: Great. Thanks, Dan.
Dan Rosensweig: Okay.
Operator: Our next question comes from the line of Ryan MacDonald with Needham & Company. Please proceed with your question.
Ryan MacDonald: Hi, thanks for taking my questions. Andy, best of luck in the future and David, congrats on the promotion. Let me just start Dan, can you talk about sort of — get it sounds like there’s some real benefits here on the automated answers in terms of cost savings and how you’re answering questions. Can you talk about the mix of, I guess, the overall questions that are being answered today by automated answers. And then perhaps what we need to see that get to as a percentage of the total questions for that to start to translate into some margin expansion relative to sort of what the first quarter guidance implies? Thanks.
Dan Rosensweig: Yes. That’s exactly what we’re looking at. So what you should be looking for what we should be talking about and what we will be talking about over the course of the year. We didn’t really know where to start talking because we didn’t know how popular automated answers would be and how quickly it became this popular. But as we articulated in the prepared remarks, already, the cost of answering a new question is 75% less, simply based on the tailwind that we’re getting from AI. We call it automated answers, but it’s a result of us taking our own large language models and developing our own answers. We have a quality algorithm. So right now, you won’t see much in the way of savings in the first part of the year, because we haven’t fully rolled out all of the different large language models.
The expectation is over the course of this year and by the end of this year, the overwhelming majority of all new questions will go through automated answering, and there’ll be fewer and fewer going through human answering. So you should begin to see margin expansion towards the end of the year as a result of — you really see a CapEx reduction first, right, which is the amount of money that we spend on content has been coming down, but it’s going to come down again this year and then substantially in ‘25. So we can answer 10 times as many questions for the cost of one question is the way to think about it. But we have to get all the large language models out so that we can and the quality of the level that students expect from Chegg that they don’t get from ChatGPT or Bard or anybody else.
And that will take the rest of this year to do it. And we’re very excited about it because it will be a meaningful improvement in our ability to pay for content and have super high quality. And as I said, also generate the flat well. The other real benefit is the more questions that get indexed, more people find us and we don’t have to pay for that traffic, and it comes in. So that’s the way Chegg was originally built when we first built Q&A. And it’s just going to be a lot faster now because just the volume of questions that we can answer. I don’t think we’ve ever answered 2 million questions in a month, and that’s just the month of January, which really doesn’t get started until mid-January.
Ryan MacDonald: Super helpful color. I appreciate that, Dan. And then you talked about in ’24 growth priorities or one of the priorities is obviously that pathway and returning back to growth. Sounds like the promotional pricing is going to obviously help with that. But as you think about the macro, how important is sort of enrollment within colleges and universities to driving that return to growth algorithm for Chegg. And sort of how do you — how are you sort of building that into your sort of assumptions for this year?
Dan Rosensweig: Yes. Look, of course, it has an effect. It’s just not the biggest one. And I know people think that it is, and I can understand why they do, frankly. But let me just sort of size to you just U.S. because outside the U.S., it is just — it’s wide open, right? We jumped during COVID, then sort of stop when things got out of COVID, then ChatGPT launched and had a beginning of an effect. But the new products and the new automated answers and the new indexing of the question, the promotional pricing has reversed that course outside the U.S. So yes, when you think about enrollment, last year or on average in a given year, Chegg will have over 10 million students in the U.S. that will register for Chegg, but not convert.
And so we have that whole universe to go after that has nothing to do with enrollment. So we think that near-term, the goal is to get those people to cross the chasm into becoming loyal Chegg customers because our retention rate is so high, because our quality is so good. And now with the immediacy of the responses at high quality, which no one else can give them. We think that’s the key to growth. So enrollment will matter over time. But right now, there’s a big group of people that are right in front of us.
Operator: Our next question comes from the line of Brian Peterson with Raymond James. Please proceed with your question.
Brian Peterson: Hi, thanks for taking the question. Well-deserved congrats to Andy and David. So just — I want to follow up on Ryan’s question. You’ve done a lot with the platform this year, and you’ve done that with R&D dollars actually being down year-over-year. I see the opportunity for automation, but also mentioned scaling to different languages and time to roll out the LMs. I’m just curious how we should think about the puts and takes to R&D intensity in 2024, maybe how that arc should look longer term? Thanks, guys.
Dan Rosensweig: Yes. Really, really good question. So hopefully, for those people like you that have been tracking Chegg for a while, you see we try to be very, very efficient with our capital and so if you take a look at OpEx, OpEx has not grown substantially in the last couple of years, and it’s been relatively flat. We expect it to be relatively flat again for this year. And yet, we have 250 people now working on AI. And that is because we have a history of reallocating resources against the things that have a priority and either stop doing the things that no longer matter as much or where they’re at scale and can survive with the number of people that we have on them. Same thing goes through with our sort of capital expenditure, which is the overwhelming amount of capital that we spent was on content.
So we’re not — like — I think there’s some confusion because we said last April that we have a partnership with OpenAI and so it’s easy to have made their conclusion posted really on us that we have big API costs. We don’t. We have moved mostly often will almost be completely off open AI shortly because there’s now a lot of different alternatives and there are a lot cheaper. But more importantly, the real cause is sort of GPU costs and other things like that. It’s not — and so the overwhelming cost that Chegg has in CapEx is always going to be content. And since the cost of content is dropping significantly, it’s already down 75% based on automated answers versus human answers. So the next big step is to get off more human answers, which we hope to do mostly by the end of the year.
So the way to think about it is you can assume our capital this year will be much better, much more efficiently spent because AI is allowing us to spend it a lot cheaper on a per question basis than we ever could before. So we are not holding back investment. It’s just we can get a lot more for our money now, thanks to AI. It’s one of the great benefits of having moved as quickly as we move. Operator, is there another question?
Operator: Yes, our next question comes from the line of Alex Fuhrman with Craig-Hallum. Please proceed with your question.
Alex Fuhrman: Hey, guys. Thanks for taking my question. Can you talk a little bit about your core customer today versus a few years ago? It sounds like the biggest change given everything going on with AI? Is that maybe some of your more casual users at the margin who might pop in for a month or two around final or mid-term exams have kind of maybe looking at other AI-powered options? Can you talk about the types of students stuck with Chegg. Are there any major call outs now versus a few years ago in terms of your core student in terms of either age or full-time or part-time status or field of study or two versus four-year schools, anything like that, that you’re focusing on a little bit differently versus a couple of years ago?
