Pearson plc (PSON) Earnings Call Transcript & Summary

Good afternoon, and thanks, everyone, for joining. Joining me today are Tom Simon, President of our Higher Education division; Gio Giovannelli, President of our English Language Learning division; and Mike Howells, President of our Workforce Skills division. Now given the recent conversations around generative AI, we thought it'd be useful for you to hear where we, Pearson are in executing on our own AI strategy. And we want to give you an overview of some generative AI product enhancements that we're developing and rolling out across our businesses. Before we get to that, it's important to remind you that Pearson has a unique range of assets supporting the diverse business with expertise in assessments and qualifications, language learning, workforce solutions and virtual schools. Together, they represent 80% of our profits alongside our Higher Education division. Just over a week ago, we announced a very strong Q1 performance with sales growing 6%, underpinned by each division growing in line or ahead of our expectations. Pearson continues to show really strong performance with our paid subscriptions growing threefold this year versus spring 2022. As we've discussed, we'll be sharing specific user and subscription numbers at our interims in July. As a company, we remain on track to achieve our 2023 guidance and are confident of our ability to meet our medium-term financial expectations. Turning to AI and specifically, generative AI. As you all know, this is a fast-evolving space. And indeed, there was a great note from City out today on the broader implications of AI. That's worth a read. That said, I think it's important and responsible to be thoughtful while executing at pace. We're focused on one thing and one thing only, how do these tools enhance the functionality of our products and services across all our businesses to better serve our consumers and customers and help people improve their lives through learning. I want to lay out 3 reasons why we're well positioned to capitalize on the opportunities surrounding this new technology. Pearson has a long track record in using AI successfully in our products. It's important to understand how the basics of current generative AI play to our unique strengths. In broad terms, what generative AI does today is predict the next word in a sentence based upon user promise. It may seem to know what is written, but in reality, it's a powerful algorithm trained to predict the likely answer to a particular prompt. As we know, given the noise in its structure, this sometimes leads to inaccurate or inappropriate answers. That's why teachers, faculty, institutions, corporate enterprises and trusted content creators like Pearson are so important. Together, this is how we help people learn. You learn by contextualizing information and by utilizing the structure of the learning content, which is built by our expert authors and creators. People have asked me if ChatGPT could write a textbook. Yes, it certainly could. But so could I. Our advantage is the combination of deep subject matter expertise, teaching experience and learning science knowledge that our authors and faculty bring to the table -- if all we had to do was read a set of facts, numbers or words in order to learn, there'd be no need for schools, colleges, teachers, employment, training, et cetera. The structures, methods and pedagogy create the learning experience. Secondly, it's important to remember that data sets involve more than just words. They involve numbers, sequences and proprietary algorithms. We're currently using generative AI and those exact proprietary data sets in our Fathom business to develop personalized career and learning pathways. Part of our investment thesis around Credly, Mondly and Faethm included the data sets and AI expertise they bring to the company. We've been engaged with AI and thinking about the future of this technology for a long time. Thirdly, we need to remember that the quality of the data inputs correlates to the quality of the output. Again, this is where our IP gives us a competitive advantage. Our data sets are derived solely from Pearson IP. That IP is uncorrupted from the noise occurring in other large language models. This purity leads to better outputs for consumers and it leads to consumers being able to trust the information from Pearson AI. Our proprietary IP not only includes the information in our learning materials, but the structure and the curation of that content as well. In addition, we also have data sets on skills and credentials through Fathom and Cradle and assessment items across a number of businesses. This all adds up to vast amounts of data, which make an unrivaled set of intellectual property across our businesses. Now I want to be clear that we're committed to protecting that IP and then we're following legal developments in other industries such as music and photography with great interest. But generative AI is a significant positive opportunity for Pearson. Over the past several months, we've been using our deep experience with AI and learning science to develop generative AI product enhancements that benefit learners and instructors alike. So let's talk about what that looks like across the company. Our workforce skills division has some of the most advanced applications of generative AI in its field. Our AI experts have developed large language models, natural language processing and advanced machine learning, which we apply to a range of tasks in support of our enterprise and government customers as well as individual consumers. This includes analyzing skill trends and proficiencies, modeling the future impact of technology on the workforce, including the impact of AI itself, improving the speed and accuracy of assessments, generating verified credentials and advanced recommendation engines for learning and career pathways. Turning to Pearson Plus. Starting this month, Pearson Plus users will have access to a generative AI tool that automatically summarizes the content of channels videos into 10 simple bullet points, helping to contextualize the material and make learning more efficient. Within Pearson Plus, our experts have taken the best aspects of AI technology and are harnessing it to support learning. In addition, we're also developing a feature within channels to provide students with AI-generated multiple choice questions in the specific areas where they need more help and practice. What's more, the generative AI output will derive results only from Pearson's proprietary and licensed learning materials sourced by our subject-related eText and videos. That's why this content can be trusted. We now have 18 academic channels live. And in the past several months, we've broadened the offering beyond academic subjects to include 31 channels focused on tech and soft skills. Adding these generative AI features to channels support our ongoing strategy to build Pearson Plus users beyond higher education. In our higher education business, we're focusing on product enhancements that help enrich the learning experience for students and make teaching more efficient for instructors. In Mastering and MyLab, we plan to launch new features that use generative AI specifically to help faculty. From the student perspective, we're working on features that use Pearson proprietary content to guide them through their problem areas. Again, we believe trust and quality of fundamental differentiators in our offering. These personalized features will be contained in a secure environment, so data stays free from the noise and corruption of web-based AI models. In English language learning, we've developed AI-based open response assessments. These are based on our proprietary global scale of English and evaluate learners written and spoken responses against a variety of different metrics, including vocabulary and grammar and the use of phrases and structures. Additionally, with a generative AI-based conversation partner in development that provides speaking practice and assist with real-world conversations to provide anywhere, anytime languages tutoring on demand. These examples demonstrate how we're focused on the opportunity around generative AI and how we're fostering innovation across the company. We have more initiatives in development, and we'll be sharing more details at our interims in July, but they're currently commercially sensitive, so we're not going to discuss them further today. The trust that learners and educators place in us mean we have a responsibility to be thoughtful and considered about how we use generative AI. We take that responsibility very seriously. While we're moving at pace to capitalize on the opportunities of generative AI, our priority is to get this right for the people who depend on our products. So to recap, we have a diverse set of businesses, unique to the education, enterprise assessment and qualifications industries that can benefit from generative AI. Our strong performance and momentum mean we can be confident of our 2023 and midterm guidance. Our trusted and unrivaled depth of IP means we're uniquely and well positioned to capitalize on the generative AI opportunity. And we've adopted an AI strategy that will embed this technology across key products in a way that enhances the teaching and learning experience with several projects underway. With that, Tom, Gio, Mike and myself will be happy to take your questions. Alice, over to you.

[Operator Instructions] Our first question comes from the line of Tom Singlehurst with Citi.

It's Tom here from Citi. It's great to see you. Yes, being on the front foot. I had a couple of questions. One on higher ed, so I suppose it sits on. And then another one, actually on an area that you didn't talk as much about, which is assessment. Starting with higher ed. We've heard a lot about how students are potentially embracing some of the opportunities around generative AI. And yes, it might be that there are issues with accuracy and numerous, I suppose the stakes are relatively low for them. Can you just talk a little bit more about faculty -- because you mentioned that at the beginning. Can you just talk about some of the conversations you've had with universities about how they view the risk and opportunity from AI, in particular, with reference to Required Materials? And then the second question is on assessment. It strikes me at least that formal assessment feels like an area that should directly benefit from some of the confusion that's shown up by the use of [indiscernible] Can you just talk about that a little bit. I don't know whether there is a way of quantifying it...

Yes. Sure, Tom. Okay. From one Tom to another Tom. Do you want to go to higher ed?

Thank you. So I think our sales teams are having a number of conversations with faculty. And I think what's clear is that they are along with their institutions, all working out what their policies are with regard to chat GPT and some of this technology, there's I think, a wide range of different approaches in terms of people understanding it, not understanding it, early adoption and so on and so forth. I think they are often looking to Pearson for our point of view on how this can be embedded in required courseware. And so we're having a number of discussions with them and working with them to think about what some of those use cases are. So we see this as a positive discussion because I think from a faculty perspective, they see things like this all the time. I mean, if you look at ChatGPT in many ways, an extension of some of the academic integrity issues that search engines, of course, over the last 10 to 20 years. And so they're looking to Pearson for how we incorporate this into our products and services, so we can provide more value to our students in terms of the learning outcomes we're looking to deliver.

Perfect. And then on your second part, clearly, well, there's a lot of focus, Tom, on the world of assessments. What is paramount there, I think, is 3 things, is accuracy, trust and verification. And you think about the breadth of our assessment and qualifications business from GCSEs, A-levels, BTC all the way through to our Pearson Test of English, the verified specifications that you get on the Credle platform, the work that I referenced that we're doing with Fathom. And then more broadly with Pearson View, the positive applications of AI and generative AI in this field, I think, provide us with a lot of opportunities. I don't know, Mike, Gio, if you want to add anything?

I think -- so we certainly see opportunities to increase the speed and accuracy of our assessments. So to Andy's point about the core part of the core business value proposition being that trust, that accuracy. We see the ability to improve that actually through these technologies, but also the personalization of the customer service aspects. The speed of return of results, something which I'm sure Gio can speak to in a moment, is a key part of what we bring to the table here. But we're also always looking at ways to utilize generative AI to evolve the way in which we can do assessments and particularly from the perspective of the workforce skills business, how we can bring that technology into and closer to the flow of work. So we have numerous R&D projects underway at the moment, we have looked at that. So there's a long way to go, I think, in seeing how we can leverage generate AI to improve what Pearson is already great including in new markets.

Yes. In the space of English language learning, the Pearson Test of English has been for many years at the cutting edge of the use of new technologies. Our test, as you know, is fully powered by AI with human intervention and human review. It has a very fast way of scoring tests and giving exact unbiased feedback regardless of the accent or the appearance of people who are taking the test, and we take a lot of pride on that. And we're continuously developing the new generation of this test working very closely to the authorities that grant us the rights to administer the test and to have it valid for both immigration and study. We feel very good about our domain of these technologies and the evolution that we're making to the test, and you will see more in due time.

Thanks, Gio. It's worth saying, and Mike alluded to it, we often talk about the technology and the impacts and the opportunities. I think one of our particular strengths as a company is the people. We have a cohort given our long-standing use of AI. And as I mentioned in my prepared remarks, part of the acquisition of Fathom, we have -- and Modality Credly and others, we have some incredible experts in this field in-house. We have an in-house teams channel that I think has got over 100 participants now. And I just -- I marvel. I spent yesterday just going through this real-time conversation and the level of understanding and knowledge and expertise we have within the company is something that should not be underestimated. Anyway, thanks for your question, Tom. Alice, onto the next question.

The next question comes from the line of Sami Kassab with BNP Paribas.

I have 3 questions, please. The first one, you talked about the role of AI and product innovation. But can you please elaborate on the impact AI and generative AI could have on your cost base? How much cost perhaps could be saved by replacing labor intense process with more automation? Secondly, to a large extent, where all of this because checks suggested that students are increasingly using generative AI to get student services instead of subscribing to them. What is the risk that students use piracy for content and ChatGPT for study services instead of subscribing to Pearson plus -- and lastly, you've argued that the purity of the data can produce superior learning outcomes to AI applied to web-based data sets? It makes sense, and I would agree. But can you provide any evidence or examples to the that statement and idea, please, Andy?

and I can take certainly the first 2. Maybe, Tom, you can talk specifically around the third question as it relates to higher ed, but I can Gio -- feel free to interject. I think clearly, there we -- everyone looks to the threats and opportunities, I guess, on the top line as it relates to AI, generative AI. As you quite rightly point out, there is also often lots broken but inferred certainly within Pearson of the role of AI in terms of creating efficiency. Some of the efficiencies actually that we utilized last year when we did the margin improvement was because -- and I said at the time, it's about the company was working more efficiently, not just -- this wasn't the case of just taking costs out in an analog sense. And so you can imagine in the area of tech in terms of coding in different areas engineering. There's a really positive role that AI can play as it relates to costs. And that's something that we're across the company, not just within the divisions, but within corporate as well, looking at ways that can further increase efficiency and drive margin improvement. So that's a very good point. As it relates, I think I don't want to get too much into the business of Chegg, but only to say that they are in a very different business to us. And we see also a very great differentiator between what Chegg or offering and what Pearson Plus is offering. Pearson Plus is about helping you learn and improve your knowledge and skills and practice that. It's not in the business of answering or helping you answer those particular questions. So I think there's a great amount of differentiation there. We're seeing amazing engagement with Pearson Plus channels across the world. It's been utilized in over 172 countries. As we start to broaden beyond academic, as I mentioned, we now have 31 channels dedicated to job skills in -- particularly in the tech area, that we're seeing a great deal of traction. As I mentioned, we're going to get into more detail about this in interim. So you'll have to bear with us in terms of -- we don't want to reveal too much at this stage, apart from saying we are very, very positively encouraged. As we mentioned at Q1, we're trialing different commercial opportunities with channels that's going on at the moment, yet to make final decisions to what we do. But you're going to see the channels product along with other products that we haven't announced that are in development across the company sort of extend way beyond higher ed and actually broadening the definition of higher education. But more on that really to come at interims -- and then, Tom, do you want to talk specifically about quality of data in relating to quality of data out as it relates to higher ed and then Mike, Gio, you can chip in.

Yes, absolutely. I mean I would just say that people have been using the Internet, obviously, to get answers for a long time on a number of questions, and that's been a known issue. But I think, Sami, as you think about the impact of AI here, we have a position of huge competitive strength. We have unrivaled content franchises, which are sort of definitive guide to concept master and lots of different subjects. So if you take something like Campbell's Biology, which is sort of the world-leading introduction to biology and content, we've got 14 additions worth of that background context history. That can't be replicated by LLM because they're picking up everything from the Internet, everything that's accurate, everything is inaccurate, everything that's got sort of all of the inherent biases on the Internet. So we start off from that opposition of very strong both content assessment learning science knowledge, all of the work that's been put into Campbell's biology, that's a specific method that's been built by faculty who have talked for a long time and we scale that to support students around the world in terms of their learning of biology. We then build on that by embedding the conversational nature of the LLM models into our core products, and that allows us to provide that secure scalable environment for students to learn. That means that we can keep our IP safe. We can provide students the ability to learn from that content franchise, and that helps us to build engagement and increase learner effectiveness. So just like I think the calculator changed the way that math has taught, -- we think that this will change how teachers are going to teach, but faculty can then get things like a transcript of a student engagement with a large language model and then evaluate the student saying questions, concept master and understanding to see how they're doing in terms of building these critical thinking skills. So we think that we are firmly in the critical thinking business, and that this is the type of technology we can couple to our deep competitive notes in terms of content expertise to build unrivaled learning opportunities for students to learn.

Great. Mike, anything you want to add, particularly in workforce skills?

Yes. So I would encourage you all to sort of see the strength Pearson has in this space in 2 different areas. The first one, as Thomas just explained and Andy's talked you through today is our content. So obviously, that's the #1 in the quality of your output is the quality of your content that you've been training your models on. The factor number 2 is the quality of the model. And what we have in workforce skills is a very strong data science and AI engineering capability, which we have used to develop and own our own large language models, which we then apply to the data sets that we use in workforce skills. So I think the combination of both the unrivaled IP that Pearson has but also our ability to hone and develop specific and tailored data models to the use cases for our customers is a really important distinction. I think you'll see a lot of companies out there in the world that are applying licensed AI models to their existing businesses, and that will have a lot of transformational impacts, but the best companies in the world will be developing and evolving their own models that meet the use cases and their customers better. So it's a combination of those 2 factors, I think, that really give us opportunity and strength as we move forward into this new world.

One thing I'd like to point out for English language learning is our proprietary global scale of English, which is a unique tool that allows us to tag everything that we have, both in the content and in the assessments to a scale, a very granular scale that measures where you are in the knowledge of a language, -- how do you progress and how you're evolving and what exact remediation you need to improve and also to take you to the next step and how you learn. And so the uniqueness that we have in the technology and AI powers is how do we tag everything that we have to that global scale of English? And how do we use that to your point, in order to reduce the cost of generating new content because once everything is tagged and it's a common repository, you can reuse that content in a very efficient way to target for different use cases for different conversational and different target audiences. So I think we're very well positioned to use those technologies on our benefit.

Our next questioner is Mr. Nick Dempsey with Barclays.

I've got 3 questions as well. So first of all, in the area of language learning, you've pointed to a conversational Avatar style model. I think you showed us that of the full year results. What advantages would Pearson have in that area versus, say, Duolingo or Babel or anyone else working in that area that would give you USP? Second question, can you tell us roughly how many technology employees with skills in this area of AI, data and analytics, broadly the Pearson has across the organization. And the third question, -- in higher education, I understand that they give an opportunity for you to improve your product. But if students are starting to behave radically different in terms of how they approach learning and professors begin to acknowledge and adapt to that. Is it not possible that the competitive dynamic for courseware could change and it will be difficult, therefore, for Pearson to maintain its current good market share?

Great. Thanks, Nick. Gio, do you want to take the first question?

Yes, sure. Thank you for the question, Nick. On the point of the conversational digital human, I think everyone who's tried to run the language knows how difficult it is to practice speaking. That's usually among the 4 skills of acquiring a language that were the one in which you need most practice. And you can do it 2 ways. One is simply by engaging in a normal conversation. The other one is what we are trying to do is by putting our pedagogical science and the breadth of the assessments and the granular knowledge that we have to service to that. So that speaking practice in what we are building today is tailored to your level of English is stable to specifically who you are to what you want to learn and also how you learn so that it can bring you into a real word conversation, which is, for example, different if you're trying to prepare for a job or if you're trying to prepare to take an English test or if you want to just have a conversational level of English to travel. And the other thing is that we can do it in a way that is safe and according to our internal guidelines, making sure that it respects our editorial policies, and it doesn't enter into any issues when, for example, a child uses it versus somebody who could be exposed to some type of buyers. So we feel very good about the developments that we have, and we think we have a unique way of addressing this problem that will help people be more engaged and learn language is better and faster.

On your second question, Nick, I mean, the -- I don't have an exact number for you, but -- and it depends on how you define but it is in the hundreds, I would see probably over the thousands of individuals in the company who are versed in terms of AI, given the history, the long history we have with using it and generative AI in particular. And I think the 3 areas that you kind of think there's clearly around technical engineering, data science is second and also -- the third area, which is -- gives us unique insights is also with our auto population. Mike, Tom, I don't know if you want to speak specifically to either of those.

So we have -- so in terms of very narrowly defined AI engineers and deep expert data scientists in what full skills we have between 25 and 30 people working in those roles. So these are people who are developing our models and our algorithms and machine learning tools that we apply specifically. And then we have lots to Andy's point, many layers of experts outside of that, so are focusing on the utilization of those tools in products and the analysis and utilization of the data outputs as well.

And then, Tom, maybe on also some pickup next third question?

Yes, definitely. So a very good example of one of our authors is a guy called Peter Lee, who works at Microsoft, he's the VP of Research and incubation. -- he's recently published a book last week with us, which was the AI revolution in medicine, GPT4 and beyond and say that's a good example of how we are able to understand what's happening at the forefront of sort of AI research on something as important as AI in the role of medicine in helping to evolve our teach nurses and doctors around many of the evolving themes in that field. And then I think if you think about higher education, Nick, I guess I would say that students are learning differently today than they were 5 years ago, 10 years ago and 20 years ago, right? And so I would say that historically, -- we've always seen adaptation in student behavior and learning. And we've seen that in terms of attention span changing, shortening the increasing role of video, the increasing changes in terms of the way that students are interacting with things like YouTube to get video understand what's going on with the world and learn very differently. And I think that, that's being incorporated in those K-12 classrooms and higher class from in multiple different ways. And our expertise really is that, that intersection of content and technology to help scale great teaching. We did that in an analog format with the textbook. We're doing that with the homework assessment platforms that we've been very successful at in terms of the MyLabs and really driving innovation in higher education over the last 20 years. And as you can imagine, we're fairly determined to do that with this new generation of technology because we deeply understand by working with faculty and those who are in the classroom who've got that great learning and teaching experience, how we can scale that to support millions of learners around the world in multiple different disciplines to learn. So we take this challenge very seriously. We are being very thoughtful about it. We are speaking to a number of faculty forces experts as well as our students. And we're taking on that board in terms of our consumer resection and what we're doing differently from a product proposition.

And that last point Thanks, Nick. That last point, I think, is very important in terms of one of our 3 priorities for the company that I laid out us this year is to be consumer obsessed in terms of access with the consumer, their behavior, their shift in their behavior. Tom referenced, I know they have live insights coming back from students, ultimately, the ultimate consumers of our product. We were all together last week. We had a very interesting conversation, which we could probably spend the next half hour on in terms of Gen Z and their retention spend. It's 1.8 seconds, -- for those of you who are interested and the impact of Gen Z as they move through education into employment. And so understanding the consumer, consumer habits, what they need and making sure that our products are relevant to the end consumer is really, really important. And I think that's something that, hopefully, you've seen a shift in -- over the last couple of years with the company. Alice, who's next?