Great. Thank you, everybody. Good afternoon, and for some, good morning. Welcome. I am Paul Sullivan, Head of Investor Relations at RELX, and it is a pleasure to welcome you to our STM Investor Seminar today, and also introduce the Chief Executive Officer of Elsevier, Kumsal Bayazit. Kumsal.
Thank you, Paul. Good afternoon, and good morning to those joining us from the U.S. Welcome to the STM Investor Seminar. Thank you very much for joining us today. I am Kumsal Bayazit, Chief Executive Officer of Elsevier, the STM division of RELX. I joined RELX in 2004 and have been the Chief Executive Officer of Elsevier for the last eight years.
Before that, I held operational and strategic roles in all parts of RELX, in Risk, in Legal, in Exhibitions, as well as at RELX Corporate as Chief Strategy Officer and Chair of the RELX Technology Forum. It has been four years since we last did a deep dive on Elsevier. My team and I are really looking forward to sharing with you the progress we have made since then, as well as the significant opportunity that we see ahead. I will start with an overview of the business and our strategy.
Gemma Hersh will then cover Primary Research. I will come back to cover our Decision Support Tools, and two of our product leaders, Max Khan and Paul Crockett, will bring these to life with product demos. We will then hear from our Chief Technology Officer, Jill Luber, and I will close with some concluding remarks and move to Q&A. Let me start by putting Elsevier in context. In H1 2026, STM represented 28% of RELX revenue and 30% of adjusted operating profit. That equates to annualized revenue of roughly $4 billion or GBP 3 billion, and adjusted operating profit of over GBP 1 billion. We are a very global business with customers in more than 170 countries and territories.
Little over 40% of our revenue comes from North America, a little over 20% from Europe, and just over a third from the rest of the world, which gives us broad exposure to research and healthcare spend around the globe. Almost 80% of revenue is subscription-based, primarily under multi-year contracts, and the rest is transactional, much of it recurring in nature. The large subscription base of the business gives us stability. It also means the positive impact of increasing volumes or new higher value products comes through gradually in our results.
A little over half of STM's revenue is in Primary Research. Here, we offer article publishing, research, and analytics platforms. Decision Support Tools, which we also refer to as databases, tools, and electronic reference, is around 40% of revenues. What we do here is help our customers with evidence-based decisions and tasks through multiple offerings.
The two parts of the business reinforce each other. Primary Research gives us the content, the credibility, the domain expertise, and the expert community on which our Decision Support Tools are built. Our Decision Support Tools increase value for our customers by combining that trusted content and curated data with advanced analytics and evolving technologies such as artificial intelligence.
We operate in large global segments with attractive structural long-term growth drivers. R&D spend is critical for nations and corporations to drive innovation, economic growth, and solve societal issues such as finding cures for diseases or climate change. This is why, in real terms, global R&D spend has consistently grown at an annualized rate of 5% over decades. In nominal terms, the growth is, of course, higher. This growth trend has been secular and has held through economic cycles, geopolitical tensions, political, and policy changes.
In any given year, of course, there can be ups and downs in spending in any one country or institution or discipline, given budget pressures or short-term fluctuations. Fundamentally, over the long term, global R&D spend continues to grow. This is one of the reasons number of researchers also consistently grows at about 5% per year. I could have shown you this data for 50 years, and you would have seen the same trend, and we expect this growth to continue for many decades to come. Technology advancements have always improved research productivity. It's another driver of growth, and generative AI is having a similar impact.
These drivers, R&D spend, number of researchers, and improving researcher productivity lead to more research, more discoveries, more information and content, higher number of submitted and published articles, and greater demand for more evidence-based Decision Support Tools to understand, build, and apply on knowledge. Looking at health, you can see very similar trends, with global health expenditure growing at 3% a year plus inflation. Number of physicians and nurses has also been consistently growing globally.
We serve a large global customer base. We work with over 10,000 academic and government institutions, including virtually all leading research organizations worldwide. We do extensive work with millions of researchers who interact with our tools each year. Our teams have deep connections with research leaders and leading researchers in their fields. We also serve over 2,000 R&D-intensive corporations and work with leading companies in life sciences and biopharma, engineering and technology, chemistry, and energy.
In health, we have solutions for most stakeholders in the health ecosystem, from medical and nursing schools to healthcare providers such as hospitals. No single customer accounts for more than roughly 1% of our revenue. In every segment, our objective is to be a trusted partner and to help customers improve their outcomes so they can, in turn, advance science and healthcare.
O ur relationships are long-term and deeply embedded in how our customers work, which is a strong foundation for our Decision Support Tools. We have three main strategic priorities. First, we expand customer value in Primary Research. We continue to deliver industry-leading research quality, article volume, and user experience on leading technology platforms at a lower effective price. Technology advancements such as AI enable significant improvements in the author, editor, and peer reviewer experience and outcomes throughout the whole editorial process.
Second, our Decision Support Tools continue to deliver higher customer value. By combining our deep domain expertise, leading evidence-based scientific content, advanced analytics, and evolving technology, we deliver innovative products that improve customer outcomes in existing and new use cases. You will see examples of that with LeapSpace and Sherpath today. Third, our global modular data and technology platform enables us to build segment-leading products more efficiently.
We have always operated with a technology-agnostic approach so that we choose the right tools for each use case as technology evolves. A current example of that is our multi-model AI orchestration layer, so we can optimize for quality, latency, and cost. We also leverage our technology platform internally to accelerate our speed and agility and to drive efficiencies. That is what allows us to innovate at pace and manage cost growth below revenue growth. These priorities reinforce each other.
At our last seminar, I spoke about how we evolved the business towards higher value, high-growth Decision Support Tools, which had improved our revenue growth from 2% to 4%. Since then, we have continued to deliver on that improving long-term trajectory, and we are currently at 6% growth. Equally importantly, during this time period, our customer metrics, including net promoter scores for products and buyers, and customer satisfaction, have been on an improving trajectory. Employee engagement metrics have also improved.
Let me talk you through the drivers of our improving revenue growth in a bit more detail. Four years ago, we said that Primary Research had been growing in the low single digits. Since then, we have delivered increasing growth in submissions, high-quality published articles with an expanding journal portfolio, and improved author, editor, and peer reviewer experience.
That volume growth has started to come through gradually in revenue growth, and we are now growing in mid-single digits in Primary Research. For Decision Support Tools, at our last seminar, we said revenues were growing in mid to high single digits. As we add more customer value with increasing pace of product innovation, we are now delivering strong growth in the segment. Our objective is to continue on our improving long-term growth trajectory by delivering gradual improvements in both Primary Research and Decision Support over the long term. Our objective is also to continue to manage profit growth ahead of revenue growth through process innovation. I hope that this gives you a good overview, STM. I am now going to hand over to Gemma for a deep dive on Primary Research.
Thank you, Kumsal . Hello, everybody. I am Gemma Hersh, Senior Vice President, Primary Research at Elsevier. I have been with Elsevier for over a decade, and I have held a variety of strategic and commercial roles. Today, I will cover what is it that we do in Primary Research, the value we create for the authors, reviewers, editors, and readers we serve, the opportunity in front of us, and what we are doing to capture it. At the heart of what we do in Primary Research is supporting researchers and their discoveries. Researchers get funding based on their research proposals from governments, funders, and corporations. They conduct the research, at the end of which they have findings. They codify these findings in an article and look for the right journal in which to share them.
Publishing plays a critical role in how researchers share their contribution with the scientific community and build their reputation and careers. Published articles also enable other researchers to build on this knowledge. We enable that process by helping researchers verify and share their findings. The process starts with the researcher submitting their article to the journal they'd like to be published in.
All submitted papers are first assessed by one of the editors of the journal. These editors are recognized experts in the field. Some work as full-time employees, others work part-time alongside roles within academic or research institutions. In total, we work with over 37,000 editors, supported by around 105,000 editorial board members. We have publishing teams who constantly recruit leading experts in their fields as editors and editorial board members. For example, over the past 12 months, we recruited approximately 8,000 new editors and 12,000 editorial board members.
Around 70% of submitted articles do not progress beyond that first assessment, called the editorial review, either because they're not the right fit for the journal or because the science is not yet sound. Papers that progress go to the second assessment, called peer review. This means editors send the article to at least two reviewers who are experts in the specific field. In total, we work with more than 1.9 million reviewers per year. During the peer review process, the author will respond to questions, improve the paper, and sometimes conduct additional work. Peer review pressure tests research findings. Expert peers scrutinize the methods, challenge the conclusions. Their feedback improves papers, sharpens arguments, and highlights gaps so the published record is stronger.
At the end of peer review, about half of the manuscripts they received are accepted for publishing, meaning less than 20% of all articles that are submitted to Elsevier are published. This process provides quality control so that other scientists can trust the findings in published papers. Clinical guidelines, drug development, and policies depend on peer-reviewed published findings. The published article becomes the version of record that clinicians, funders, and policymakers can rely on and other researchers can build on.
After the article is accepted by the editor and before it's published, it goes through the production stage. It's critical that the paper does not have any mistakes and all images and data are accurate. During this stage, we also add metadata that enables the extraction of billions of data points. These data points also fuel the quality and value of our Decision Support Tools for customers.
We distribute the article via our research platforms. ScienceDirect is the biggest, with significant reach, 20 million unique visitors monthly, which drives broad visibility and readership. This is critical for researchers as it enables other researchers to build on their findings, helps drive collaboration, funding, and ultimately supports researchers' career goals and their impact on society.
We spoke about this in the last seminar. It's easy to launch a journal. It is much harder to build a mature one with a trusted community at scale and with global reach. On average, it takes four years for a new journal to get indexed by a trusted indexation service, and it usually takes much longer than that to become a leading journal in a field. Kumsal touched on how technology advancements have had a positive impact on research productivity. AI and generative AI are great examples of this.
They speed up the research process and make it more efficient. That means more quality findings, more quickly, and more submissions to publish. However, it also means more research integrity issues and AI slop, which makes trusted journal brands and trusted peer review processes even more valuable. Journal quality is a key driver of submissions, and citations are a proxy of that. Journals that are the most highly cited develop reputations for being impactful and high quality over time. We have the largest and the highest quality portfolio, and scale and quality both matter. For authors, publishing in a high-quality journal raises the visibility of their work. We look at quality in three ways. First, share. We publish 18% of all articles and receive around 30% of citations, which means the research we publish is read and cited more frequently.
Second, Field-Weighted Citation Impact, which normalizes citations for differences between scientific fields. The industry average is one. Elsevier's is 1.46, well ahead of other large publishers, as you can see on this chart. Authors, funders, and institutions use measures like these when deciding where to publish, what research to fund, and to measure the impact of that research. The third lens is quality tiers, where journals are ranked by Field-Weighted Citation Impact.
We participate in every tier because our objective is to find a home for all sound science, but our portfolio is weighted to the top. We publish about 40% of all articles in the top 10% of journals by this measure. This was 32% when we presented to you in 2022, so we've grown our article share in this top quality tier. For our authors, that means a wide choice of high-quality journals in their discipline.
For customers, it means a subscription that gives access to a portfolio weighted to the highest impact research. This is a virtuous cycle. A high-quality portfolio attracts high-quality submissions because all the authors want to publish in trusted journals in their discipline. That content, with our share of citations, is also the foundation on which Decision Support Tools are built. Quality is more important today than it has ever been, with increasing research integrity issues and AI slop, and we expect it to continue to be even more important in the future. Let me now turn to the opportunity and then to how we are capturing it. Today, total industry articles published has grown at a compound rate of 5% a year since 2015, and we expect that growth to continue.
We covered the drivers of this growth: continued long-term growth in R&D spend, number of researchers, and advancing technology, such as AI improving research productivity. Zooming in on Elsevier, our goal is to continue to grow faster than the industry. Let me start with submissions. Elsevier submissions have grown from 1.3 million in 2015 to 4.2 million in 2025, a compound annual growth rate of 12%.
Growth was over 20% in 2025 and over 20% again in the first half of 2026, so momentum remains very strong. We expect continued strong submission growth for many years to come. Why are authors choosing Elsevier? First, we have a very extensive portfolio with over 3,000 journals and industry-leading quality. Journals like Lancet have been a leader in global health for over two centuries. Journals like Cell in biology and Blood in medicine are field-shaping.
Concerns about research integrity may also be leading more authors to choose established, trusted brands with a long track record. We are also continuously improving user experience for authors, editors, and reviewers. We give authors choice in how they publish, so they can publish in a journal that meets national or institutional policies that they need to follow. Published articles have grown from 400,000 in 2015 to over 800,000 in 2025, a compound annual growth rate of 7%. Over any multi-year period, our published articles grow more slowly than submissions by design. We are selective with a rigorous quality-focused selection process across all our journals, and that is important to protect research integrity and our quality position. We have three ways of growing published articles while protecting quality. First, we are actively extending our portfolio across all disciplines and tiers.
We launch approximately 70 - 80 new journals a year to fill gaps in disciplines or quality tiers or in emerging fields. One example of this is Chem Circularity, one of our new Cell Press journals, which advances closed-loop solutions across chemistry, materials, and engineering. In addition to launching new journals, we also focus on successfully growing and maturing our existing journals.
Second, we support authors in trying to find a home for all articles that are submitted to one of our journals, even if they do not get accepted in their first journal of choice, but are suitable to be published in another journal. We call this Article Transfer Service. It takes a lot of effort to facilitate this at scale. We use technology and automated processes, as well as increasing editorial capacity to help authors find a suitable journal to publish their article.
It is important to highlight that it is always the author's choice on where to submit. Third, we are constantly improving our submission and peer review systems to improve the author experience and develop tools to support reviewers and editors to make the best use of their time and enable them to make better decisions more efficiently. This lets us increase our volume without compromising quality or research integrity, and I will come back to this in a minute. Now to the value customers receive. Our strategy in Primary Research has been consistent: higher quality and a lower average effective unit price in the industry. The two primary payment models in Primary Research are pay to read and pay to publish. Both models are available as a subscription or on a transactional basis to our customers.
We provide both payment models for our services, as well as combinations of the two, to support our customers' diverse needs and preferences. As our article submissions have grown by double digits, our volume of published articles has grown by mid to high single digits, and our Primary Research revenue has grown below this. Revenue per article, which is a proxy for average industry spend per article that we publish, has decreased consistently for many years. As you can see on this chart, we have continuously delivered more customer value for every dollar spent while increasing our revenue growth for the last decade. This is our strategy: deliver more customer value at a lower effective unit price and maintain high levels of quality. This ensures customer trust and revenue growth.
Let me show you now how we add significantly more value to Primary Research by leveraging sophisticated analytics and advanced technology like AI. The first value is safeguarding research integrity and maintaining trust. Integrity issues in research range from honest mistakes to deliberate misconduct, AI slop to fraudulent manuscripts, including organized paper mills. We screen every article from before submission to after publication, with many active detection signals covering misuse of AI, policy compliance, authorship manipulation, citation abuse, paper mill activity, and data fabrication. We continuously increase the number and quality of the signals, and we draw on our fraud expertise from our Risk division, giving us richer integrity capabilities. The more papers we see, the more we can identify new signals of misconduct, and so the better our research integrity solutions get.
Our new tools bring screening, investigation, and decisions into one system, so every case moves from identification to resolution consistently, and every decision is transparent, traceable, and defensible. We expect this dynamic of AI slop and research fraud to continue to be an arms race, like fraud in other industries. We will continue to mature our tools and drive innovation to ensure that the scientific record can remain trustworthy.
The second area of value is to continue to evolve editorial and peer review processes to drive higher quality, faster turnaround, and lower cost. We have already made good progress and are seeing measurable results. Let me give you two examples that are quite advanced in implementation. The first one is a vastly improved submission process, which we accelerated the rollout of about a year ago. The result is faster, better quality submissions.
Median submission time by an author is down 60%, and 90% of authors find the new experience both useful and easy to use. Editors benefit too, because they receive better quality submissions. This has been rolled out to nearly two-thirds of our 3,000 journals. The second example is a fundamentally redesigned peer review experience for which we also accelerated rollout this year. The result is more transparency, less friction, and more positive peer review outcomes. User preference is around 85% among authors and reviewers, and the acceptance rate of peer-reviewed manuscripts has consistently improved. This also has been rolled out to more than two-thirds of our 3,000 journals. The implementation of these new tools is a significant change management process, with tens of thousands of editors and millions of authors and peer reviewers.
Let me give you three examples that are in testing stage and that we plan to gradually roll out. For editors, we are launching AI-enabled tools to support them in their work so that they spend less time gathering information and more time making decisions. Over time, we see an opportunity to automate more of editorial triage so editors can focus on decisions that genuinely require human expertise.
Another example is from the acceptance to publication process, where we prepare accepted papers for publication. Over the last decade, we cut the time for this process by more than 50%. We aim to decrease the time it takes by another 75% over the next three years. This will drive quicker access to the newest research, a more positive author experience, and decreased costs. We are also building AI agents to do significantly more enrichment of the published content in this last mile.
That drives richer analytics capabilities in our decision and task support tools, as well as lower cost to serve for analytics use cases. We are also in the process of rolling out LeapSpace capabilities, such as Claim Radar, which Max will share shortly, to our editors, authors, and peer reviewers. This is a very helpful tool during the editorial process to assess the paper. Across all these stages, human expertise remains central. Editors, reviewers, and subject matter experts make the decisions, and technology gives them better information, makes the assessment easier and quicker, and frees up time to apply human judgment. Let me hand back to Kumsal.
Thank you very much, Gemma. Thank you. Okay, let me sum up Primary Research. Long-term segment growth continues to be driven by volume growth, with increases in number of researchers, R&D spend, and continuing improvement in researcher productivity. We are growing faster than the industry because authors value our quality, reputation, and reach. Our strategy is to ensure that customer value increases continuously.
High quality and trust, more research, faster time to publish, and a better and more efficient editorial process at a lower effective price. We have already gradually improved revenue growth from low single digits to mid-single digits. Our objective in Primary Research is to continue to deliver gradual improvements in growth over the long term. Let me now cover our Decision Support Tools. We provide a broad range of Decision Support Tools to help our customers improve decision-making, increase productivity, and achieve better outcomes.
Collectively, they account for approximately 40% of our revenues. Research solutions are just under half of our Decision Support revenue, and it is a very global revenue base. The use cases here focus on research tasks and management, and we serve two broad categories: tools that help institutions manage and evaluate research and research funding, and tools that support scientific research, publishing, and corporate research. Customers are research institutions in academic and government segments, as well as R&D-intensive corporations. Within this category, roughly a third of our revenue base comes from corporate customers. Nursing solutions are around a quarter of Decision Support revenue. These tools support the whole nursing journey, from helping students build knowledge and prepare for licensure exams, to supporting them as practitioners. This business is primarily in North America.
The remaining quarter covers medical education, physician care, specialist use cases, and patient engagement, split pretty evenly between the U.S. and the rest of the world. Our objective is to increase customer value by improving existing use cases, expanding into new use cases, and embedding where customers work. This improving customer value translates to an improving revenue growth trajectory in the following four ways. The first is by increasing the value of our existing high-value tools. We are doing this by adding an AI layer to many of these products. Scopus AI is a good example of this. Revenue growth increases as more customers opt into these higher-value capabilities, and there is a spend uplift when they opt in, usually in double digits. The second is when we launch a completely new product.
These offerings extend to new use cases as well as go deeper in existing use cases by drawing on the full strength of our capabilities. LeapSpace is a good example of this. These are new propositions that drive measurable value and improved outcomes for our customers. They also reinforce the value of our existing products and drive greater usage of both our partners and our content. These new products are priced by customer segment based on the value they provide. When our customers decide to buy these products, the additional new business improves our revenue growth. Third is when we bring together different products and capabilities in our portfolio for a specific user segment and make these capabilities interoperable, easier to use, and easier to integrate using AI. Our nursing education suite is a good example of this.
Revenue growth improves as customers buy more products from us, as they see additional value in the combined interoperable suite. Fourth, we're also increasingly integrating where customers work. The combination of customers' data and our own content and data create higher value. For example, in corporate R&D, we integrate our products into our customers' agentic environments. These are machine-to-machine value propositions, similar to our Risk business, and are priced based on value.
As we deliver one or more of these value drivers for our customers buy more of our products to capture that value, and our revenue growth improves. We provide higher customer value by leveraging four key capabilities. The first is deep customer understanding and domain expertise. There are several factors here: decades of close partnerships with our customers, deeply ingrained expert communities of editors, editorial board members, and peer reviewers across discipline.
Many of our thousands of employees are researchers, PhDs, doctors, and nurses, so have in-depth understanding of our customers' use cases. Finally, human oversight is ingrained in everything we do, from content curation and integrity checks to expert-led evaluation of machine outputs. The second is leading content and data. As Gemma just shared, our research content is of the highest quality, and in addition to our own content, we collect, license, and curate large content and datasets, including abstracts, full-text articles, patents, grants, policies, preprints, and books.
We continually expand the breadth and depth of our content in science and healthcare. Third is sophisticated linking and analytics. We have extracted billions of data points from leading scientific and medical content sets. We have dozens of taxonomies and ontologies that create structured knowledge graphs and a semantic layer supporting scientific reasoning and evidence.
This has been built over decades with deep domain expertise, and it would be hard and costly to replicate. This third capability enables us to provide benefits to our customers with outputs that have more transparency, more provenance, and more consistency. Given we've already extracted billions of data points and pre-computed responses for a lot of different use cases, it also increases speed of response for our products and reduces costs, including token costs.
The fourth is advanced technology and global platforms. Our architecture is flexible and modular, so we build once and reuse everywhere. We can ingest new content and data and plug in new AI models seamlessly. We're technology agnostic and multi-model. We have used extractive AI for more than a decade and now apply generative AI across our solutions in a way that supports the determinism and the accuracy that science and healthcare requires.
Transparency is built into product design, so outputs are more reproducible, more comprehensible, more defensible, and linked to verifiable sources. Our solutions are private and secure, and we do not use customer data to train models. You will see each of these capabilities come alive in today's product demonstrations. While each capability is critical on its own, it's the unique combination of the four that creates the customer value. I'm going to hand it over to Max now to provide a demo of LeapSpace. Over to you, Max.
Thank you, Kumsal, and hello all. I'm Max Khan, Senior Vice President, responsible for our Research Solutions. I joined RELX in 2007, and over the last 15 years, I've had the privilege of serving our customers at Elsevier. We support the entire research journey by helping funders and policymakers decide where to invest and track outcomes so they can show taxpayers the value of their investments.
Academic leaders shape their strategy, attract funding, manage faculty recruitment, and serve their communities. Researchers develop research questions, review the evidence, find collaborators, plan and conduct studies, and write and publish their findings. Corporate researchers apply research and R&D to support their innovation. Our solutions help people we serve across the entire research ecosystem get to better outcomes faster and in ways they can trust. As Kumsal described, our research solutions improve outcomes for existing tasks and support new ones.
LeapSpace is one example focused initially on researcher tasks. It helps them complete a wide range of tasks in one workspace, including developing ideas, planning research, and writing up findings. LeapSpace is Elsevier's AI-enabled research workspace. We launched it in January 2026, building on ScienceDirect AI, which we introduced in March 2025. We added significant new capabilities in June and September and will continue to expand them rapidly. LeapSpace helps researchers by improving outcomes for tasks such as exploring ideas, planning research, finding funding, and collaborating, all in one workspace, drawing on comprehensive full-text content from Elsevier and other publishers, Scopus data, funding data, and content that users can bring in themselves, letting researchers check the evidence behind outputs, judge how much confidence to place in them, and decide what to investigate next.
We will work closely with researchers to keep improving LeapSpace, adding capabilities, and supporting more tasks and types of users over time. I'm now going to demo LeapSpace. Let's follow a cancer researcher through three tasks: understanding existing research, comparing approaches for an experiment, and strengthening a paper before publication. I'll explain what's happening on the screen for each of the three tasks and how LeapSpace helps the researcher get to a better outcome on each task.
In the first task, the researcher wants to understand a treatment pattern for a particular cancer. They enter that question into LeapSpace and can also add their own materials. Here, the response brings together evidence from full-text peer-reviewed content from Elsevier and other publishers, as well as data from our Scopus database. LeapSpace is private and secure, and we do not use customer data to train models.
Customers have told us that they value LeapSpace for bringing comprehensive peer-reviewed content and responsible AI together in one place while supporting research integrity, transparency, and confidentiality. Now that LeapSpace has provided a response, the researcher can check the evidence behind a particular finding. This panel, which we call a Trust Card, shows the study, who carried it out, how it relates to the finding, and the exact passage behind the finding from a 10-page research paper.
Researchers tell us that Trust Cards help them judge how much confidence to place in the finding, where to dig further, and how to use it. As they consider the response, a researcher might want to check whether the wider research supports a particular claim before building on it. This view brings together studies that support a claim, show mixed findings, or contradict it, along with a summary explanation.
We call this Claim Radar. Gathering these studies can be time-consuming. Here, the researcher can go straight into examining the evidence. Researchers tell us that they value being able to check references and understand conflicting findings. This supports their critical thinking and scientific judgment and keeps them in control of how they use the information.
Both Trust Cards and Claim Radar draw on scientific content and data we have already organized and linked to knowledge graph, connecting scientific concepts and evidence. This helps researchers trace findings to their sources, while reusing the work already done helps us respond faster, more accurately, with lower and more predictable processing costs. In the next task, the researcher wants to plan an experiment, for example, to test an approach that targets cancer cells. The response brings together approaches used in previous studies, so the researcher can begin comparing how those experiments were carried out.
Here, the researcher can compare studies side by side, including their methods and test conditions. This helps them spot differences that may explain different results and decide which approach and conditions to test next, or how to best set up their own experiment. They can also add study details for closer comparison. Researchers tell us that comparison work, which used to take days or weeks, can now be done in minutes.
In the final task, the researcher is writing a paper. They need to explain how their finding fits with existing research and make sure their argument is well-supported by evidence. First, LeapSpace checks the researcher's argument against literature. LeapSpace then shows where each claim in the argument is supported and where evidence might be missing or needs further consideration. This helps the researcher decide which claims might need more support before publication.
LeapSpace can then help strengthen the argument with further evidence. The researcher chooses to accept or reject every change. LeapSpace never silently rewrites the manuscript, and the researcher stays responsible for their work. A researcher once told us this writing support helped her strengthen her research proposal, surfacing papers she had missed and highlighting arguments that needed more support. Her proposal has been approved, and she's now writing a detailed plan for carrying out her study. LeapSpace integrates with Microsoft Word, so researchers can check claims using LeapSpace while they write. This is how we bring our tools into the places where customers work. Researchers must still follow their journals' and institutions' AI policies. As Gemma mentioned, we're also in the process of rolling out Claim Radar and other LeapSpace capabilities to editors, authors, and peer reviewers.
This helps them examine the evidence behind the paper's claims and identify any points that need closer review. You have now seen the four capabilities that Kumsal described working together. Deep customer understanding and domain expertise shape the tasks we focus on. Leading content and data provide the evidence. Sophisticated linking and analytics make that evidence traceable, transparent, and explainable. Advanced technology and global platforms let us choose the right AI models for each task and bring these agentic capabilities into one workspace and into tools such as Microsoft Word.
Together, they help researchers get to better outcomes faster while keeping scientific judgment with the researcher and the researcher in the driving seat. We will continue to work closely with researchers to improve how LeapSpace supports their work. We are seeing very positive feedback from customers on the value we bring in three primary areas. First, better outcomes on important research tasks.
Researchers tell us LeapSpace helps them strengthen their work. Almost 90% reported help in formulating or refining research questions and hypotheses. Second, greater efficiency. In a survey of users, 97% reported time savings, and more than half reported time savings greater than 50%. Researchers described reinvesting that time in other tasks such as ideation, deeper reading, and actually doing the research. Third, trust and control. There is a trust gap for researchers using AI tools. Our industry surveys show that more than 80% of researchers use generative AI, but only 22% trust it. You have seen how LeapSpace bridges the trust gap by allowing researchers to inspect the sources behind the answers, examine the supporting evidence, and calibrate their confidence. More than 85% of researchers using LeapSpace tell us that the answers are supported by robust evidence.
The way that customer value drives our revenue for LeapSpace is that we sell it as a standalone product, primarily a new subscription to institutions. In academic and government, pricing is based on research intensity, the number of researchers, and geography. For corporate customers, the pricing is value-based. LeapSpace was only launched a few months ago, so it is still small compared to our overall revenue.
Academic institutions typically have long sales cycles that are usually tied to 12-month academic calendars when budgets are set for the upcoming year. As you know, LeapSpace has not been in the market for a full calendar year yet. Corporate customers typically have shorter sales cycles and faster decision-making processes, and we are seeing faster take-up. Across these customer segments, we already have roughly 1,000 institutional customers who have subscribed, and monthly active users and usage have continued to grow rapidly in these accounts.
In addition to our paid users, we have been focusing on building awareness. We are fortunate to have millions of touchpoints with researchers via the editorial process and our destination sites such as ScienceDirect. Since June, we have been surfacing LeapSpace capabilities and features to showcase the value in the right moments for researchers. Through these activities, over 1 million users have already engaged with one or more features on LeapSpace. That helps build awareness and extends the feedback on the users and our product experience. Despite the early days, we are very excited about the future opportunity. Now over to Paul for the next demo.
Thanks, Max. I'm Paul Crockett, and I'm the Senior Vice President for our Healthcare Education business. I've been working in education publishing and technology for over 30 years, and prior to Elsevier, I was the Chief Executive Officer and founder of Authess, an ed tech startup acquired by Elsevier in 2020. Our nursing solutions serve educators, nursing students, and nurses throughout their careers, providing solutions to support nurses as they build foundational knowledge, prepare for licensure exams, and finally, supporting them when they move to practice applying clinical skills at the point of care. Our nursing business is primarily in the U.S., and there are close to 5 million nurses in the United States. With increasing demands on healthcare and a projected 8% shortage of registered nurses by 2028, nursing is a critical education pipeline needed to support labor demand.
Today, I will cover our nursing education portfolio, which is a bit over half of our nursing solutions revenue. We provide a suite of connected solutions that help drive better outcomes and greater efficiency in nursing schools. Foundational to our solutions is our content that is built and continually updated by experienced nurse practitioners and educators. We utilize over 15,000 lessons, 500,000 adaptive quiz questions, 100,000 licensure assessment questions, and more, all built upon our decades-long brand and content strength of nursing education books. These assets ground our AI-enabled teaching solutions, which help students build knowledge, apply clinical judgment, and assess and review their learnings as they prepare for exams. We also offer a suite of predictive analytics that help faculty understand a student's progress and their likelihood of passing the licensure exam.
These are based on proprietary datasets built over many years, and our analytics give teachers an early view on where and when to intervene. It also creates a powerful flywheel. The more of our products customers use, the better our data and predictive analytics become. This helps us deliver better results for the students and a more comprehensive view of a student's learning and performance to educators. This suite of solutions delivers measurable results for students. Students who used Sherpath with our Elsevier Adaptive Quizzing product scored higher on our predictive exams more than those who did not. And students who score above our testing benchmarks on our predictive exams pass the licensure exam on their first attempt 98% of the time. Higher pass rates help nursing schools attract more students, more faculty, and more funding.
Equally important, three-quarters of students who use our simulations improve their clinical judgment skills as they prepare for practice. Sherpath is our leading courseware solution, which faculty use to deliver content and assignments. Two years ago, we launched Sherpath AI. We will continue to build Sherpath AI to be an agentic learning tool that utilizes user data, context, and analytics to anticipate user needs across precise, personalized nursing education use cases. Let me walk you through a demo. First, let's experience Sherpath AI as an instructor. Most nursing instructors are practicing nurses who then became teachers. They understand the nursing, but pedagogy, assessment writing, and lesson design are some of the areas where Sherpath AI acts as their teaching partner.
Sherpath AI is purpose-built for learning, and the first example of this is that an educator could start with one of these three prompts based around their typical needs instead of starting a new conversation from scratch. Let's plan a lesson. Within seconds, we will receive learning objectives, key concepts, and activities, all grounded in Elsevier content.
But it's domain expertise that calibrates the AI to where these students are in their program. Sherpath AI is built by nurses and educators who know how nursing is taught and tested. Hours of work just became an effortless starting point for lesson planning, just as one example. Now, on the student experience, students' goals are to succeed in their courses, pass the licensure exam, and become practice-ready. Licensure preparation is where Elsevier excels. Sherpath AI also has three starter prompts tailored to how students actually learn.
Students can even filter by the books that they use in class so that the chat's answers stay aligned with their course. Content is where we have a real advantage. Answers draw on our signature textbooks, such as Potter Fundamentals of Nursing, which is the standard for nursing students, not the open internet, and every answer cites its source on the page.
Let's break down the concept of how type one diabetes affects insulin production. The response is authoritative and accurate because it's sourced from our trusted content. Each citation links to the exact chapter, section, sentence of the e-book where the response was sourced from. As we scroll down, you'll see that responses are supported by our Osmosis videos, as well as relevant images directly from our textbooks. Explanation, video, and supporting content from one query, all of it is uniquely Elsevier's.
As perhaps one of the best examples of how Sherpath AI is purpose-built for healthcare education and differentiated, we've designed learning tools directly into the application. To show an example of this, I can request flashcards to help me study this topic from follow-up questions. I can take the content with me or work through the flashcards directly in the app. It's the same thing for quizzes. Analytics is where all of this compounds. Ours span Sherpath, Elsevier Adaptive Quizzing, and HESI. So a quiz tells us where a student struggles and what will help. Concept understanding, flashcards, quizzes, study plans, and more. The study loop flows without ever leaving the product. Students will often paste a quiz question in before answering. Sherpath AI recognizes the format and doesn't hand over the answer.
The student works through what each option represents first, and only then can they reveal the answers the evidence most supports. That's responsible AI built into our product behavior. So far, this demo has been showing what we call study mode. Tutor mode moves from ask and answer to a Socratic lesson. It gives the student something to think through, then turns it back with a question, all grounded in Elsevier clinical content. Again, concept, video, flashcards, practice, and guided reasoning all in one session.
We are now integrating our AI chat and capabilities across all of our learning systems, including Sherpath, personalized learning plans, and Elsevier Adaptive Quizzing for a more connected and consistent experience, and to increase value for our customers. Each integration reaches a student at the moment they struggle. That signal improves the next recommendation, which drives more use. It's a flywheel.
The increasing value comes from the combination of educational content with proprietary usage datasets and sophisticated predictive analytics in a suite of products that cover all the critical use cases across the learning journey, leveraging evolving technologies. Our roadmap is about further value enhancement, context-aware AI that gets to know the learner, combined with our leading trusted content. Sherpath AI is a core component of our nursing education offering, and we have seen strong adoption. In 2026, we have already hit 100,000 students, 20% of all nursing students in the United States. As customers see strong value, they buy more nursing solutions. Institutions are choosing Elsevier's products to support all student learning needs across the program. We are seeing a steady increase in sales. Here is an example.
A leading U.S. university used a limited set of our products course by course with no Elsevier contract, and they spent roughly $200,000 with us in 2025. Now, after seeing strong value, they moved to buying the full suite of our modular products on a three-year contract, which is expected to bring in around $1.7 million in revenue next year. This growth does not come from price increases. It comes from expanded relationships as customers see greater value and better outcomes from using a combination of Elsevier products. In our top tier of expanded relationships, we see spend per student more than double. I hope that gives you an overview of the nursing education suite of products. The other half of revenues here comes from solutions for nursing practitioners.
I am not going to cover that today, but we have a similar approach there and have recently launched a specialized nursing AI solution that supports nurses in hospitals at the point of care. With that, I will turn it back to Kumsal.
Thank you, Paul. I was watching a Sherpath AI session with a student a few months ago, and at the end, he said, "I love you, Sherpath AI." Let me now summarize the key points for Decision Support. Growth in global research drives increased demand for trusted evidence-based tools. Our strategy is to increase the value of our Decision Support Tools by improving existing use cases, expanding into new use cases, and embedding where customers work. As we add more customer value, there are four primary ways we translate that into faster revenue growth over time. First is adding AI to existing high-value tools, such as Scopus AI, usually for a double-digit spend uplift. Second is launching new products, such as LeapSpace, and driving new business.
Third is building interoperable suites, such as the nursing education suite that Paul just shared, that leads to higher take-up of component tools within the suite. Lastly, it is integrating into customers' own agentic workflows in a machine-to-machine way, priced on value similar to our Risk business. Our objective for Decision Support is to continue to improve customer value and to continue to deliver gradual improvements in our long-term revenue growth trajectory with the increasing pace of product innovation. We drive innovation through the combination of our four key capabilities: domain expertise, leading content and data, sophisticated linking and analytics, as well as advanced modular technology platforms. Now, each of these capabilities are critical, but it is the combination that actually creates the customer value. Now, I am going to hand it over to Jill, who will bring these capabilities to life.
Thanks, Kumsal. All right, the fun part. Good afternoon. Thanks, everyone, for joining us. My name is Jill Luber. I am the Chief Technology Officer at Elsevier. I joined Elsevier almost five years ago, and that is after spending 20 years in technology at our Risk division. I worked in our data engineering and entity resolution space, where I helped create identity linking technology, which allows us to accurately link billions of data sources.
I now work closely in partnership with our Chief Product Officer, Jamie Buckley, who, after a decade with Legal, also took an STM responsibility over a year ago. I am going to start with the role technology plays at RELX. You saw this slide in a May Risk seminar presented by Vijay Raghavan. I want to start here to reiterate that technology is a key enabler of innovation at STM.
At RELX, we have more than 12,000 technologists worldwide. RELX spends $2 billion on technology annually, and we are able to leverage these resources, capabilities, and infrastructure at STM. We help build new products, as well as data and technology platforms that fuel all data and analytics solutions. We ensure these platforms are reliable, scalable, and secure. We use machine learning and AI capabilities to improve our offerings, enhance customer value at scale.
We extract context and insights at scale from scientific content. Billions of records of structured and unstructured content come into our infrastructure. We use AI and other advanced methods to extract entities and relationships. It is not unlike what we do at Risk, but it is more complex. Most of our data is unstructured, and we have dozens of entity types. It is a number that continues to grow.
This creates a powerful evidence-based layer on which to apply generative AI. I am going to highlight some of the key capabilities that are driving customer value. I am going to start by zooming into our content and data layer for a minute. We have the leading scientific and medical knowledge base, proprietary content of the highest quality, as well as decades of aggregating, licensing, and curating highly technical databases.
We extract billions of data points from that content, citations, authors, research concepts, chemical compounds, substance reactions, tables, images, and all of this creates a rich layer of derived data that fuels our analytic solutions. We then tag, link, and validate that data using expert-informed taxonomies, ontologies, so our entities are normalized, disambiguated, and put into context. That is what matters in science and health. We then build semantic relationships across disciplines to connect concepts, evidence, and knowledge.
That connected context feeds our multi-model architecture. Specialized agents handle discovery, synthesis, analysis, and Decision Support. It is orchestrated across workflows with human validation, transparency, and feedback built in. We have built foundations that assume constant change in enabling technologies, in models, in retrieval, and in agents, as well as change in the customer expectations.
We keep improving without starting again each time as the technology moves. The impact for customers is better outcomes, outputs that are more accurate, relevant, and verified. Let me drill down just a bit deeper on how this works and why it is important in a generative AI world. Generative AI is probabilistic. Our customers need answers which are deterministic. They operate in regulated and critical areas, scientific research, drug discovery, patient care. A single compound may have multiple names. Almost identical names can be completely unrelated.
A single data point has different relevance depending on the context. It could be toxic in one context and not in another. AI agents can answer science questions competently, but a scientist can only act on an answer they can check. Being probably right and different every time you run a query is just not good enough. The answers need to be accurate, verifiable, and have clear provenance. We add curation, standardization, and traceability that makes the answer checkable.
This slide shows how we do that. Let us take a chemistry or biomedical question. First, we resolve. We align identities, synonyms, registry numbers, and nomenclature, so the system knows when three compounds are actually the same name. Then we retrieve the relevant chemistry, pharmacokinetics, safety, and regulatory evidence from our specialized databases, Reaxys, PharmaPendium, and Embase, so the data always comes into context. Then we harmonize.
We normalize units, parameters, ontologies, protocols, and study context. Finally, we deliver structured, analysis-ready content and data. We deliver evidence, not answers. We do not decide whether a compound is safe. We make the evidence easier to inspect and leave the judgment to the scientist. The customer value here is consistency, transparency, and traceability, one scientific task delivered the same way every time. We deliver this value in workspaces, such as LeapSpace, as well as point solutions. The same approach is used when we embed into customer environments.
As an example, we support our sophisticated corporate customers who use language, machine learning, AI to build high-value applications, knowledge graphs, synthesis planning, toxicity predictions, integrated into their systems similar to Risk. This allows customers to combine their proprietary data with ours, unlocking new customer value. As corporate customers adopt generative AI and build agentic platforms, this becomes even more important.
A recent webinar, a leading pharma customer described how using our solution allowed them to achieve 25%-30% better quality and precision. Higher precision also means lower costs, so they were able to reduce their context token cost by 30%. I spoke about the importance of our content and data, and now let me bring in domain expertise and our AI approach to complete the picture. For the quality science demands, a simple retrieve and summarize approach is just not enough. We need to work in three stages. First, we understand the question, its intent, concepts, and constraints. We retrieve using lexical, semantic, and structured search. This includes citation graph and ontologies. We search across text, abstracts, tables, figures, and equations. We can then answer with claim to source citations that explain why the evidence was chosen.
That explainability and repeatability come from our taxonomies and databases, not from the model. The same approach is applied to both insights and tasks that need an evidence-based approach. The next step is agentic. Rather than one pass, the system plans and splits the question into sub-questions, retrieves across the data sources, scores its own competence, names the gaps, and can re-query up to three rounds. It will not respond until confidence thresholds are met. If the evidence is inconclusive, that will be transparent to the user. In scientific and health use cases, an honest but inconclusive answer is better than a confident but oversimplified one. This is a deliberate design choice. We build on frontier models from several suppliers, including Anthropic, Google, and OpenAI, as well as other open source and open weight models like Mistral. Models move fast.
Providers are leapfrogging each other and prices shift, so we've architected for change. This week, we have more than 120 approved models across the major providers and choose the right one for each task on quality, latency, and cost. This slide shows the loop. We benchmark, measuring quality, latency, and cost across suppliers, model families, and hosting environments. We optimize. Prompts and configurations are generated for each model family, and they are versioned, traceable, and repeatable. The prompt layer is where our domain knowledge is encoded, and we monitor in production with observability, traffic routing by customer, product, and if there's an outage, we fall over to a backup model. Drift and regression feed the next benchmark, and the loop reruns automatically when a new model is released.
The business point is that the model choice becomes a commercial decision rather than a technical dependency, and switching suppliers is a configuration change, not a rebuild. On Sherpath AI, model and prompt optimization decreased token cost by nearly 50% this year, while usage and users had a strong increase. We also pre-compute based on our knowledge graphs and already extracted data points. For example, LeapSpace Claim Radar is a pre-compute feature that enables us to manage cost of tokens and have predictable costs for our solutions. We pair this with expert human evaluation. This year, we have already run more than 20 evaluation cycles with real domain experts checking outputs. We have a hybrid approach of automated benchmarks and human in the loop, and that's why we talk about precision AI, not just AI.
Technology also helps us operate more efficiently, which is how we manage cost growth below revenue growth. We've done this through continuous process innovation, and increasing the use of AI is speeding this up. In technology, AI-powered spec-driven development is what improves our code quality and shortens our development cycles. Our reusable modular platform is AI-enabled, and we launch products faster while improving our security posture. Over the last five years, I've seen a significant improvement in our agility, and we will continue to accelerate going forward with new processes, methods, and technologies evolve. In operations, automated reusable content pipelines let us process growing volumes at lower costs. AI-assisted evaluations of outputs keep accuracy and quality high as we scale. In customer support, AI agents improve response speed and consistency. AI-driven case routing and prioritization make the team more efficient.
In sales and marketing, AI analyzes customer conversations to surface insights and improve conversion, and also supports the right channel mix. In internal use, we take the same multi-model approach that we developed for our customer solutions, giving us the ability to optimize quality, speed, and cost. To summarize, there are four things I hope you walk away with.
First, we are innovating and developing products faster. We build once and we reuse everywhere. Second, we provide evidence-based answers. Precision AI, Trust Cards, and expert evaluation keep outputs more accurate, traceable, and consistent. Third, our technology flexibility. Our agnostic and multi-model approach means no lock-in, and we move to the most appropriate models as the market evolves. Fourth is cost efficiency. AI-assisted development, pre-computed content, and prompt optimization reduce the cost to build, run, and support our products. I am going to hand back to Kumsal.
Thank you, Jill. That was the fun part. Okay. I would like to leave you with a few key takeaways as we wrap up. We operate in large global segments with attractive structural long-term growth drivers. We have segment-leading positions serving our customers for decades, and in some cases for centuries, with critical and complex use cases. We deliver increasing customer value through the unique combination of domain expertise, leading scientific and medical content, sophisticated linking and analytics, as well as advanced technology platform. Our objectives continue to be to deliver higher customer value, improving long-term revenue growth trajectory, and manage profit growth ahead of revenue growth. Thank you very much for your time today.
I hope that my team and I have highlighted the exciting progress of the business since we last did a deep dive four years ago, as well as the significant opportunity we see ahead of us for further progress. We now look forward to your questions, and Paul has already joined me to help with the Q&A.
Great. Thank you, Kumsal. We will now go to questions. I think for those online, if you can please write your question in the box provided. I think we will start in the room. If you would like to ask a question, you need to pick up the microphone, which is buried in the seats beside you or in front of you. You must press the button, otherwise you won't be heard. If you could please give your name and organization. If you can get all that, who would like to go first? Right, we will start off on this side. Joe.
Excellent. Thank you very much. Joe from UBS. Hopefully I can show people how to use the microphone. Firstly, just kicking off on LeapSpace, obviously you're seeing very rapid usage growth, but its revenue contribution is obviously not yet material. What needs to happen for the current product adoption to flow through as a material revenue driver? Is it just time and the sales cycles, as you alluded to? When can we expect that to be more material as a factor? Secondly, when you negotiate prices with your customers around analytics and decision tools, I assume the ROI improvements are a major driver of that. Can you help us understand how those conversations go and where the funding comes from? My point here is, this isn't like law, where the marginal hour of a lawyer has a value.
How does that change that dynamic?
Great. Thank you very much. You have two questions. The first one is LeapSpace user and revenue trajectory, and the second one is how we actually show ROI on our analytics products. I'll take the second one, and I'll ask Max actually to take the first one. In terms of the ROI, I think we have highlighted to you today, with the LeapSpace illustration as well as our nursing suite illustrations, how the ROI differs by customer segment.
ROI also differs on whether it's an academic and government customer or a corporate customer. Some of the ROI we deliver is measurable, and Jill, for example, alluded to an example with a corporate R&D customer, where we actually integrate our capabilities to their agentic environment, and they could see 30% improvement in precision as well as 30% decrease in token costs. Some of it is really measurable.
Others are not measurable. It's time saved as well as quality of outputs improving. In each case, I think what we do is continue to show the customer value through better answers, better outputs, faster process, and lower cost to serve. The ROI is sometimes measurable, sometimes it's not measurable, but actually obvious, and based on that, we price the product. Max, do you want to answer the question on LeapSpace? Thank you.
Joe, on your question about LeapSpace and adoption and engagement, there's a couple of ways that we think about it. Firstly, we have almost 1,000 customers across corporate and academic and government using LeapSpace, and we're driving rapid adoption in those customers, good engagement in those customers. Then the second thing is the awareness point. We're using our touchpoints in our editorial systems, in ScienceDirect. There's millions of researchers interacting with our surface, and we're using those touchpoints to surface the value of LeapSpace to researchers to let them taste the value so they're aware of it. And we've got more than 1 million researchers that are now aware of LeapSpace, and we'll keep driving that activation. That's our objective.
The other thing to say is in, I think it was June, July, Erik talked about the monthly active users doubling in a 90-day period between March and June. And we've seen the kind of new monthly active user addition every month continue at a similar rate, if not a bit quicker. Again, the feedback from customers is extremely positive. The engagement is robust and rapid, both across institutions and awareness. The thing I did say is that an academic cycle can take 12 months, and you need one or two of those cycles to really have any confidence in the patterns and the trends. That's kind of where we are at the moment. Very excited about the potential, but still very early days.
Thank you, Max. To summarize, rapid awareness building, and then that actually means more customers, paid customers, and I think it'll be gradual, which is usually how our segments behave.
Ciarán and then Nick. Yeah.
Thanks, Paul. Thanks, everyone, for the presentation. Couple questions from me. Ciarán Donnelly from Citi. One just on the Decision Support Tools market size. You talked about corporates and academics. If you could just give a sense of maybe, I know it is nascent, but maybe market sizing, market growth, just to help us contextualize the opportunity there. Separately again on Decision Support Tools, you talked about the use of proprietary and licensed data. Could you just give us a broad split across the Decision Support Tools in terms of use of proprietary and licensed data? Thanks.
Okay. So your first question is Decision S upport market size, and the second one is Decision Support proprietary versus licensed data mix. I will take both, but my wonderful team here will jump in if I missed anything. In terms of Decision Support market size, I think what makes it a little tricky is we have a very large portfolio of Decision Support Tools. You have seen that. It is about 40% of our revenue. About 50% of that revenue comes from research solutions. About a quarter comes from nursing solutions. Another quarter comes from medical education, patient engagement, physician point of care.
So, the market sizes really differ based on which segment we are serving. I think in all of these cases, what we are finding with our Decision Support Tools is when we add customer value, we are able to actually translate that into revenue growth, and we do that in four primary ways.
One is when we add AI capabilities on our existing high-value Decision Support Tools, we are usually able to get an upsell. The second one is when we launch a brand new product like LeapSpace, that actually generates new business for us that turns into revenue over time. The third one is when we actually bring together a suite of solutions, similar to our nursing education business, where the suite itself, the integrations around it, and the interoperability drives more adoption of the individual component product. The last one is integrating our capabilities into corporate agentic environments. So, that is the way we think about delivering customer value and turning that into a higher revenue growth trajectory instead of sort of market sizing. So hopefully that helps a little bit.
In terms of our Decision Support Tools and how much of it is proprietary versus licensed data, it is really a mix based on which segment that we focus on. If you focus on nursing education suite that we showed to you, that is all proprietary. That is all our content. If you look at LeapSpace, that is a combination of proprietary and licensed content, both through Scopus database with abstracts, as well as full-text articles that we have from more than 1,500 publishers so far.
I think what is really equally important is not just the proprietary and licensed content combination, it is actually the insights we drive from them by extracting billions of data points and linking them through our knowledge graphs, our taxonomies, and our ontologies, which creates an evidence layer that is also an equally valuable layer. Thank you. Did I? Okay.
Yeah. Nick Dempsey from Barclays. I have three questions, please. First of all, if your Primary Research is growing mid-single digit, then I am kind of backing out 7%-8% or so for the rest, the Decision Support, the other 40%. Has that stayed the same in recent years, or is it slowing a bit as it gets bigger? Second question, I guess with the abstracts in Scopus, you had originally permission from all the other publishers to create abstracts of their articles, and it is your content. But now that you are getting extra value out of that in something like LeapSpace, is there a chance you have to go back to all the other publishers in order to renegotiate that, or will they want to do that? Because it seems to me like a different layer of value now.
Third question, across the different areas of Decision Support, are you starting to see new AI-native competitors or adjacent players emerging? Are your customers having to split their budgets across a load of new offerings as well as yours, even if those do not replace you?
Great. Three questions. Primary Research and Decision Support growth, Scopus abstract values as we use abstracts across new products, and then the last one was about new entrants and competitors. I will take the first and the third, and Max, I will ask you to take the second, if that is okay. Let us start with our growth profile. When we actually presented four years ago, we said that our growth had improved from 2% to 4%, and now our growth has improved from 4% to 6%. That is driven by improvement in both Primary Research as well as our Decision Support portfolio.
If you look at Primary Research, it is really high-quality volume growth that actually delivers growth there, and our objective is to continue to gradually increase the long-term revenue growth trajectory in Primary Research as volumes continue to increase.
On Decision Support, when we presented to you four years ago, we said Decision Support was growing in. Going back to Primary Research, when we actually presented four years ago, we said Primary Research was growing in low single digits, and today we said it is actually growing in mid-single digits. Going to Decision Support, when we presented to you four years ago, we said that growth was in mid to high single digits, and today we said Decision Support portfolio is actually growing strongly.
Our objective for Decision Support portfolio is also to continue to gradually improve the revenue growth trajectory there. If you look at that improvement from 4% to 6%, it has come actually improvement from both our Primary Research as well as our Decision Support revenue base. I will answer your third question, and then I will give it to Max on Scopus.
In terms of new AI competitors, I think there is a space for a lot of different products in the segments that we serve. There have always been lots of different providers, including startups, traditional competitors, and we see that today as well. We have not seen that impact our customer spend with us so far.
Maybe I'll answer the licensing question in a concrete example of LeapSpace. In LeapSpace, we have two primary data sets that we're effectively building the value on. One is the full-text content data that Elsevier has, and also other publishers, but also the Scopus data. We basically have continued to expand the amount of content that we license from other publishers. Where the conversations are is really licensing the full text from other publishers in order to use it in LeapSpace. Since we started, Elsevier's share of content is around 20%, and its share of citation, which is a measure of impact in science, is around 30%. In LeapSpace right now, we have content that represents 56% of the impact across science right now, and we've been doing that by licensing content, full-text content from publishers.
Again, we'll continue to have active discussions with publishers and bringing content into LeapSpace. Our goal and our objective across RELX and also across our solutions is to grow revenue ahead of costs and grow operating margin ahead of revenue. That's kind of how we think about that.
I think it's important to say that we always want to deliver value to our content partners. We've been doing that with Scopus for a number of years, and now we're actually delivering more value for their content through LeapSpace, which means more visibility, more usage, more people actually using the content in their scientific research.
I'm just going to go to a couple of questions online before we carry on in the room. The first question is, or the first topic really, is about if we can talk more about evolving customer value in Primary Research and the sustainability of growth, and how quickly we can improve growth in Primary Research, particularly as the model continues to migrate to open access. Then another question or questions around our approach to M&A, in light of discussions around certain assets potentially becoming available, and would it change the competitive dynamic if consolidation happened elsewhere?
Great. Excellent. So, I am going to give the evolving customer value in Primary Research to Gemma, and I'll start with approach to M&A. I'm sorry, I keep changing glasses because I'm at that age where I can't see far and I can't see close, so I need all kinds of glasses. So, our approach to M&A, as you know, at RELX, our strategy is through primary organic development. That's always our preferred methodology. Our strategy at Elsevier is also through organic development as the primary vehicle. In our Primary Research business, from time to time, we'll do small acquisitions to actually fill a gap in a particular discipline or a particular tier, but our strategy continues to be driven very much by Primary Research.
In terms of whether further consolidation would change anything from an Elsevier perspective, I would say though, because we have really large scale is important in Primary Research, and we already have very large scale. Primary Research, over to you, Gemma.
Thank you. So, we think about customer value in Primary Research, there are a couple of lenses which are mutually reinforcing. So, if you think about the researcher and what they care about, the researcher cares about being able to submit to a journal that they can trust, that delivers high quality, that provides a home for sound science.
We also think about that and the ways in which we're continuing to enhance the author experience, the way we're building out our journal portfolio, the quality differentiation between us and other publishers, continuing to deliver value there. But of course, that piece around research integrity and the importance of research integrity. We talked about the investments that we're making in screening papers, pre-submission, all the way through to publication, and that importance of both quality and trust to deliver value to researchers.
And then in turn, the flip side of that for our customers who purchase content from us, who license content from us, that's around being able to feel confident that they're accessing the highest quality research that they can trust and have confidence in going forward.
Thank you very much. To add on to Gemma, as you know, we've shown you today that R&D spend has consistently grown at 5% for decades. Number of researchers have consistently grown at 5% for decades, and researcher productivity continues to increase every time there's actually a significant technology improvement. All of those drive more research, more findings, more knowledge, also the need to be able to understand that knowledge more quickly and to stay on top of it. So that drives both volume in our published articles, numbers in Primary Research, as well as demand for Decision Support Tools where you want to be able to apply that knowledge into the different scientific cases you're working on, whether that's in a corporate setting or an academic and government setting.
Will, and then Steve.
Hi, it's Will Packer from BNP Paribas. A couple from me, please. Firstly, the performance of the Decision Support Tools is very impressive looking backwards. Going forwards, are we entering a new era as the AI labs verticalize with products such as Claude Science, which will compete for spend there, and could that weigh on growth? On the other hand, could you help us think through the extent to which your view is that they won't enter the space or that your products will be superior? Secondly, like other info services providers you're talking to, content differentiation is an important moat. For STM, there's a particular type of content, right?
Does the history of political scrutiny of STM, where governments and universities heavily fund both the research and the journals, bring a different dynamic, and stopping the integration of journal content with third parties would be more challenging? Any perspective there would be very helpful. Thank you.
Yeah. I think I am going to take both of your questions and combine them, and the team can actually add anything I miss. In Decision Support, the customer value we deliver really is actually about combining four key capabilities that we have. The first one is domain expertise, where we have real deep ties to scientific and health communities, and we understand their day-to-day work. That is what enables us to actually deliver use cases that really are applicable to those customer segments. So Trust Card and Claim Radar are examples of that. That is really important.
The second one is content and the breadth and the depth of our content. We have, again, decades of having proprietary content as well as licensed and curated content, and we have the largest scientific and medical content base in the world. We are leading in that case.
The third, which is equally important, I hope that came across in Jill's presentation, is the analytics we have built on this content base over decades. So that is extracting billions of data records from the scientific and medical content base and really creating an evidence layer. Then the last is our modular approach to technology, which is technology agnostic.
So when we focus on use cases, all of these capabilities are critical, but it is the combination of the four that enables us to deliver customer value in science and medicine, where we can actually give better answers, better transparency, better consistency, better repeatability, better reproducibility, as well as a lower cost to serve with predictable subscription pricing, because we are able to pre-compute a lot of different use cases and also optimize our prompts based on the domain expertise. So that is really the value that we are delivering.
LLMs, I think, and AI has been really great for science, and they are going to continue to actually deliver more science activity, and that is a good thing for us as well. For our business, that is a driver. In terms of content differentiation, we have both proprietary as well as licensed third-party and open access content. So when content is available for different use cases, we are also able to take advantage of that.
But it is not just the content aggregation, it is the sophisticated analytics layer that we have built on it that actually has created the linking across all the different, whether it is chemical reactions, whether it is claims, whether it is assertions, all of that sophisticated linking actually is what drives a major part of the value as well. So I hope that gives you a sense. Thank you very much.
Time for a few more questions. Steve?
Yeah. Hi. Steve Liechti from Deutsche Numis. Just on open access, can you just talk about that in terms of the growth rate of Primary Research in the last four years? How much of that growth is the shift from subscription to open access? How much of that is your ability to leverage article processing charges? I just wonder if that might flip the other way, and we've seen some pressure from the U.S. commentary in terms of article processing charges there, so anything there would be useful. You may or may not want to answer this next one, but what's your definition of strongly in numbers?
In terms of definition of strongly in numbers, I will leave that to our wonderful investor relation teams that can give you the lexicon, I'm sure, over drinks. In terms of the Primary Research growth, I think what I'm going to ask is Gemma to help with that question and tell you how Primary Research growth is actually driven in our business.
Open access makes up just over about a third of our articles today, but we are business model agnostic. What's important to the researchers that submit to us is that they can trust in the quality and rigor that we provide. What's important to the customers who purchase from us is that they're buying a corpus of content equally that they can trust in and have quality around.
We work with our customers on a range of models, pay to read, pay to publish, a combination of both. To the question on different things that happen in different markets, be they policy or regulatory, we work in 100 of our 170 countries. In any given year, there's inevitably different things that pop up, but we have always been really well-placed to navigate those and don't see any issues with navigating those going forward.
Thank you, Gemma.
Henry, and then George.
Yeah. Hi, it's Henry Hayden from Rothschild & Co Redburn. Two questions from me today. The first is on the corporate opportunity. It seems very attractive given the quantum within R&D expenditure and the growth profile there, but it still represents a relatively small portion of your revenues. How do you think about the positioning of the portfolio to address that market? Are there any shifts that would be needed to tap into it more aggressively? The second question is on Primary Research. You mentioned the objective to go publish articles ahead of the market, which seems to have been working particularly across the last few years. Is this primarily driven by a flight to quality? If so, how do you think about the durability of those market share gains? Thank you.
Thank you. I am going to give the flight to quality question to Gemma in Primary Research, and then I will take the corporate question you had. Corporate is a very attractive opportunity for us, and I think I mentioned today that within our research solutions, about a third of the revenues are coming from corporate market. Our strategy is to continue to pursue that opportunity organically, primarily, but if we do find acquisition opportunities that are good add-ons that actually fill our strategy, as always, we would consider those opportunities as well. That is really primarily an organic strategy for us in corporate. Gemma, over to you for Primary Research.
Our published article growth is ahead of the industry, but you will notice also that it is below our submissions growth, and that is really by design because we have that rigorous process in place and are very selective and have things in place to make sure that authors can feel confident and comfortable that when they are publishing with us, there is quality and integrity there. We talked about the fact that with the rise of AI and generative AI bringing many benefits for science, there is also that increase in research integrity issues, AI slop. There is potentially there the opportunity for authors to think even more carefully about which journals they choose, and they will look at markers like our Field-Weighted Citation Impact and where we sit.
In terms of thinking about the long term, I think Kumsal talked about the underlying growth drivers of the business, again, in terms of R&D spend, number of researchers, researcher productivity, all of which give us thinking optimistically about the future.
Yeah, I think in Primary Research, our strategy is very much around high-quality volume growth at a lower effective price unit, and we have a real long-term view here. Science is a long-term game, so when you work with researchers who are serving the scientific community, you need to have a long-term perspective as well.
George.
Yeah. Hi, it is George Webb from Morgan Stanley, and thanks, Kumsal and team, for the presentation. I guess two last questions. Firstly, when it comes to embedding AI in your Decision Support Tools, and I appreciate it is Decision Support, not automated decisions, and you have talked about the different ways you enrich the professional with the different aspects of that to make them or help them make the right decision. But how do you kind of manage the residual liability that AI might represent, and how do you think about who that liability might fall on within that kind of ecosystem of parties? Is it the model provider? Is it you? That would be interesting. Then secondly, just more of a straightforward question.
On the Primary Research portfolio, where are you on turnaround times at the moment, and do you have an objective for what you think those can get to over the coming years? Thank you.
Thank you. I am going to give the turnaround times in Primary Research to Gemma. On Decision Support, the residual liability is managed in two ways. One is actually through our technology approach, so I'm going to ask Jill to talk through how we continue to improve the precision of our answers with the technology approach we have. It's also managed through contractual agreements. We have been in the business of providing information to researchers and physicians for decades, and there's always contractual. In our agreements, there are contractual. If I can just think of the word. There are contractual clauses that actually make sure that the physician and the researcher is in charge of the decision at the end of the day.
That's sort of a contractual approach, but obviously, we never want to put the physicians or the scientists in that position, and to make sure that our precision is incredibly high level, we have a technology approach for that in addition to our content, and then Jill can help you with that. Yeah.
Sure. The key to our monitoring precision on our answers is in the harness that we build, coupled with domain expertise and experts who are looking at the results that we provide. We couple both automated benchmarks as well as domain-specific and domain expert results. The way we do it is, I talked about this harness. We optimize the prompts because you need to understand the prompt going into a model so that you get the most out of it. That is optimizing for accuracy, is number one, and then latency and cost. We do that for every model family as well as across any model hosting. It's the ability to take the benchmark amongst some subject matter experts and really measure against things like accuracy, completeness, correctness.
In our health space, one of the things we measure with actual MDs is, would this answer actually do harm to a patient? That is one of the top benchmarks that we want to make sure we do not drop in our accuracy on, would this answer actually do harm. So we are taking that then curated domain-specific benchmark, automating it, so when we make any kind of model change, we are modeling against that benchmark again for completeness, correctness, accuracy, doing no harm.
Thank you, Jill. It is also product design because you have seen the traceability that we provide so that the users are always able to check the primary source to make sure that they understand where that information is coming from and be able to calibrate the strength of the evidence and how much they can trust it. Then Jill also mentioned that we rather give an inconclusive answer versus an oversimplified yes or no answer. So we also address that through our product design as well. Oh, yes, Gemma.
Speed is important to us because it is important to our researchers. They want their findings to be published, and they want that to happen as quickly as possible. But we are really careful to do that while at the same time upholding those quality and integrity standards that our brands are known for.
Some of the things we talked about are the investments that we are making in the peer review experience, in the author submission experience, but also some of the tools that we are going to be rolling out for editors, all of which are designed to save time in editorial decisions so editors can be freed up to actually look at and do their job, and we can automate the things that can be automated. So it is that balance of speed, but also making sure we do that while maintaining the quality and integrity that our customers appreciate.
Thank you, Gemma. Our objective is absolutely to increase the speed of the process, and we have done that historically as well. We really look at that submission process, then the editorial triage, then the peer review process, and then the production. Each of those processes have different time frames, and we try to cut days out of that consistently.
Now, not because I like to have the last word, but we have had a late flourish online, so I will ask you a couple of questions to finish off.
Okay.
One from some friends in Canada. "Will adoption and usage of LeapSpace be broad-based across the research community, or will it be concentrated at the high end of research?" That's the first question. A number of people have asked us to expand upon the machine-to-machine sort of risk-like integration that you talked about in Decision Support. Can you provide some examples, and is that a meaningful driver of growth going forward? Finally, could you just talk a little bit more about how AI changes science and how we benefit from this changing dynamic, and does that extend the opportunity beyond research for us?
Okay. I will ask Max to pick up the adoption of LeapSpace question. I'll take the machine-to-machine opportunity in corporate as well as how AI changes science, which is one of my favorite topics, and what that means for Elsevier. Let me start with the machine-to-machine opportunity. That is a very exciting opportunity. It's early days, and I think you're all familiar with our Risk business. The best way that I can describe it to you is we have a lot of different data sets that we have built over the years. In Risk, we have attributes, and in these data sets, we have what we call connectors.
These connectors make it easier to actually integrate some of the AI precision and the data sets we've built in chemical reactions, in different assertions, et cetera, to an agentic environment and improve the precision as well as the cost to serve for our customers' agentic environments. It is early day. It is an exciting application. It's also early days for our corporate customers where they're building these agentic environments, and they're using it in their R&D processes. More to come on that. In the early days, similar to LeapSpace, we're very excited about the opportunity. Jill Luber, would you like to add some more to that?
I do. Yes, please. I did spend a lot of time in the Risk business. I just wanted to call out because I did have a conversation with a customer CTO. I was talking about the machine-to-machine opportunities. I know that this was a customer in the life sciences space that was getting direct data from us, full data feeds, because they were subscribing to a journal. The CTO actually said, "You know, it would be more helpful if you gave me, like you said, this when you've already got your knowledge context on top of it and your solution came machine-to-machine with that knowledge, that would really help me out. I don't want to have to process the data. I don't want to have to build that knowledge graph on top of what you already know.
It's more valuable if you can give that to me directly into our workflows.
Great. Thank you very much. Hopefully that creates some color. In terms of how AI changes science, let me first talk about the business side of it and then talk about AI changing science. From a business perspective, the four key capabilities we have, which is the domain expertise, the leading scientific and medical content, the sophisticated analytics and knowledge graphs that we build on this evidence base that actually improves decisions, as well as our modular technology approach, are, I think, equally valuable in a machine-to-machine environment, and we're showcasing that in Corporate already, as well as in a machine-to-human environment. From a business perspective, we're excited about AI and the changes that it's going to bring. We'll continue to serve the use cases that we are able to serve with our asset base going forward as well.
In terms of how AI is changing science, we're really at the early stages of that. When I talk to scientists around the globe, there's a lot of different thoughts that are going around. The first one I would say is, if you look at historical scientific breakthroughs, a lot of them have come from actually betting against conventional wisdom. This goes back to Galileo, which was published by the Elsevier family, when he said that it's actually the Earth that moves around the sun, and it was quite controversial at the time, to recent history when, to Nobel Prize, finally Nobel Prize-winning scientists said ulcers are actually created by bacteria, not by stress, and this is sort of the '80s.
A lot of scientific breakthroughs comes from going against the conventional wisdom, and whether AI models, which were probabilistically, will be able to actually pull that, going against the conventional wisdom and drive the breakthroughs, remains to be seen. There is also, I am sure you have seen, for example, a frontier model has actually had a great breakthrough in a maths problem, in one of the Millennium Prize Problems.
The issue is, we still, even mathematicians, I certainly don't, but even leading mathematicians don't exactly understand how that problem was solved. That's a problem, because when a math problem is solved, it's not the solution that actually drives the scientific process. It's actually understanding the steps and being able to apply that to an engineering problem or a material science problem that actually drives scientific progress. We are still at the early days of AI use.
It's clear that it actually helps speed up the process. It's very, very helpful in computational areas with speed and efficiency. But I think we are living through a fascinating time, and if AI can actually accelerate science, that would be an incredible benefit to all of humanity. I mean, I will give you an example. Statins, which has saved millions of lives. We don't have heart attacks in 50s and 60s anymore because of statin. Statin was actually first discovered, the first step of statin discovery, from the first discovery to the drug was about 20 years. And the very first discovery was in a tiny little Japanese lab, with a researcher who was really, really interested in mushrooms. And over time, many researchers built on that, and we had statins two decades, which now saves millions of lives.
If we can actually squeeze that to be 5-1 0 years, it's really exciting time for both the scientific and the medicine community, but it's very early days for the technology. Sorry, I can talk about this as you can see, for a long time, and my team can talk even more about it. Do you want to add anything, Max? Oh, LeapSpace, yes. Come on up. Yes, we had the LeapSpace question.
On LeapSpace, we are relentlessly focused on driving customer value. And you saw from the demo just three examples of tasks that we are trying to improve for researchers. I mean, there are hundreds of tasks that people do in LeapSpace, and for each one, we are trying to drive a better outcome faster and in ways that they can calibrate their levels of confidence and have confidence in whatever output we are giving. That's the first thing to say.
Secondly, as we have launched it and started to drive adoption, what we have seen is, as you know, LeapSpace is resonating with corporate researchers, and the value is also resonating with academic researchers. And then when we look at things like discipline mix, we see a broad range of disciplines. The value is resonating with a broad range of disciplines and also different seniorities of researchers.
We see the market potential as broad in terms of the value that we can deliver, in terms of the tasks that we can focus on, the disciplines, and the segments of the market that we could serve with the value proposition.
Thank you, Max.
Great. Thank you. I think we will draw formal proceedings to a close. If you do have any follow-up questions, you know where to reach me, James. Do not hesitate to reach out. For those in London, please join us outside for a quick drink. Thank you very much for your interest, and thanks again for coming. Thank you.