Providing data and feeds to third parties has become a major revenue stream for the London Stock Exchange Group (LSEG) over the years and with the rise of AI, that’s set to become even more lucrative, according to David Schwimmer, Interim Group Head of Data and Analytics and Group CEO:
Looking out, our continued investment in high pace of product innovation position us well to strongly compete and continue taking share. An example, in Workflows, we displaced almost 100 users at a major commodities trading firm. That was driven by our superior market data, news and commodity-specific research. In Data & Feeds, we won a highly competitive process to provide a global hedge fund with data and insights powering new AI-driven trading models. Central to this success was our leadership in news, both the breadth of our offering and the investment we’ve made in making it machine-readable and AI-ready.
LSEG is already a global provider of financial markets data and infrastructure headquartered in London, England, as well as owning the London Stock Exchange (on which it is also listed). While the Group has a wide tech portfolio on offer, it’s clear that AI is going to be a dominant force in the coming years:
You have to think about what financial markets participants want from their workflow. Natural language search and the ability to automate a lot of time-consuming activities with agents is very attractive. We’re all headed in that direction. But they want that and all the things they get today from an advanced desktop, curated news and alerts, portfolio tracking, live charts, trading capabilities. So the future is AI integrated into a desktop, not AI replacing a desktop. And that desktop should combine financial markets content, insight and workflow with enterprise workflow.
A couple of other really important aspects to this. Customers also need absolute certainty around data, trust, accuracy and compliance, including the outputs from AI models. And of course, they want simplicity and value for money. LSEG lines up against these needs very well.
Data foundations
It all starts from the Group’s firm foundations in data, argues Schwimmer:
On data, our position is very strong. Our starting point is leadership in real time, unrivaled depth and breadth of data and significant trust in that data, given the rigor of our processes. And then our approach to applying AI to this data is differentiated as well. The truth is that even for relatively simple prompts, accuracy levels across the industry for AI model outputs in financial services remain below 50%. I’m confident this will improve rapidly, but that’s the situation today. As a result, we’re taking a rigorous approach to testing and evaluation to improve and refine AI outputs. And our overall commercial strategy is an important tool.
Through access agreements, we offer lower cost of data ownership from major customers. On top of that, we are more liberal in our contracts and allowing customers to train their own models on our data. Others do not take this approach. Our position is clearly differentiated from others in the market, and we like and have a lot of confidence in our position. We’re also confident that as the world continues to change rapidly and in particular, as Agentic AI functionality improves, we will remain at the forefront of that change with our data at the core.
AI is being incorporated into LSEG’s wider portfolio of products and processes, says Schwimmer, with a focus on how people work:
We have over 20 live use cases in our business today with a further 100 in development. On products….we’re putting AI to work from advanced dealing and the analytics API to the new Workspace Teams app and soon in Workspace AI and in various Agentic AI tools. For our processes, the addition of AI tools is also making us more efficient and agile.
I’ll give you a few statistics. But please keep in mind that while we are moving quickly down the path of widespread AI adoption, it is still relatively early days. More than 80% of customer queries are now resolved using AI customer support tools, that’s already helped us significantly improve resolution times and there’s further to go.
We’re also deploying AI tools to ingest data more quickly and accurately. The sourcing failure rate on our automated data retrieval has decreased by 95%, significantly reducing the need for human intervention.
As for the Human Intelligence side of all this:
We believe it’s important that we in-source more of our engineering talent and train all staff to operate in a modern AI tooling environment. Not only does that ensure we keep winning in the war for talent, but it will maintain our agility and reduce our time to market. And there is a direct line from this transformation to what you are seeing in our margin improvement. We are becoming a modern, more efficient, more skilled and scalable business.
In all of this, LSEG is being essentially LLM (Large Language Model) agnostic. Schwimmer explains:
We work with a number of different models. We’re not tied to any one particular model. We work with open source models. We work with the pay models. And then the second aspect of that is that we also already provide our data to many users for their consumption in their models. And in fact, we have, I think, a more liberal approach. We’re very comfortable with it from a legal perspective and from an economic perspective, but we provide our data in a more liberal way to our customers for their usage in their models. Part and parcel of this is making sure that our data is packaged and AI-ready. And we have that in very good shape in a number of our data sets, and we are improving it in some of our other data sets so that, that is a core part of the value proposition within our Data & Feeds capability.
Race to the bottom
Of course, when data is a commodity, the risk arises of a race to the bottom in terms of consumption and pricing. Schwimmer is phlegmatic on this point, countering that quality will win through:
With respect to the AI ecosystem, there are three legs to that stool. There is the computing power, and that includes the data centers and the chips, and we’re seeing enormous capital and investment go into that, and you’re seeing some potential commoditization over time there. Then you’re seeing the models themselves, and you’re seeing enormous investment go into that and lots of competition and some open source models and some cheap models coming – or cheaper or smaller models – coming out of China and other jurisdictions. So [there is] competition and potential commoditization there.
Then you have the data as the third leg of the stool, and it is hard to commoditize the data. The data quality is incredibly important in terms of informing, training, feeding the development of the AI functionality in these models. You can have synthetic data, but synthetic data is of the same quality. If synthetic data is built off of, I’ll say, commoditized cheap low-value data than it is relatively cheap and low value. If synthetic data is built off of high-value data, it has higher value itself. So it’s challenging to get to a race to the bottom or a commoditization of high-quality data. And that is what we have.
Those questions around cheaper options aren’t going to go away, he admits:
We’ve been getting this question really since ChatGPT first entered the public consciousness in November ’22. Questions about or concerns about whether people would just go to a sophisticated chatbot and get all the information that they need, that’s clearly not the case.
Those kinds of tools and models can be pretty cool, they can be fun, you can get a lot of information from them. But especially when you don’t know exactly where they are sourcing that information and especially when you get into financial analysis and market analysis, it’s well documented that they’re not even close to capable of providing the kind of certainty and quality and accuracy that our industry demands.
So I think that if anything, that’s sort of the fortress of quality, credibility, accuracy that has to exist within our industry, while we’re going down this path of using more and more AI functionality. And it’s something that we take very seriously. We are investing in AI capabilities. We’re doing it on our own. We’re doing it with Microsoft. We’re doing it with other partners. But we’re very focused on making sure that we have that kind of quality control.
My take
An organization well-placed to make the most of the need for a solid data foundation to exploit AI’s potential in the best way. It just needs to keep banging the ‘quality counts’ drum as hard as it can to be heard about the din made by the Altmans and Musks of the world.