Aidan North: Alt Data in an Agentic World

In this episode of Selling Signals, we’re joined by Aidan North, Commercial Lead at Orbit. Orbit sits at the intersection of unstructured data, AI and institutional investing, helping transform unstructured information into formats that can be consumed within investment workflows.

We discuss how hedge funds are adopting AI, Orbit’s position at the forefront of agentic use of alternative data, and what has changed since the early days of pre-LLM systems. Aidan also explains Model Context Protocol (MCP) and why it is becoming an important part of the data ecosystem, enabling smaller funds to access and integrate datasets that were previously out of reach.

This episode is essential listening for anyone building or selling data products into the investment industry, or trying to understand where AI is having real commercial impact.

Transcript

Tempo: 120.0

SPEAKER_02 0:02

Welcome to Telling Signals, the podcast focused on how businesses actually monetize and sell data. Each episode, we interview an industry insider to hear their experiences and lessons learned.

SPEAKER_00 0:11

The series is powered by Valcist, the company that transforms your data into investment-ready intelligence products. If you enjoyed the episode, please subscribe from wherever you get your podcast.

SPEAKER_01 0:22

Today we're joined by Aidan North, commercial lead at Orbit, a firm that sits right at the intersection of unstructured data, AI, and institutional investing. Aidan has spent the last five years helping hedge funds turn messy real-world information into structured investment grade data sets, long before large language models made it fashionable. Orbit has worked with some of the most sophisticated investors in the world, building both bespoke and scalable data products that feed directly into investment processes. Given how quickly AI is reshaping the way data is sourced, processed, and even monetized, Aidan's had a front row seat to what's real, what's hype, and where the market's actually heading. Aidan, great to have you on.

SPEAKER_03 1:02

Yeah, thank you very much, guys. Very excited.

SPEAKER_01 1:04

You're welcome. We should uh probably all apologize. We're all quite ill recording this. So uh apologies if we all sound a bit husky.

SPEAKER_02 1:11

Yeah, all the chamomile tea in the world has not been enough to get rid of this lost voice or to regain my voice.

SPEAKER_03 1:17

I've actually got like a hydration drink in front of me as well.

SPEAKER_01 1:21

So I should have bought that. Yeah. Um also why don't we jump into who Orbit is? Um to not all listeners may be aware of you guys, but you you kind of have some really interesting products and um offerings.

SPEAKER_03 1:35

Yeah, so I guess at a high level, so Orbit have been around since 2015. Um, we essentially automate the stages of investment research, but very specifically targeted towards the financial services industry. And as you said, we've we specialize in unstructured data. So that is really, to our definition, all of the you know, corporate filings, transcripts, internal data like broker research, share drives, but anything which is really difficult to extract that information from. And I guess that's where like large language models will come into play in how we automate that for um institutional firms. And I guess, you know, the term AI, like it's been thrown around a lot recently, and it's a big buzzword. But again, it's like we're not just a simple chat bot system. We're something a bit bigger than that. So this is where we start to dive into the systematic workflows and actually kind of like bringing in that connectivity part as well of how unstructured data can really be combined with large language models more effectively. So that's kind of who we all are in a nutshell. But I guess that point of being around since 2015, it's seeing the changes from when the large language models come into play, how we used to do it with natural language processing and tweaking models, and then even from a data perspective of you know, processing that data and what that then looks like for the end user. Um, so yeah, we've been around and sort of done the hard graph, which a lot of firms are doing to date.

SPEAKER_01 3:09

And I know you touched on a bit there, but being at the coal face of AI and investors, how is the adoption going? Where where are where is AI winning for investors today?

SPEAKER_03 3:20

I think it's quite a funny one because when we when we sort of talk about AI or kind of anything within you know finance, when something new comes out, there's a lot of hesitancy at first, whether that's AI or even like you know, servers being put in like locations when the cloud came out, there's always hesitancy. And I kind of look back on that that part where you know I wasn't necessarily in the finance industry then, but I've heard stories about you know, I want the servers to be on my location because I want to be in control of you know all this compute, all this usage. And now people are a bit more flexible in terms of, hey, I'm just gonna save everything to the cloud and use AWS, and it's a massive thing. So I think AI is going on that journey as well. So when large language models came out, you know, don't use it, there's hallucinations, don't use it, there's security risk, like compliance, and so on and so on. But I think you know, you move on a few years and now everyone's integrating their own clawed code and using Chat GPT Enterprise and now even looking at external providers because building it internally is one thing and giving it a go. But then there's a lot of complexity with even just data ingesting, or you know, how do I, you know, prompt engineer this use case, or you know, what is the right process for my investment teams and kind of bringing that in as an enterprise solution. So there's a lot of um, I guess, complexities when people say that they want to build internally, and there's a lot of like road hurdles there, I guess. Um, and I guess that's where we come in because we've done all that hard work on the data, and now it's a lot easier to use the large language model.

SPEAKER_02 5:01

You you mentioned um uh cloud and you know, originally physical servers in the office versus uh when you move to the cloud. And I think even I believe Azure did particularly well early on because they were able to say we can actually tell you which box your data is being uh your data's being processed in. On the AI side, probably what's analogous to that would would maybe be locally deployed large language models like a like a Facebook's Lama 3 versus a Chat GPT, uh a Claude. Um to what degree have you seen success with those local models, at least from my perspective, um a large amount of the discussion around them tends to be for often personal use cases, etc., as opposed to maybe more enterprise grade uh use cases. Has that been your experience or or yeah, so what what's what type of uh LLMs are you using, or is it a mix of of both?

SPEAKER_03 5:57

So we we kind of are like model agnostic. So it's all the work that we've done on the data so that any large language model can then come into play. So I think the traditional ones, I'd say like Claude is being widely adopted. I think ChatGBT was kind of first, but now Claude is, you know, it's doing a lot of different use cases in in one model. Um there's a lot of more visualization in Claude as well, in terms of like the graphs and different things that you can populate. But if we look at like a hedge fund or an asset manager, you know, some of the ones with I guess the resource there, whether it's like an NLP expert or um a head of AI or something like that, I think some of those are using like Claude predominantly. But then they're also developing their own models as well, which it might be an open source model and tweaking it. It might be, you know, just actually developing that model internally. But again, it's one thing to kind of create your own model and then like actually kind of keep it in like maintenance because I guess you know, for firms, they're not they're not uh a data vendor kind of bringing in all this data, and they're not someone that's gonna kind of like always want to build that model themselves. Like there's a lot of models available on the market, so I guess it's like a time thing. Um, I think that I actually see a lot of the smaller hedge funds kind of doing their own models and tweaking it more so than the big ones, which is um I'm not sure like I guess the reasoning behind that, but there's I think the some of the smaller ones are building internally as well, which is quite a yeah, it's quite a change of pace because I guess before they couldn't build internally, but now they can do it in a much more cost-effective way.

SPEAKER_02 7:40

Interesting. And I I guess to go back then to um orbit and your products, um, you mentioned the kind of natural language processing 2015 to 2020, say days, and then the the post-2020, where we start to see the large language models really, you know, uh not only be released, but then start it start to become obvious that there are serious enterprise use cases for this. Uh it'd be interesting to almost like walk through an example of a kind of product of yours and maybe how that has evolved since uh large language models have come out. Perhaps to take an example, you mentioned company filings. Um so I imagine you have a product which is going through 10Ks, 10qs, uh consolidating income statement, balance sheet uh and cash flow statement data, but also, and this is kind of where the I think it becomes even more unstructured, uh, is often in the kind of industry-specific KPIs. We're talking same-store sales growth, for example, in a in a retailer. Firstly, am I correct in thinking you're nodding? So I'm assuming you do have a product like that. Uh and then the second question, uh, the main question is how has the that product evolved over time?

SPEAKER_03 8:55

Yeah, I think like if we kind of like take a step back, I suppose, it's like the I guess the saying is like rubbish in, rubbish out, right? So one of the core components of using any AI, whether it's a use case or just using it off the shelf, is the data that's always fed into it. So from a data level, um, that's kind of like one of our core USPs. So we already cover all the global companies that's that's listed. So that is Japan, Hong Kong, Taiwan, China, all the hard-to-reach regions. But then also the Western side, like you know, you've got US and Canada and European markets. So from our perspective, it's like we're data hungry. So we will collect any different disclosure that they'll make. So annual reports, 10Ks, 10 Q's, exactly what you just mentioned, but all the niche ones in between as well, which we typically see a coverage gap on the market, because you know, all the firms are wanting AI ready data, but maybe some of the even some of the major vendors out there don't cover specific types of documents because there's there's quite a lot available. So we've basically processed that into the like machine readable block level data so that when you are using a large language model, you can extract the content more easier, you can you know digest it quicker, you can like consume uh a lot less compute because you're not feeding like a big document into Chat GBT, you're actually just feeding these blocks off to the model instead. So that's like a really core part that I guess lots of firms are trying to do today, but then it involves data mapping, security, entity masters, like all the sort of fine details that just take a really long time. And I guess from that angle, the large hedge firms of the world or asset managers will consume all that data and kind of do what they want with it internally across the board, whether it's use cases or just better, better quality data for large language models or whatever. And I think then it kind of dives into the the other angle. So the second stage, I suppose, from our perspective, is not just using AI for like a chat system, which is put a answer, put a question in, get an answer out, and happy days. We like to consider the like a genetic orchestration layer. So what that means is any type of workflow that a researcher or an analyst might do over a routine basis. So it might be something like I look at X amount of companies every single quarter, I look at revenue growth, CapEx, expenditure, um, sustainability metrics, and it just takes a really long amount of manual effort. So, from our use cases, I suppose, in particular with you know those types of users, it's like I want to automate that part of my work so I can focus more on the end result of my work. So I don't want to take three weeks of the month out of you know just doing this manually, downloading documents and and so on. I want to automate that across my portfolio, for instance. So then it then it gets to the stage of what do you want to do after that? So once I've automated that workflow, I then want to scale it. And the reason why you want to scale it is because I don't want to just look at 20 or 30 companies manually. I now want to use AI to look at it across 20,000 companies because I want all the information in front of me, and I want to be able to find more investable companies, which I've probably never been able to do because it it's it's a manual effort all the time, and you you you can't always focus on maybe companies from China because you've you've never been able to dive into it before. Um, so those are the that's typically like how it works. So there's a data wrangle and a and an automation piece. But I guess some of the core use cases we see is one, can you automate my work, save me more time? And then the second stage is like screening. So screening like formatically, where I want to look at maybe all 50,000 companies, and I want to track which ones are mentioning AI or Gen AI or large language models in their transcripts or filings. And from that perspective, you can then whittle down a universe. So I'm looking for every single company globally, but then I'm whittling down the universe into like, say, 400 companies where I know that they are mentioning AI more often. I'm seeing a trend, and I can invest in those companies, maybe in my like global AI fund, for instance. And that is that's where we're actually seeing alpha then come into play. Because then I can actually invest in those companies, my fund grows, and yeah, over time I see more performance.

SPEAKER_02 13:43

Just to make sure it's very clear in my head, then it sounds like if I was to based on two of the kind of buckets of products you've mentioned so far, there's taking unstructured data to structured data, which might mean um structuring it for traditional programmatic use cases, i.e., a fund building, a dashboard or running a um some sort of strategy based off of the data. Um but there's also structuring it specifically for the consumption by large language models, perhaps not even done on your end. That may be being done on the funds themselves end. Uh and then the other uh basket of products, shall we say, is how, and this is perhaps more on the fundamental side, the research side, how can we automate a traditional workflow that an individual analyst would do, but but do that reliably in its scale so they don't have to do it, and then also start surfacing ideas for their research. Is that a fair kind of high-level summary?

SPEAKER_03 14:41

I think so. And I think like there's tons of use cases because once the data's kind of ready, you can do any use case over the top. But then it just involves the firm kind of creating that use case and us kind of helping out with that and like, you know, suggesting better ways to do it, suggesting better ways to run the prompting, or even like, you know, do you want to increase the frequency? Because if you're only doing it four times a year across transcripts, do you want to start doing it weekly across news? Do you want to factor in more data into that? Do you want to start doing it across your internal data? So there's a lot of angles to kind of bring all of this unstructured data all together and form a very, very large workflow, um, even across data that you might not have looked into before, which that's the edge of AI, I suppose. It's doing the the the sort of hard graph work on the data to then uncover more ideas, investment opportunities, um, more data points, maybe even, you know, creating data sets which aren't available on the market, because from a filings perspective, you know, firms are only just sort of diving into some of this textual data, but AI just enables it now much more easily. Um, I was actually chatting with a firm at an event not too long ago, and they said, you know, I'm traditionally buying, you know, a data set with some sentiment over the top, which, you know, it's quite a I guess a black box view, because that sentiment against that data set is only only a part of that. But they're like, how can I surface more ideas and put my own sentiment ideas because I have tons of ideas that I want to put into this. And this is where you can start creating your own data sets and apply your own sentiment against any sort of financial document. And that for them is uncovering new alpha because they're able to do it across any type of document now.

SPEAKER_01 16:38

And that and that seems more of a quantitative research process strategy. For the fundamental teams, I imagine you know you're covering unless you're covering 10 names a sector. You probably are reading the filings and and ingesting that information as a human. Have you seen any games there for a fundamental team to more systematically get to insights? Is it just a speed to insight? What's the the play there on the fundamental side?

SPEAKER_03 17:09

I think it's allowing them to not just be fundamental anymore. So like we can now move them to like quantum mental. So you're probably seeing a lot of firms now kind of making that shift. And what it basically allows them to do is one, not just extract information and be, you know, very factual. So like this revenue is growing and so on. Um, I think from our perspective, when we see these sort of sentiments being populated and then aggregated, it allows them to quickly see if something shifted over time, whereas before they would have just had to sort of go after the information that's kind of put in front of them. So we allow them to put like flagging systems in in place as well. So like, you know, green, amber, red, like only track if a sentiment's decreased on like by 75% or something like that. So like we allow them to put this, I guess, prompt engineering in place so that it pushes the information to them rather than them hunting for the information. And that's where like I guess all this new idea generation comes into play and scaling their research and yeah, allowing them to kind of look over time series information as well. Because I guess in the past, I mean, even just digging out all this information from your share drive or whether it's stored in your data warehouse, or even involving multiple people in the company to kind of like it might be like a however we we kind of see it, so like a like a data scientist and then a quant and then the fundamentals being like they all do have a little role in play where like you know a new investment idea comes into place, but now it allows them to kind of get to the information much faster against their desired use case, I suppose.

SPEAKER_02 18:55

You mentioned that uh sentiment uh use case and the change of sentiment. Can listeners take that to mean something like year one, um the CEO on an earnings call is asked about issue X and the sentiment score from the AI is he's 80, 90% positive. And then he's asked the same thing maybe the following quarter. Um, and the AI classifies that actually sentiment around his response, i.e., how positive he, or I should say she is on um uh on uh doing the um answering the question has gone down to 50%, say, then it's that's an amber or that's a red. Um and the analysts are able to define those parameters themselves. Is that a fair description of an example of that?

SPEAKER_03 19:43

That's exactly it. So like management tone is actually one of the ones that comes out quite often. Um obviously many more data points around it as well, but it's it's allowing people to kind of run their own ideas over this data, whether it's management tone or just you know, extract this information. And it's it's kind of allowing them to almost be their own data vendor in a way, because they can create these new data sets on the fly. And for them, that gives them a massive edge because if no one's looking at that information in this way, and there's no data set available, then surely you've got this new data set which you've created, which then has alpha once you've done your backtesting or something like that and found it. It's I think it's just a new market that we're all going into, which I would anticipate maybe in the next few years there'll be new data sets being created by the firms which are smaller or mid-size and not just those big ones which can kind of handle all that work internally. Um, and I guess with more and more people kind of running their own AI internally as well, we're just going to be seeing like I think one of the things I heard recently was eventually it will get to the point where it's the data's obviously really important, but it's like selecting which firms that have that sort of like golden source, I guess. Like it has to be processed correctly, like which ones are actually leading the way in that angle. I think in the past it was, I'm not gonna say like a tick box exercise, but if we just look at filings, for instance, we we do know that there is a bit of gapy coverage on like APAC regions, like it's quite hard to get. And now it's now it's come to the time of like large language models where gappy coverage is no longer acceptable and bad quality data is no longer acceptable. So which firms are providing this best in class of data for large language models? And I think that's where we're seeing a massive shift on even just data consumption um from our our perspective as a data vendor as well, because firms need better quality data and it's just not out there.

SPEAKER_01 21:56

I want to take the conversation towards MCPs, because when you talk to Data providers who have I don't know embraced AI and agentic systems, the term MCP comes up a lot. But for those who maybe are nascent in the AI space that are listening, what is an MCP and why is it important to orbit?

SPEAKER_03 22:14

Yeah, it's quite uh I guess it's a new term. So obviously we've all heard of APIs, but I guess a really simple way of putting an MCP is so the term MCP in general stands for model context protocol. And at a very simple level, it's a way of standardizing how an AI model, like a large language model, um, can connect to external data and different sources and different tools. So I I kind of see this picture always in my head of like a USB, and then that USB has lots of different API calls in it. So it's a way of connecting data and tools to different workflows internally. So if you're getting like a hedge fund that's building internally and they want to consume our filings, but maybe they can't store all those filings on their site, it's very, very costly. So this is like an enabler to them to actually start utilising things like Claude and uh Gemini and um OpenAI or any sort of model that they're using in the same way that they've used it before, but just over the better quality data. So maybe to kind of expand on that, it's like if you think about how APIs are standardized and how applications talk to one another, an MCP is doing something really similar for AI. So it allows models to securely access all this data, a workflow, or anything that's custom built. And where we're starting to see people encounter MCP is in these agentic systems. So AI agents don't just answer those questions, but I guess they actually go off, retrieve the data, run the process, and then come back with that uh that output. So for us, from a data vendor perspective, it's like it's becoming very important to have these MCPs because you've kind of got a new source of people that can actually access your data now. So it's a new market for the smaller funds that are trying to access textual data. So it's I guess it's making data accessible, queryable, real-time, that that sort of perspective. So that that's what an MCP is.

SPEAKER_01 24:27

And the this is going to be a really dumb question for anyone who has any idea what they're talking about. But does this sit on the provider side or on the the end user side? Is a firm creating their own MCP server that sort of a provider is plugging into, or is I, as the provider, creating like an API feed, an MCP feed that sort of fits into that what the end user workflow?

SPEAKER_03 24:47

So an MCP can actually be lots of different things, I guess. So like we can wrap an MCP in whatever way we want to. So we have a standardized MCP where you can kind of if you're using Claude, there's a really, really easy connector where you can literally just connect directly through Claude, all it is listed on their sort of MCP marketplace, and you just connect with it. And once you're connected with it, it will do like a task. So that task might be if you run a any prompt like Find Me, Microsoft's revenue growth over the last couple of years, it would run that. And I guess it just has a limit of like five companies that you can run that against. And but then you can create MCPs based on a workflow. So I think back to the start where like firms are starting to create these workflows against thousands of companies, and you've really like tailored it for their use case, you can run an MCP over that individual use case as well. So it makes the data accuracy even more powerful and more accurate. And you don't have to go outside that. I guess the the big the big value here is if you're using Claude or Gemini or anything like that off the shelf, you probably will see hallucinations and you probably will have elements of pulling in web data which is out of date and and so on. And my my question is always like you can very, very easily drag and drop a document in and run a prompt over it and get an answer out. But one, the data hasn't been processed correctly because you're reading over the PDF and not the like the machine readable. And then the other factor to always consider, especially for like firms that are trying to build internally, is do you want to do that every time for thousands and thousands of companies and run the prompts every time? And like it just becomes impossible. So that's where we then see ourselves in that position of running it autonomously against their own latency requirements and like you know, scaling up basically. So yeah, an MCP is is quite important, I suppose, because it allows them better access to the the quality data.

SPEAKER_02 27:04

And if you're able to share this, just so that um listeners can get an idea of where the m like where the market is in terms of its transition to usage of LLMs, what proportion roughly of your customers would you say are are keen to connect to orbit via MTP as opposed to more traditional means?

SPEAKER_03 27:22

Yeah. So I think I would say the larger firms, like I mean, I won't mention any names, but the larger firms of the world, they will just take our data and they will run it, and they're the systematic multi-strategy sort of quant firms, and they'll just do what they want with that. And everything's built internally, done. I think the mid-size sort of firms, they usually have a use case, but they don't have an infrastructure to handle it. They probably have different data sources being fed in as well, and and so on. But those ones would typically use our like no SaaS ecosystem where they can build their own agents or let us build it for them. And then the smaller firms actually, um, I think I think those ones are quite interested in MCP because it's like I don't have the infrastructure, I don't have the ability to bring in the data anyway. But this now enables me to use my off-the-shelf model, whether I'm building my own or not. But it now allows me to like consume that data, which I probably never would have been able to before. Um, there's always budget constraints um for the smaller firms as well. So I think that enables them to use it um much more effectively. But I think I would say across the board, the term MCP is just of interest. Like, I'm pretty sure if you mentioned this is our MCP, this is our product, this is what it does. I would guarantee that that probably gets you more of a meeting than saying, like, here's our SaaS solution sometimes, because you know, there's lots of SaaS solutions, there's lots of AP like AI products on the market, but it's quite hard to create an MCP. And I do see like a lot of MCP marketplaces out there as well. Like there are there's quite a few. Um, and it just depends on where you list that MCP. But I I would imagine over time what would happen is you have all these different MCP marketplaces out there, and the firms just go into them a bit like a data vendor uh marketplace, but instead of accessing like a trial and giving them the historical data and so on and so on, it'll get to a point of well, I'll just connect with the MCP, I'll try it out myself, and oh, I like this quality. Um, I'm gonna now speed the firm to onboard their data or something like that. I just think that that will definitely help the vendors in terms of like the trial process.

SPEAKER_01 29:46

To that point, you mentioned you're listed on Claude's MCP marketplace. Commercially, how does that work? Are you giving part of your data away for free? Is there a tokenized type of no?

SPEAKER_03 29:59

So um I think there's there's different routes across different like large language models. So I guess there's different like revenue options if you get to that stage. Because I know with I think with Claude it gets to the point of there's like an official like listing area where like anyone can see it and anyone can consume that. Like a lot of the big firms are kind of listed on there as like a stamp of approval, but you can actually anyone can create an MCP and actually list themselves on it's almost like a separate marketplace, and for then orbit to kind of go, this is how you directly connect with that MCP. And so I guess anyone can create an MCP, anyone can list one, but there's a couple of different tiers of like your visibility. Um, I don't think we're quite at the visibility where it's like anyone can like have a look at it, but we are listed on it if we can like share the details with you and you can connect with it in in that way.

SPEAKER_01 30:56

And let's say I I find orbit, I click on orbit and I ask some questions around some 10K filings or whatever it may be. Um what what is the fee structure for that? Again, is that is that a free to use thing or is there a paid element?

SPEAKER_03 31:12

So, in terms of like, it's like a um a consumption-based model. So essentially when you connect with it off the shelf, um you can expect like a free number of queries, basically. So those queries you can run, you can test, you can have a little look. If the quality is good, then it you can reach out to us and sort of expand on that trial or purchase data or whatever that looks like. But I think in terms of like the fee structure, we typically say, look, here's a base fee to just access the MCP and start using it. It has all of our data included in that, X, Y, and Z, and then it just moves to a consumption-based model. So like every time you query, it's X credits being used, and it's calculated on like uh the model consumption, basically.

SPEAKER_01 32:00

Yeah, the modern consumption's uh the consumption model uh pathway seems quite interesting. I know Carbon Arc are quite big in that space. Um, and I quite like Kirk, the CEO's kind of view on it in that you know you you pay gas per mice you know, per litre, cost of fuel, uh uh you know, gigabyte for uh debt transfers, all of this, that it kind of makes sense to to go that way. Um and I I wonder if AI will force the hand when you think about all of the SaaS platforms that if we do end up going the MCP route and a lot of the foundational models end up in the hands of marketers and product teams and they can access all of this data um through these models and ask the questions and then get the analytics that they need. Whether that makes some of the analytical platforms a little bit more redundant, I don't know. Uh, and that forces it to go to a consumption model.

SPEAKER_03 32:55

Yeah, I think so. I think over time, like we'll see that sort of shift. And you know, it just depends on if people just I guess it's like personal preference. Like if I'm happy with using Claude and I'm adamant that I'm using it and there's no change, then you know, MCP is absolutely fine and that's where the consumption comes in. I think there are select people that do want the stats environment still, and that environment's kind of I guess it's there because the UI's been built, the connectivity's there, all the data's under one roof, it's more visual, and it's very off the shelf in terms of let me use the data, let me build an agent, so on and so on. So some people like that approach, and then there'll always be the big firms that always want the data.

SPEAKER_01 33:40

Yeah, and and to that point, I think you mentioned, and I categorise them as AI native hedge funds that seem to be taking on more of an MCP approach. But I think historically there's been a problem in our industry that there's not enough new buyers coming to market. And that's probably because being uh a fund that buys a lot of data and alternative data is a very expensive pathway. Uh and I wonder if what's driving that the smaller funds being more AI native, is it it's sort of like a gateway to being able to ingest more alternative data. Um but yeah, talk me through some of the conversations you're having with those types of funds and whether you're seeing sort of more smaller funds be able to compete with some of the larger incumbents.

SPEAKER_03 34:28

I think the adoption of like unstructured data for firms and alternative data for those smaller ones. Um, and when I say smaller, like it could be anywhere after that admin stage is kind of all set up, they kind of hit that one billion mark. I'm even seeing those users kind of come to me. I think there was one the other day, it was like 960 million, so not even quite at that one billion mark yet. And they're interested in MCP because they can't consume the data, but they are using Claude and they are using um Copilot and they want to ingest that data in some way. This now allows them to do that. So I I feel that you know, for the larger firms, there does tend to be a bit of like a there's a bit of a selection period. Like I'm chatting with so many different vendors, there's lots of different alternative data like products on the market. Which ones do I have for my budget? And it tends to be a long sales cycle, but obviously once you get the name, it's it's great. But then the smaller ones, they they want the edge. So you would probably anticipate more of them kind of coming from this alternative data route. Like almost data that they've never ever ever been able to touch in their lives now be fully accessible through Claude. Like if everyone creates an MCP for their product, I'm not gonna say it's like the complete gateway, but it'll be a gateway into the way that firms actually just be able to see your data set is is probably the best way of putting it.

SPEAKER_02 36:02

And so from the sounds of it, you would say setting up um your data to be accessible via MCP opens up essentially uh a new part of the market for for many uh data vendors, probably, as you say, in that kind of one billion to five uh five billion range, because that's exactly the way they're thinking about utilising data at the moment, and as they haven't been able to in the traditional sense before.

SPEAKER_03 36:26

Yeah, and I think like if you look at sort of like again, won't sort of mention the names, but if we look at like the common market in terms of even the events that there are out there, which there typically is like large funds attending those alternative data um conferences or whatever, it's a great like door opener to them. And you will always get the conversation, you'll always get like more leads and so on. But I would just imagine like this giving probably the events industry and also the the data industry an ability to go after however many thousands of fundamental hedge funds that there are out there and the ability of providing this data to them. And I guess not just always being consolidated into the the big AUM funds, which yeah.

SPEAKER_01 37:19

I'm gonna ask an awful question that you can say please don't. Uh but if let's say we do go on this route of uh MCP becoming in your access, let's say for Claude, um how does compliance work for uh do smaller funds really just have much less of a care about um uh about data governance and compliance? Because but the hoops you have to jump through right now are are are tough with the bigger funds if you're coming into the market or even an existing provider. Um did you have any insight on on how that might actually work as a if I'm sitting in a compliance team right now and in a$1 billion hedge fund and they're about to plug in 10 data sets from this random MCP, how can you manage the risk?

SPEAKER_03 38:10

It's quite a hard one, I suppose, because I'm not on that side of it in terms of like how they would manage that. Because I guess when everything becomes off the shelf, there is there's a risk, that is always going to be a risk. But then I suppose if you're signed up to whatever enterprise large language model, it I don't know if the risk kind of already been evaluated by using that large language model because they will know that you can connect to MCPs via that source, but then again, like even through MCP, like we would ask, do you need a trial agreement for that? Um pretty simple setup, like this is how you can access it. Here's the links, here's the information, and I guess they evaluate it themselves in terms of do they need to go down a big route of long processes or not?

SPEAKER_01 39:02

But I I yeah, I've given this zero thought into into you've said that, but even to the simplest aspect of like if a fund gets audited, uh maybe a dead set they've been using is classified as material non-public information, insider information. Uh how are funds gonna be able to control for that? And I don't know.

SPEAKER_03 39:22

Yeah, I don't know. Not sure. I think from our perspective, it makes it easy for us because I guess our knowledge base is all the publicly available information anyway. Yeah. So for us, it makes it very easy. But yeah, if there's other data sets which are a bit more like of a compliance risk, then I can see why yeah, you might want to sort of get a bit of an agreement in place or compliance checks and so on. But I don't know, even with a smaller funnel, you would probably anticipate that process going a lot faster because how many alternative data MCPs are out there and selling to these types of firms anyway. So maybe it'd be quicker.

SPEAKER_01 40:00

Interesting. You um you mentioned there that you you're talking to a lot of funds between you know billion dollar to sort of five billion dollars. Talk us through in terms of how you identify that being a sweet spot.

SPEAKER_03 40:13

Um, to be fair, I think it's more like just how we've been chatting to people, and like as we all know, like all the things I've mentioned then about some of the larger funds, all the different processes that there are in place, and then you know, we didn't chat some of the smaller ones of like one to five billion, um, even some of the mid-size. But I think identifying it more like even if we step back into like a business development role or a salesperson outreaching to them, you would anticipate that there's less people reaching out to a CIO of a one billion fund, and your response rate would probably actually get a lot higher than reaching out to the CIO of one of the top funds of the world, because everyone's going to be reaching out to a CIO or a head of AI or something like that, and it just becomes like overwhelming. So your message just has to be really, really clear. Whereas when you're chatting with someone you've never met on LinkedIn and you're outreaching to some of these top performers of the world, but on the on the smaller side, I think that response rate is just so much faster. Like I think that's just through trial and error. Um in terms of like, I guess, my own approach and outreaching to people. So I know the types of people I want to reach out to, but if I can speed up the process of my response rates and getting them into a meeting faster and still sell the same products to them, then that's kind of like my that's why it's the golden sweet spot, I suppose, for me.

SPEAKER_01 41:45

That's really interesting. Because you're you're essentially saying that in a crowded market, go and find somewhere where people aren't crowding around. Yeah. Do you find that that erodes the pricing power that you you can you can sort of quote? Or no?

SPEAKER_03 42:00

I think for us it's like we focus on we obviously want to get the product in and we know that they're going to be either building internally with AI or buying it externally. And for us, because it's like a flexible ecosystem, we can handle any part of that. It's just kind of finding the use cases and their progress and what do you want? Do you want a SaaS? Do you want an MCP? We normally know that they probably won't consume the data part. So I guess for us, if we sold a date the whole data set into one of the top funds, it kind of stops there because it's like I've bought your data set now, I'm going to do what I need to do with it. And you can't really upsell past that point. So our view is if we can get in front of some of those smaller funds in terms of selling the SaaS and the more consumption-based element, then surely more use cases will come out, more data sets will be created, and it just becomes easier to kind of do that upselling, I guess.

SPEAKER_02 43:01

It's been a bit of a theme throughout the conversation, actually, but it sounds like um there's a reasonable amount of kind of services or indeed consulting that that happens on your end. I'd be interested to know when traditionally we when you hear from data vendors, they try to make it very clear that they don't do that. And it it sounds like you you guys are obviously clearly a very successful data vendor, but you've lent somewhat into the consulting, the services aspect of things. Is that correct? And I'd be interested to hear uh how that decision was made and whether you think it's a really key part of what you do.

SPEAKER_03 43:39

I think it is a key part because lots of firms, even lots of people out there, don't really know their plan, I suppose, for AI. And I'm not gonna sort of there's not a negative spin, but like a new AI product will come out on the market or a new model come out, and then it's like, oh, what do we do now? And is there a strategic way of thinking about AI? AI? Are you thinking about the bigger picture? Are you just using it for one use case? Are you just using Claude off the shelf? Like there's so many different routes, and I suppose it's like an ever-expanding market, isn't it? It's like there'll always be something. There'll always be something new. There will always be something flashy. And I think even if you look at like a map of all the different AI companies that are targeting financial institutions, there's there's tons and there's new ones being created all the time. But then it's like from a consultative element, we're not just an off-the-shelf product and you buy it and you use it and off you go. And yeah, maybe you decide on a different route later down the line. Like we want to be kind of like embedded in that process. So for us to be the knowledgeable ones on how to use the data, how to use AI, I think that becomes quite valuable. And I guess firms are always wanting to know what use cases are being created. And sometimes those use cases are things they've never thought of in their like ever before. Because I guess AI from our perspective is like everyone has a view of what AI is in their heads. And maybe rightly or wrongly, when people start to use Claude or Chat GBT when it came off the shelf, I think there is an element of hey, I should be using a tool off the shelf for everything to do with AI. And I think that's a bit of a wrong way of looking at it because there's so many elements of like creation and tailorizing and kind of thinking past let me use it and try and get something out of it, because then you know you might get hallucinations, you might get all these issues that crop up, and then that's just a one-way of thinking. Whereas we kind of like to bring a like the technical elements to it and parts they've just never thought of like ever in the in the time of using AI.

SPEAKER_01 45:56

On some of the round tables I've sat on with explicitly the buy side, uh in the earlier days of the rise of AI, um there was a narrative that if anyone was sophisticated enough to create anything meanerful meaningful in the AI space, uh, they would be on the buy side. Um I think maybe that's probably a an unf definitely an unfair narrative to have today. But I wondered in your sort of go-to-market service element advisory part of that that sort of strategy, whether you kind of have friction against that sort of mindset?

SPEAKER_03 46:35

I wouldn't say friction because we because we are this open ecosystem, it even if it was a buy-side firm like creating something like incredible, because even some of the use cases that we kind of work with hedge funds on to date, like formatic screening or large-scale data extraction or sentiment generation, like these firms have been doing that for years in terms of NLP almost going along the same processes of what we've done to build our product, and now they've just built their own internally. So I guess our own friction point is we kind of know that we won't sell the AI element to them, but we will sell the data to them. And yeah, I think there's it's not really a friction point, I suppose, for us in terms of selling our products, but I think because they've built internally, like we we we can't really go down that route. So it always has to be the better quality, like large language model data. Um, yeah, I don't know if that kind of answers the question there, but no, it does.

SPEAKER_01 47:36

And and one other question I had was um uh I I think where a lot of crowding happens, going back to your point around crowding around the bigger buyers, they're obviously way more visible in the market. How are you finding these sort of smaller funds? Have you got sort of a a cheat sheet of uh of a d identifying someone that kind of falls in that ICP for you?

SPEAKER_03 47:59

Um I suppose there's different ways of doing it. So we well, I mean, we've actually built something internally to um identify and it is using AI as well. So yeah, so one brand. Yeah, like summarize this company and like identify the key people in it, and you know, all those things that we kind of value from an outreaching or like you know, meeting standpoint, but then there's loads of tools out there anyway. So if you look at LinkedIn, there's ways of you know segregating the amount of employees in a company in the financial services industry, and I tag it based on head of AI or something like that. And you would probably think that the employees that I guess the the companies that don't have as many employees, you would probably think that they're the lower AUM ones, whereas the ones that have like thousands and thousands would probably be the you know the bridgewaters of the world or something like that. Um but then again, it's it's looking at other sources. So on LinkedIn, you'd probably see the major firms on there all the time, and quite a lot of employees on there. And usually for the smaller ones, you'll see like a founder or a CEO, because you know, and maybe 10 other individuals, but not all of them are listed on LinkedIn. So you kind of have to go externally to some of those. I guess there are um what do I describe it as? Almost like a vendor for selling what firms and what AUM and pre-Quin and stuff like that. So all those types of dashboards out there. So they're quite good. If it's in um, you know, like HFM, I've got them. So they sell that type of um functionality to kind of search for those funds, but it is quite difficult because there are less events, there are some cropping up now as well. But that's kind of how we do it. I think internal tools, LinkedIn, anything that can kind of provide a bit more value there.

SPEAKER_02 50:08

Are we waiting for the the LinkedIn uh MCP around the corner? Hopefully.

SPEAKER_03 50:15

Yeah, maybe, maybe. I think to be fair, I'm surprised they probably wouldn't do that already. But yeah.

SPEAKER_02 50:21

I have no idea. That was a terrible joke, but that was not a piece of analysis.

SPEAKER_03 50:25

No, I literally have no idea. Um I'm to be fair, LinkedIn's quite accessible anyway, and I know you can download a lot of the data off there. There's actually tools that you can use to um like you connect into it, and it's like its own web scraping type tool, which yeah, there's a lot, there's a load of them out there.

SPEAKER_01 50:44

There's also a load of them that have been killed by LinkedIn. I know they LinkedIn can kick off some of those companies off their own platform.

SPEAKER_03 50:50

They are very, very strict on that sort of thing. But I mean, you can also use like automation tools that there are out there to do your like LinkedIn campaigns, and there's big tools, small tools, it just depends on how much you want to spend, I guess.

SPEAKER_01 51:07

Awesome. Conscious of time, uh maybe to sort of move on to the concluding question. I think with more and more corporates uh and more businesses coming to this the alternative data industry or monetizing their data, I think um there are more people coming to market with minimal finance knowledge, minimal data monetization knowledge. Um, I think this was the same for you five plus years ago when you joined. So, what would be your advice to a new entrant into the into this market?

SPEAKER_03 51:39

Um I would say even at the beginning, like focus on one thing and do it really well. So there's so much knowledge out there in terms of the financial services industry, and it's very easy to get lost and confused in all the jargon, the terminologies, and and so on. But I guess if you can like simplify that jargon, that makes it so much easier. It takes it did take me like a good couple of years to like get fully embedded and learn different processes and different, you know, different ways of doing things. But the I guess the simplest way to look at it is from a sales or a business development point of view. One, get out there as much as possible. FaceTime is really important, like AI and even like COVID when that all happened, like it just eliminated FaceTime. And I think there is a bit of a tendency now to go, let's hop on a few Zoom calls before we even have a copy. Like, but yeah, go into events, learn in as much as you can. I think the biggest thing for me may be like focusing on if it's a product, if it's a service, like learn that product, like really, really learn it, like distel yourself in it. Because there are experts out there in terms of the roles that they play, but you be the expert in the role and the data that you're selling, and everyone wants to learn, don't they? So, and you can't be a master of everything. I think when I came into the industry, I was worried about not being a master of everything and not having the right answers or so on, but it it's quite a good networking like industry, like people understand as well. So, yeah, I'd say like don't worry so much is is probably my biggest thing.

SPEAKER_01 53:24

I I think you raised a really good point there. There was some research done by Gartner that owned quite a few sales research businesses that suggested that, at least in the SaaS world, the predictor for a sales performance, or the worst predictor for a person's sales performance was their product or industry knowledge. I don't think that's as true in our industry, where you're talking to really sophisticated buyers and you're trying to fit something that you have into their process. And I think that requires a much higher level of product and industry knowledge to be successful. So I think to your point of making sure you know your data inside and out, and I think back to the conversation we had with Michael Tyndale way back when, when he was talking about making sure you know where the pitfalls in your data are at least, and being open and transparent about that, it kind of feeds into that same point. So yeah, I completely agree.

SPEAKER_02 54:20

It also comes back to something Aidan was saying earlier around orbit's obviously right at the frontier of the consumption of data using AI. And your and a large part of when I asked the consulting question, you said it's sort of we bring that education with us as well that we can provide to customers, which in a way is a kind of way of bringing them onto your data set, that that level of uh being the experts and really understanding not only your data, but also how your data can be used in context with AI, adds a greater degree of validity to the data itself.

SPEAKER_03 54:51

Yeah, exactly. And like I think maybe the key takeaway is like everyone wants to learn, everyone wants the edge, everyone wants to get more alpha or more automation, and there's so many things out there. But it's like if you can be the expert in your own product, your own service, they'll want to listen to that. And assuming you can articulate it, and if there's a challenge or something, or there's uh you know, someone else that does a similar thing to you, like and if you can really like educate them, then that's probably half the battle when you're speaking to them.

SPEAKER_01 55:23

Agreed. Appreciate the time, Aiden. Thank you for for for coming on and joining us. That was an awesome conversation. Yeah, that's great. Thank you. Thanks very much, Aidan. See you soon. Yeah, thank you very much, guys.