Cover information services companies here at RBC. I'm excited to host Jigar, CTO of MSCI. Jigar, thanks for giving us this opportunity.
Thank you, Ashish. It's great to be here again.
Thank you. Jigar, so you joined MSCI, I believe, in July 2018, and we've seen significant technology transformation at MSCI under your leadership. So I was just wondering if you could provide a brief overview of what changes, from a technology perspective, have you seen at MSCI, and what's your vision for the future?
Yeah, thanks, Ashish. Yeah, it's amazing how time flies. It's been six and a half years. And we went through a pretty multi-layer transformation of the company from a technology perspective. First was to lay down the foundation of shared engineering and shared services and DevOps and CI/CD and so on and so forth. On top of that, we built the cloud partnership with Microsoft in 2020 to start the projects we wanted to build on the cloud to accelerate the pace of innovation and client needs and so on and so forth. And then we launched the idea of and with the platform of Investment Solutions as a Service. And in Investment Solutions as a Service, we looked at our headaches from 1959 when the first indexes were launched.
And from that point on, we kind of democratized all the code and the data through 1,000 APIs, Data Explorer, Developer Community, Index Builder, and many, many products we launched after that over the next several years. And that was more of a platform-level new change. And then we launched MSCI One, which was an integrated platform for all product lines across MSCI. And that has grown quite a lot since then. And then so that was another layer on top of the platform. And then we started doing data distribution through Snowflake and another partnership with Google for building the data acquisition platform, which we did a lot of innovations in how fast we can collect data. We'll talk about that when we talk about AI.
Since then, you know, in the last two years, all the focus has been moving further up the stack with the power of GenAI.
No, that's a great segue into my next question, GenAI, right? That's the question of the day, question of the year. What are you doing on the GenAI front? Can you just talk about what are the key initiatives and the key categories of initiatives for GenAI at MSCI?
Yeah, so there are hundreds of projects going on across the 6,000 people of MSCI with GenAI. But if you want to simplify them, you can categorize them in three buckets. One is the complete transformation of how we work, right, driving massive efficiencies and operational transformation of the company. So, whether in that category, let me finish the three categories and I'll get to each of them in detail. The second category is the whole company is about data and data production and models and so on and so forth. So how do we use generative AI for creating a lot more data, scaling much faster, much cheaper, much larger coverage, and much better accuracy than we ever done in the past? So that's another huge area of investment for AI, and the last, but probably the most exciting one for me personally, is the client experience, right?
It's not just AI is not about simply driving efficiencies, even though that's not a small enough. It's a massive contribution to the company, but also client experience. Everything we have been doing in the investment industry for all these decades around performance and risk and portfolio construction and indexes and sustainability, private assets, public assets, how does generative AI transform their client experience and make them become a lot more efficient, make better decisions at a much faster pace? If we were able to do that, and I have several examples I'll tell you about where it's going to be pretty game-changing for the industry.
Okay, why don't we go into examples then? That may be a good segue just drilling down further on those examples.
Sure. So the first category was workforce transformation, right? So let's start with developers. There is so much excitement amongst developers. We are building products like index product center was something we built in a matter of a few weeks on our website. So all our indexes are, you can go view it, and it's a new modern website is coming soon as well. Now, in that example, the entire code architecture, the test cases, we heavily use generative AI for doing all of these things. And that entire product, and there are three other such examples in the company over the last 12 months where things were moving at a much faster pace than ever we have seen before.
The other example I would give is massive amounts of code, which is written in, let's say, Ruby, and our developers are more Java experts.
Now they are looking at generative AI to help them understand this code and manage it better. Huge amount of code we want to upgrade for, for example, vulnerabilities and cybersecurity threats. All of that work is going so much faster. Even in this framework in security or MITRE Framework, which is more advanced techniques of protection against cybersecurity threats, we're able to use generative AI to predict these threats and also look, analyze our code base and show us where the improvements need to be made. And in fact, even do it for us, right? So that's just an area just in developers where we are seeing massive transformation. But with things like Custom GPTs, every function, you look at our legal team, look at our marketing team, they are building their own Custom GPTs.
We have had so many hackathons across the company, 350 AI champions in every single function. They are teaching everybody what to do, how we can use this thing, and there are a lot of these things are no-code tools, and so the awareness and adoption of these tools is increasing dramatically. I look at the statistics of the Custom GPTs created in marketing or sales or servicing among engineering itself, obviously, in finance operations. It's everywhere. Everybody's excited, and it is taking away the grunt work from what everybody's doing, so on the one hand, you are able to not deal with the grunt work, and you save a lot of time, and it provides you with a lot of efficiencies that you can invest back in the company, so those are some of the examples. Customer service is another huge one, right?
I mean, the amount of answers we get through Ask MSCI done, so it cuts down the time where clients can get their own answers themselves without having to contact us, especially in the case of sustainability and climate as an example. In the second bucket, which is data production, simply put, in the last couple of years, we have doubled the throughput and at a 25% lower cost or slightly even lower than that. And the quality is getting higher. So we are able to scale at a much faster pace in data collection. We built the Google partnership, and the entire data collection process, data acquisition platform was built from scratch on the entire data stack, which is very heavily AI-infused. So we are able to use the best models in that area and able to produce data at a much faster pace.
One exciting example I'll give you is our controversies product.
We have a very rich product to share controversies data for the issuers and the companies we cover, and normally, three years ago, you had these extremely complicated documents which explain our methodology on how to determine what kind of controversy is it? Is it related to sustainability, or is it a governance issue, or is it a scandal of an executive? How severe is it? Does it have a material impact on the stock price, or is it just something on the side? There are these documents you would have to read. What we have done is a multi-agent debate architecture we have created where we train three or four different agents in different aspects of these controversies, and they debate amongst each other.
They come to a conclusion of what is the classification of this controversy, what is the severity of this controversy. Finally, when it produces the result, then a human gets involved. Even the prompt generation is getting done for these kinds of things using AI itself. We are at a level which is unseen two years ago before ChatGPT and all of these things came out. Data collection is a completely different world right now. If you look at our data analysts and you look at the demos they're doing, it's all about AI at this point. The last and probably the most exciting to talk about is what is the client impact, right? I'll share three examples of that. Last year, we launched the Geospatial Explorer. We launched AI Insights for portfolios.
We are working on things like Ask MSCI, which you can ask a lot of questions about private companies or sustainability or even indexes going forward. We're doing quite a bit of work on private assets data collection as well. I'll give you one example a little bit more detail, right? Over the years, we've had a large amount of institutional portfolios with extremely complex instruments, thousands of them with billions of dollars of assets in them. Six years ago, we would give you data files with all the risk and performance and go do something with it, figure out what's going on. We built the Insights product to say it's a Snowflake data warehouse. On top of that, there are Power BI dashboards which tell you all the insights about your portfolio. We added the generative AI layer.
Now, this is not something you'll get an answer from ChatGPT or Anthropic or whatever your favorite AI tool is because you have the portfolio context, you have the decades of model of risk and performance and investment models that our research team has created. And then you combine all of that with the latest news reports on the markets, the economy, the industry, and the company-specific events. And then you can simply, in very simple words, you can chat with your portfolio. Ask really complicated questions. Show me a chart of the biggest tracking error from the top five companies. Where are the biggest downgrades in ESG ratings for the assets I own, those kinds of things.
In the Geospatial Explorer product, for example, you can ask questions about various different assets like a data center or physical offices for the companies inside your portfolio and ask questions about where are the risks, and it shows you on a map in a very simple manner. Imagine the amount of processing and the data analyst time that is cut down from our clients. That is really adding unique value to the clients, and it is not something a hotshot tech company can do without the context of all the investment models we have, all the portfolios we have, and the highest quality content we have in the investment industry.
That's very helpful, Jigar. And there are a lot of questions that I want to follow up on. But before we go there, you mentioned obviously OpenAI, Anthropic. I was just wondering if you could share. You obviously talked about Microsoft and Google partnership as well. I was just wondering if you could share which LLMs do you use, how do you expect this LLM landscape to evolve, and then also if you can talk about some of the disruption in the LLM industry like DeepSeek and your view on that front.
Yeah, the whole industry is driven by the AI scaling laws, right? The more, basically, it's very simple. The more compute you throw at it, the more data you throw at it, and the bigger the model is, the number of parameters. The bigger the number in all three areas, the better, smarter model you'll create. And that is the war everybody, all the AI players are on. They're running very fast on it. Now, the general consensus was the more hardware and the GPUs from NVIDIA you throw, hundreds of thousands that you saw with Grok and what Elon Musk's team did in a very short amount of time, the more money you spend, the better it gets. And then to your point about DeepSeek, they came up with some infrastructure innovations to do it at a lot cheaper price.
There is also a little bit of advantage they have for being late to the party in some ways, so there is. I'm not here to comment on exactly whether can you in one-time training, they just spent $6 million on that, for example, right? And those kinds of things, but from our perspective, all of this, we don't pick winners here. We try out everything. We have an AI platform inside MSCI, which wraps APIs from OpenAI directly with Microsoft partnership as well. Gemini from Google. We have models from Hugging Face, which are open-source models. We are working with Anthropic because they have some unique value propositions there. For internal tools, we talk to companies like Glean. We talk to companies like Perplexity. We're looking at that. Many different tools and enterprise-level GPT, which we can use, which we are partnering directly with OpenAI on.
We are discussing things with Databricks and especially Snowflake. We are distributing our data through Snowflake. How can we enable more AI directly on top of our data on Snowflake? So there are so many elements of where the LLMs are going. Now, the interesting question for me is, yes, there are these AI scaling laws, and everybody's talking about it in the industry. How does it apply to MSCI? What are the scaling laws at MSCI? If we think about the next 10 years, how are we going to multiply our revenue and sales and number of products by a certain amount of number based on the power of generative AI? For that, our three parameters are what is the context we have, how do we get more portfolios and more custom indexes built on our platform, what are the richest models we can create more and more of.
We're using AI in our research very heavily, so creating better and better models using AI. And lastly, combining all of this with how smartly we apply the latest technology at the great price point and performance and put it all together at a fast pace. And that's how we will scale. So we have to watch out for that. And Ashish, to your question about these partnerships, see, these partnerships have evolved, especially with Google and Microsoft. Initially, it was about buying capability, cloud and storage and all of that. And now it's going more towards higher-level conversations that how can you help us add top-line growth? How can you help us solve the real problems we have? For example, with Google, we are discussing performance attribution on fixed income. How can we do it at a much faster pace, not 20% or 30% faster, but 5x faster?
Can we prototype using their GPUs that they use for AI to run these things? The cloud players are also interested in helping add value to us at the top line as opposed to simply providing cost optimization. I think these partnerships are maturing. We're also creating some cool projects like Geospatial Explorer that Google invited us to the Google conference last year, and we did a presentation. Our engineers did a presentation of what we're doing. They also want to highlight that it's not about AI for the sake of it. How does a company like MSCI, a premium provider of investment tools, use their AI, and what do we do with it? That's unique value we're adding for our clients. It's also unique and exciting for the cloud players to show off what we are up to.
That's very helpful, Jigar, and I think the way you explained it, it definitely feels like MSCI is ahead of the game compared to most of the companies that we speak with. Maybe a question on agent AI, agentic AI, right? That's another one which is a lot more in focus lately. I was just wondering if you had any views or thoughts on AI agent or agentic AI.
Yeah, I think, look, I think it's in multiple levels. So if you look at the workforce, right, you'll have agentic AIs who can do some of the work that a financial director does, for example, or what a data analyst does or a developer does even, right? So what happens? Do you think AI is going to replace the knowledge workers and the cognitive work they do? It's a changing landscape because what knowledge workers like us do is also shifting with the power of AI. We'll get out of the grunt work we are doing so we can do more higher-value tasks. So it's a moving target to some extent. So I'm not too worried about these conversations about will it replace humans or not. At the moment, we're in a growth phase.
So if they can replace half our humans, we'll use those humans to do a lot more innovation, a lot more products and enter new markets and new segments and so on and so forth, right? So we're focusing more on what more we can do with the existing people we have. Regarding agentic AI, look, there is an advantage there. We have built MSCI One with a great new interface and integrating all the product lines into one place. But we still have a lot of rich heritage applications like RiskManager, BarraOne, Barra Portfolio Manager, and so on and so forth, Hedge Platform and Aegis and so on and so forth.
How do you rewrite all these things? Millions of lines of code, heavily used products, and old code base. Well, with agentic AI, you may not need to, right?
If you can ask intelligent questions about your portfolio for performance attribution, for example, or risk in a particular portfolio, you don't need to rewrite these complicated pieces of UI. If you look at the entire SaaS industry, I was the first developer on Dynamics CRM many, many months ago, spent five years building a CRM system for Microsoft. All these SaaS applications, like what is their future? There is a database, there is business logic, and there is UI on top of it. This UI is extremely old and extremely complicated. No salesperson likes the UI in Salesforce, nor in Dynamics. So what happens if a salesperson or a servicing person gets a quick answer for what they need to do? What happens if they can just converse with the platform, agentic AI, which says, "Do this for me," and done, right?
So if you look at deep research that is coming out, as an example, from OpenAI and Google, and in these areas, the way we think about it is we have a roadmap for our application platform strategy. We obviously do a lot of distribution through Snowflake, but we also need these tools to be having a longer-term strategy of convergence. And agentic AI is a big boon to us because a lot of these complicated processes running on top of extremely intelligent engines like the RiskS erver or the Index Platform, common index factory we're building with Foxberry and foxf9, we can provide a very simple experience. In agentic AI, you ask questions, you get the answers, and you get rid of all this complicated UI.
So for us, it's actually really, we see it as a boon to our industry, to our company, because there is too much complication in lots of acquisitions we do and a lot of, and even if you look at our client landscape, they have their own platforms. Nobody wants to have 15 different platforms. Everybody is trying to sell and install them all, and everything has to be, now people have to figure out how to use these things. With an agentic AI approach, we can show up with our content and our insight wherever you are. You are sitting on Teams, you're sitting on Bloomberg, wherever you are, we can give you the answer you need from us, so we are investing very heavily in this area.
That's great. And the way I understood it, you're willing to disrupt yourself in order to make sure you're ahead of the game. There are three buckets that you talked about. I just wanted to go back to the first bucket, improving developer efficiency. You mentioned a lot of, you provided a lot of good examples there. I was wondering if you could talk anecdotally how you think about improving efficiency or improving time to market and maybe elaborate further on how you've been able to leverage that to drive faster product developments.
Yeah, I think that's a very good question. I think everything new we are doing, it's much easier to drive faster pace of innovation. And the recent projects we saw, one in private assets, one in index, we are looking at, we have historical data, we have Jira tickets, like how long does it take for a storyboard to get completed, number of features written by a developer, how many check-ins do you make, how many commits you made. All of that data exists for the last several years, okay? So you can easily compare when a new project gets started, there's a detailed accounting from finance and from product perspective, how many engineers are working on something. We are able to show that we are actually moving at a much faster pace than before.
Now, you have to remember, a lot of people are working on massively old code bases and very complicated millions of lines of code as well. Now, in those areas, you may not immediately see the pace of innovation getting faster, but you will see immediately is the pace of maintenance, things like security updates, quality checks, automation, modernization of the code base, refactoring of the code base. Those, if you have 20,000 lines of code in a file and you need to, and you're new to this language even, it could be intimidating, but with the power of generative AI and GitHub Copilot and Cursor and other applications like that, we're able to tackle those things with a lot more confidence. So there are a lot of ways we are able to measure this.
The other example I'll give you is that in areas like data collection, we are now, we had, you know, every time we became more efficient with data collection, we would reinvest those humans to be applied to new products, right, and it was difficult to see what was going on, so now we became very deliberate. We tell the product team and we tell the finance team that every quarter we'll show you how many data analysts, for example, are now freed up. It's efficiency.
We have multiple options, right? You could say, "Okay, we don't need them." Very likely that is not the answer.
Or we say, "No, we deliberately put them on the new product." And the reason to do it very explicitly is then you can have accountability to show that truly we became X% more efficient in data collection.
So we are now also going to each function and saying how marketing, sales, finance, HR, all of them will become more efficient through the power of generative AI. We're looking at the most high-value tasks they are doing and helping them find a way to automate those things. And we are starting to track these things. So these are financial discussions going on about how do we track these things to make sure we are accountable to ourselves. That's one reason. Second is to justify the cost of AI tools. When we partner with these vendors and we spend billions on these kinds of tools, we negotiate with them if we don't see the right kind of efficiencies. Or we use the right tool based on where we really are seeing the ROIs there.
Yeah, actually, that was going to be my next question, and I'm glad you brought that up. Can you discuss what your investment strategy is for investing in cloud AI, and how do you measure those ROIs? What are the key metrics that you track? Because Gen AI obviously can be quite expensive. So.
Yeah, so I think, Ashish, if you look at our cloud spend, our modernization spend with AI or anything, there is no differentiation between that. And let's say you want to spend on building something in the wealth market or something for fixed income. We simply look at what is ROI, okay? Now, in some areas, you show cost avoidance as an ROI. If I don't use this piece of AI, which let's say costs me $100,000, then I'll have to spend these many people to do this task, and that will be more expensive. So cost avoidance, but it has to be done very rigorously and shown very clearly. The other example is to say you compare the cost of data collection for a million data points year after year after year, and you show how is that dropping.
And then for many areas of investment in AI, it's very simple. If you're building products like Geospatial Explorer, Ask MSCI, or AI Insights, then these products are either generating new revenue of their own in case of, let's say, AI Insights, or they are utilized as a tool to increase the prices when we are renegotiating a contract, when a contract is up for renewal. And that happens at a particular period. And these are very large enterprise deals. So different deals have different ways of getting compensated for the value we are adding. So when you take that into account and you say, "Okay, how much new top-line growth came out of these things?" And you compare that with the cost we had for the human capital involved, the cost of the models and the compute, and then we track all of that.
We do talk to ourselves, and sometimes we look at some part of the product in MSCI One, where we have very world-class investments in insights and AI. If the usage is not there for various different reasons, we discuss we should shut it down, right? We are very actively looking at these things. Nothing is constant. We validate what should we shut down, what should we keep going, what should we double down on based on very detailed financial tracking. If we don't do that, this will be pretty expensive.
That's very helpful, Jigar. You talked about, obviously, a lot of efficiency on the data generation process, your ability to scale up your data collection process. One of the adjacent topics there is synthetic data. I was just curious, if you generate or use synthetic data, how do you think about the evolution of synthetic data, if there is any role in the investment industry?
Yeah, there is a role for that, for sure. For QA, for example, it helps a lot to create synthetic data. For in things about like prompt generation, creating synthetic data to test out a few things, it's very helpful. But it is limited to those areas in my mind. We are not using it for any of the real core models creation, for example, or index production or anything, or ESG, sustainability, or climate data. In those areas, we are not heavily leaning on synthetic data. But when you want to do some anomaly detection and testing and QA, and we want large amounts of data, there is pretty heavy usage for that.
That's helpful. You mentioned, obviously, the M&A. The company, MSCI, has done a lot of tuck-in acquisitions, some big acquisitions, transformative acquisitions like Burgiss with the private capital data. How is the technology team involved? And some of the tech transformation that you've done helps accelerate the integration of those products or those offerings?
Yeah, that's a great question, Ashish. So if you think about Burgiss, Foxberry, Fabric as three examples, generally, they overlap with the last two years of generative AI investments and the big boom in the industry. And everything we had invested in building something like Ask MSCI, you can apply to a real estate or private asset side. The investments in AI we have done can be utilized towards when Foxberry gets fully integrated into our custom index platform, and we are very, very close to that. Then you can ask questions and use AI to create a custom index. They don't have to build that capability from scratch. For MSCI Wealth Platform, which is what used to be Fabric, same example.
If you want to create world-class search with auto suggestions and AI elements infused with it, all you have to do is connect the data sources that are powering that application and blend that with our AI platform, and you can turn these things on at a much faster pace than if you had to do it from scratch. Burgiss, in particular, what you asked, the GenAI platform of data collection we built on GCP was almost perfectly timed with the acquisition, the LP/GP data collection we are developing, and some of our clients want us to provide them as a service to them, it is going at a phenomenal pace.
We would have otherwise looked at some acquisition three years ago to buy somebody for quite a lot more money than we wouldn't even think about doing that now because we have all the AI models and AI tools we need from data collection, in this particular case from Google, and we're able to apply that to Burgiss and drive tremendous efficiencies in the way they used to collect data. Because remember, a lot of this was built years ago, long before generative AI was a big thing.
That's great, Jigar. You highlighted several products which are AI-driven or tech-driven, like for example, the AI Insights. Can you just also talk about your product roadmap over the next, let's say, three to five years? How do you think the products will evolve?
I think it is going to evolve around mass personalization and customization. It's going to evolve around more going towards real-time as opposed to batch processing. It is going to evolve towards more agentic AI that you were talking about earlier. It is going to evolve toward much faster scaling across either its markets or coverage of the securities or the amount of data you need in some areas to go from millions of data points to billions of data points. Once we create this capability horizontally across all product lines, it will power a lot of these vertical scenarios. What do we do in wealth? What do we do in private assets? What do we do for total portfolio foot printing? What do we do with climate?
All of these questions, the business strategy could be different and the go-to-market plans could be unique, but the capabilities needed for all of these things are tied with the entire conversation we had over the last few minutes.
Yeah, no, that's great. And maybe the last question would be just what are you most excited about?
You know, the thing that is most exciting is if I think about the next 10 years at MSCI, I keep reflecting on the scaling laws of AI. But if you think about what are the scaling laws of MSCI, we are writing a document to explain what are the different vectors in which we need to go 10x in data here, 1,000x in here, 300x in this area. How can we use the entire platform we have created on technology? Six and a half years ago, I came in, bulk of our content was distributed on FTP folders and SFTP folders. Today, we have AI-driven insights running on top of Snowflake and MSCI One. We have come such a long way. Now it is time to reap the benefits of all the investments we have made.
In the next 10 years, as we enter new spaces and new product lines, I'm very excited about the profound impact of generative AI and all the investments we've made in technology and data over the last six plus years. It's exciting. It's funny. It's going to start now.
That's great. Thank you. Thanks, Jigar. Thank you, everyone.
Thank you.