Jason Mirsky: Hi, I'm Jason Mirsky, Head of Data Solutions in Security Services.

Ashley Peterson: And I'm Ashley Peterson, Global Head of Fusion Sales.

Jason Mirsky: So Ashley, how long have you been with J.P. Morgan? Where were you beforehand? What were you doing? How did you come to be here?

Ashley Peterson: So I joined J.P. Morgan in 2023, actually working on the Fusion product. Previously, my experience had always been in markets. So I spent 13 years in markets covering institutional asset managers and hedge funds and saw a shift around the time that MIF had happened that clients were taking some of their spend and putting it towards data. So that started me on this data journey. And when I met with the Fusion team and they were talking about the innovative process and the journey that they were on, I got so excited to join the product team. And then shortly thereafter, I joined the broader data and analytics sales team, which is where Fusion sales sits.

Jason Mirsky: So as a kid growing up, it was data?

Ashley Peterson: Originally it was being a vet, but I guess that's very data orientated too.

Jason Mirsky: Close enough.

Ashley Peterson: What about yourself?

Jason Mirsky: Before I came to J.P. Morgan, I've been at J.P. Morgan about three years now. I joined in 2022, so just a little before you. And before that, I spent 23 years at an analytics firm. Analytics and indices is what they're known for. And so most of my career has been in analytics.

Ashley Peterson: So data management is critical to our clients. How do you think about the data management journey and why is J.P. Morgan making the strategic decision to go into data management?

Jason Mirsky: We'll go to the data management origin story. Every superhero or super villain needs a good origin story. And so I'll let you choose whether we're the hero or villain.

Ashley Peterson: Hopefully hero.

Jason Mirsky: Hopefully hero. But in terms of the origin, we are doing this because this is what our clients need help with. And they're all on their own data journeys. A lot of them have decided they're moving to the cloud in some way. And a lot of the ways that security services deliver data is by FTP, secure email or other channels like that. And our clients were saying, "Well, could I get my custody data directly into Snowflake or my fund accounting data directly into Databricks or AWS or Google BigQuery or Microsoft?" And so we want to be able to put the data where our clients are going to use the data.

That's what really started the journey. But then once you have the data where you want it, you realize there are other problems that you need to tackle. It's not just about the pipelining of data to the destination. You want the data in a similar shape and format, you want common identifiers in the data, you want the ability to express your opinion on the data. And so that starts to really get into the heart of data management and what we've been building with Fusion by J.P. Morgan in terms of an enterprise data management platform.

So you've been speaking to clients a lot. What do you see as problems that our clients have and are there differences in terms of problems in different regions?

Ashley Peterson: Absolutely. We've seen demand across all the regions, which is nice to see, but there's different market maturity in each region. So APAC is one that surprised us the most. Originally going in there, thinking about the technology journey, there are a lot of asset managers and asset owners that are very sophisticated, but what surprised us is that there was a good chunk of clients that were still slower to adopt some of those new technological advances. The EMEA region, the European region seems a little bit further along in that data management process. Our first solution was an ESG solution. So obviously in that region, there's a lot of traction there with the regulatory reporting that's required.

Jason Mirsky: Sure.

Ashley Peterson: And then lastly, in the Americas, it's a bit different, very similar to APAC where it's fragmented. So we have Canada, we have the US and we have Latin America. And what we see is that there's a large incumbent of data management platforms that exist. So typically with a data management platform, you're not shifting your data management platform frequently. It's usually every 5 to 10 years. So with that, where there are incumbents, we don't see that shift happening as quickly.

Jason Mirsky: Right. So the destination used to be just to get into the cloud, and more and more now that destination is the cloud and AI. So what does that mean for clients?

Ashley Peterson: So all clients come to us with challenges around flexibility, scalability, data governance, data quality. So we hear that from both asset managers and asset owners. We like to think about it in a triangle. All clients are asking for better, cheaper, faster. And when we think about better, so that's basically how we're doing operational efficiencies for clients. When we think about cheaper, that's cost management and focusing on that. So most of the time is spent around those two angles of the triangle, but now with AI, the faster component is being driven very quickly.

Jason Mirsky: Faster, better, cheaper. I love it.

Ashley Peterson: So one of the other challenges we see is around private assets. So data being trapped in term sheets, PDFs, emails. So extracting that unstructured data and making it usable. So a lot of clients who are investing both publicly and privately want to make sure that data's extracted and then they're able to see that whole of portfolio view.

So regarding clients in their cloud journey, we see clients typically sit in three buckets. One, they're either on-prem, two, they're in a hybrid, either on-prem and on the cloud, and then finally on the cloud. So from that perspective, what do you view as success in that journey and those transformation steps for the client?

Jason Mirsky: It's a good question. When you think about success in that journey, you are not just trying to get your data into the cloud, you want the data easily discoverable. You want the data easily accessible. You want the data entitled so that only the people who are supposed to have access to the data can get access to the data. You also don't want to impose a lot of work on the end user after they get access to the data. So there's a process in the industry called shift left. Can we do the normalization of the data upon entry into our data ecosystem rather than after exit from the data ecosystem?

Another aspect is a robust encompassing data model. So can you model all of the relationships that are within the data in the data architecture itself? That way it makes it very easy to extract that information with the relationships intact. And so when you think about that as a cloud set of goalposts, it naturally flows into AI goalposts in that when you have that data accessible, discoverable, and titled appropriately in a robust data model that's normalized, then you can start to think about exporting the semantic layer, the ontology of that data along with the data so that your LLMs can make use of it and generate high quality results.

Ashley Peterson: And I've heard you use this analogy before, but it's very similar to changing the wheels on the bus while it's being driven. So it's something that's very complex to solve for a client while making sure not to disturb their day-to-day operations.

Jason Mirsky: Absolutely. It's really difficult to change while data is in play. And so really what you need is the in-house expertise or the third party expertise on people who have been through that journey before in order to do it.

Ashley Peterson: And I think that's where J.P. Morgan is a very strategic player. We have great assets, but let's talk a little bit about how J.P. Morgan can strategically use those assets for something like this.

Jason Mirsky: Yeah. I mean, when I think about what we make use of in terms of the assets of J.P. Morgan, it's really what we call SMEs, subject matter experts. We have subject matter experts who work in our custody business who help us design our data model, subject matter experts from fund accounting who help design the data model for fund accounting, middle office trading services. And so we're pulling from the expertise of the bank when we're designing the data model that goes across the data within our architecture.

The other thing we do is we subscribe to a lot of data. When you think about all of the big data providers out there at Bloomberg or Refinitiv and ICE, like we have all that data at J.P. Morgan, we're creating our own security master at J.P. Morgan. So we are familiar with that process of mastering data from multiple sources. And while we can't give away the J.P. Morgan data to our clients, the data that they're licensed to receive, we can distribute to them. And so we make use of that expertise of handling and merging and mastering all of that data on behalf of our clients.

So after the data is normalized and modeled, how are the clients using it?

Ashley Peterson: So typically we would bucket it into two different categories. There's the technical user and the business user. So the technical user, those would be the data scientists, the data analysts, and they are using it for very specific, unique use cases to support the overall business. And if you look at leaner organizations, they might not have the capacity to go out and hire so many technical users or employees that are native in APIs, Python, Jupyter Notebook, et cetera. So we also need to cater to that business persona. So making sure that the tooling is there, whether that's the UI where you can access the data in a low code, no code way, or making sure that they have the dashboards and the visualizations to do that.

So one of the largest challenges we hear from business personas is the linking of that public and private data. So can we dive into private assets and how data management can help solve that?

Jason Mirsky: So before we get started, within this same market series, Montserrat, Larry did a great episode on private markets. I suggest you watch it if you haven't yet already. But going to the problem with private markets data, inherently it's private data. So you just can't subscribe to any vendor and get your private markets data. The data is hidden away in GP portals, it's unstructured, it is often in PDFs. It's often in emails. And there are technologies out there, particularly with gen AI, that are starting to extract that data from unstructured documents and make it structured. And so we make use of some of those technologies. But the second problem is that often your private market investors are using more than one fund services firm. And so each of those fund services firms is going to provide fund accounting data in a different format with different data points.

And so the second problem is aggregating that data across fund accounting firms. And then the third part of that is aggregating across public and private data. So even within a firm here at J.P. Morgan, we have a separate department that does private fund accounting versus public fund accounting. And so clients need to aggregate that to get whole of fund reporting. And so the data model that we've built up with Fusion helps to aggregate the data consistently across fund accounting vendors, but also across public and private.

The next problem in the private markets is entity resolution. There's no real identifiers for securities in the private markets. Often all you have is a name. And so we've built out a whole lot of machine learning to try and match names. So you can think of it as a whole complex set of rules where we take data from one vendor and data from another vendor, and we just try and match the name. That works sometimes. Let's capitalize the name on both sides. Now we get a little more matches. Let's expand out all abbreviations that are within the names. Let's remove punctuation.

And so you do more and more rules until you can come up with high fidelity names. And for those that you can't match, maybe there are other points like country of incorporation or revenues that you can extract from the datasets that'll help you in that matching. So we use that same entity resolution matching process across all different kinds of entities, whether that's securities, companies, counterparties. And so the matching process, what we call spining, is a key element to our data offering.

So look, we've talked about private markets. What about regulation? How's regulation affecting our clients and how's it driving their data needs?

Ashley Peterson: Regulation's always a hot topic.

Jason Mirsky: Always.

Ashley Peterson: So globally, we see a lot of countries looking at data leaving their country. So in Europe, we're looking at GDPR, same issue in Saudi Arabia where they want to make sure that when data either can't leave their country or when it does leave their country, that it still abides to the same regulation of where the data originated.

We can also think about regulation in two separate buckets. So one is where the regulation is driving the client. So an example of that is the new reporting regulations out of Australia. So we hear a lot of Australian clients looking for datasets around modern day slavery and supply chain because of those regulations. And the other bucket is where the client's trying to set their own internal goals before a regulation is rolled out or potential audit. So we have a client in Singapore that looked at their data management platform, wanted to create a more robust governance and audit, and that's what drove them to change their data management platform.

Analytics is your bread and butter. So maybe we pivot from speaking about data management specifically and talk about how, now that we have the data managed, how we're leveraging it from a Fusion analytics standpoint.

Jason Mirsky: Sure. So the most important element is high quality data to go into your analytics. But our clients are asking, if you think about security services, we start, we custody the assets. Once you know what assets you have, you want a valuation on those assets. So we strike a NAV with our fund accounting process. And then the natural question is, well, what's the performance of my portfolio? And so we help our clients with performance, but also performance attribution. Why is my performance in a certain way vis-a-vis these benchmarks?

The same is true on the risk side. We help our clients with ex-ante risk, where we can understand what's the value at risk. We can stress test those assets. We also do risk attribution. We help our clients with liquidity analysis, understanding how many days it might take to sell a particular asset or portfolio, or what's the cost if I needed to sell within one day, like those sorts of analyses.

As you mentioned in Australia, there's ESG regulation in Europe, there's also ESG regulation. So we help our clients with ESG analytics, creating pillar scores for their portfolios, carbon footprints for their portfolios. We do post-trade compliance to make sure they're not violating any of their policies or any of their regulations. And lastly, when we think about the private assets, we want to do look-through analysis. So understand what's within the portfolio underneath that fund name and be able to aggregate that across funds, both publicly available companies, so mutual funds, but also private funds.

So in addition to analytics, what other things are our clients asking for with the data?

Ashley Peterson: Even though we're talking about a post-trade world, clients still want to access their data at speed. It's not just the accuracy, but getting that data right and in a timely manner. And second to that, we get a ton of demand for dashboards, right? So that business persona, even the technical user, wants to make sure that they're able to visualize all of that data in one place. And then lastly is really that audit trail. So making sure when there's a change to that piece of data, that they could go back, see who made the change, if it's a different team, who's approving that change, and that audit trail's becoming increasingly important for clients.

Clients are asking to backtest their data. So we want to make sure that we also give that data in a bitemporal way. But I would be remiss if we were having a data conversation and I wasn't asking about AI and what you're seeing from clients from an AI perspective.

Jason Mirsky: Yeah. I mean, that's what everybody talks about today. And since ChatGPT was released, it's very easy to put together some sort of pilot using an LLM. And a lot of clients have embarked upon this and they see initial success, but what's difficult is doing that at scale in a repeatable, robust manner with good governance and the proper entitlements so that you're not releasing data you're not supposed to be releasing. And so having a data platform that can work side by side with your governance policy, your entitlements, and allow you to create AI at scale in a robust, repeatable manner is really the goalpost here.

Ashley Peterson: That makes a lot of sense. Fusion sales had over 500 client conversations last year and every client that we spoke to has released some sort of ChatGPT or LLM suite type functioning within their organization. And what we hear and talk about with clients is J.P. Morgan, because of the breadth and the depth of our business, we are looked at as thought leadership. So I know Jamie's mentioned that we spend two billion in AI a year. So while the buy side and the sell side might approach AI very differently, they're looking at J.P. Morgan and also our chief data and analytics office to kind of guide and shape that conversation.

Jason Mirsky: Yeah. And when you think about it, there's really three pillars here. There's the data, there's the platform, but then there's the operating model. And people neglect the third one, the operating model, but that's the critical element for success in your endeavors.

Ashley Peterson: Which goes back to the bus analogy.

Jason Mirsky: So one of the keys about having a successful AI project is having high quality data. High quality data generates high quality results, so you need to have the pipeline, have the technical data quality, the business data quality, what's called TDQs and BDQs in the business. And you have to have that data normalized. You have to have consistent identifiers. You need a data dictionary associated with that data. You need a semantic layer and you need to be able to deliver that semantic layer out to the LLM so the LLM can understand what the data means, and that will generate higher quality results for you.

Ashley Peterson: So what's something that is not standard now that we expect in the next 5 to 10 years to be standard, and where is the puck going?

Jason Mirsky: When you think about the three phases of AI that we see, the first is just play with the LLM itself. The second is, can we give the LLM a knowledge base to work off of? It's called RAG, but the end result here is that you want the LLM to be working off of your data, your data as a strategic asset. And then the next phase after that is agentic AI. What I see really in five years that we're going to take for granted is that chat for data with your own data, your own strategic asset will just be ubiquitous.