Accelerating New Revenue Streams Through Innovation​ (Moody's)

Dive into the strategic realm of tech innovation with Luis Amador, GM of Moody's CRE business, in an engaging session exploring a compelling approach to new product growth amidst challenging market landscape. Discover how leveraging technology (including GenAI) as a strategic differentiator can propel the development of new and thriving lines of business, offering invaluable insights for navigating the evolving landscape of commercial real estate.

Transcription:

Louis Amador (00:09):
Okay, great, thank you. I just want to start by saying thank you for joining me. My name is Louis Amador, I'm a General Manager for Moody's Analytics, and I started with Moody's Analytics back in 2006 when the company started looking at the ratings part of the business. We all know Moody's for ratings and we split the company into an analytics side of the business and a rating side of the business. And since then we've been able to grow our business to about 3 billion a year in revenue. And over the years we've always looked at ways to accelerate growth. Obviously we've grown a certain extent, but what I wanted to do today is really show a case study of how growth can be tied to leveraging technology in a pretty big way. And many of the panels that you've been to at this conference, I imagine there was a big focus on gen AI.

(01:17):
gen AI is also going to be a big focus here as well. One of the ways that I wanted to try and show you what I mean about how technology could actually be pretty disruptive and aid you as you help grow your business in this case study is by actually just doing a quick demo. How many of you have folks that are involved in commercial real estate lending or CNI lending? Just show. Okay, show of hands. Okay, good. So we have the right people in the room. What I'm going to do is show a quick demo of something that is being leveraged in the market today, leveraging gen AI, and we actually worked with McKinsey in the business build process of this starting in 2021. And one of the ways that I'd like to show it is just by showing, just by running it for a few minutes and then finding out what you think.

(02:17):
We've all been there, we've walked into our offices and news comes out, and the news actually maybe may have an impact on your lending book, on your portfolio. And ultimately you have several or many smart people running around trying to find out what is the impact of that news to your lending book or your portfolio or your business. And typically, depending on the state of your data and technology infrastructure, that could be pretty tedious and time consuming and highly manual. So I am going to simulate the process of coming in the office and going to our new system here, and I get an alert and the alert says Rite Aid is closing stores, and a gen AI assistant appears. So a gen AI navigator appears and this gen AI navigator, the gen AI navigator mentions that there are store closings in this alert. The date was 1-17-2024, and there's an article that shows the information on those store clothings.

(03:41):
I opened the article and says 45 stores are clothing, clothing, and obviously I want to know as a commercial real estate lender, what could be the impact of those store clothings to my loan book. So this Geni portfolio assistant goes from looking and looking at news in the open market and then mentions to you interacting with me and says, based on the article, based on the reference article, your portfolio may have an exposure to Rite Aid closing stores. Below is a list of loans and properties that are backed by properties in your portfolio that have Rite Aid listed as a tenant. So here I'm interacting with a portfolio assistant and it says four loans are potentially impactful to me based on what this article is saying. So I look at these four loans and I quickly analyze some of the data that the gen AI portfolio system is displaying.

(04:45):
And I notice that loans one and loans three, Rite Aid is the only tenant that's backing that loan. So obviously it's a single tenant scenario and that's more impactful to me as a lender and I want to find out more about it. And commercial real estate, obviously the lease end, the in place rent is important. The net operating income coming from that property is important and the ultimate value of that property and the debt service coverage ratio on the LTV. Now I am mentioning terms using commercial real estate, but everything I'm going to show you would apply to any other asset class or sector. So the way that this was developed, it's completely asset class agnostic. I'm purely using commercial real estate as an example. So the gen AI assistant tells me, look at these four loans I do, the analysis and loans one in three are more concerning.

(05:43):
So I'm going to type in based on the information above, I'm concerned with the two locations that represent a hundred percent of the rent loans, one in three. Please assume that the in-place rent drops to 85% of market and that the tenant leaves and provide me a new NOI assume a seven and a half cap rate in this submarket and infer me a new value, give me a new DSDR and a new LTV. So the platform will now isolate these two loans and the gen AI assistant is responding by telling me that based on the assumptions that you just gave me for these two loans, assuming 85% of market for the rent, I have a new DSCR and a new LTV for this loan, assuming a new cap rate. I've isolated these two loans and I'm actually going back and forth with a gen AI assistant.

(06:49):
And I also know that these metrics are inputs into the credit models that I use for risk ratings and other internal probabilities of default calculations. So I'm going to ask the Geni assistant to now go and run a credit model. So please take the updated values, run a baseline scenario of interest rates. It could be your interest rates or interest rates from a company like ours. Run our credit model or yours. Ours is called CMM. Use the most recent forecast for baseline interest rates and provide me a new implied rating and new expected loss. Now I scroll to the right and for those two loans that have isolated, I have a new expected loss for those two loans. And remember that all of this started at the very beginning with an alert that came in when I came in in the morning, and it was based off of an article that Rite Aid is closing 45 stores. And this interaction with this gen AI navigator would usually take multiple days and a lot of code writing potentially to make sure it's accessing the right data, the right models, and ultimately I was able to do that analysis, the alert based on the article, the back and forth interaction with the data, and then ultimately running a credit model in five minutes.

(08:25):
Now for many of you in the room, how long do you think that would take a scenario like that happening with your portfolio managers? I'm just curious if anybody has anything, any input on that longer than a few minutes. Longer than a few minutes. And typically in my experience, it takes a couple of days and what's happening now with gene AI is extremely interesting. It is not only able to parse with its large language model parse information that's out there in articles or in documents, but now we're able to take that information and then use the output from that analysis and have it feed internal models, do math, and typically someone can create a report or a BI tool, but I'm just interacting and asking it any question that I want because anytime that you do try to build a report, ultimately you would want changes to that report and the changes and the changes that mount up is what makes it very difficult.

(09:37):
Now I'm going to also look at a second alert. Now the second alert tells me that there's available space in available and that this available space may be a property that's backing one of your loans. At Moody's, we made an acquisition a few years ago where we have 24,000 brokers that are listing properties into our marketplace. So in this platform you can see when there is a broker listing a property and we know if that property's listed is backing one of your loans. And this scenario here is alerting me that essentially for one of your loans, there's an office listing, there's a sublease situation, and I can go in and basically run the same type of analysis again based on an alert that came from a listing of a property at that point. Now the first example was going through an example more on the comprehensiveness or the amount of work it would take to run analytics here.

(10:46):
The example is more on the real time nature that the alert can give you. This is the second that a broker may list a property that you are alerted that it may impact your loan book, right? And that's also pretty interesting using gen AI. The most common scenario for you all is a regulator in the building and they have a question about your data or your portfolio. And the ability to answer that information for quickly can mean the difference between having an MRA or not, right? Depending on the ability for you to surface that information quickly. So here the example highlights a regulator saying, well, give me your 10 worst office loans, industrial backed loans, and sort them by give me the worst loans based on LTV, please sort them from highest to lowest. And again, in seconds I can get that information because it's connected to your loan book, it's connected to market data and ultimately the gen AI assistant is what you're using to actually navigate through this process.

(12:07):
If you want to also then create a pie chart of this information, a scatterplot of this information, it's just a matter of just asking it now versus again coding like we were doing for the last 20 years of doing this type of analysis. So this for me is extremely different. The way that we're operating is changing. I'm literally just going back with a gen AI assistant asking it to create a chart, asking it to do math, asking it to run some analysis, run a credit model, look at news. The last example I'm going to show before we go into how we actually put this together with McKinsey is a natural disaster scenario. So let's say there's a natural disaster potentially or a hurricane potentially coming through south Florida. The alert comes in immediately. This portfolio assistant will come in and notify you that there are several assets that may be impacted in your loan book.

(13:13):
And I may want to ask the gen AI assistant a question about these assets and I'll ask it. I'm most concerned with the properties that are one mile in of the coast. Please calculate a stressed NOI assuming a 30% loss of income from those properties that are one mile in from the coast. Assume that I lose 15% for assets that are a mile and a half from the coast and then assign a new cap rate because given the conditions in the market, you'll get higher cap rates in that scenario and then run that analysis on those assets. And then give me the new metrics. So again, there are multiple alerts and multiple scenarios, and the interaction now is just completely different in the way that you work with data and using gen AI within an institution. And I'm going to flip for a second on the case study and how we got here.

(14:14):
So back in 2021, we worked with McKinsey because we knew at Moody's we had a plethora of assets internally that we can utilize for more use cases like this. And we chose the commercial real estate market as a market to innovate in. Because of the manual nature of this market still, you still have spreading done manually. You still have quite a bit of digitization that's needed. So what we focused on is looking to find ways to eliminate those manual workflows. And then the market started to change, obviously at a higher interest rate environment, higher inflation and analyzing those risks became more and more important. And you also have looming loan maturities that are our customers were more concerned with. And it just made sense that the solution like the one you just saw would be needed in the market. And the gen AI variable is what really made it a game changer.

(15:23):
Once we layered in genai, it really changed the way that the trajectory of these types of solutions, what we did in this case study at Moody's, and remember I mentioned that in 2006 we had just started Moody's Analytics and we were looking to grow our business rapidly over the next few years is that we picked asset classes or segments and we told ourselves, can we leverage more of our data? Can we work with customers to not only drive our revenue but drive customer revenue solutions that we can bring to market that if you were using it, you can drive your own revenue and also reduce expenses. And if we line those things together, then we would consider it a major success. What we did in the case of commercial real estate is we worked with McKinsey to bring all the pockets where we had analytics and data and workflow solutions into one operating unit with McKinsey.

(16:26):
We designed a playbook in that even though we were growing at Moody's at a high single digit every year level, we knew that there were areas that we can grow faster. So we developed a playbook that allowed us to continue innovating and making sure we have a pipeline of innovation coming in every year and ensuring that that's happening across all of our operating units. And that muscle of innovation was important. We also made sure that for the pockets that we wanted to grow, that we were hiring the right people, that we were looking at our go-to-market capabilities, looking at everything from talent strategy to product design. And we did that alongside our McKinsey business partner and we also wanted to make sure that that culture stayed within the organization. So again, making sure that single digit growth wasn't good enough, that we were looking for growth that was more larger, but also looking to make sure that that same mindset was in the products that our customers were using to enable their growth.

(17:45):
What we produced in 2022 was an ecosystem of products and starting with commercial real estate. So I showed you an early warning platform at the very beginning, but in this ecosystem of products, we also developed loan operating systems. We expanded our commercial real estate data to 8 million properties covered. We also expanded our model output and again, leveraged gen AI in all of these solutions. The biggest challenge that we had is obviously we have 2,500 banks that we have as customers. The biggest challenge that we have and the biggest response that we got is from our customers is how do we get the data in how do this looks nice, but how is it that you can get your data into these solutions to perform the analysis that you just were able to perform? So we leveraged AI at the very beginning on the data ingestion side.

(18:47):
We make sure that when we went in and started to implement that, AI was used to parse rent rolls, to parse operating statements, to spread financials. And by using AI in those ingestion tools, we then made it possible for the data to come into these tools, whether it's a loan operating system or portfolio management system or a credit model. And not only were our bank customers happy because their data was more structured, but also they can run the type of analysis that you saw before and do it in a new way. Every prompt that I actually went back and forth in the tool with is code that someone would've had to write using an API in order to get that metric and that response back. So now with gene ai, that code is automatically written. So we focus on data ingestion. We focused on leveraging gene AI and making sure that gene AI taps into not only the market data, but taps into your portfolio data and taps into the analytics capabilities to give you this type of functionality. And for us at Moody's, we're pretty excited because several of these solutions have been implemented at the top banks around the world, including digital banks, and we see GenAI as the enabler right now in all of this. Not only are we able to save on writing code if we enable our products, but you are as well. So I wanted to get a sense as to what you thought. I mean that interaction back and forth with a product like that, is that something that you've seen before?

Audience Member 1 (20:50):
Probably what? What is the default probability of a

Louis Amador (21:19):
Right. So we've been in the early warning space for many years and our traditional early warning systems do that. So we use our probabilities of the default to calculate the probability of that happening. Where the innovation came in is tapping into events and data real time and having the impact basically analyze real time when it actually happens. You can forecast anything you can, but the event and when it happens and the work that you have to do is critical.

Audience Member 1 (22:00):
Are you getting questions from some of the larger, sorry?

Audience Member 2 (22:04):
Are you getting questions from some of the larger banks on how they can incorporate this application into their existing workflow applications as opposed to a standalone application like this?

Louis Amador (22:15):
Yes. Yeah, that's a good question. So on this slide, the ecosystem concept that we have is designed around the aspect that you don't have to use the whole ecosystem. So everything, and especially with the work that we did with McKinsey in 2021, we developed a data foundation, a data lake foundation, and a single set of APIs that powers every part of this ecosystem. So if you only wanted to use the early warning system or you only wanted to use the data or only wanted to use the loan operating system on transactions, then you could do that based on knowing and being comfortable that everything is coming from one foundation and you only have to in the minute that you want to go to another part of the ecosystem, you've already integrated an API, that's a single API from us, and you can expose new variables if you chose to go into another part of the ecosystem. So I just walked away from a bank, walked into a bank where they said, look, we're using, we like your loan operating system, but we're not ready for it. We like our loan operating system, but we want to use your early warning platform. And so they're only going to use that.

Audience Member 3 (23:40):
I guess just reversing the arrow of that prior question, have you thought about or is there a capability for clients using your system to incorporate their own data in some capacity that's secure?

Louis Amador (23:55):
So obviously we have a large data business, so the focus for us was to make sure that was secure. So the same security and protection that we put on our own data is then applied to you as well that you can leverage and know that your data's safe. But what I showed you at the very beginning is leveraging the customer's data, right? Yeah. So that solution that I showed you at the very beginning knows the gen AI assistant knows your loan book, it knows your loan book specifically, and all you need to do is provide it, the addresses, the rent rolls, and the operating statements, and the rest is digitized by the AI agent. But the first step is ingestion of your portfolio in a secure, secure way. So

Audience Member 4 (24:57):
Data potential to compliment, basically extend features of the alternative.

Louis Amador (25:13):
Absolutely. Yeah, absolutely. I mean, the large language modeling that we do looks at structured and unstructured data. So learning new documents has been something that depending on the asset class is already happening. So I showed this for commercial real estate, but we have examples internally where people are working with servicing data, all the data that needs to structure before it hits a loan operating system or early warning system or an analytic system data from your core system. So the very first part of the implementation is to talk about what you've done internally and what you're able to surface, whether it's through a data lake that you've developed or that you need help in making sure that you're more ready to do that. And you may just give us documents. So depending on the institution, it just depends and depending on the asset class. So there are many more documents beyond this that the technology is used for, and we're open obviously to work with you on that.

Audience Member 4 (26:31):
You have a large

Louis Amador (26:35):
Large language model. Yeah,

Audience Member 3 (26:40):
You're taking the portfolio data and then you're extracting from that and you're using gen AI. Are you also taking from other large, where does that data go and how do you extract from it? Does it go into the broad ecosystem of data?

Louis Amador (26:54):
No, no. So the way that it's designed, because we have large asset management customers and banking customers, is that that data sits in its own instance specific to the entity. And in many cases, for example, we've got the complimentary data market data, the models, whatever you want to use in the ecosystem, but your data is not going in any process that enhances somebody else's dataset or environment. And because we have a large data business, we don't want our data going anywhere either. So those same practices are used when we work with customers.

Audience Member 4 (27:52):
So are you able to view this across an entire enterprise organization in the sense that if I'm looking at, let's say a Citibank and there's a natural disaster in some country and we've got a commercial relationship, a retail relationship, I can look at every aspect of each business and the overall impact that that can have on each individual business in those portfolios?

Louis Amador (28:13):
Yeah, I mean depending on the type of lending that you're doing or investments, that parent child relationship and that lineage is something that is done during the implementation. So every example from whether or not Rite Aid was backing a loan or Rite Aid was backing a bond in a securitization or there's a CI loan every scenario at this point. We've seen so many scenarios, obviously with our architecture over the last 15 years. So that part of it is covered. Yep.

Audience Member 3 (29:03):
Sorry, is the data bundled with this product or do you sort of charged for all these things separately?

Louis Amador (29:13):
The data is bundled. So the philosophy that we took again in working with McKinsey early on is we made a decision that every product should have integrated data analytics and potentially workflow. So there's one price for the solution.

Audience Member 3 (29:36):
Does that licensing relationship provide just access to, so if I ask a question, it might include your data as a part of the response, or is it a general licensing agreement where the client can build things off of your data?

Louis Amador (29:52):
The client can license just the APIs or the client can use it in a way where it's just specific to the portfolio only, or the client can use it with their portfolio and our market data. So there are settings in the solution that just allow you to set that on your own. And obviously we would entitle everything on the backend during the onboarding process. So for example, some customers do not want any interfaces because they're happy with their interfaces, so they just license our APIs.

Audience Member 5 (30:28):
So you mentioned that this is modular and you would, when you demonstrated the gen AI application, it was mostly as a chat assistant and extraction tool or to query in the research and insights in the monitoring and signaling. How is it determining what is the relevant information for you?

Louis Amador (30:50):
For each asset class? There's an infrastructure diagram that I can share with you. Essentially, for commercial real estate, for example, there are 800 articles that are reviewed every day by hundreds of operations folks as well. So there's the Jenny I interpretation of articles, but if the operations team sees any duplicates or anything that's not relevant to a commercial real estate transaction or a bankruptcy or a host of 20 relevant events that may impact what we call life events for a lender, those are calibrated at the very beginning. So right now we look for 20 types of events, but we work with the customer because the customer, essentially, some of the customers we implemented with, they wanted to bring in, for example, the assets that were the comparables at the very beginning when they were underwriting. So we have different permutations of this where you can run it with your subject, your loan book, but also if anything happens with the comps that you use during that underwriting process, you'll get alerted as well for 20 life events. And those could be expanded as we learn more and work together.

Audience Member 5 (32:16):
And it's modular. So you can subscribe to one component of this and not

Louis Amador (32:21):
The one component or, and since we have 2,500 bank customers, undoubtedly most customers have at least one bubble on average, we have customers with one and a half, one and a half to two.

(32:43):
Now the gen AI aspect, the way we look at the chat box, it goes way beyond the chatting. The navigator part where we worked and invested a lot of resources was making sure that we can have that assistant be almost like someone that's sitting next to you at your office. So if you need to run a model, you ask the assistant, if you need to do math and compare two numbers together and give you the deltas, it will do that for you again. And the way that you type could be the way that I type or the way that you type. So all of that is where the innovation has gone in over the last four years of this. And the beauty of it again is that it is a solution that we're seeing where if a customer has hundreds of portfolio managers or hundreds of analysts, those analysts could, the use of their time could be better put to use somewhere else. The key is that for these solutions, and when we worked with McKinsey at the very beginning is how do we innovate faster, but how can we drive not only revenue for us, but drive revenue for the customer? So in using this solution, you should be able to do your cost benefit pretty quickly depending on how large your book is.

Audience Member 6 (34:24):
What sources are you pulling from for the early warning system? And so what your demo with CRE, do you have the same for CNI?

Louis Amador (34:37):
So CNI is the next release. So we started with CRE just because the market demanded it, given the volatility in the market. But the infrastructure that we put together and the way we built CRE is like the news aspect is something that's asset class agnostic, right? The way that it looks for news in the market, the way that it looks at your portfolio, the way that it runs a model, it would just run a different model. So the next release of CNI, and after that it'll be ag lending. It is already designed in a multi-asset class way. CRE was the first one that came out. The loan operating system came out over two and a half years ago. And that has AI for credit memo generation for data ingestion. So the folks that are using the loan operating system have an even easier transition to, because the data's already in the loan operating system

Audience Member 1 (35:41):
From outside of your systems.

Louis Amador (35:45):
So there's an ingestion capability, and that ingestion capability could also be licensed on its own. So let's say you only wanted to use the ingestion capability, and that's it. That is designed to ingest, structured and unstructured data, and you do that on your own without us. And if you need it to also have us handle the exceptions, we have teams that can handle the exceptions for you as well when there's an exception and the data that's pulling in. So essentially, I believe what I'm showing here is pretty disruptive in the way that that lending institutions are actually operating their teams. What we're seeing is cycle times decrease. We're seeing higher precision when it comes to analytics and less manual errors, right? Because they're using AI across every part of the ecosystem. And the alerting is more real time. So we meet with, many of our banking customers have portfolio managers that have great relationships with their asset owners, but depending on the institution, there may be five PMs or 500 PMs and getting information for sure real time is something that the customers that have onboarded the solution really appreciate.

(37:29):
And the alerts vary, right? The alerts vary. The alerts vary from someone listing a property, a broker listing a property, and that being connected to your loan book to standard calculations that you run every day, right? So if you wanted to use this the way that you were using it every day for what you do today, I didn't show any of that because the assumption is that you realize that you can do that in a platform like this, right? If you wanted to have a report that basically shows you nice pie charts and scatter plots based on stratifications of your book, that's something that you can design here and save the template and it's always there ready for you every time that you go into the system, right? So I purposely ignored the standard use cases for today's case study. And really the big thing was the point that I made earlier on that in 2006 when we started at Moody's Analytics, we knew that we can grow pretty fast internally, but where the light bulb moment came is that we needed solutions that were leveraging technology faster. We needed solutions that were aligned from a revenue and margin perspective with our customers. So if it wasn't a solution that wasn't going to help you grow your business, we weren't going to put it on the roadmap. And that's what we did in our business build journey with McKinsey.

Audience Member 3 (39:04):
Oh, sorry. Do you have any clients that are on the asset management credit research teams side?

Louis Amador (39:12):
Yes. Especially with the rise of private credit, we probably have over a thousand asset managers as well. And whether they're asset owners or asset managers, they're leveraging that ecosystem as well, either to make loans or to monitor their portfolios From an asset management perspective, either backed by insurance or backed by pension fund resources, we have them in our client book. So we have about 2,500 banking customers, another thousand asset managers. Now what you're seeing there is relatively new on the early warning. So that just came out, but the ecosystem already has the other three bubbles already has 2,500 customers. The data, the analytics, and the loan operating system. Right now we're onboarding folks for the early warning platform, which has every bubble integrated with it. Alright, so thank you for your time and if you have any questions, feel free.