The Status of Tech and AI Incorporation into the Muni Space

Transcription:

Lynne Funk (00:09):

Hello everyone and welcome to our panel on technology, the status of tech and AI incorporation into the muni space. I'm Lynne Funk, Executive Editor at the Bond Buyer, and I'm delighted to have this wonderful group of panelists up here to discuss all of these important things for the industry. To my left, I'm going to go left to that direction. I've got William Kim, who is CEO of MuniPro. I've got Tyler Traudt, who is CEO and Co-founder of DebtBook. Next up is Gregg Bienstock, Senior Vice President Group, Head of Municipal Markets at SOLVE. And the next up is Justin Land, CEO of AG Analytics. And lastly, Dan Silva, who is CEO of Adaje. So welcome to my wonderful panel and we'll get into it. So I think when we prepped for this panel, we talked about the discussion really just turned about what our panelists are really providing as solutions to problems.

(01:10)

And that means different things to different market participants, issuers face different challenges than underwriters, then do FAS, then do lawyers, et cetera, et cetera. So that's not to say there is an overlap. There definitely is, but the end goal is oftentimes the same, right? Make the industry function better, make it more efficient, make it more effective and transparent. So we're going to touch on how each of our experts see where they provide solutions to the various problems and challenges in the municipal market and how and why everyone should approach adapting to and evolve with the technological opportunities that do exist. We're also going to go with a bent on AI, how AI is, what AI is and is not in the muni market and how it's being implemented and how it's expected to impact the industry broadly. But I want to start off with issuers. Oftentimes they are the group that perhaps struggle the most from staffing levels to market expertise to cost limitations. So I would invite the panelists to talk about what exists today for issuers that did not say even a few years ago, and I welcome whoever wants to kick it off there.

Tyler Traudt (02:20):

I'm happy to go first. Excited to be with everybody. My name's Tyler Traudt. I spent 10 years in public finance investment banking. I was a municipal advisor as well before starting dev book five years ago. And I'll say that when I started the company, I really didn't understand exactly what the issuers were doing all day. To be really clear, my experience with an issuer, with a treasurer on the team was when they needed to borrow money and I'd raise my hand. I'd say, great, let me help you with that. I can be very valuable to you and your team. And as we've gone and grown and spent more time on the treasury side working with our nation's governments and hospitals and colleges and universities, we've learned the job that has to be done in the treasury function is really pretty tremendous. We have debt managers obviously who work to help with post issuance compliance and new issuance.

(03:10)

Those folks have a peer, they've got a cash manager sitting right next to them. They have an investment manager sitting right next to them and they probably report to a treasurer. They might be doing all of these jobs. You might be working with a deputy CFO who is acting as the treasurer and oversees all of those areas. So there's a tremendous amount of work that needs to be done. There are a significant, really unfortunately, a tremendous lack of existing tools. I think five years ago to answer the question, these folks are generally stuck using Excel spreadsheets, an internal access database or software that was built in 1985, which is fine, but as many folks up here will attest to, there have been tremendous advances in the tools that are available and the general ease of use of software now. And so just to quickly grab it today, there are a number of companies that are out there focused on helping the treasurers, helping their teams do their jobs all day long.

(04:06)

I didn't know that payment processing was an issue. I didn't really understand year end financial reporting and how the treasury team had to work with the controller's office to go and take care of those things. We have customers now who lose a cash manager and all of a sudden the debt manager is a cash manager. And we're all, as a former investment bankers and financial consultants, trying to figure out why it's so darn hard for me to get good information, maybe in some of these cases. Well, there's a tremendous amount of work that's being done inside those organizations. So I think especially as we broaden into AI generally bringing automation to these teams to enable them to do more work, faster work so they can go spend less time being operationally focused and more time strategic, it's going to be super, super valuable. And we're certainly one of those companies that's having a lot of fun trying to help them do it.

Gregg Bienstock (04:52):

Yeah, I'll just add to that pre what Tyler just described here is one of the things that we were involved in continue to be involved in is the idea of issue or disclosure management. So your point about the cash manager leaving and all of a sudden someone else having to pick up the role, imagine someone's responsible for disclosure obligations, the person leaves and nobody knows what the heck they're required to disclose. So that's one piece of the puzzle. The other way that we've been helping issuers over the years, and this gets to me, gets more to the point of a theme that I always tend to speak about, which is everybody requests more and more data and that's great, but if you don't have the tools to process the data, right, you need data to become information. And that's really important. So we have for example, a pricing platform.

(05:37)

And one of the ways that our issuer clients are using it is number one, they just want to see how their bonds are trading in the secondary, right? It's really important for them to understand. And we had a client who said to us, I have two people who spend Friday afternoons calling around to see, and now we can do it in 10 minutes. And the other is, I want to have the same information. I want to have access to the same information to understand when my or my banker is sitting down with us and we're talking about where my deal should price, I want to be able to run my own analysis and identify comps in the market and do those types of things. So those are the kind of efficiencies that we've looked at. And similar to what you're talking about Tyler, about bringing to the issuer side of the market.

Dan Silva (06:19):

Thanks Greg. Am I on over here? I don't think so. I don't think. Just talk louder. We go. There we go. Alright, so hi everyone. Dan Silva. I'm with Adaje. We are a debt structuring and modeling tool and management platform. So one of the things I found interesting is just to piggyback off of what Tyler and Greg were saying, there's a wealth of new tools out there that help with data transparency, that help with productivity, where they're really great multipliers in terms of what you can do with these functions. You all are doing so much with respect to capital planning, with respect to projecting out what this new issuance is going to look like with staying on top of refundings, staying on top of all these limitations that are also imposed on you in terms of coverage, making sure the rating is in a good place.

(07:10)

And what I was struck by is there was not a good tool for folks to do this. On the issuer side, we mainly compete with something, you all probably might be familiar with DBC finance, which is great for its day and age, mid nineties software, but believe it or not, there's a lot of room for improvement. We're taking processes that folks would take weeks to do in terms of the capital planning side, projecting out what their bonds are going to look like and doing it in seconds and minutes. So I think it's really interesting to see that there's all these different processes that all these different functions where it would take a ton of time and we're seeing that whittle really, really shrink down.

Justin Land (07:55):

And I think that's important. I mean, I come from the kind of other side and AG Analytics is a startup of assured guarantee building software to scale credit analytics. I think Greg's point is everybody wants to ingest all this data and they don't have great ways to do it at scale. And in the last panel they talked about the growth in the separate account business and how complex that is, that you're not running 10 billion in one fund, you're running 10 billion in 20,000 accounts and every single state and you have tiny issuers and big issuers, analysts are expensive. How can you leverage all that knowledge and have the computer do most of the grunt work, for lack of a better term? So the analysts are doing things that are better for their clients and investors. And I think in the long run as more of this technology has taken on, it's actually great for issuers, especially smaller issuers. I know a lot of large asset managers who won't look at issuers who have less than two or 300 million in debt because it's not worth the time to research. But if you can digitize that with good data and they don't need that human interaction, they can get involved in deals they might never have touched for the last 20 years.

William Kim (09:16):

I think. Can you hear me? Sorry. So Wil Kim, CEO of Muni Pro. The way we look at using technology to help issuers really depends on what their needs are at the time. So when an issuer is thinking about doing a pricing pre or post, we're looking at trading data comps like Greg said, but we're also looking at option pricing pre a deal so that they can evaluate different coupon and call structures. And then throughout the year, not only do you have regular debt management, which is now on the web, like Tyler mentioned, we also do a lot of forecasting. So folks can look at their capital plan over the next five years, next 10 years, 30 years and help structure their deals and their capital plan to meet their financial targets. So I think there's a lot of opportunities for issuers to layer in technology.

Lynne Funk (10:09):

Great. So what about, let's talk about disclosure by issuers a little bit more broadly. Have they advanced transparency? When you talk about data, how have issuers kind of grown in that capacity or have they and Tyler maybe start us, kick us off.

Tyler Traudt (10:26):

I'll just share. I'll just My personal experiences as a former municipal advisor, sweating profusely, getting ready to file QIPs and disclosure information on Emma for my customers, I was absolutely scared to death. I was going to miss one and the next time they went to price, they were going to be flagged by the underwriter and I was going to look like the biggest idiot in my firm. And so disclosure obviously feels way too hard. It feels way too hard if you are highly educated. Sometimes I feel like understanding and performing disclosure at a really high level requires you to be a lawyer in many cases. I personally remember all the frustrating evenings working my way through bond documents, trying to piece together exactly what this organization has to share, how to do it appropriately. And so I couldn't imagine being a debt manager with no legal background.

(11:17)

Maybe somebody who's wearing multiple hats inside a governmental or nonprofit finance team having to do that work really well. And of course we work with some amazing professionals on the outside to help them do it. To me, fundamentally it gets back to resource allocation for these teams. I think everybody, if you talk to any organization, any CFO debt manager or treasurer, they want to disclose really well. I mean, nobody is saying it's not important. Everybody knows that it's really important. It's just about how many hours in the day you have. And there are tremendous hiring challenges that are being experienced by, I'm sure everybody in this room, no matter your role, but especially in the US public finance workforce, national associated state treasurers put together a really great workforce study. And you look at the distribution of employees by age range and the regular US workforce, it doesn't look like the public finance workforce does.

(12:12)

And so my perspective on disclosure, yes, we need to make it easy to understand what are my requirements, what did we file last year? When does this due exactly, but additionally, giving folks time in their day so that they can focus on some of those higher order more strategic assignments can be really important. And I think you'll see a lot of great tools and folks here will speak more in depth about disclosure in particular, but just allowing these folks better access to information, better access to information, because they're using modern tools to do the work every single day. That information can be a byproduct of their use rather than having to go into scrounge through a bunch of spreadsheets to pull together a chart to stick into a document that goes on to Emma that some municipal advisor in Charlotte, North Carolina is sweating to death trying to figure out if it's accurate or not. There's got to be a better way for us to build a better foundational system.

Gregg Bienstock (13:00):

I'll just say that I think if you look back from time of MCDC, I think things have gotten a lot better. I think that's a reality. I think your question and what Tyler is saying leads, its right into FDTA. Right? And to me is an organization that's part of what we do have done over the years, whether it's for issuers or MAs or underwriters. In doing the 15 CT 12 analysis, I am waiting to see what comes out from Washington, and I'm thrilled that the industry has taken such a strong interest here. But the thing that scares the heck out of me is whether or not they've truly understand the complexity of what they've bitten off here. In terms of the most important baseline component of we're going to do here is a taxonomy. And I had a conversation with someone recently about this and I said, I'm going to open up to you a bit of our database and share with you that the top 10 employers is defined at least six different ways.

(14:00)

I kid you not. And so it's very easy or it's relatively easy, will would say very easy to use technology to pull that type of information out. But if you're not comparing apples to apples, that's a real challenge and that's just kind of the tip of the iceberg. So the complexity you spoke about is real. You all know that if there was a way to simplify it, that would be really beneficial, I think, to the industry. But I think it's a very complex subject and peeling back the onion exposes lots of issues as well as opportunities.

Lynne Funk (14:37):

Yeah, the FDTA is definitely on my list to ask you all about. So please anybody else?

Justin Land (14:42):

Sure. So I'm on the working group for the NFMA, for the FCTA. And so we've engaged with regulators, the treasury, SEC MSRB, GASB, all that. I think the taxonomy is the crucial piece. If you dumb it down too much, which may help the issuers make it easier to do it, I think investors, it's really not going to help them. However, you don't have necessarily the same accounting standards across states. The normalization of data, I think is going to be one of the biggest challenges beyond solely the taxonomy. Again though, even though there's going to be this transition and pain to get through it, I think it's a positive because as we've seen in virtually any market, the more data that is transparent, the more liquid that market gets and that's better for issuers, that's better for investors. It's just going to take a lot of work by a lot of people to get there.

Tyler Traudt (15:51):

Just quickly, I'm sure that most organizations around the country look at FDTA and yes, there's time until we have to implement, but they look at that. I'm sure they all want to do that. Well, I don't think there's anybody who says, I don't care about making money, information accessible, readily accessible. But when you talk to teams, when you talk to finance officers, whether you're talking to the largest organizations in the country that may be in this room, but there are a lot of medium-sized smaller organizations that are going to have to deal with this presumably. And you talk to them about how it's going at work, we hear all the time, if I lose one more person, my team is going to break. I can't recruit and retain people. We just did an ERP implementation. We just put Workday in, I'll give them a shout out.

(16:32)

We just put Workday in and we've been working so stinking hard trying to upgrade our internal systems. If I lose one more person, I'm going to lose even more people and we're not going to be able to do the work that's necessary. And so disclosure is obviously incredibly important. It's got to get fundamentally easier for borrowers to do it at a really high level. And services I think have a big part to play in that. I think technology obviously has a big part to play in that as well. And hopefully as the FDTA comes, we will been able to make significant progress and they'll be really easy to use software tools available to borrowers and their professionals to knock this down and make it a success.

William Kim (17:08):

I think operationally, the way I look at FTTA is even if we could solve all the technical issues, I think Gregg mentioned this at a previous conference, but there's an issue with keeping it all updated, right? Let's say there is an accounting change and suddenly there are new fields throughout the year, but the taxonomy only gets updated once a year. You've got to have everyone coordinated where it either is coming out monthly or have some system to update this to deal with issuers from small to big across all sectors. And I don't think people have really looked into the details and that's where significant costs might arise for issuers. And so we're hopeful that implementation will be smooth, but I think there's a lot of challenges to address before that.

Gregg Bienstock (17:58):

Well, just so you know, I've evolved from that point and actually I treated as now an orchestra and you need a conductor who can actually make all the pieces and all the parts move in unison throughout. That's the next analogy for you.

Justin Land (18:09):

There we go. And I'm not sure we have to recreate the wheel. I mean, there are states doing this already, like the Texas Mac or the North Carolina LGC that are collecting all this data from all their issuers. So while it seems overwhelming to a lot of, especially smaller issuers, I mean, my parents live in a town of 3,500 people in Georgia, and I asked their friend who's the city manager, what the FTTA was? She's like, what are you talking about? Not surprised, I think even obviously the bigger issuers in the states. But I think one of the best ways maybe to look at models like Texas Mac or something like that, to have an overriding workforce that helps all these small issuers comply as opposed to everybody doing it on their own and maybe not doing it right.

Dan Silva (19:05):

And the only thing I'll add is on Justin, you hit the nail on the head, there's solutions for this already. This isn't necessarily a new problem. And you look at, we're just catching up what's going on in the corporate space, right? Annotated data has been in the corporate space on ICC filings for a very, very long time. So applying this to the muni space, I think there's good technologies out there for it. We probably need more nuanced technologies for the muni space, but annotation of data and then mapping when taxonomies change and getting the taxonomy right, of course in the beginning is critical, but it's going to be costly. But there's solutions out there I think that are going to be applicable.

Lynne Funk (19:47):

Great. So you brought up data. We talked about the issuers we're talking about when you look at it from perhaps the buy side, how do you think this has helped the buy-side digest credit? And there's a lot of data out there, right? How much of is it good? How much is it useful? And yeah.

Justin Land (20:12):

Okay, I'll go. My background's a buy-side, so I'll Sorry, Justin, I'll take this one. Yeah, I think one of the things we have to remember, and I'm going to insult myself and many people in the room, is that my whole career has been in the muni business. And we sometimes just think that the way we did it 10 years ago is the way we should keep doing it forever. And I think that really comes out in the last 10 or 15 years when you've seen the rise of algorithmic trading in Munis where now a significant portion of the daily volume of trading is done with little to no human intervention. And you have firms that can bid most of the market in less than one second. That's starting to flow in the asset management industry as well. The way that works is if they have enough data that they're confident, if they're having to pull through PDFs and pulling out these top 10 taxpayers or whatever, that's not scalable. So I think the FDTA and any of the technologies that make it easier for an asset manager to allocate capital quickly and efficiently is good for the market. It lowers borrowing costs and it increases liquidity.

Gregg Bienstock (21:33):

Sure. So I'll add just to a slightly different take on this. So our firm was acquired by a firm called Solve a year and a half ago. And one of the things that solves does their flagship product is around parsing, being able to take pre-trade messages from multitude of sources, unstructured messages, and being able to parse that and turn that into information. And so whether that's on a bond or bond basis for an asset manager or an asset class basis, for them to be able to look and identify where they think the best value is is also another tool now that's in the toolkit and is an important piece of the puzzle. I think for the buy side. James, who was up here early, talked about no longer being a price taker and it's another tool in there to all of a sudden have greater insight into the market, whether it's parsing from your own messages or getting market data that's much broader.

(22:30)

Taking a step forward from that and thinking about, I know we're going to talk a little bit about AI, but one of the things that we are involved in right now, we just launched beta, is something called solve price. It's a predictive price. And the idea is that where is the next trade going to occur? It's not a valuation, it's not a baseline of evaluated price, but it's more about where that next trade is going to occur with a degree of certainty. And I think early on it's going to be another tool candidly for buy side, for sell side to be able to get a sense of where the market is, right? We have unique data that we're able to utilize. We have underlying AI models that we can talk about later, but the idea is to create other tools for additional insights. Maybe it's against their own models initially, and as they gain confidence in what we're doing, they'll use it for even greater purposes in that regard. But I think there's more and more out there for the buy side to have greater transparency, but then also to have greater insight and prospective perspective insight as well.

Tyler Traudt (23:36):

Can I dream for a second? Is that okay? Please. So one of the things that I think is personally really cool and interesting is having spent 10 years pulling financial statements down, extracting the data out, put it into a spreadsheet and then go to work, every piece of financial information I've ever received from a borrower is static point in time information. Got it man. We're going to be able to convert that into machine readable format and I'm going to go be able to get it and apply it and do whatever I want to do faster. It's going to save me time. I'm hopeful that there's a future state where a combination of a treasury management system like Debt Book is working to build right now for public sector organizations combined with their ERP, their primary accounting system. You could take the live data that's happening every single day and there are tools out there now.

(24:19)

There are tools that organizations are using now to compile all the financial information from their ERP into an audit ready format to make the auditing process much simpler if that data fundamentally lives in that ERP already, if you could have a company and my investors hope it's debt book, but if you could have a company that says, great, we want to be a treasury focused function. We love debt, we love cash, we love investment, but we are not an accounting solution with integrations, could you pull the data into a format that would create a snapshot, a real-time snapshot of financial information? Yes, it's possible. Hard, yes, obviously yes too. But my hope is that by the time that wherever I am in 10 years or beyond, we're really at a point where software has really done its job and treasury functions in the office of the CFO inside of our nation's public finance organizations have gotten together, are pulling live data out of that accounting system, out of their treasury system with their loan profile, their cash trends, those types of things, investment profile and can share a live view any day. Here is a current read and if that happens, then you can go and share great financial information and it might not be audited, certainly not going to be audited and perfect, but maybe it's great financial information that can be used by organizations that are looking, investing into these securities and get a better read on it rather than just waiting for a disclosure filing to hit Hema.

Justin Land (25:46):

Can you make that happen in the next year?

Tyler Traudt (25:48):

That absolutely great. 60 days.

Justin Land (25:50):

60 days okay. Alright.

Gregg Bienstock (25:53):

I think your has got the average time for financials down from 200 days to like 190.

Tyler Traudt (25:56):

I said dream. Okay. I did say dream at the beginning of this.

William Kim (26:00):

I think on the buy side and sell side, we interface with the credit research analyst. And so one thing we've been doing to help them leverage technology is we're using, we're combining a simple tech search to let's say look for a certain credit term and documents to get a list, let's call it 10,000 issuers, have this type of credit term and then flowing all those documents through to AI to say, okay of that I'm looking for this very specific term. Can I get getting that down to a list of maybe a hundred good candidates? And so that really takes a process that would used to take a year maybe and gets it down to a day. So I think there's a lot of opportunities for better credit analysis beyond just the trading of bonds.

Lynne Funk (26:46):

Okay. So what about, can we talk a little bit about when you think about the sell side, how has technology aided workflows for both the primary and secondary markets? Is that something that we can get into?

Dan Silva (27:01):

I can talk about the primary since that's basically what we do. I mean our entire focus is on making from the ideation stage. When you're thinking about a new money deal or a funding all the way through, you're structuring the deal to the point where you're going to offer it and have a final transaction is basically automating as much of that process around the modeling as possible. And that workflow is incredibly complex, right? You have all these constituents that you need to run these numbers by. You might have a finance committee or folks that you need to present to about what this capital plan might look like, and these things go through tons of iterations. It's incredibly time consuming. So what I'm get excited about is just the ability to shorten that timeframe when you're looking at different plans of finance, when you're looking at different scenarios, being able to very quickly pivot with what market changes that are occurring.

(28:02)

That's where some of the data integrations that we're doing is interesting around getting live information around what's happening with M and D or some AAA scale or what's happening with secondary pricing, which matters for evaluating tender offers. And can we quickly update those analyses and quickly execute on them. Right now, those processes are pretty slow. I mean a lot of you, probably some of you might do some of that analysis internally on the issuer side. A lot of you probably rely on your MAs and bankers to do that as well. But those MAs and bankers in here also know that that is an incredibly time consuming process. You have to have hordes of analysts work on it. And so just on the workflow side, on the sell side, there's a ton to do in terms of removing data entry. We're doing things like pointing to official statements instead of for old transactions, having to enter that data manually.

(28:57)

You just point to an old official statement and you have a fully structured deal in a second when we're talking about sensitivity analysis, when you want to understand what's going to happen around changes in rate environment or maybe what things will look like if you have a different size capital project, instead of going through those iterations, we'd do 25 iterations in two seconds, things like that where we're pulling the monotony out and automating a lot of that monotony away and letting folks focus really on the end results. What does this plan finance look like? What does it mean for me in terms of metrics that I care about coverage, my max debt service, what my refunding statistics look like, and being able to view all the possibilities with respect to the way these plans of finance, the shapes these plans of finance can take. So I think there's a lot to do on that sell side workflow.

Lynne Funk (29:49):

Okay.

Gregg Bienstock (29:50):

Sure. So someone used the words before productizing processes and someone else used the words that there was an arms race in technology to serve this market. And I think they're both really applicable to what each of us are offering when you think about the market, right? So when Lynn posed this question at the outset, I was like, well, it's kind of interesting. We do something similar. We do kind of a debt analysis, we debt service calcs and things like that, identify callable debt and it's like one piece of the puzzle. And when we have clients who say to us, well, you save us 90% of the time we used to do on that. So we're productizing a process that you all engaged in next part of the deal transaction or is the 15 C to 12 analysis someone has to do that? That's something that we've been doing and whether it's us or another firm, but a process that as you alluded to Tyler, that took a significant amount of time for people who don't do this every day to all of a sudden put a process in place, build database around it to make the market more efficient and let you all be more efficient.

(30:56)

Moving to pricing or kind of thinking about the idea of creating indicative scales, right? The ability to identify comparable transactions in the marketplace within seconds and be able to deliver that to market participants. These are the types of tools that the respective firms that we're offering. And the idea is not to eliminate jobs, but it's to with limited funds. Rick talked earlier about how much are we getting paid on deals now it's a lot less. So you got to do more with less and your clients are asking to do more. And I think that's what each of the different technology components that we're offering and we're talking about here, that's what they're about. It's productizing something that is a process where technology can step in now, but you do more interesting stuff.

Justin Land (31:44):

And I think on that to the arms race thing, I think one thing that's important because how the buy side works is not the same as the sell side is not the same as the issuer side, whether it's us up here or anyone in the room. There has to, for us all to be more successful, there has to be better communication about what's the best way and the most efficient way for an issuer to deliver it. But that also needs to be digested by the sell side and the buy side. And if the issuers are doing this over here and the buy side is doing this over here and the sell side over here, I think there needs to be more collaboration about the ways to make efficient workflows between all the constituents.

Gregg Bienstock (32:29):

Justin, to that point, I'm sorry, just one quick. I was going to say just on that point, we developed software. One of the things I think over the past probably 18 to 24 months that I've seen more of than I've seen in the 14 years since we started my old company is the request to, Hey, can we take that as a data feed? Can we take that as an API? Because we have our own internal system and that's something that's highly important when you think about workflow. I don't want to go to another application, I only have so much screen space hear it all. But that's a critical aspect also of what we're doing.

Tyler Traudt (33:06):

I just want to echo Dan's point for a second. Coming up through investment banking and financial consulting way too much of my personal identity was tied up in being able to operate DBC without touching my mouse. I mean it was just way too important to me personally. When I got to Citi, they said, welcome to Citigroup, and they ripped the mouse out on my computer and I told that story a thousand times. I thought it was so cool. I was eight years into my career and I worked on a financing for five years and I remember finishing it, scenario number 127 D, and I was like, I've worked two thirds of my professional career on this financing. Like this is ridiculous.

Justin Land (33:45):

We all have PTSD from that.

Tyler Traudt (33:48):

So been there and totally, totally echo it. And that's the only comment I wanted to share.

William Kim (33:55):

For our sell side customers we're all about getting scale, right? Having a banker from who used to cover 10 or 50 clients just because they were constrained by the amount of work their analysts could do to now covering a hundred or a thousand clients and really getting out there to do better for their one in the sense of covering more clients, but the other is to do better work for their clients because all of that almost drudgery the kind of basic work of producing a debt profile and making sure it's verified by credit we do. And then you go ahead and layer on your particular financing ideas and what makes you special and differentiates you from another bank. So that's where we see banks really adopting our technology and really being excited about it.

Lynne Funk (34:39):

So what about in the secondary, can we talk about liquidity in the marketplace? How has technology aided technology AI aided in this space in the market in liquidity particularly?

Justin Land (34:55):

I mean I think it goes to the rise of Algo traders and electronic trading. I mean you can have a small staff, they can do thousands of trades a day. I was just at a conference in December and the head of systematic trading for BlackRock spoke, and it's not in Munis, but it will come. 80% of their fixed income trades globally are systematized. We're going to get there in Munis maybe in 60 days I think. Yeah. But I think the more trading we can do and whether it's better pricing or future pricing, liquidity begets liquidity and asset managers, the more they're comfortable with liquidity can start to do things they might not have done in a less liquid environment. And I think the numbers are bearing it out. I mean, I don't have it in front of me, but the MSRB puts out their cost per bond trade statistics, and I think a hundred bond trade now is the same bid ask as a million bond trade like five years ago.

(36:03)

And it's always fits and starts, but it's definitely happening. And again, I think the nice thing about a lot of the Algo traders is they didn't come from a muni background and that lack of bias actually made them very good at what they're doing. Like the old headlands tech, which is now td, they're like dark pool equity quants and commodities guys, they didn't know anything about munis. They just said, here's a math problem, we're going to figure it out. And they did. So we have to have that beginner's mind of thinking about not the way I traded munis when I started in the nineties, but what's the new way? And I think to the comments about not being able to find staff at issuers or anything. I think Greg talked about a few years ago at a conference, the silver tsunami coming in the meeting market two years ago. I'm going to rip you off, but I think to get more young people who are coming out of school with Python skills and all these things, you got to make it a little more exciting and not say, just make this how I've always done it. Let's make the new mass trap.

Gregg Bienstock (37:17):

So I'll just add to that in terms of trading activity over the past couple of years has been through the roof in Munis, right? Compared to where it was I think in 2019. I know the MSRB just came out with data on that. And so part of that is rate driven, part of that's Algo driven. Part of that is there's more information and information on differentiating from data. There are people who are taking data, there are people taking parsed messages and turning them into information. So there's a greater ability to understand, it doesn't matter if it's a hundred bonds, now I can go out into the market and look for it. I think the other is prior panel talked about SMAs, the growth of SMAs and Rick talked about the idea of you need to look at them as institutional, but what are they buying?

(38:05)

What pieces are they buying? And that's going to be different. So I think that's a component aspect. I think the idea of pre-trade price transparency, as I mentioned before. So we built an AI model to solve price that's out there. That's another tool that's going to be available. The idea is to create it, whether it's an Algo or it's a predictive price or something else, it just makes the market more efficient and be able to be responsive to the environment. They talked earlier about volatility and those are things that people are going to respond to when they have the ability to, as opposed to when there's a lack of transparency and a lack of information out there.

William Kim (38:45):

I think technology has also changed the nature or the composition of liquidity. SMAs, that's the growth of SMAs, I think is enabled by all these tech firms who are helping the large SMA providers. So you have folks like IMTC and investor tools who will have a ready-made presentation so that individual investor can actually access this. But while that improves the depth of liquidity for higher quality names, you see as the funds move from the mutual fund complex to SMAs that expertise at the mutual fund level, there might be less liquidity for the lower rated credits out there. So I think technology is overall increasing the pool of liquidity, but certainly there's spots where there's inconsistent growth. But at the same time, overall, folks, I think last week or two weeks ago at the muni forum, parametric mentioned that they have a huge program using technology to automate tax loss harvesting. And I know Andy Cate, who should be in the audience, has incredible paper about it. So there's a lot of different moving parts. Overall it's a great thing, but I think certain there'll be winners and losers.

Lynne Funk (40:01):

I think I forgot to, sorry, I glazed over a mutual fund question, which obviously I was going to ask you all if anybody else wants to comment just on the growth of SMAs and ETFs and how has that been shaped maybe by automation. Anybody else wanted to opine there?

Gregg Bienstock (40:17):

I think it was covered at the earlier panels, right?

Lynne Funk (40:20):

So we're good.

Justin Land (40:23):

I agree. I mean Gregg said it, it's a different type of institutional investor and learn how the SMAs act differently. I mean, there's an article today in the Bon buyer about this subject, and SMAs used to be very plain vanilla, AA, AAA, one to 10 year ladder. And now you have completion funds to SMAs with credit focused, you have more interval funds coming out to deal with some of these liquidity issues. So the market is going to find different ways to productize it. The better liquidity and technology gets, I think about the equity market and now direct indexing is just a massive business and it's fantastic. I mean, you get S and P 500 returns with one to 2% tax alpha a year. I mean that's great. If parametric or any of the others come up with something like that, it's going to be a huge hit. And we just didn't have the capability to do that 10, 15, 20 years ago.

Lynne Funk (41:32):

I didn't ask you a specific question about artificial intelligence, which if maybe perhaps you could each give just where you see AI fitting, and I know it's sort of nebulous, but how do you all see AI fitting in the muni market or how is it already being used from your seats?

Dan Silva (41:51):

So we're seeing AI pretty much everywhere. I think all of us are doing something with some level of AI. And I mean AI, all it is is compressed representations of data that's out there training data, and we're using that data to predict something. So it's just an interpolation from the training data, albeit very interesting and useful, but at best what you're going to get is some representation from that training data. It's not going to come up with something entirely new. It will sometimes look new, but it's going to come from that training data. And I think there's so many applications, whether that's streamlining, getting rid of data entry like we're doing where you're pointing to existing documents and just being able to have an AI that knows, oh, that's a call option. Oh, that is a security feature. And being able to structure that in a way that is searchable in a way that is helpful in terms of structuring bonds or whether that's pricing like Gregg's doing where you have data around previous prices, similarities, and you're able to extrapolate what this price should be. So there are just a ton of applications and it's going to be interesting to see where it goes forward. But there's limitations to it too, right? It's never going to be perfect Again, it's just going to be its best prediction. Guess from data that we've, the training data.

Gregg Bienstock (43:26):

I'll take a slightly different take on that. So I think interpolation extrapolation are methodologies that have been known forever. I think true use of AI where you're building models that are constantly learning based on the inputs associated with those models are going to come up with more refined solutions over time. I do agree with you though. I think there's a trust but verify component with many aspects of AI where people talk about it's extracting information from documents or you're using chat GPT to create your next email, read it before you send it. But I do think when I think about truly how AI can impact our market, I think there are a lot of things that can be done. I think as you are on the working group, I think as we get through the FDTA, I think there are things that are going to emanate from there.

(44:18)

I don't know what they are yet. We don't know what it looks like just yet. I think on the pricing side, I think some of the stuff that we're doing, I think there's a lot that can be done there. We used AI to fill in gaps on creating scales where there was no trade data, there were no comps in the market. There's nothing there for that maturity for that type of structure. So we have experience using it, but I am fascinated by what's been done so far and the things that are talked about outside of our industry and start to think about how those things can be applied to our industry. And I wish I was smart enough to know what that next thing is. We think we know what it is when with predictive pricing, but I think there's so much more that will be part of our ecosystem in the very, very near term, and I mean the next three to five years.

Dan Silva (45:05):

But it's never magic, right? There's going to be training, even Chat GPT, the thing everyone's talking about now. All it's doing is predicting the next word off of a massive training set of the Internet's language, all the language on the internet. So it's drawing from that training data always is now the outputs we're getting from it do sometimes look magical, look incredibly creative and infer a certain understanding, but we have to have the data, the data's core.

Gregg Bienstock (45:39):

Oh, absolutely.

Tyler Traudt (45:40):

I'll probably sound like a broken record and I apologize for that. Our thought around it right now is just automation. It's just process improvement. It's just less time, more efficiency being delivered. And so not so debt focused example of this for many of your customers in cash management, you've got a lot of different employees internally that are pulling a lot of bank transaction data down and they're trying to parse through all of the information and analyze it, whether it's looking at fees that they're paying to their local banks based on their balances or if it's looking for instances of fraud. And you're going to be able to automate a lot of that work and help those team members do their jobs faster. So there are going to be a lot of really nice ways for AI to come in, but it's just a little piece of a broader solution for those folks and we'll get better over time. Chat GPT. My name's Tyler calls me Taylor every single time, so it's not perfect.

Lynne Funk (46:31):

Will, I'm sorry, go ahead Justin.

Justin Land (46:33):

Yeah, I think I just echo the garbage and garbage out is super important. I think the trading side will adopt machine learning AI a lot faster, like solve pricing or other products out there. I think for credit research, a lot of that data is still locked up in PDFs or it's behind paywalls and you need better training data. And now it's not just looking at numbers, it's understanding different sorts of liens by state or is litigation material or not, or all these other things. So I think it'll come, but I think the trading piece, because the data is so much data is already there and I think it will have a bigger impact quicker by going after trading before credit. Because we all know most ratings don't change from one year to the other, so we can't get the bang for the buck versus I can save two basis points on every transaction. I can monetize that very quickly.

William Kim (47:38):

We're seeing AI, I mean historically we've used AI to extract data. That was our main use of it. We've been hired by a few clients to use it in a hybrid model approach. And you're seeing that from folks like Google and others who are using AI and layering in other systems. So we're using traditional tech search, like we talked about, going through all the documents, the credit terms that you're talking about. We're doing that today, right? We're taking those text feeds, we're going ahead and asking AI to look for specific credit terms and for the composition of escrow is a composition of a swap document. So you can look for different financing ideas. And so we've gone from not just a source of data, but also helping you actually execute on financing ideas and again, visualized. And so garbage and garbage out is always a problem. That's why everything we do, we have a human person verifying and checking it. But there's a difference between one person doing the whole process where they can maybe go through one document a day versus one person going through a thousand documents a day or 10,000 in terms of the space that you're evaluating. So I think there's a lot of opportunities for scale there.

Tyler Traudt (48:48):

Just a broad point, technology's not perfect. Everybody knows that it's not going to be perfect if you assume that the software vendor you're working with is going to magically make all your problems disappear, that's a bad assumption. It's not going to happen. It's really, really hard. We have this in our own business. We're implementing Salesforce right now. We drew up Salesforce, our implementation to solve all of our problems. And guess what happened? None of our problems were being solved. It was way too complex and we had to rework our thoughts around how this piece of software was going to help. We were going to put people and processes and tools around it, we're going to simplify the software so it performed one key function for us so that we could be successful as a business. And that's exactly the way that I think folks should approach it. If you think that you're going to work with a vendor and bang, everything's going to be perfect, that is not the case. Be flexible in how you are thinking about these solutions and how you're going to bring them into your workflows. But you got to have people, you got to have processes around it for sure.

Justin Land (49:41):

If it doesn't work, you just turn it off and turn it back on again.

Tyler Traudt (49:46):

At your computer, right?

Lynne Funk (49:48):

That's the truth. Right? We actually, we went a little over so we are taking a break after this. So are there any questions or should we.

Audience Member 1 (49:58):

One quick question. So away from the paper documents that you may be surveying using AI, what type of sources for data are you using in your analytics and in your processing?

Lynne Funk (50:14):

Go ahead.

William Kim (50:15):

We use a multitude of sources. So we cover munis, but there's also corporate CUSIPs out there. There's a lot of corporate bondholder data that the SEC has. So we'll take in every feed that's publicly available, the entire SEC database we check every single day, right? We download that and that's part of the FDTA question, how much data's available. We ingest all of that and layer it into all of our products. So all the data feeds come through our models and outcomes. The end product for our clients.

Gregg Bienstock (50:47):

Yeah, I would just say it is publicly available. There are private data sources that we are engaging as well, and then there's proprietary data that we're bringing as well. So if I looked across the platforms, there are well over a hundred different data sources coming in that are databased and just on the muni side alone, everything's mapped against an obligor database, which is a critical component, whatever it is that you're taking in, when you're starting to aggregate and bring that data to the client.

Lynne Funk (51:17):

Alright, well thank you so much everyone, and thank you to our wonderful panelists. Thank you.