Dean Curnutt: My guest today on the Alpha Exchange is Kumaran Vijayakumar. He is the co founder and CEO of datadoc Solutions, a fintech firm working with the sell side to really understand flow dynamics and the value of different clients. Kumran, it’s great to have you on the podcast.

Kumaran Vijayakumar: Great to be here Dean. Looking forward to this. We worked together at this point, I guess almost 20 years ago which is kind of scary but look forward to catching up.

Dean Curnutt: Yeah, I think we’re going back 23 years. 2002. Perhaps the world has changed, technology has changed, but what hasn’t changed is the risk taking process for broker dealers facilitating client trades. It’s one of these businesses where in some ways you don’t know the cost of goods sold for a while and so a lot of this is going to resonate throughout the conversation. Let’s get our discussion started and just have you tell our listeners a little bit about you and your career history leading into the founding of datadoc.

Kumaran Vijayakumar: After going to business school at Wharton, I started my career at Goldman Sachs initially as an exotics trader but very quickly sort of morphed from that to being an index options trader and really was there through the Long Term Capital era, the financial crisis. That financial crisis in the late 90s left Goldman at the end of 99 to go to a hedge fund that had started a tech hedge fund that had started right in that boom and that immediately went under. So I went to Citibank. At Citibank again was trading index options and was running part of the ETF options desk before coming to bank of America and working with Dean, initially running the index and ETF options business, eventually running global derivatives going into the financial crisis. After the financial crisis a good part of the team exited bank of America and I went to MF Global which was also an interesting experience and started that from greenfield. As you can tell, those last two experiences during different financial crises soured me on that. So I started looking at different things and different ways to be an entrepreneur and maybe get away from that cycle.

A few years later Tom Wadsworth, who is someone we also worked with at bank of America and I started datadoc Solutions. The initial idea was that Tom at the time had been at Goldman Sachs and and he had seen a much more data centric approach to really how you make decisions and our thought was can we bring that to everyone else and also can we advance that ball even further. We in 2018 started Datadog Solutions as a software company. I have a computer science degree from A long time ago, but had never really done that as a career. We had always worked very closely with technology in each of those different banks and thought, how can we sort of bring that to everyone else? So that was a real initial thought of founding datadog Solutions. One of the things that we realized very quickly was the banks have a lot of technology. It’s not that it’s just getting the data and control because basically, even when you look back at 20 years ago when we worked together, data’s actually just expanded and expanded and expanded. So in a lot of ways the goalposts keep moving.

So as there’s more and more and more data and the noise gets greater and the garbage in, garbage out issue gets greater, it’s almost as if the things that we were trying to wrestle with 20 years ago have become bigger issues and harder data problems, not easier ones.

Dean Curnutt: It’s super interesting. I go back to those B of A days and you and Tom having started Datadoc. Henry Schwartz launched TradeAlert. Christian Houff started Quantitative Brokers with Robert Elmgren. So there’s some good DNA that has been put put into fintech Solutions from those days. You said a couple things there. So you said you started as an exotics trader. Of course, those are OTC derivatives. They’re not very price transparent. Usually the buyer is the one that perhaps might be on the wrong side of it. But there’s plenty of instances where the sell side winds up mispricing correlation or divs or something. But you don’t know it initially. A lot of times these are longer dated trades. And so the modeling error turns out to materialize after some seasoning of the trade with the listed space where you and I were leading those businesses together. You kind of know almost immediately if you were wrong. You know, if you got the delta wrong or if you got the volume, you get that feedback from the market very, very quickly. And trying to understand those businesses and understand the value of clients, the value of different salespersons or traders, it’s not an easy task.

And I think that’s really what datadoc sets out to do. Give us a little bit more of the impetus for launching the business. How you were thinking about it back.

Kumaran Vijayakumar: In 2018, Goldman Sachs was definitely a leader in this space. They had actually gone through what they called the development of SecDB, which was really getting all their databases in line. Every way they touch a client in one database. And then once they did all of that, which is a very large project, they then went about setting up a whole Strat group and Strat language on top of it to build tools to actually analyze that data and actually impact businesses. We realized that that was not a model that most banks can really approach because basically most banks and any other kind of financial institution is not going to be able to get all their data in line. So our thought was basically, how can we get the value at the end of that process and think backwards to what you actually need to get that value and develop that rather than waiting for all of the data to be in line? It is true that the transactions are quick in listed space and you do have a sense, but there are all the biases that you have that we all have, you know, sort of a recency bias, there’s a big trade bias.

There’s sort of an anecdotal approach to all of that stuff where you do have a general sense. It’s a good client, that’s a bad client, that’s a good trade, that’s a bad trade. That guy is always right on delta, that guy’s always right on volume. But even those are somewhat deceptive because all of us have memory holes. But also, when you look at the data perspective, you look at all the trades ranging from the tiny little ones all the way up to the big ones. Do it on a data centric approach and really sum it all up. You find some of the things you find out are exactly what you thought, but you find a lot of things that you didn’t think. The basic thought was, what do you actually need? What do you actually need is you need all the data. The ways a firm touches their customer, the ways that they hedge themselves, the ways that they take on proprietary risk, the ways that they capture whether it’s mistrades or trades that are shows. So how do you take all of that data? First of all, how do you get it over to us in a way that is comfortable for banks?

So how do you anonymize it? How do you make it something that can be secure, meet compliance? And once you get it over, how do you actually take that and model it out? Simulate every behavior you could have had, every knob you could have turned, every salesperson, every trader is doing a million things. What are those things? Let’s simulate them all out and then basically tell you back what the different approaches could have done and really do that at the trade level, all the way at that granular level, so everything can be kind of summed back and looked at at any cut you want to. So you can look at on a customer level you can look at a customer trader level. Basically everything can be sliced on dice any way you want to. And our thought was basically, how do we do that? And let’s just From Greenfield, in 2018, we said instead of thinking of these kind of big projects of how do you get all your data in line, let’s think about how the data is at banks and let’s basically start from scratch and say, if you took that data, how would you develop these results?

How do you find these answers? How do you find out which clients are good, which trades are good, which trades are the right efficiency and risk wallet, the right return on risk, all those kind of things.

Dean Curnutt: Let’s just make sure we haven’t jumped ahead too far and made assumptions about the listeners. So let’s go through it. Just a basic transaction is you’re a sales trader at a large bank, a light lights up and you’ve got a voice on the other end looking to do something. And it’s a very compressed time frame for a lot of things to happen at once where a lot of risk is transferred. A lot can go right, a lot can go wrong, but it happens very, very quickly relative to the size and the risk profile of the transaction. The salesperson takes in the order and verbally typically is going to shout it over to whether it’s an index trader or a sector trader, what’s looking to be done, and a price is going to come back and perhaps there’s a very quick round of renegotiation and then a trade is either going to happen or not. It happens, all things considered, in the blink of an eye. So let’s kind of go through the types of information that are capturable through that interaction and how the banks are getting that information to you and how you’re curating it.

Kumaran Vijayakumar: The first type of information is basically all the historical trades you’ve done with that counterparty. How have those trades gone? When I say how they’ve gone? It’s a really broad statement, but basically at that point you, as you said, the sales trader basically needs to know how to approach the trading desk about this order. Going back to our time together, you used to have the list of things that every sales trader needs to do before presenting an order. In a lot of ways, those are the things that sales traders first doing. How are they presenting this order? The trader is then making decisions as to how to price this order. And again, in a broad strokes, is this customer good? It’s customer bad. Is this customer difficult? But they have to make some decisions basically Based on the customer’s history, they have to determine customer’s history. Is this particular asset or this particular type of trade, is this something they want to do? If they do want to do it, how do they want to price it in terms of pricing? How are they pricing their future actions? It might be a different price. If you’re thinking of this as something that you’re in a warehouse of risk versus if you basically just want to find out what the market price is and want to get rid of this risk off your table, those are all decisions you have to make.

And then that price will then, as you said, go back to the sales trader. Then there’ll be a back and forth. That back and forth is also informed by what this customer is. For some customers you might say you want to make a better price because you need to do this trade. Or there’s other things that with this customer that make it more valuable for a different customer. You might say this is the price and we stick with it. Once the trade is done, now the trader has to actually realize this expected value. So you priced it where you thought you price it. Now the problem is with derivatives is that this is not easy because what ends up happening is you do the trade now, you’re going to do a lot of hedging transactions. You’re going to, you’re going to take risk off risk on. All your trades are going to kind of be put together in kind of a soup. Now that soup, you might know in a very broad sense how profitable that soup is. But it’s really hard to take the components of a soup back out. To really say this particular trade gave me this much profitability or this much risk.

What did it actually do to my profile? Is this a trade I want to do more of, less of? So what we try to do is take those granular trade levels and really simulate out every behavior you could have done. You can actually go back historically. So a customer onboards with us would bring several years of data in the form that they have it and they’d upload it. That data of how they interact with customers, what they did in terms of hedging, what they did in terms of laying off, what they did in terms of missed trades, all that data comes in. We then simulate everything they could have done. One of the conceits of our approach is that we do everything on a trade by trade basis and we do everything on a simulation basis. So this is not so much about the traditional banks, where the banks have always said and over time call up a customer say, I lost a million dollars in this trade, and it’s about what I actually lost. And there’s something sort of about that that always begs the question where the customer’s immediate reaction is, well, you shouldn’t have if you knew how to trade better.

And our approach is different from that which basically says we have a way of metricing you and that metric lost a million dollars. And that opens up the conversation completely to a different approach.

Dean Curnutt: I think that’s really interesting and I’ll make this statement and maybe you can run with it because I always thought about this. You’re always going to make and lose money. It’s a big business. Stocks are volatile, things happen. You’re trading convex instruments, but you want to make or lose money for the right reasons. If you are on the wrong side of a dividend increase that you had no idea was coming, that doesn’t feel like the right reason. If you took down a big trade and you lost some slippage on the delta, but overall you got the hedge on perhaps that’s losing a little bit of money for the right reasons. I think that to me seems like some part of what you’re trying to measure is when you talk about the quality of the flow, there’s some part of the losses are explainable in the context of things you can get comfortable with.

Kumaran Vijayakumar: Absolutely. One of the things that you know, going back to again, our time together, one of our clients that you’ll immediately recognize was one of our biggest sources of volatility. That client, if you put those trades into in most markets other than the really volatile times, most times, volatility trades above where it’s realizes. So that client, if you modeled their trades almost every year that we work together, would have been a big money loser because we would have basically been buying volume from them and that volume would have realized less than what they sold it for. But what that client did was provide us the bullets to basically trade with almost all the hedge fund clients who are largely buyers of volume. That was basically capacity for us to trade more. That’s the kind of stuff we model as well, because sometimes it’s not as simple as is this particular trade in a standalone good or bad. It’s also whether does this trade give me the ammunition to do more trades that are more prevalent in the marketplace. So, and I think you know the client I’m talking about. But that client was our biggest debt source of volume, probably our favorite client.

Even though on the whole, if you just looked at them as a sleeve, you’d have lost money just doing the trades.

Dean Curnutt: You’re effectively on the other side of the volume risk premium, but you’re leaning into the receipt of that volume into the book. How would something like that, which as you said, because of the vrp, even though their client is perhaps overriding, they’re not strictly trading volume, but they are selling it above realized. How does the value of that client, how is it numerically quantified?

Kumaran Vijayakumar: Again, when you have all this data, we would simulate that client on an individual basis. You would be able to see that loss of I’m buying vol above what’s realized. And you could see that on a realized basis. You might be able to see a smaller loss when you compare it to what you could actually offload that vol in just general markets. And that might say, well, more of a break even business. I’m buying vol and I can sell that right back out. It’s not that price sensitive. That’s kind of good. But then you can go one step beyond that because we actually have a module in our system that does the third thing which basically says, hey, what are all the flows that I’m seeing from all my clients and how does this client look as an offset to that? If this client basically risk increasing or risk decreasing on a portfolio basis? We have several ways of scoring that, but those scores can really give you a lot of insight as to whether this client is beyond their individual simulated P and L, beyond their simulated p and l vs offloading into the marketplace.

There’s a value beyond that and you can see that. You can see the opposite too. You can see clients where their trades don’t seem that bad, but they’re basically always going the same way that everyone else is going. In a way, it almost becomes a flood where they’re always participating along. It’s almost, it becomes like a thing where even if their trades individually are okay, you don’t particularly want to do a ton of them because they’re doing the same things your larger clients are doing along with you’re probably not getting the actual benefit from doing it. So you see both, but you often do see that the same kind of thing. You can also model things like lots of micro things. You can model things like what’s the probability of a customer rolling. I remember talking to you about this literally 20 years ago, I think Dean, where we talked about in some ways a commission is not just a commission, it’s you should look at the expected value of the roll commission. It’s Almost like a SaaS model in derivatives, the right client that is constantly rolling or unwinding almost becomes a recognizable revenue stream.

And that stream is very worth jumping into. Much more so than in Those days the 3 cents would pay and these days maybe the penny would pay.

Dean Curnutt: There’s the quality of the flow. Everybody wants to get triple Q option orders not held. And then there’s deal stock special sits with funky distributions and illiquid options with no price discovery that you’re in some ways implicitly forced to stand up to. Those are. The underlyings are tricky, but the setup very much matters. So we’ve seen auctions. Give me your best one of three or four quoting the winner’s curse can be a real thing there. How is that client behavior or the way in which the client chooses to present his or her order, how is that modeled?

Kumaran Vijayakumar: We have a lot of modeling for that too. We also have a pretty robust group of theoretical values. We do a lot of analysis also. One of our most recent modules is analyzing how the trade in question behaves and how theoreticals behave half an hour before the trade, five minutes before the trade, a minute before the trade, minute after, half an hour after, all the way through the day and obviously days forward. And that I think gives a lot of insight. It’s actually been a recent module that people really taken to, which gives a lot of insight to just that some of these credit funds used to send out batches of orders where everyone would see the same thing. So you get a batch of orders, you often see that movement half an hour before the trade. You see the movement in volatilities already. So we do have some analysis to do that. We talked about a customer having edge in Delta or having edge in volume. All that also we think of in terms of time frames, almost no one, there are a few exceptions that were right in all time frames. A lot of times when people on derivatives desk say someone’s really right on Delta, what they’re talking about is a very micro view.

They’re talking about people being really right because either they themselves are causing kind of a fast market or there’s participating things that are moving very quickly and you can’t quite keep up with it. So a lot of times people are talking about those windows and those same windows can reverse. A lot of times customers are right for five minutes or not right for five days and not right for 15 days, not right for 50 days and not write to expiration. And like you said, there’s also a completely different world where people are trading on events, there are event risks that we try to capture, we try to capture the actual event times and capture also the historical participation of events. Clients who trade events are often going to have big swings. But there’s a difference between the normal expected distribution around those things versus it’s something different if someone has someone in the courtroom when an event’s happening and that event reflects back and that jump is always that way. That’s a client you just want to trade with. And again, some of those things you can actually feel without doing this analysis. But our analysis pins that down.

You can see exactly what’s going on there. And you’ll see that for example, when you take an average client, even a very good event driven client, the distribution makes perfect sense and it is still tradable. You have to have the right pricing for it, but it can be traded. And then you’ll see other clients where it’s just you are the sort of the last destination and you’re getting the very difficult toxic flow and that’s something you just want to avoid. But for the most part our tools are less about the avoid the client totally or do all of this client’s business because that’s almost the easy part. Those almost everyone gets right. It’s more everyone else really. The middle 80%. How do you trade with them? How do you think about.

Dean Curnutt: When I think about derivatives, I think Delta Gamma Vega for the most part. So the Delta hedge, you talked a little bit about getting that in and trying to understand whether there is some optimal thought process around being more or less aggressive in terms of the client’s potential edge on that part. How about the volume side? One of the things about the listed space is you certainly don’t get a price discount for quantity demanded. In other words, the trade gets big enough, the Vol goes up two or three and you never get it back. In that sense, of course, understanding the client, what’s behind the trade is a critical piece of information the salesperson needs to be able to communicate. But the trader’s got so many decisions that he or she could make, whether it’s to try to lay something off immediately, whether it’s a different month, a different strike. Talk about the Vega part of the risk management and how Datadoc tries to quantify some of the trade offs there.

Kumaran Vijayakumar: A lot of hedge fund clients especially which tend to dominate these businesses are not actually vault traders per se. And a lot of times the Vega can actually be relatively mean reverting. It’s more about balancing flows. We help our customers basically understand their clients in terms of how right they are on Delta. As I said, in all these different timeframes, our simulations often involve remarketing, rebalancing, remarketing, rebalancing. So you see that realized component and then the Vega is in there too. And one of the things we do for everything is attribute everything across all those. In Vega especially, there’s really different trading practices about whether people are right on a short basis, a longer term basis, or how everything is done to beat sleeves. There’s often not one answer either. Because like you described, that Vega movement can do different things depending on what the asset is. So you might have a really different approach to something like index options or a lot of the big names, even the biggest clients might have an impact. But at the end of the day, that impact is relatively short lived. And often what the pattern you see is if you can live through that initial impact, if you’re pricing it and trading it correctly, you actually can find edge there.

The more challenging things is if it’s a name where they are the entire name and they’re dominate the market. Obviously that movement may never come back. So you can break those things into sleeves and look at them. One of the funny takeaways is a lot of times some of the clients that are viewed as the worst by our bank customers are the ones who have immediate Delta impact. The trader kind of focuses on that. They’re like, I lost $400,000 yesterday on this move in the stock. And reality, when you take all of their trades together, you see that VOL is actually not their strong suit. It’s not even what they’re trying to trade. If you just can absorb a couple of those P L’s the back end volatility really comes in or goes up whichever way they’re trading. But the P and L is there. So that analysis really does come in very handy. It is true that often, like you said, things are very quick. Other thing that we really help with is that because we’re simulating these P Ls over long periods of time. And the salespeople have access to this too, which is something people never had before.

Most desks, they just hear anecdotally from a trader, oh, event happened, I lost money for the trade you did three weeks ago. But it’s not something they’re actually watching. But these simulations are available for the salespeople too, so they can actually follow their trades. And again, it doesn’t become this thing about looking at the actual trading books and second guessing or adding sort of a layer of confusion that can happen that way. Basically it’s almost like a paper trading world you can live in which the salesperson can also see these trades and kind of live these trades. And what we find our clients is that the salespeople become much more informed. Traders are like, wow, that’s great. The salesperson actually was following this. They actually came to me and said this was happening because they saw it on datadoc’s site versus me saying, hey, you know that trade from four weeks ago, I just lost a ton of money last week and you never even came to talk to me about it. Some of those little things which are not even great data analysis, it’s almost like a low hanging fruit. But it’s amazing how much of that is still not always there.

There’s just too many moving parts and people just don’t have those in front of them most of the time. Yeah.

Dean Curnutt: It’s an incredibly complicated set of data points and you’ve really got three constituency. You’ve got the trader, you’ve got the salesperson and you’ve got the client. Each of which in some ways has a unique agenda. I’d like to think that the salesperson and trader are pretty aligned and that the salesperson and client can be aligned as well. But oftentimes, at least that potentially doesn’t work out that way. So you’re trying to kind of measure and manage the data and be able to make decisions based off the data. A very complex data set.

Kumaran Vijayakumar: What you mentioned about the salesperson and trader being aligned, it’s a good thought. And one thing I’ve realized is that it hasn’t changed that much since we were working together. Our desk did a really good job being aligned, but many desks around us we knew didn’t. It’s still the same. What we can often provide is that language for salespeople and traders to talk to each other in a singular language where they can actually talk about, for example, P and Ls in an intelligent way or return on risk, return on wallet, hey, this is an opportunity cost. Trading this means that this couldn’t be traded. We could have traded more of this if you didn’t trade as much of this. Those discussions that are often really difficult to have, easy to yell about, but hard to have the discussion and having a common language and a common system to look at those really brings that to the forefront. Because really at the end of the day, as much as the market’s advanced, as much as technology advanced something that not changed, we’re still often trade happens and Then there’s a lot of complaining about bad trades and there’s a lot of defensiveness about customers.

And some of that’s healthy. That’s why people often have these different roles. That’s meant to be a sort of healthy discussion. But having more language, having more data in front of you never hurts. It’s just a better discussion. Doesn’t stop all the times that there’s conflict. But the conflict at least is centered around actual numbers and actual decision points and decisions that can be made differently, that can be discussed intelligently.

Dean Curnutt: So there’s the old saying which you can measure, you can manage, and if you can manage it, you can improve it. So we talked a little bit about the trading side, whether it’s the ways in which hedging might be implemented, timing around delta hedges and so forth, movements involved. Talk to us about the sales side of things. So the types of metrics that you’re able to capture. One thing I’ll just kind of lead into is there were certain smaller accounts that provided a lot of value and there were certain giant accounts where when you really just looked at the totality of it, it just didn’t add up to much, even though the top line was enormous. So what about measuring sales effectiveness? Talk about that.

Kumaran Vijayakumar: One of the things we provide also is it’s an agnostic system. Even when firms have some of these or try to do some of this stuff internally tends to be sort of loaded. It’s a trading desk trying to make a point to sales desk. We make this all sort of one set of numbers. Every calculation can be kind of backed out. You can see all the work that went into it. So you can see it all together from the sales side. Absolutely. Where you can actually follow trades, you can actually think about the different decisions you can make. You can also think about how are you missing, are you missing gracefully? Are you losing the call option on future business by not missing gracefully? Is there a probability of roll? Is there a probability unwind? Can you be more proactive about the roles and proactive about unwind? So there’s a lot of things that we bring to the salesforce. But one thing you mentioned is definitely one of the main ones which is that it is true that the big banner headline names often get over focused on and those are the ones that every other bank is covering.

Everyone does talk to. And it’s not that those businesses aren’t worth doing. Of course we all dealt with some of the biggest hedge funds and that’s a great business as well can be. But it’s absolutely true that a lot of times there’s so much underneath and so much idiosyncratic to each bank where they have a set of relationships, a smaller relationships or mid sized relationships. And often that stuff is hard to deal with because it is smaller. Maybe you don’t have the manpower or the technology to really address it, but one of the ways you build that technology or build the ability to address it when you see what the P and L is with it. Almost every bank that we’ve worked with has found a sleeve of business that was basically being sent out to the pipes or sent out to layoff accounts without a lot of thought because they were all small trades. And that stuff often does add up to a lot more. And especially if you can start cutting and slicing that into pieces, you can really find these gems. There are diamonds in the rough where you can find a lot of money often and at the same time also benefit your clients.

Because a lot of times that a lot of money is going through pipes and being handled by layoff accounts or handled by market participants. And you’d actually be a better price than that too. So it can be all of. So it can actually benefit all three parties that you mentioned can benefit sales, benefit trading and benefit the customer all at once.

Dean Curnutt: You said something there. I just run with it a little bit. I think it’s super interesting and certainly something I used to really push, which is there’s a little bit of a beauty contest here. In other words, yeah, we’re going to be really aggressive. That needs to be a part of getting the business. It’s a very competitive, price focused business. But you want to get credit for coming in second and at worst third. Sometimes you don’t want to be first or last. Consistency ought to matter. What might result from that is getting future looks. So how do you quantify or capture the ones where you didn’t want to win, but you did a good job of coming in second even though you really didn’t want to win?

Kumaran Vijayakumar: Most of our customers capture missed trades. That’s one of the data sets we take in from them as well. And one of the things we do with that is we actually take what they showed. We go find it in the marketplace, what traded, we put it together for the probability of trade, can figure out how likely it is to have traded. Did it really trade? Did it not trade, maybe trade it smaller, whatever. Put those things together from numbers and, and basically model not just what they traded but what traded in the marketplace. So you can kind of look at what if it had traded where I showed it? What if I traded with a competitor? Would I wanted to do that. If a salesperson who’s asking for that price from a trader at the same time the salesperson and trader have in front of them all the previous experiences with that customer, not only what they showed, whether they did show it, did it get missed? How was what they showed versus what the competitor, what it traded in the marketplace look like, what’s that difference look like? And also how does it look like on a P and L basis if it had traded where they showed it, where the competitor showed it.

And sometimes even people even mark down things like where the customer told them it traded as a third number to metric off of as well. Because that can be different from what actually shows up in the marketplace. So take all of those. First of all, modeling those so you can actually do the analysis. Deep dive, saying this, looking at a trade, just having that pop up in your screen, having all those things in front of you so you can kind of say for trading. Also, desks are big sales. You might cover with a pod of three salespeople. So maybe you didn’t see this last experience and the trading side, maybe this customer is great on a lot of other sectors, but they’ve come into you for two trades that are terrible in your sector and you think of that as a different way. Or maybe you don’t know that we missed by a mile the last two trades and you’ve generally found them to be a good customer. So maybe you want to make that extra penny. But all those things sort of come right in front of you. But again, part of it is all just sort of modeling the different ways and taking a step back and saying, this is where it trades.

Do I want to trade it there? But absolutely that difference, we have a lot of tools to basically calibrate that missing across the board. Unless it’s a customer that’s so difficult, where you just can’t find a price. You’re just so concerned that you want to be wide on purpose. And that does happen. But that’s a very small sliver for the most part. You want to be close because you want to be coming in. If not second, you want to be coming in third or fourth in a relatively small area because you do want that future optionality. You do want to see that flow. We have a lot of tools to kind of calibrate that.

Dean Curnutt: There’s an old saying, I think Paul Britton said this to me. You want to be in the moving not the storage business. Brokering is just moving stuff around. Even if you have a deep base of capital to take the other side of trades, you, you got to be good at knowing where you can move stuff as well. A question I have is just around an order comes in and having a sense as to whether to lay it off, where to go for that we all know about the folks in the layoff community have different specialties. Even hedge funds can be quasi layoff accounts. We’ve seen some of our ex colleagues be very good at knowing where things can land, tell us a little bit about that part of things and maybe trying to capture that data.

Kumaran Vijayakumar: So one of the things we do for all of our customers is measure a few things. One we actually put together what the layoff was. You mentioned those behaviors like are you laying it off? That’s part of what the whole system’s built around. So basically our system is built around this kind of idea of you’re taking in this risk, you’re going to hedge the risk in different layers. Delta gets hedged right away, then you take this in, then you sort of are going to rebalance that, but you’re also going to have some unwind of the optionality, different versions of it, whether more aggressive or less aggressive. And that’s kind of sort of the core of our system, doing those kind of things and trying to match what your desk is like. And different desk can be different ways. A higher risk tolerance desk might have a different approach where they’re warehousing more, keeping it for longer, and hedging it in a more careful way. A desk that has less risk tolerance might be taking it on more quickly, more aggressively, getting rid of it and laying off more. So there’s definitely things that calibrate to each thing in terms of layoff space.

One of the most important things is that we help our customers model what the layoffs are actually experiencing. So they have a much better sense of what are the P&Ls that layoffs would actually experience. So they have a better sense of one. Are they giving a lot of value out whether they need to or not? It’s not so much about just saying, hey, I have a lot of value, so I’m just gonna keep it all in house. That’s nice. But more, none of us have infinite capital. But the thought is more along the lines of if I do give things to layoffs, first of all, I want to make sure that I’m giving to layoffs who give me value in other ways. The difficult prices are getting the easier, more valuable things. And also, do I really understand what’s happening there? And even managing with the layoff, we have tear sheets where our customers will actually print out exactly when they go sit down with XYZ layoff account, they can actually sit down as you go through trade by trade. And again, because we’re living in a simulation space, this is not about them saying I experienced this or experienced that.

It’s just saying this is how I metric this. I sent you trades that I think are worth between 400 and $600,000. It can be a simple discussion. A layoff account that wants to have no bill. For some trades, you sit down and say, can’t do no bill because I’m sending you this. You’re actually providing me enough value. I value this. So there could be a lot of just conversations. Modeling what the layoffs actually do is important. So you actually have a sense of what that is. Rebalancing and load balancing with the layoffs. Basically knowing that you’re giving the good stuff to the people who are making the hard things and so constantly sort of improving the amount of liquidity you have from layoffs. And also having a sense of. One of the challenges with having multiple counterparties in a trade is that they can actually change the behavior of the trade. One of the questions that can also help address is, is this something where, for example, especially on things like Delta, are you causing your own losses because what you think is risk reducing is actually causing more counterparties to be racing against each other?

And how do we capture those two things and decide maybe cutting 30% of risk is not helping me because I’m actually losing the 30% of P and L out of the gate with information leakage or information leakage or just racing. I mean, like, there’s just a sense of, especially with large trades, it’s just a question of whether the juice is worth the squeeze. Are you getting enough value? Because if everyone’s trying to get Delta, if everyone’s trying to sell long term Vega, that becomes this kind of thing of just who’s first, who’s fastest. And it’s not always optimal. We also model people’s OTC trades and one of the things that comes out from that is that often you can see in the OTC trades with the same customer, you can see different behavior versus listed trades. And you can analyze that. You can often see that very thing, which is that information leakage. Because the OTC trade doesn’t have that information leakage. So you’ll often see the Same customers, just better on an OTC basis and often especially on the Delta, much better.

Dean Curnutt: Now every one of these banks seems to have their sector books set up differently and I think we had tech and financials under one book at one point. There was some old economy books. Of course each of the stocks can be very different depending on the sector. Techvol is a lot different than GMVOL circa 2005 with all the debt and all the volume, SKU and so forth. All traders are different. Some really think about volume from a I’d like to be long it, I’ll pay it away. Others are more carry oriented. But what you want is consistency. And so I’m curious just around the valuation tools of the information that pops out of datadoc and the way in which someone running a trading desk, not on risk management but on trying to create an even feel across the sectors. Is that something that the data can provide?

Kumaran Vijayakumar: It can definitely help it. We do do this on a sort of a granular level and we do all the simulations. Whenever a customer says I wish you could model this, what we say is let’s get a whiteboard up, let’s figure out what you mean by this and let’s model it. Let’s add it to the toolbox of different simulations. Because there’s a point where it comes data overkill. But in our mind we model everything. We’d like to model every single little micro behavior you can have. Where you get that down to consumable data is on the UI front where you can basically find the ways to cut that so you actually get the value to actually impact. But start with basically simulate everything the behavior out there simulated. And so what you describe is true. And it’s not just across sectors, it can be across traders, it can be across even salespeople, trader pockets. That trader works a certain way and doesn’t work that way over here. So our system can often pull all that out. And yes, there’s things you can actually definitely pull out where there’s a lot of analysis that we’ve helped our customers do on some of those things.

This works over here. Can you replicate that? The answer might be no. The answer might be this is biotech over here. And honestly we’re not going to trade it like we trade this. We can help with that answer too. There are analysis you can do that help you find solutions like that. And if you do see something that trades where you’re trading the tech book in a way that works better with this customer or across customers or anything and you can apply that to another area. We can help you do that because again, the difference between the two knob settings of what’s happening might give you some insight into maybe I want to take this knob and move it over here.

Dean Curnutt: So you’ve got a lot of different metrics, one of which, and I love the acronym Bravo. Talk to us about Bravo and what you’re capturing there.

Kumaran Vijayakumar: Bravo is a way of basically capturing, at some point you have too much. How do you actually distill that down to knowing what a customer is? One of the first metrics we did is something called Bravo is Business Risk and Volume. That’s the acronym. What we do is basically take your customers and for you, look at them on the business side, look at how much commissions are they doing, how much revenues are they generating against one of these simulated environments or multiple simulated environments on a risk basis, do you have to cross against them? How much vega per trade are they taking up on a volume perspective? How often are they trading with you? How much are they trading with you? Those kind of metrics. And again, take them, and those are all very interesting distributions. It can be very complicated. And what we’ve done is set up a curving system. So for each of your metrics for your own client base, and every customer of ours has very different client bases. They look very different from each other. In some ways, it doesn’t matter what’s out there, because you really care about what you can address and for your own client base.

Is this someone who is good on the risk perspective, good on the business proposal, and really take that down to sort of a curved grading schedule, and then take that grading schedule, sum it up, and aggregate it. So one of the things you can see is not just what you think of your customer on those metrics and analyze that, but also break down what your customers are like over time. One of the things I know that we worked on a long time ago was not just seeing customers, a singular photo, but it’s really a constantly moving and changing thing. How do you think about your customer’s momentum? How do you think about whether your customer is actually someone who has had a terrible trade? Are they paying you back? Customers grade might drop, but the grade might come back very slowly in a good way, and you might be able to capture that. So we take those numbers, really process that. We have an entire sort of dashboard that comes out of that that helps people get a feel for it. But the idea is to basically give your customers scores along these different axes.

It’s sort of a Living, breathing thing where that score is constantly changing, that score is constantly moving. So you get a sense of where your customer was, what your customer is, what your customer could be.

Dean Curnutt: There’s a couple of follow ups there. So first is we’ve been in situations where a large customer is imposed upon the desk by another part of the firm. It happens. So you have to stand up in some ways to support another business. It’s unclear what the payoff is to the desk, but you want to give the sales trader a lot of credit for handling an especially difficult and dare I say, dangerous client. There are clients for whom you kind of have to do the business, but by virtue of the nature of how that client chooses to interact with the desk, it can be quite dangerous in some ways, at least for me. I always thought I want to capture the salesperson’s alpha in being a protector of the desk. Maybe that means coming in second and doing gracefully, as you said before.

Kumaran Vijayakumar: I think that’s a very good notion. That’s part of what we try to do there. And we have an entire sort of module of sales alpha that comes on the back of this. It’s based on some of those things, like I said, some of the things like roll probability, unwind probability. Are you actually being squeezed for price? All of these metrics are kind of useful. I think the actual Bravo metrics themselves do give a sense of that. Because one of the things I think that you also mentioned is that it also gives a sense of up and down your client base. Because a lot of our banks have, these clients are smaller that are actually probably a client. And then these bigger clients, they may be a range of grades, but also those might be less affectable in terms of their behavior. You may not have much to do with that. The bottom half of your client base, often you have a much more interactive approach with them and you can actually guide their behavior and actually make them better and better clients. And one of our things that our tool allows you to do is have those measures of momentum of clients.

How are things gone last year versus this year, this quarter versus last quarter, all of those things. So you identify these customers you want to do more with and improve them as you go and get that acceleration going. And to some extent the biggest clients, the ones you all trade with, in some ways that’s going to be a little bit more guided by what they’re going to require you to do. And even things like a loss being paid back. One of the things I love about our system is you can actually see it which is really cool because I talk to people about this all the time. We’d always talk about whether the trade was hard, whether customers paying back. Everyone have their own opinions. The traders always think they weren’t. Salespeople might think they were. Who knows? Now you can actually see it just pop up off your screen. Sometimes it just jumps off the screen in both ways, good and bad.

Dean Curnutt: We talked a little bit about investing in clients and having that data. I’m wondering if you took it up a level. And if we think about the volume community, the long short equity community, the crossover trades that come, let’s say, from the high yield investment community that uses hyg, if there’s any way to derive trends or information content that says, hey, you know what? We’re doing really well with long short funds who just have found a way to make it a relatively fair and profitable business for us and we want to devote some resources there to maybe hire a new salesperson that specializes in them.

Kumaran Vijayakumar: These are things can be very difficult. Internally we basically say, tag your data any way you want. We have these transactions, we have probably 100 tags for most transactions. We could have a thousand tracks. We have 10 million tracks. It makes no difference if you can take an account and say this is this kind of account, but it’s also this kind of account. The more layers are better because again, working in the side pocket of data where you’re taking all this data out and stripping it down and just working with it for this financial data intelligence versus working in your primary system, where just changing a sector can be relatively painful. Here it’s sure if you want to have customers, you can even have the same customer pool cut up seven different ways. Where you might have long short versus this, but on a different level, you might think of like your focus accounts versus your middle market accounts. And in different world, you might think of accounts with high momentum versus low momentum. You might take your McLaughlin rackings and basically put those into a layer and say, you know, basically I want to focus on the accounts that you might take your McLonga Ranking Change Year over year and say, I want to focus on the accounts that move down in ranking and I want to move them back up.

You can do any of those things in this datadog side pocket, adding more and more or different layers of data tagging is super easy. And once you do it, the system just allows you to sort of cut and slice along those tags.

Dean Curnutt: So our conversation is obviously very listed. Option, equity, derivative, centric. That’s our backyard. That’s the starting point for sure for datadoc. And of course these transactions hit a tape. They are visible in the light of day, so we can trace them. But other markets are now literally traceable as well. And I’m curious if you can talk about future developments, how you are contemplating working with other asset classes, what you’re seeing from your client base in terms of how you might help them on that front.

Kumaran Vijayakumar: It’s one of those things where there’s an ease and listed space where there is all the transactional data and everything is on the tape. And that definitely helps. The funny part is in my career I also managed convertibles groups quite a few times. And I always think of convertibles when I think of this because convertibles is a less visible marketplace. It’s often traded more on a transaction by transaction basis. But in a lot of ways there was more inefficiency there and more to actually capture when you do capture it. So we find the same thing. So like I described, listed where we started, list of derivatives, OTC derivatives became a very quick jumping bot point variant swaps, other things we model. Again, some of these things you don’t have as much data as to exactly where traded or where it hit the tape, but you can definitely trade it. You can have different behaviors, you can have marks that make sense, you can have unwinds that make sense versus some liquidity metric. So you can do all those things. So that came up next and then we’ve really expanded now. We were never built a company we thought of as a druidist company, we built it for derivatives because what we knew best, where we know the people the best, we built a company for financial services, capital markets decisions that involve data.

That’s really what we are. We want to be a financial data intelligence company. So in my mind I picked the stuff that I knew best as a starting off point where the customers should do it. But the same thing applies across asset classes, across regions in credit space. For example, credit’s going through the steps of becoming electronic fight that we went through 25 years ago. And it’s still a work in progress. And some of it will always be a little different. It won’t follow the exact same pathway, but we’re actually working with one of the largest areas. They basically facilitate trades in the credit space. And we’ve been working with them to basically look at all the different ways that they can handle trades. And like you described before, kind of applying what we do with Bravo, we’re kind of saying how do you actually take the Trades that come in optimize how they set the settings to send those trades out. And also really look at how do I send this to my grade A counterparties? How do I figure out what my grade A counterparties are as a counterparty? How do I look at the people who I view as most valuable?

So credit’s become a really big space for us. We really look forward to expanding across all asset classes over time. And I think that some of these asset classes that have actually less visibility could be even better. They are a little more challenging at the get go. But there’s actually more edge to come from it because the analysis really shows you things that just aren’t known in those marketplaces where they haven’t had that in front of them. They don’t know. For example, the idea of just, hey, I do a credit trade and I want to model how that credit trade on the other side gets put into a portfolio or made into an etf. And I want to actually know how that correlation is being handled by my counterparty. Something that has not really happened in that marketplace. We’re looking forward to that.

Dean Curnutt: Last question. 2018, you guys launched the business. The acronym LLM is not really a part of the lexicon. And you’ve talked to me about this term agentic. With the incredible pace of advancement in data and inference, how is that interacting with your business?

Kumaran Vijayakumar: I’ve always been a little bit of a skeptic in terms of buzzwords. My approach to these things is always there’s a lot of cool things, but there’s no reason to do these things for the sake of doing them. But there is something very specific here, which is that even in this conversation we talked a lot about all these different data and all these different things that come about. What we’re trying to address is the original question. So salesperson says this order came in. Who should I call to lay this off? That’s the question they’re asking. And we have a bunch of tools to answer that question. And we have a lot of ways to use those tools and we try to make them easier and easier and make them more and more available pre trade. And that’s all great, but some world in the future, and we think in the relatively near future and we’re working on these tools already, that question should be asked as that question and the answer should come back answering that question. All the data is here. And our idea is not so much about this sort of very broad, can I have this truly AI conversation with this?

But more for what our Customers really need, which is they have these real questions. So trade comes in. How should I hedge this? Trade is a real question. If I find a layoff, who should I lay it off to? Should I look at this versus other things? We know what the questions are being asked on the desk are we know how that dialogue’s going and we can show how to do it now. It’s a relatively short leap and one we think we can deliver in the next couple of quarters where you can actually ask that question and get that back. And that’ll obviously be a work in progress. It’ll take time, it’ll constantly evolve. But it is something where there is a real need here, where there is just simply too much data and there’s too many ways to use it. And really distilling it down to that impactful. Here’s what I want to do. The data shows me this is what I should do and I do it. And getting that really down to that simple stretch is what I think of as the LLM component of this. And I really say it’s agentic because not really.

The models don’t have to be that large language in our space. It’s a more restricted set of things that people really need to be able to ask, but they want really accurate answers. As you know, on a trading desk, if you ask that question, you get junk. Even if you’re getting it at an 80% hit rate, it makes it basically unusable. So what you need to do is strip down a little bit the wow factor, but make it be able to answer the questions that are being asked really accurately.

Dean Curnutt: Such an interesting and fast paced transaction that comes to be in a moment’s notice where tons of risk is transferred, lots of decisions are made in a hurry, and there’s so many pieces that can ultimately be picked up in terms of compiling the information. It seems like that’s really what you guys are doing. It’s been great to learn more about.

Kumaran Vijayakumar: The product, great talking to you and really, really appreciate your time. There are many, many times when we’ve been going through the last six years building these tools that I’ve thought about things that we would have liked on the desk, in conversation me and you had, which we would love to have. And a lot of it’s been about how can you bring those things forward that we would have liked to have 20 years ago and weren’t even close to having in terms of technology or in terms of data sets. And now it’s a lot, lot closer.

Dean Curnutt: Looking forward to tracking the progress and the growth. Thanks for this.

Kumaran Vijayakumar: Absolutely.