The Massive Company that’s Trailblazing like a Startup with Desikan Madhavanur, CPO & CTO at Capital One Software

(Intro Narrator at 00:00:00) Today, we're talking to Desikan, CPO and CTO at Capital One Software, about how they're democratizing cloud data management. You're listening to Joel Beasley, Modern CTO.

(Joel Beasley at 00:00:17) When I saw the opportunity come up to get to talk about what's going on over at Capital One and their technology, I said, "Yeah, sign me up." That's where I want to start. I'm just curious, what is going on in technology at Capital One?

(Desikan Madhavanur at 00:00:30) Yeah, you know, first, let me do a very quick intro so people know who I am. Desikan Madhavanur. I'm the Chief Technology and Product Officer for Capital One Software, which is the B2B enterprise software business within Capital One. So that's what I know. And yeah, lots of stuff happening at Capital One, right? You know, as a company, we're always breaking forward. We're leaning forward. We take a look at new technology as they emerge. We are the first to the cloud. We are the first to adopt a lot of data-driven decision making before most other people did it. So you can expect us, in that way, to be leaning in in lots of different domains, from user experience to data. We're all over the place here, moving ahead and experimenting constantly and pushing new innovations. So yeah, it's another interesting year at Re:Invent next week, I have to tell you.

(Joel Beasley at 00:01:23) What's taking up most of your time right now?

(Desikan Madhavanur at 00:01:25) So I'm specifically in the software division, right? So for me, I look at the world in two ways. Right? On the one hand, I look at Capital One and what we have done in terms of building assets to manage our infrastructure in the cloud, our data in the cloud, standardizing so that we can leverage cloud without creating bottlenecks. And then I look around and see my customers. I talk to a lot of companies in various industries, not just in fintech, but I talk to people in the healthcare business, in pharma, in consumer goods, in e-commerce. And I see them struggling with things that we have solved over the last three, four, five years. And for me, the excitement is, how can I translate what we have done into tangible assets that can help these companies move forward? And how do we turn that into a commercial business if the opportunity exists?

So I'm on the lookout. So on the one hand, it's very interesting. I come from the startup world. You're building new stuff from scratch, and here I am sitting with a bunch of assets that I know perform at scale within the bank. And I have exactly the same problems out there in the industry across so many different companies. And for me, it's like, how do I take the most valuable thing that I have? How do I do it in a way that is truly commercial, truly enterprise software that works across customers? And how do I now turn that into something that I can go deliver and implement? So that's the exciting part of my job. I take a lot of pride in doing that every day.

(Joel Beasley at 00:02:56) Do you have something yet?

(Desikan Madhavanur at 00:02:58) Yeah, we have one product already in production. It's called Slingshot. So just to give you a background, we've been a huge user of Snowflake. We're one of the earliest users of Snowflake, started in 2017, probably still one of the top three biggest consumers of that technology, to the point where our scale, probably—if I can tell you—it's over a million queries a day from hundreds of different applications. 6,000-plus analysts hit it on a daily basis.

Now, how do you run this effectively? How do you run this efficiently from a cost perspective, from a performance perspective, from a governance perspective? So the last four years, given Snowflake itself was, you know, growing and creating functions, it was not the most well-managed out of the box as a new product in the domain. So we built a whole set of technologies to optimize, to make it work in a way that you can actually onboard hundreds, thousands of users safely and cost-effectively. Now it produced tremendous success for us in terms of savings, in terms of time as well as dollars. And we productized that capability as Slingshot. It's now available to try from the Snowflake Marketplace, and we have that out there. We have multiple customers in production. And so that's our first product that's already in production. And now we're working on the pipeline for the next set that need to come out as we get deeper into managing data in the cloud.

(Joel Beasley at 00:04:24) Are you going to spin these out into their own companies?

(Desikan Madhavanur at 00:04:27) No, at this point we are not contemplating that. I mean, you know, there is a lot of things that we have to do. Right now, there is a big advantage for these things to stay here within Capital One as a business and deliver them. Because, for instance, a lot of the tech we leverage—internal platform—even for building the software that we ship, whether it's the DevOps side, whether it's the security side, lots of stuff is built in building the technology that is proven in Capital One. That's number one.

Number two, Capital One is probably the biggest tech brand name that's not purely a tech company that's in the industry. So what we do, we have a huge advantage as Capital One because the software we build is proven to be cloud-first. It's proven to be highly scalable. It's battle-tested at probably the largest scale you can find. So there is so much goodness for this business to be a Capital One business that we don't contemplate anything else at this time.

(Joel Beasley at 00:05:24) So how do you handle the executive structure of it? Like, who's running sales for Slingshot? How does that integrate in with what you guys are doing?

(Desikan Madhavanur at 00:05:34) So it's a business unit. So in every sense, Capital One Software has its own sales team. It has its own product management and engineering team, right? It receives funding like any other business unit does. Right? So without going into too much organizational detail, right, there is a lot of independence and recognition within the company that this is something that we are building not to serve ourselves, but to serve other customers. Right? So that is a clear recognition that it'll need a different clock speed and different set of talent.

So, you know, the reason I'm here—I've been here 10 months—is because they brought me in because I have a lot of experience in the enterprise software space. So there's the recognition of that. So yeah, it is a business unit in the true sense of what you would call from a business perspective. So we have our ability, our go-to-market, our engineering, our product management. We manage our scope, our roadmap, and we deliver our customers the experience that they seek, obviously operating under the Capital One overall umbrella and organization. So that's the way it works. So there's a lot of synergy that we get benefit from, but where we need the ability to make decisions, we have that as well.

(Joel Beasley at 00:06:48) How do you figure out if there's enough of a market to justify it?

(Desikan Madhavanur at 00:06:52) Oh, we use the standard—you know, Capital One as a company prides itself in being very data-driven. So there's a lot of customer testing, validation, strategy development that happens everywhere within Capital One. So we actually benefit from that. So within Capital One Software, obviously, the way it manifests itself is we take something that's worked really well, and we apply—we go meet customers, talk to customers, get feedback from customers to see how does this translate. Then we come back and look at what would it take to take what we've got to make it this enterprise SaaS product. Is this going to work?

And there's all kinds of constraints there. There's a product fit, which is what you would think, but there is even other things. When you have a broad array of things you can do, you have actually a lot of choice. So there is actually a process of prioritizing. Is this something I want to do right now? Is this a pressing problem now? Does this have wide market appeal? So we do a lot of that before we decide where we go, and then it goes to the normal software development process we have for customers. We get feedback. We go through phase gates. It's almost like funding from a startup perspective. Do we fund this thing into a beta? And then it goes to beta. Do we fund this thing into production, to a launch?

So that's the phase gates that we use just like anybody outside. But this is—you know, we have the benefit of taking advantage of very well-developed software development life cycles that we have within Capital One across the board. So normal customer validation, normal product market sizing from a strategy perspective that works in card, that works in insurance, works here as well.

(Joel Beasley at 00:08:30) How do you learn how to do all this?

(Desikan Madhavanur at 00:08:32) How did I learn how to do this? I've spent about 20, 22 years in enterprise software. I've had, you know, it's an interesting journey. Like, I started, and my first big win was part of a startup called i2. And this is before the hyperscalers were real, so I had to handle the cloud stacks, you know, in the private cloud world. And so I cut my teeth there. It was solving how to optimize supply chain networks now that we have the internet and we can actually connect supplies and customers together.

So up until that time, you optimized factories, you optimized warehouses independently, right? You never could take advantage of the fact that there is data that's changing in the demand that should impact supply because these are not connected. They were sitting in different silos. With the internet, we're able to connect these things. So this software optimized the network. That was the first thing we did. Led to a successful IPO on NASDAQ in 2011.

So there, I was able to understand the use case and how technology and changes in technology make things possible, right? And how do you put that together? And obviously a lot of engineering around what makes a good, manageable cloud stack, right? You know, we had big, big customers, so I did that.

Then in my next journey, I went to a unit, part of Computer Associates, now Broadcom, that focused on cloud management because the experience of doing the use case translated into how does the technology help promote broadly. So I did that and that was a good experience. Then I went back into the industry with a company called Blue Yonder most recently. And there, that company, we moved it to the cloud. And beyond that, it is the advent of AI/ML, of big data and crunching. So here I was using modern technology around demand forecasting and other areas, predictions, to solve supply chain problems, right? Again, back in the industry.

And as I finished that, we had a great exit. We got acquired a year and a half ago by Panasonic, so that was a great exit. When we finished that, I was like, okay, now there is massive opportunity to take these data infrastructures that we worked on and make them more broadly available. So there is a pattern here of use case driving, you know, innovation in tech, and then democratizing that tech, then new use cases are driving more tech.

And now for me, as I look here at Capital One, the reason I'm here is it's an unbelievable opportunity to democratize data management in the cloud, which is going to be the biggest problem. Cloud solved a lot of problems in the space. It's created—or several problems that we thought we solved a long time ago have reemerged. And data management, doing that right, is going to be absolutely critical. I think I'm super excited to be here. And this Slingshot is a first example of just cost optimization, right? It's just one part of that, performance optimization. There's so much more to do.

(Joel Beasley at 00:11:21) What's a problem that we thought we solved, but it's reemerged?

(Desikan Madhavanur at 00:11:25) I can give you a few, right? Let me tell you. The first time we had massive opportunity to use data to make decisions, like I told you before, was when the internet came about, and all of a sudden you could connect silos. And now you could optimize networks rather than nodes, right? That was great. That led to problems of integration, problems of non-standardization, problems of data quality.

And the way we solved it at that time was locking it in with RDBMSs, right, with Oracles and DB2s. Standardize the schema, put a central IT team in place. Basically, we created central procedures, teams, and policies that mandated that everything be standardized. That way we got around it for a little while.

Now with cloud, it's changed, right? The cloud stack is very different. Here, you have—if you think about it, that paradigm doesn't work anymore because you have to federate. First of all, relational databases as a technology doesn't scale to the volumes you're talking about. So that's an architectural problem. But if you put that aside, if you just look at the management problem, every team within enterprises have the flexibility to go leverage technology that is custom-built for that particular use case.

If you have a fast streaming use case, you can get streaming databases. If you have a large data crunching, predictions, forecasting, you can go get the best big data stacks, right? So now what's happened is flexibility and freedom is what's coming in as a way to solve in the cloud. It's reemerging the exact same problems that the relational databases thought that they solved.

Number one, you have no single—very few companies have standardized completely in a single public cloud, which means you have data sitting in legacy systems inside the enterprise. You have data sitting in multiple clouds, and now you have a very weird architecture trying to integrate this, you know, really bespoke architecture to integrate that, which goes back to the time before standardization. That's one problem. Integration problems.

Number two, the thing that we thought we absolutely solved was data discovery. If you go into an enterprise and you want to know the MDM, master data system that could tell you where the data was—today, data is everywhere and you really—the data discovery problems reemerge as a big thing. Many companies struggle from not knowing where data is.

Let's say you can get that data. So let's go to the next one. Today, there's multiple copies of data that teams create in different stacks for the same set of elements, right? There's so much replication, fragmentation happening, which means you do not know if the copy that you're looking at that you discovered has no errors, is clean, has no latency, is current. None of that is obvious. So which means you brought back the data quality issues that, again, we thought we had solved is again reemerged.

So if you look at it, and then finally fragmentation, right? There's so much fragmentation of data and tools that governance is back to being a problem. Compliance is back to being a problem. So if you think about it, cloud made certain things super easy in the data space. If you want to be a data-driven organization, it's really simplified. You can get storage at the click of a button. You can get compute at the click of a button. You can get visualization tools at the click of a button, transformation tools at the click of a button. Everything is easy and everything has more than one choice.

So it made building data-driven applications super easy. What has it led to? The reemergence of the biggest problems we thought we had solved—data quality, data identification. Where is the data sitting? You know, that is a big problem. Data governance is a big, big problem, right? So we have got all these problems that we thought we solved coming back in force with cloud. So it's a big advantage, cloud, for building data-driven applications.

(Desikan Madhavanur at 00:15:05) It's produced the biggest challenges that companies are now facing. A lot of inefficiencies are sitting in every enterprise. That's precisely the opportunity I want to take advantage of here.

(Joel Beasley at 00:15:16) Does Slingshot address all of these or just a couple of them?

(Desikan Madhavanur at 00:15:19) No, Slingshot is very focused. Right? Slingshot looks at one thing. It looks at the fact that in this process, when you onboard several users and several applications into a Snowflake ecosystem, now with that flexibility and freedom you've given to the teams, we have tremendous inefficiencies.

(Desikan Madhavanur at 00:15:35) We have warehouses that are not sized correctly. We have queries that come from applications that are not written well. So all of a sudden, this federation of responsibility in a Snowflake ecosystem means you're running serious cost overages, you have governance challenges. What Slingshot does is solve that problem. It goes in and optimizes both the warehouses and the infrastructure itself that you're running through schedules.

(Desikan Madhavanur at 00:16:00) How often do you need a large warehouse? Can you turn this into a small warehouse and you don't need the large warehouse? Can you actually put it in a mode where it's not costing much? Right? And then you have serious SLAs. You have performance requirements. Can we actually right size it and go hit the performance SLAs when we have to? So all of that, that is on the infrastructure side. And then on the workload side, queries, the applications they bring in. Why are you doing full table scans? What are you doing that you should actually be doing in memory that you're not doing correctly? So it right sizes the workload. So it essentially takes cost inefficiencies and governance problems out of the Snowflake ecosystem that customers have. So it's a pretty limited first solution, very effective but limited there. Now you expand it to all ecosystems, you expand it beyond performance and cost to all sorts of management problems. Now there in lies the big prize to go solve.

(Joel Beasley at 00:16:52) Well, sounds like there's a lot of opportunity over there. There's a lot of these problems. As you're doing this and you're building teams and you're growing people and you're solving these difficult problems, when you're looking at the next generation of leaders that you can invest into, like, these people that are showing signs, what qualities, what signs do you look for or behaviors exhibited in people that you think to yourself, you know, I should invest some time into them. They're going to become a great leader one day.

(Desikan Madhavanur at 00:17:24) Yeah. And this is a very, very important question and one that we ponder. A lot of times, you know, in my, I'm guilty of this, you rely on expertise. You want to bring in people who know something really well. Right? The thing that we have been taught as a collective group, and I have learned this lesson multiple times, the average life cycle of a tool, a system in today's data-driven world is measured in ten, twelve, eighteen months, perhaps three, four years before it gets replaced by the next thing. Right? That is how quickly the thing switches. So I look for people who demonstrate clear problem-solving skills. Like, you know, from the ground up, they can go solve problems. And judgment, the ability to understand, is this going to be a market? Is it not going to be a market? So those things take precedence over knowing one domain or a single product or a single ecosystem really well. And just in the last ten to twelve months, I've done two pivots. Right? I move around. Like, this seems like a more important idea. And if I don't have the ability of leaders to quickly catch on, make those judgment calls, turn, and then start working, it becomes problematic. So, you know, obviously, we bring in people who have strong skills in specific areas, like whether it's coding skills or whether it's automation skills. Those are all very important. The true leaders that we look at and say, oh, these people are going to really make a difference are ones that exhibit their ability to have quick judgments and the ability to pivot quickly and not have a big learning curve. So that's what we look for continuously. How easy is this for this person to go from problem A to problem B and be effective right away? That is something that we look for. And I think that ability is going to create the next set of leaders because there's not going to be, we're not going to have an Oracle generation where entire generation works on one, two, and goes into retirement. I think that time is behind. We're going to have the people who have to get trained, get educated, move on to the next one, and drive innovation there. That's the generation we live in, so that's what we look for.

(Joel Beasley at 00:19:36) Yeah. I've noticed that it used to be a lot like that with the Oracle, and then people almost shifted that to stacks. Like, oh, I know this stack. I know these set of tools together. I'm like, what's the next abstraction?

(Desikan Madhavanur at 00:19:49) Yeah. Yeah. That's exactly it. Yeah. And, you know, these abstractions themselves are changing. Right? And then depending on the use case, they're changing very, very, very fast. So I am with you on that, man. I mean, it's, this is evolving in front of our eyes. I think, especially in the future where a lot of the work that we do today, which is specialized, become commoditized, judgment, decision-making, understanding the user experience, translating that into product need. Now that is going to become the primary skill because pretty much everything else is going to get commoditized. So yeah.

(Joel Beasley at 00:20:24) Say that one more time so everybody can take that in. What's the primary skill?

(Desikan Madhavanur at 00:20:28) The primary skill is going to come down to the ability to understand customer need, turn that into real experiences. Because the actual creation of those experiences, the grunt work that we today call fresh life, that's going to get commoditized.

(Joel Beasley at 00:20:42) Yeah. I remember when I realized that I'd spent the first, you know, seventeen years of my career writing software.

(Desikan Madhavanur at 00:20:49) Yeah.

(Joel Beasley at 00:20:49) And then I realized, I'm the construction worker. Yeah. You know? And I was like, it's important. Absolutely necessary. But, like, writing the code all day, you're just laying, you're laying the infrastructure. You're laying the groundwork of what is the thing that's valuable and it's completed state of bringing value to the market. And so I just climbed up the stack. I was like, alright. If I really like this discipline, but I like more helping people with problems. And if I can apply all of my knowledge of this discipline to helping people with problems, then, you know, it's like what you're doing. It's a nice pairing.

(Desikan Madhavanur at 00:21:23) That's like we were doing two sides of the same coin. That's exactly what I'm here trying to do as well, which is problems that are solved in certain places. How do they translate well into emerging problems that are coming out there, and how do we pivot and solve those problems? And that is going to be the name of the game here. You know, people who can't have that clock cycles, that turns in an environment where an ecosystem, even a full stack, doesn't last more than three, four years and gets completely replaced by the next thing, you're not going to make it. So that becomes very, very important. That ability to pivot, the judgment, and understanding customer need, market fit, turning that into real experience. So that's going to be the next big thing.

(Joel Beasley at 00:22:02) Tell me about the pivots that you made on Slingshot.

(Desikan Madhavanur at 00:22:05) Well, there are a few things that we did. Right? For instance, let me tell you. Usually, you're very customer-driven. So you go after the first set of customers. What do they need? That's the journey that we take when we have the money coming out.

(Joel Beasley at 00:22:14) Yeah.

(Desikan Madhavanur at 00:22:15) Exactly. Right? You've got to get new products. You've got to come in. You've got to get people to start paying. So Slingshot attacks the problem of optimization. Now optimization is a very loosely used term. It means different things to different people. Right? For some people, optimization is let me operate at the lowest possible cost. Right? So we went and did this. But the more you climb to mission-critical problems that are being run by Snowflake now, optimization is not just that. For some people, in fact, cost doesn't even factor. Optimization is how can I optimally deploy capacity to meet all my SLAs so I meet all my customer needs? So there, it's a very performance-driven journey. And in most cases, what we realize is a trade-off. So the big pivot that we have done is not only to add performance-based optimization as a primary requirement, just as important as cost-based, but then realizing the trade-off is not really a pure deterministic decision. You have to have what-if scenarios. You've got to tell people if you added more capacity you could meet these SLAs. Is that more important? So there is now a scenario planning capability. So these are all things just within the optimization model within one ecosystem you learn in a matter of six months. You go from, hey, I'm going to help you reduce the cost to, oh, no, no, I'm going to help you optimize the performance to, well, I'm going to give you the tools so you can make the right decisions, and then we'll learn from it. And then next time you come in, we'll tell you what's the most likely solution that you should go after. So this is a learning process. I wouldn't call them massive pivots, but these are areas where what we learn has gone in and fundamentally changed how we build certain aspects of our software. Right? So, you know, a couple of quick examples.

(Joel Beasley at 00:24:00) Yeah. Yeah. I mean, it's like, what? You mean to tell me spending time with your customer makes a better product? Right?

(Desikan Madhavanur at 00:24:06) Yeah. No. I love it, though.

(Joel Beasley at 00:24:08) I'm learning a lot here too.

(Desikan Madhavanur at 00:24:11) Absolutely. Yeah. I mean, customers, you know, most of the times that I've made decisions in my life, not just here, and I look back and said, I should have done this differently, that was a customer feedback that I overlooked. So, you know, that is, we take this very, very seriously. We have a real process. Even in our early stages, we are very driven by having good processes, and one of those processes is how do we deal with customers? How do we get feedback? How do we incorporate that feedback? And how do we pivot quickly based on that? So we do this in a clock cycle that you'd be surprised that we can actually do this. You know, we deploy every two to three weeks, and we get feedback every two to three weeks. And we get out, you know, that it's a pretty aggressive cycle of production deployments, especially in our industry. Right?

(Joel Beasley at 00:25:00) Let's talk more about leadership. Typically, the people that are listening to the show are either CTOs that want to hear from other CTOs or VPs of engineering or first-time managers or people that want to become managers. So definitely people who are interested in technology and leadership. And one of the biggest pieces of feedback I get as far as guests coming on and sharing stories is that it's really, really helpful when people share mistakes that they've made as a leader. Yeah. Maybe one that they've made multiple times when they've had to learn the lesson, and then they finally picked up on it. Do you have any of those?

(Desikan Madhavanur at 00:25:34) I mean, we all have our share of mistakes. You know, the two things I'll tell you. The first one is the quicker you can learn from them and then the stronger the surround team around you, the easier it is to deal with mistakes and learn from them. So I've been fortunate throughout my career to identify the mistakes pretty quickly, as well as have a group of people around me that are pretty straightforward and come to me and tell me, hey, look. This is going sideways. Right? So, you know, for instance, in my very first company that went public, right, about halfway down our path, AWS was beginning to become mainstream. Right? So I had a true choice. Do I switch over to the public cloud or do I continue with my private cloud journey? Right? At that time and in fact, it was proven for the next three, four years when AWS went through certain cycles and there were certain challenges in managing that stack and their uptime and everything, availability, that my initial decision to go stick with the private cloud with all the instrumentation I had hand-built in the private cloud would be the right way to go. And I was pretty certain about it. When we went IPO, I told people, look, it happened because we had control over our infrastructure. But looking back, right, as a company that is still there, I think that probably put a lot of pressure later in that public cloud journey. So today, I look at those decisions with a lot more care, a lot more long-term view in mind. Right? Even here in Capital One, when I look at what visualization tool am I using, or what stack choice am I making in terms of my rollback access to some stuff. And I look at those things. I look at not what I know for sure in the next two years maybe, but I look more in terms of optionality in the long term. That's a lesson that I learned that, you know, you cannot optimize. I mean, the deficit that you build, which becomes debt over time when you make those decisions which you think at the time are right, get to become a problem. And the biggest, a lot of people say, hey, the quick learner, the right way to, for me, I've, that's worked for me is surrounding myself with people who are smart and who are not afraid to pull me up and say, have you thought through this? Right? So that's a big lesson. So I surround myself with people. I hire people. I work for people that do that all the time, that question me, that make sure that I have the right motivations in mind. And even if it's suboptimally, we all know where the suboptimality is so we can go back and fix it. So I mean, I would tell you that I probably should have jumped on the public cloud sooner than I did. It would have might have delayed the IPO, but it would have been long-term good for the company. Right? So I learned from that obviously. My next journey with BlueJeans, which I talked about, we're all in on Azure. I knew right up front that I am not going to go and try to do any of these things on my own, and that turned out to be an amazing, amazing partnership with Microsoft. So had mistakes in the past, but thankfully, nothing that was huge in terms of my career or company itself. And most of which I overcame with good people around me.

(Joel Beasley at 00:28:57) That's brilliant advice. Let's wrap up a little bit here. Is there anything else going on at Capital One that you want to get out there to the world, or do you want to tell people, are you guys hiring? Are you looking for people for your team?

(Desikan Madhavanur at 00:29:10) Of course. I mean, we are always hiring in this space. You're always looking for talent. This is a talent game. At the end of the day, right, software is you're arbitrating people's brainpower and skill for, you know, that's what you're doing. Right? You're just bringing people. Without people, you don't have software. So we're always looking for that, and it's an exciting space to be in. We have a very strong, powerful roadmap for Slingshot, what it's going to do in the Snowflake ecosystem with partnership with Snowflake that's coming up.

(Desikan Madhavanur at 00:29:38) And there is new products that are going to come out that's going to help us get broader into the data management space. I would say stay tuned. We'll be out there in the ways, letting people know what's coming out. And if you're excited about this problem, if you want to solve complex data management challenges in the cloud, we are here. Please reach out and we'd be happy to have a conversation.

(Joel Beasley at 00:30:02) Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it. And if you have topics that you'd like to hear discussed on the podcast, either add me on LinkedIn or send me an email [email protected]. Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.