Latest / The Longest View with Dez Fleming / AI Agents for Private Equity with Rohan Parikh, Co-Founder of Keye
Transcript
- [Desmond Fleming] (0:00 - 0:25) I would love to start the conversation. We'll talk about Keye. We'll talk about your journey, but I would love to just dive right in and talk about the need for deterministic software within private equity and how they think about obviously adopting LLMs, but some of that tension in order of being able to deliver precise and accurate answers for the industry. [Rohan Parikh] (0:25 - 2:54) Yeah, totally. Well, thanks for having me, first of all, Dez. Uh, it's, it's very interesting, right? Like private equity as an industry, if you look at their core function, you raise money, you have a fund, you're looking at 10, 12 investments in that fund, probably more or Dez, depending on how big the fund is, we're looking at every investment, almost being like five to 10% of your portfolio, anywhere between $10 million to billions of dollars. And when it is such an important part of your portfolio, understanding what is the reality behind the numbers, understanding the numbers themselves to the T, you're not looking to invest 5, 10, 15%, you're looking to buy the company out, majority ownership, where you will be responsible as a fund, uh, keep the management team operations, take the decisions, et cetera. And so when you look at what's at stake, it is really, really high stakes. Pretty much so much that your one investment goes wrong, like you're wiped out of any of the returns of the IRR, right? Like one mistake. Yep. And so like when it's not deterministic, the things at stake are, it's, it's massive implications throughout. And so understanding not only if this is the right metric, the right calculation, this is the right health of the business, understanding how you arrive at that. And the industry has been going on since over 50 years at this point in time. And people have done this over and over again. Nothing, none of this is rocket science. And so there's a process set to it in terms of how deep you can go, how deep you want to go. And so not having deterministic outputs, not having an understanding of like, how did you arrive at this number is massively uncalming for anyone in the seat of a private equity professional when you have to make a decision that actually could change the company, the employees, your own fund. [Desmond Fleming] (2:54 - 3:03) Yeah. Yeah. Your career's at stake with some of these decisions or your career is at stake, depending on the underwriting you do, you have to live with it. [Rohan Parikh] (3:04 - 3:33) And it's massive. And so when you're making a decision, I mean, that's why due diligence cycDez could go anywhere from three months to 12 months. And so when you're doing such an intense due diligence, not having a deterministic way of understanding, not having a really deep understanding on the numbers, that's like, if you say something like that, that's like an excuse, which you cannot go over. [Desmond Fleming] (3:33 - 3:33) Yeah. [Rohan Parikh] (3:33 - 3:40) And so determinism for us was core to why we build Keye. [Desmond Fleming] (3:41 - 3:46) Yeah. And let's talk about what is, what is, what is Keye and what is your story? How did, how did we get here? [Rohan Parikh] (3:46 - 5:31) Yeah. So Keye, we're a solution that helps private equity professionals in their due diligence process, especially making sense of the numbers and understanding the true reality of the business, the health of the business, using their customers, their employees, census statements, et cetera, bringing out the, I would say the depth in the numbers. So everything related to numbers in the due diligence process, when you're going through massive amount of data in your data room, whether it's your customer informations for consumers, softwares, et cetera, whether it's your employees and how they're performing, we'll cut through those sometimes millions and millions rows of Excels, PDFs, et cetera. And then go deep and tell you like where we think the health of the company is. Now as a fund, how you use that, whether you're an opportunistic fund or you're a, you're a more, I would say, looking at stable businesses, organic businesses, et cetera, that's where you decide whether this information is something that you want to take forward or not. So we help you almost to get to 90 to 95% of those insights, like very quickly. So that you can say no to a deal earlier in the process that helps you save time to actually look at the deals, which actually would matter to you. And with the same selectivity rate, you're probably able to do a lot more deals. [Desmond Fleming] (5:32 - 5:37) Yeah. What's the biggest file that has been ingested into Keye? Like, do you even know? [Rohan Parikh] (5:38 - 5:49) Uh, yeah. I mean, when a big file comes in, the team's definitely excited because I think that's one of the biggest things. I mean, Excel is the bread and butter of everything. [Desmond Fleming] (5:49 - 5:49) Yeah. [Rohan Parikh] (5:49 - 5:56) Um, and it's like, it's, it's amazing how Excel hasn't improved since. [Desmond Fleming] (5:57 - 6:03) Well, maybe that's just function of, or expositive of how strong Excel is because it's so, so modular. [Rohan Parikh] (6:04 - 6:24) Yes, it is definitely, but not so much for private equity because now they're dealing with millions and millions of rows and Excel has its own limitations and that's why you have a lot of power pivots, et cetera, which, which kind of work on that. Uh, but what it does, it restricts the ability of an investor to cut through that data. [Desmond Fleming] (6:24 - 6:24) Yeah. [Rohan Parikh] (6:24 - 6:42) Um, and so we've had times where sometimes like hundreds of MBs, even gigabytes sometimes, which could be like millions and millions of rows, which Excel would not like, forget about even open that, it cannot open that in Excel. [Desmond Fleming] (6:43 - 6:43) Yeah. [Rohan Parikh] (6:43 - 6:57) That was the thesis behind it that you need determinism, you need a hundred percent accuracy up, and then you need the ability to take the analysis to the next level that actually brought about Keye. [Desmond Fleming] (6:57 - 7:12) Time and speed are important, especially in competitive deal processes. So that's Keye. You're building a, you know, kind of AI enabled analysis and due diligence processes for private equity. Who is Rohan? How did you get here? Why did you start Keye? [Rohan Parikh] (7:12 - 8:10) Yeah, no, uh, I, so just a little bit of a background. So born, brought up in Bombay, India, born to a family of entrepreneurs, like mom, dad, everyone in my family started their own own, started their own businesses or own firms. So a lot of ups and downs that failed quite a few times, then did a successful business. And so always were kind of entrenched in the, in the entrepreneurial ecosystem from the very start, maybe not on the driving seat, but at least on the backseat. Uh, to always wanted to start something on my own, mostly did not have the right idea right out of school. So decided that wanted to experience how corporate world works. And so was good at math, had an inclination towards numbers. And so decided to join an investment bank on the Texas, which is right here at 50, 50th and sixth. [Desmond Fleming] (8:10 - 8:12) Is that like a European, Japanese? [Rohan Parikh] (8:12 - 10:47) Yes. It's a French, it's a French bank. It's the second largest. They have big operations in New York. Um, and so joined them and then, um, worked across various different departments, so worked in global markets, structure financing, then banking. So I've gone through like the entire Excel, whether it's like modeling, whether it's like processes, like stochastic, et cetera. So worked through the bank across, uh, whether it's analyzing public companies, whether it's analyzing private companies, had a really good run in there. Had a really good team, a great boss. Went from the ranks of an analyst to director in Dez than five years. And then finally decided that, you know, it's my calling to start something on my own and didn't see the potential to be in the industry for the next 15 to 20 years with the same exponential growth that we had before, decided it was time to call, call it quits from there and start something my own. Keye actually started and we can talk about that later, but, uh, it was a more so B2B information services business before. Then, uh, in January of last year, I think, uh, we came across a crossroads where AI was obviously doing rounds. Uh, the contextual ability of AI was massive for this industry. And so we knew that that would change just contextualizing data, understanding what it means, uh, that would change entirely. And so private equity was actually primed for that change. And we're seeing that day in day out where like now even partners, senior partners are taking massive interest in how to go about digital transforming their own funds. And so kind of picked up on that theme. A lot of companies were trying to obviously bring AI to private equity, but saw a massive gap in the industry where a good amount of these companies were heavily focused on trying to build like the next version of chat GPT or trying to build like an advanced version of search and summarization tool, but for finance or private equity. Now as investors ourselves, we realize search and summarization capabilities obviously are great, but they're not a typical workflow for private equity investor. [Desmond Fleming] (10:47 - 10:58) Yeah. And also in a world where now everyone has chat GPT, the pro or maybe the free plans or like the, the Delta between you and chat GPT, like you don't want to just be a rapper. [Rohan Parikh] (10:59 - 12:46) Yeah. And we, I mean, we can't compete with deep diligence or like deep research. And so we came up with the thesis of let's not take a technology and try to push it into an industry. Let's really understand what is the pain point that we're trying to solve. And just because we've gone through that ourselves, my co-founder worked at Vista equity partners, and then before worked at Goldman. So we've kind of had 14 years in this space, in the due diligence space, as well as deal making space. And so we're like, what is the pain point exactly? And the pain point was that 80% of your insights come from your numbers and not from like your SIMs or legal documentation, et cetera. And we did not see a single tool, which was actually focusing on the numbers. And that's also because numbers were the toughest things to do. Like still getting a hundred percent accuracy in numbers in any of the models is you can't say that for us with certainty for us, like it was the pain point, which triggered is like, Hey, we need to solve this pain point. And the pain point was simple. Private equity professionals are stretched thin. They only look at about 10 to 15% of the deals that come onto their platform. And even in those, they cannot go deep inside because there's just so much you can do with the data that you get. As a result, they end up spending time on the wrong deals. And that obviously takes time away from the right ones. And so for us, it was like figuring out what is, what is the toughest things for investors to do? And that's making sense of the numbers. [Desmond Fleming] (12:47 - 13:03) Talk to me a little bit about, you all had this initial thesis, felt like the gap was in quantitative analysis rather than just another search overlay. How did you go from thesis to initial product? Walk me through that process. [Rohan Parikh] (13:03 - 13:32) I mean, we had our own evolution. We're like, I mean, everybody was talking about AI. So we're like, Hey, well, why not try AI? And then we went in and then we had like text to SQL, all of these things to start making, like you could have, you could have AI code stuff, et cetera, but we constantly stumbled upon the same, I would say roadblock is like two things. You cannot guarantee a hundred percent accuracy. [Desmond Fleming] (13:33 - 13:33) Yes. [Rohan Parikh] (13:33 - 13:44) And if it was right, by chance, let's say five times, right. You as a founder could not figure out how the software went from point A to point B. [Desmond Fleming] (13:44 - 13:46) Oh yeah. In your own audibility. [Rohan Parikh] (13:46 - 14:06) In our own audibility. We just don't know. And I think this is very, very interesting because if you ask private equity professionals how they build out their analysis, like the some ifs are so critical for them that they see another formula, they're going to get tripped up. [Desmond Fleming] (14:06 - 14:06) Yeah. [Rohan Parikh] (14:07 - 14:21) And so getting from a raw transaction based data to analysis or a final product, then they need to see the exact same things that they've done. [Desmond Fleming] (14:22 - 14:23) Yes. Yeah. [Rohan Parikh] (14:23 - 14:24) Yeah. Every step. [Desmond Fleming] (14:24 - 14:32) You could obviously calc, you could get to the same output as the some if with a different set of formulas, but if it's a different set of formulas, they're like, no, what did I don't trust it. [Rohan Parikh] (14:32 - 14:35) Yeah. Because now you have to trust and verify. [Desmond Fleming] (14:35 - 14:35) Yeah. [Rohan Parikh] (14:36 - 14:39) And for them to do that as like, okay, this is going to take me time. [Desmond Fleming] (14:39 - 14:39) Yeah. [Rohan Parikh] (14:39 - 14:40) Might as well. [Desmond Fleming] (14:40 - 14:41) I just do it my old way. [Rohan Parikh] (14:41 - 14:45) And so that was very, that was a big insight for us. [Desmond Fleming] (14:45 - 14:45) Yeah. [Rohan Parikh] (14:46 - 15:02) It's not only about accurate results. It's about how you get to that accurate results. And can you, as a founder of a company that's actually giving them a guarantee of accurate results, know how internally is it getting built out? [Desmond Fleming] (15:02 - 15:19) Yeah. It's, it's the stylistic thing, right? Like, are you matching the, are you matching the general style of private equity as an industry? Which you only know if one of two ways, you either were in the industry and know, or you do a lot of extensive research to sit with people and be like, okay, why did you use that formula in that scenario? [Rohan Parikh] (15:20 - 15:20) Yeah. [Desmond Fleming] (15:20 - 15:28) Why did, why did you start it? Uh, why did you start your cohort with month zero versus month one? Like, what, what is that? And it's like, oh, it's just industry standard. [Rohan Parikh] (15:28 - 15:59) Yeah, it is. Yeah. And I think that was very critical for us is we knew that we could not have anything Dez than a hundred percent and complete determinism to, to even get it in the hands of people. And so when we saw that a lot of tools were talking about 99% accuracy, I mean, it sounds great, but when you don't know where that 1% lies. [Desmond Fleming] (15:59 - 15:59) Yeah. [Rohan Parikh] (15:59 - 16:01) Is the big problem. [Desmond Fleming] (16:01 - 16:01) Yeah. [Rohan Parikh] (16:01 - 16:12) Because as I mentioned to the start of the conversation of what is at stake, if that 1% compounds into the final result. [Desmond Fleming] (16:12 - 16:12) Yeah. [Rohan Parikh] (16:13 - 16:13) Like. [Desmond Fleming] (16:14 - 16:15) It changes the vector. [Rohan Parikh] (16:15 - 16:40) It changes the vector completely. Yeah. You are not confident about the numbers. So for us, I think there were two parts of it is obviously AI had massive contextualization ability. But you also needed to provide the system or the infrastructure that AI has built on with the right data and massaging it the right way. [Desmond Fleming] (16:40 - 16:40) Yeah. [Rohan Parikh] (16:40 - 16:42) So they can contextualize the right things. [Desmond Fleming] (16:43 - 17:06) Describe for the audience, what components of Keye are purely deterministic, what components of Keye leverage LLMs and even how do you think about the platform? Like when, if you were onboarding a new product manager, how would you describe to them the core components of the product or even a new user that might actually be a better framing? [Rohan Parikh] (17:07 - 19:23) Yeah. So I think for us, it's, it came down to, let's actually take the entire process of an investor. When investors get data, they will clean or scrub the data. They will then create multiple different views or cuts of that data. And then finally take those, combine those into building your cohorts or retentions, price volumes, et cetera. And then extract deep level insights from it. Now, if you look at this entire cycle, you'll probably be like you'll yourself identify where is there no scope of any error whatsoever. So for example, the first one is when you're cleaning and scrubbing the data, we use AI, we use all software engineering, but it's much more about trying to give you ability of like, okay, your client IDs are different or your numbers are different. That doesn't have to be, I would say a hundred percent accurate, but because you're giving, you're having the user verify if this needs to be cleaned or not. We use AI for that. Um, and then when you pass that across to building out your numbers, that's where we tried out like 12 to 18 months back, building out things with LLMs and then realized there's no determinism, no 100% accuracy, and we could not, we could not play around with that number. So decided that we're going to make that completely deterministic. So none of that is, um, is driven majorly by AI. It's software engineering. And that's because we have a really deep understanding of the nuances of how things are built out. And so, as I mentioned, it's not rocket science, there's a process to it. And so for us, it was like, if we can put in all of our knowledge and configure this in different ways where it becomes customizable for funds, like we're actually building a tool, which if AI wants to get there and if AI gets to a hundred percent accuracy, it's going to be at the same level as us. [Desmond Fleming] (19:23 - 20:18) Yeah. But I mean, I've seen the product a few times. I'm still even shocked that you all are able to do what you do because the data inputs can be so complex, so varied, so messy. And if I'm hearing you correctly, part of the sauce, or at least part of two parts of what you all have innovated on is you'll use LLMs and maybe some other, uh, uh, computer vision, uh, machine learning models to transform unstructured data, to get it into a structured format that's, that your system understands. But then even honestly, the hard part is your system having all the context of here's how you can cut these different points of these different data points. Like that, that is, I know it's a, it's a lot. Cause as someone who manually does, takes unstructured data and then structured it, like you never know what you're going to get. [Rohan Parikh] (20:19 - 20:33) Yeah. And the thing is like, I think it all comes down to, we've just done this over and over again and understand, like, you might think like, we get this question a lot of times is like, oh, but how do you do things for different industries? [Desmond Fleming] (20:33 - 20:34) Yeah. [Rohan Parikh] (20:34 - 20:44) Like every industry has its own metrics, et cetera. But actually, if you sit down, there's a relatively, there's a relatively few and I can literally count how many. [Desmond Fleming] (20:44 - 20:51) Yeah. It's just the investors don't live, like the enterprise investors know all their metrics, the consumer investors know all their metrics. [Rohan Parikh] (20:52 - 20:52) Yeah. [Desmond Fleming] (20:52 - 20:53) They don't always know each other's metrics. [Rohan Parikh] (20:53 - 21:01) At the end of it, the base is numbers. You're just taking those numbers and placing it in different ways. [Desmond Fleming] (21:01 - 21:01) Yeah. [Rohan Parikh] (21:01 - 21:04) It's a derivative of that product. The numbers don't change. [Desmond Fleming] (21:05 - 21:05) Yeah. [Rohan Parikh] (21:05 - 21:12) And so for us, it was just figuring out, like, do we know all the nuances of like, oh, retention could be calculated eight ways. [Desmond Fleming] (21:12 - 21:13) Yeah. [Rohan Parikh] (21:13 - 21:17) Do, do we know all the nuances in how data is massaged? [Desmond Fleming] (21:17 - 21:18) Yeah. [Rohan Parikh] (21:18 - 21:46) How data is created, how data is manufactured together. Yeah. And we realized we like knew 90% of it. Yeah. Um, and like, we're not saying like, Hey, we will like just automate everything. We're saying like, we'll get you 95% there. And so either as a fund, what is your thesis? You will use that and put it into your thesis and say whether this is a good decision or not. [Desmond Fleming] (21:47 - 21:47) Yeah. [Rohan Parikh] (21:47 - 21:57) Or at the same time, like if you have a very niche industry, we're not trying to be like, Hey, we will try and get you to right at the end and you don't have to do anything, right? [Desmond Fleming] (21:57 - 22:12) Yeah. There might be some random, going to your point on niche industries, there might be some like random, uh, benchmark metric that's used for like nuclear energy consumption. And you're like, look, we don't know, like, okay. Yeah. I mean, yeah. [Rohan Parikh] (22:12 - 22:20) Okay. You, you want to, you want to know and cut data for your nuclear facility in 10, 10, 10 ways we'll do that. [Desmond Fleming] (22:21 - 22:21) Yeah. [Rohan Parikh] (22:21 - 22:26) Now how you use that information, um, as an investor is that's also completely dependent on you. [Desmond Fleming] (22:27 - 23:40) Yeah. How much automation you think is going to occur within the private equity industry in general? Cause one and correct me anywhere you disagree, but one reductive way of thinking about Keye is you are taking a component of the associate or analyst kind of analytical work. And you're saying, Hey, we can do the four hours of analysis and data cleaning that it would take you. We'll do those four hours in 10 minutes or whatever it is. And there's still obviously a lot of other work, a lot of additional work that goes into seeing a transaction through end to end, but there's certainly a strain within Silicon Valley and VCs in general, that they believe that platforms like yours are not going to stop at just the four hours of work that an analyst or an associate does. They're for this industry. There's, let's argue in a sec, they're going to go after all 40 of the hours. And I'm, I'm just curious as to how much do you agree with that? Do you think that's where you all with Keye want to go? Do you think even the buyers even care about it or are thinking about it that way? I'm just, would love to get your take. [Rohan Parikh] (23:40 - 23:50) Yeah. I think that's a really interesting question. Um, how we see it is our main incentive is not to save hours. [Desmond Fleming] (23:50 - 23:50) Yeah. [Rohan Parikh] (23:50 - 25:33) I mean, we do end up saving a lot of time. Our main incentive is just the density of the data is to be able to give you information that you otherwise would have overlooked and like that happened so many times, because when you have such dense data rooms, et cetera, is this impossible for you to cover every nook and corner in there, every nook and corner has some kind of information related to it. And sometimes your alpha could come from there. And we've seen so many times where like as individuals, capacity, limitation, resourcing, limitation, human resources, et cetera, people, when they're staffed on like five deals, just end up doing like guessing what is the 10% most important things let's go through that and take a decision. Now we're actually trying to change the dynamic, not only in times of, in terms of saving time, or also in terms of being extremely exhaustive in your full analysis, looking at every nook and corner, which might, if you correlate things, might inform you about a deal that you did not have information before, which actually could be your alpha for that. And so we're trying to, trying to go to a path of let's actually give you information that you might have overlooked because we are able to do that a lot quicker and actually go through every permutation combination. And coming back to your question, 40 hours can be saved, but like what private equity investors would absolutely love is when you show them something that they didn't look at a deal before. [Desmond Fleming] (25:33 - 25:34) That helps them win. [Rohan Parikh] (25:35 - 25:38) The system was able to identify and that's the North Star for us. [Desmond Fleming] (25:39 - 26:00) Yeah. Yeah. It's in other words, saying who cares about the 200K. I don't want to be flippant, but who cares about the 200K labor savings? If you can help me bid, you know, a half turn higher on a deal I really want that turns into the three X, you know, fund returner for me, I'll pay for that. You know, day in and day out. [Rohan Parikh] (26:00 - 26:05) Yeah. Because still the point of time where you're actually bidding, you've already spent millions of dollars on that. [Desmond Fleming] (26:06 - 26:06) Yeah. [Rohan Parikh] (26:06 - 26:11) And so like, if your bid doesn't go through, yeah, it's painful. It's really painful. [Desmond Fleming] (26:12 - 26:12) Yeah. [Rohan Parikh] (26:12 - 26:16) And you've spent at least a few months, if not more. [Desmond Fleming] (26:16 - 26:16) Yeah. [Rohan Parikh] (26:16 - 26:17) I'll work on that. [Desmond Fleming] (26:17 - 27:07) What are you hearing from the private equity industry in general? Like, you know, there's all this interest in AI, you know, there's probably at this point, billions of weekly active users, if not monthly active users of chat GPT, everyone wants to future-proof their business. Um, but what is on the ground? Like what are private equity buyers, the MDs, the people who run these firms, what tech are they actually looking for? And then the second part of that question is, without naming names, can you walk us through how you got your first customer? Like that is one of the most difficult parts of getting the momentum of a startup is having your first customer that you win, go end to end, they love you, they're working with you, and you know, that helps you get your second, which helps you get your third, which helps you get the fourth, fifth, sixth, whatever. Um, so curious about both of those. [Rohan Parikh] (27:08 - 29:07) In terms of, I think the first question that you asked is how are people using it? What were partners looking for? Um, I think it comes from like the very first thing that they started is their portfolio companies. Like it was a change that happened across and especially private equity, because you're not looking at extremely tech savvy companies or like high growth companies, which like we see is usually capitalized on. And so a lot of these companies were on a brink of like, if you don't change, you're going to lose, uh, the battle over here. And so the first things that they, that we saw was portfolio kind of optimization in terms of technology. Um, and then in terms of due diligence, if you're asking, I think there's still a big belief of going bottoms up, uh, because a lot of the groundwork in terms of like building out your numbers, cleaning your data, et cetera, which have set processes is done by the associates or the senior associates, you still see that on the VP principle or the partner level is quite more relationship based. And I still do feel like, yes, AI can definitely help because it makes you a lot more informed. We just had a client, uh, recently who, who said that the ability to speed up things made their VPs when they were going for a management meeting in crunch time, a lot more informed about the deal. Yeah. They were able to actually question a lot more and get a lot more insight into the company, which kind of trickDez through. So I think it will capture the entire stack, investor stack soon, but it's coming from a bottom up approach where like, Hey, let's make sure the associates, senior associates. [Desmond Fleming] (29:08 - 29:16) But bottom up in the sense that associates are the ones finding you and being like, we want to use Keye or bottom up in the workflow. [Rohan Parikh] (29:16 - 29:17) In the work. [Desmond Fleming] (29:17 - 29:17) Yeah, yeah, yeah. [Rohan Parikh] (29:17 - 29:57) More in the workflow. But I do feel like associates are also extremely excited about this because at the end, if you see like crunching your data and I said, the processes are the same, you're pretty much using the same formulas, looking at the same data, the same way, et cetera. When I was in banking, I didn't enjoy building out spreadsheets. What I enjoyed was, Hey, once I build them out, like what kind of insights can I actually get from it to inform my decisions and tell my VP? And so that's like the, that's what everybody's aiming for. And if you can get that quicker, yeah, I think that that's a lot more useful. [Desmond Fleming] (29:57 - 30:26) Investors want to think about the various what-if scenarios. They don't want to have to do the brain damage to think about stitching all the data to get this amount, even though that's a component of the job. And like it compounds, but okay, cool. So, it's a, it's a bottoms up motion, both from a workflow perspective. What was the story behind your first customer? [Rohan Parikh] (30:27 - 30:52) Yeah, I think the story behind the first customer was two things. I think we identified a gap in the market, which was like, Hey, we literally asked the, the first funds that we worked with is like, do you even look at legal contracts before you understand? Like even in venture capital, you're not going to be like, Hey, met you for the first time, can I see your legal documents? [Desmond Fleming] (30:53 - 30:54) No, no, that's crazy. [Rohan Parikh] (30:54 - 31:09) You're going to be like, okay, how are you performing? What's your ARR? Like how much are you retaining, et cetera. And then at the end is when you like look into your legal documents, et cetera. And so we were quite puzzled of like what was happening out there and like what was given importance. [Desmond Fleming] (31:09 - 31:09) Yeah. [Rohan Parikh] (31:10 - 31:36) And so for us, like the entire mode was like, Hey, nobody's looking at numbers. This is because it's the most, like this is the hardest thing to do. And so when we went with that approach, and that also comes from the fact that we've been in the industry. And when you combine those two things and you're pitching, I think our early, the most early, I would say users or adopters, like clearly understood the, the problem statement we're going after. [Desmond Fleming] (31:37 - 31:54) Yeah. And so is it more of a, okay, in theory, you can help me speed up my diligence, the quantitative analysis, but I want you to show me that you can do it rather than just tell me, so did it flip to like talk to a few people and then it's get into the pilot stage or like, how did that go? [Rohan Parikh] (31:54 - 32:31) Yeah. I mean, we had design partners the same way that everybody had, and we worked with them week over week, made progress week over week, realized text to SQL doesn't work, week over week realized LLMs every now and then give you wrong answers, and that was the learning process for us. And so for us, the path to determinism wasn't just, we believe in it. It was more of like, we believe in it and we cannot get to determinism with the current set of tools. [Desmond Fleming] (32:31 - 32:52) Yeah. So anytime there's a, you know, GPT 5, 6, whenever it comes out, anytime there's a new model release or an update to the models, do you guys like frantically check to see now, has it flipped into these models have reached the parity for deterministic systems? I'm just curious. I have no idea. [Rohan Parikh] (32:53 - 33:33) No, I think it's great because as soon as that happens, like we have the knowledge to actually plug and play with that into our entire ecosystem. And so we've built that out. It's just like, I get really excited to see a new model. And it's like, how close does it get? How close does it get? How close does it get? And can we plug and play it into our workflow, which requires determinism? But like, unfortunately that's not been the case so far, but we will definitely be there. And so we're waiting for that moment. So we don't have to sit down and like recreate everything. You just swap it in and out. [Desmond Fleming] (33:34 - 34:28) Yeah. Um, talk to me a little bit about, you know, what's your ambition? What do you want to, what do you want to do with Keye over time? When you think about the success function 10 years from now, what does that look like for you? One, um, headwind or negative commentary for, uh, uh, building in financial services is that there's not a lot of like workflow platforms that are at scale. There's a lot of data platforms that are at scale SMP, to a certain extent, you could frame Aladdin at BlackRock, uh, FactSet, et cetera, but in terms of like core workflow, a lot of these systems get tucked into like an SSNC where they own, uh, interlinks, right. Which is Dez workflow, more like data hosting, uh, or, or the same thing with like deal cloud being owned by InTap. But just curious about how you think about the 10 year version and where you see the opportunity. [Rohan Parikh] (34:28 - 37:10) We are not an AI tool for finance. We do not believe there can exist one AI tool for finance. We are focused on private equity and we feel like there's just so much end to end to your question of like processes, like everything is a process, whether you're going from, you know, collecting data, sourcing companies, to due diligencing, and then even going to like portfolio, like creating a hundred day value creation plan, et cetera. If you look at that, like that itself is massive and like by itself starts, like the amount that is spent just solely on due diligence is like upwards of $50 billion a year. And so for us, like, I feel like if we stay true to our motive of like trying to help investors go from point A, which is getting some data to point B, which is informing them, not only about the due diligence or diligencing a company, but eventually portfolio, being able to use Keye through your entire end to end when you acquire a company to when you actually sell a company in between you monitoring also, and there's just like so much you can do in this space. And so for us, we want to be a big part of the tech stack change in private equity. And also to your question of, you know, there's a lot of platforms out there like interlinks, et cetera. There isn't one platform which has like a system of record, which actually can go in there. It's just so much, it's like, I'm very shocked by like how hedge funds like Citadel, Jane street, et cetera, have been able to really mine their data and like go to a next level of like analysis. Like they will, they'll use like financial engineering, et cetera, and really like advanced techniques. It's like, then you see private side, obviously the data is not that available, but it still is like highly dependent on numbers. I do feel like that will change in the future. But I do feel like all of these analyses or just like launching data will become extremely table stakes. [Desmond Fleming] (37:10 - 37:45) Yeah. There's like no central system of record or data store for the company to understand. Like, I doubt that there's a single private equity fund that could write a query and ask. Out of all the companies that we've looked at over the past 10 years that were over a hundred million of revenue, what were the leading indicators of success for them? All that information is stored in like, it's like, it's like tribal knowledge for any, any of the firms and it's not all captured. [Rohan Parikh] (37:46 - 38:36) You can even go dig deeper. And so this is where I think where we also had a thesis of like the data that's provided to you. I mean, that's obviously the most critical, but you need to create like an alternate source over and above that for systems to understand what their data means. And we did not see like a single platform creating that. And that's where like converting your data into like your retention numbers, understanding what is the price volume. Effect, et cetera. Like that's what we're creating. We're creating that mid-layer, which has so much of information, which otherwise is not picked up by itself. [Desmond Fleming] (38:36 - 38:49) What would be the advice that you give to another first-time founder, whether they want to build an AI platform or whether they want to build in financial services writ large? [Rohan Parikh] (38:49 - 39:44) Start with the problem. Try not to look at the technology. Start with the problem. Nothing more than if you start with the problem and then figure out what technology works, you're literally starting off with actually helping the users versus taking a technology and trying to drill that in. And I feel like that, and that's where like, I feel like that adds up to the old adage of like focus on a problem. Make sure you crush that. And then expand. And for us, it was just like numbers in private equity. Again, not trying to boil the ocean over here, not an AI for financial services by any means. And so try to solve problem first. [Desmond Fleming] (39:45 - 40:05) I love it. And then my last, very last question for you is what is something that this audience can do to help you? One of the things that I love about the tech ecosystem is that it's very open, very willing to kind of give back, very willing to pay things forward. So if you were to ask to kind of the people that watch the show, like what would it be? [Rohan Parikh] (40:06 - 40:39) I think there are amazing founders building amazing things. Like every day, like I see some companies where I'm like, wow, okay. Did not even realize this could exist. Which some of the reaction that we get from our users is like, wait, did not realize we could do something like this with numbers. And I feel like, yeah, I mean, would love for people who are like in this industry to reach out, would love to like riff off things. There's just so much to do in this industry. Like you, the entire ecosystem is going to change. [Desmond Fleming] (40:39 - 40:40) Yes. Yeah. [Rohan Parikh] (40:40 - 40:41) And it is. [Desmond Fleming] (40:41 - 40:41) Yeah. [Rohan Parikh] (40:41 - 40:56) And so it's like really exciting to be a part of it and to work together to kind of push that because it's not a battle between companies, it's a battle between an ideology which existed before. Yeah. And now that's going to change completely. [Desmond Fleming] (40:56 - 40:56) Yeah. [Rohan Parikh] (40:57 - 41:03) Given that technology is going to start being front and center for like an archaic industry like private equity. [Desmond Fleming] (41:03 - 41:07) Yeah. Which is wild given the amount of control private equity. [Rohan Parikh] (41:07 - 41:08) Exactly. [Desmond Fleming] (41:09 - 41:20) Exactly. Well, Rohan, thank you. Thank you so much for coming on. This was wonderful to talk about. A lot of ideas stirring through my head, a lot of stuff for me to think about. Uh, so appreciate you so much. [Rohan Parikh] (41:20 - 41:25) No, thank you so much for all the questions and, uh, yeah. Thanks.