Hi, everyone. I'm Jill Hall, Head of U.S. Small and Mid-Cap Strategy here at BofA Global Research. Just wanted to welcome everyone to day one of our virtual mid-cap event. Excited to hear from corporates across the small and mid-cap space, across sectors. Great breadth of coverage here by our analysts. They cover about 1,000 small and mid-caps in the U.S., so I'm happy to bring nearly 20 companies today. Feel free to reach out to me if I can help with the schedule or getting you signed up for any additional sessions or if you're interested in broader small or mid-cap research or some of our compilation emails, we send out on the fundamental side. I'd like to pass it over to Alec for this session to introduce the company. Perfect. Thanks, Jill. As Jill said, my name's Alec Stranahan. I'm Senior Analyst covering biotech here at BofA. I cover around 30 stocks, ranging from $40 billion all the way down to $400 million in market cap. One of the more interesting names, I would say, is Recursion, and it's my pleasure to be joined by Ben Taylor, who is Chief Financial Officer and President of Recursion U.K. I would say speaks just about as well about the science and AI aspects of Recursion as he does about the finance side. Ben, really happy to have you with us. Thanks, Alec. I always appreciate the intro and the conversation. Yeah. Great. As Jill said, I'm going to run through some questions here with Ben, in a fireside format, but hope to keep the conversation topical for those dialed in. If you do have a question, please utilize the raise hand feature via Zoom, or you can email me separately, and we'll get your questions asked. Ben, maybe just to tee things up for investors maybe newer to the Recursion story, and maybe AI drug discovery as a whole, what is that? Why is AI needed in the drug discovery process, and how is Recursion maybe blazing the trail here? Yeah. I think it's really good to level set because AI, especially, currently is waved around like it's a magic wand, which it absolutely isn't. The way that we really think about it is, it's a better analytical system. So, it's more similar to the evolution of starting to use computers or starting to use spreadsheets, how those things have changed the way that we do business and look at data analysis. I think AI is another step up in that. So, where we have been able to apply it to drug discovery and what makes us different, is it allows you to both produce and analyze data in a different way than you ever could before, and also do it in a more multi-parameter way than you've ever been able to do it before. By putting together the data with modeling systems, with the ability to compute, we can really get to a different outcome than was historically possible. So the foundation of the company is actually around changing the probability of success and really trying to unlock new parts of biology and chemistry rather than efficiency. But we have also been able to do it much more efficiently and we've published some of the statistics on that showing pretty dramatic reductions in the time and cost to be able to get to those differentiated outcomes. So, there's a lot of different pieces at play. I think the nice thing about being in our shoes is we're actually at the point where we have multiple clinical programs. We have partnerships that have been running for multiple years. So you don't actually have to understand all of the AI, just like people didn't understand drug discovery and biotech for many years. You just have to understand the output of it. So that's really what we're focused on. Okay. Maybe along those lines, and this is a question I get asked a lot, which is for those paying attention, ChatGPT, you could see it coming, but if you were most of the population, it was just one day we didn't have LLMs, and the next day we did, and the world feels like it's changed. Is there a ChatGPT moment in drug discovery? Is there a moment where the entire industry, you think will just one day be AI, in terms of the back end on the drug discovery side? Or is it maybe a little bit different of a situation? No, I think it's absolutely the same, but rather than it being the entire world sort of figuring it out at once, I think what you've seen is more of layers. If I go back to the days when we were a private company, literally no one, the large pharmas, the investor base, no one was really using AI to evaluate drug discovery or try and build some of the models that we're doing now. Now you look at pharma, there have been multiple large pharma who have announced billion-ish dollar investments towards building out AI and investing in that. That's because we're actually getting better results over and over again. It's repeatable. It's not just one-off. We're doing things in a better way. We're more efficient and achieving things. All of our partner milestones, we've achieved well over a dozen partner milestones. All of those milestones where they paid us millions of dollars were things that they couldn't do internally. It was seeing us demonstrate you can actually get to that different outcome. I think that there's still another level of having it be more generally accepted. So inside of the industry, people are absolutely using it. It's funny, some of the biotech companies that are coming up now, they don't talk as much about the AI because it's such a hype-y thing to talk about in biotech. But the outputs that they're doing, if you look at Gravitas, there's a lot of AI that was involved in how they achieved that outcome, and it was a great outcome. So, I think on the industry side, it's already occurred. I think on the investor side, there's a lot of speculation about when and how and who. We haven't quite got to that moment yet, but hopefully soon. Really, on the investor side, I think it's a matter of demonstrating that the products really make a difference, and we're very cuspy on that, it feels like. Yeah. I agree. Are there maybe one or two examples that you think are proof points for how the AI-driven approach or what Recursion is doing specifically can produce better medicines? Or is it really kind of clinic, or is it maybe just all wrapped up into that? Well, it's funny. Hopefully, you'll respect this. Coming from a data-driven company, we don't like anecdotal examples. We really look for an accumulation of evidence that something is changing. How do we know if our biology AI is working? Being able to target new ideas that weren't in the literature, they weren't commonly known. We've seen multiple examples of that, like two clinical examples for us with REC-4881, which we'll talk about later, and REC-1245, so in FAP, with MEK1/2, and then with RBM39 as another target for DDR in other areas. Those were novel biological insights. What we just saw with the Roche collaboration milestone is not only had Roche opted in on the biology maps that we created around neuroscience, so neuronal cells and microglial cells, but now we've started to take completely novel programs from those maps and transition them into the design phase. That's actually saying, "This is something that wasn't in existence as a neuroscience target or known biology, and now we're transitioning it into something that can be a drug because we've done the target validation work on it and experimentally validated." I think those are three points that all point in the same direction of finding new connections in biology that didn't exist before. One of the things that always blows my mind, if you look at all of the drugs that the pharmaceutical and biotech industry have created over their entire lifespan, with all of the good people and all of the money that we put in, the approved drugs only cover about 3.5% of the genome. If you add on all of the drugs that are currently in development, we think of this massive pipeline of drugs that are coming through and all the innovation that's going on, you're still only covering about 13% of the genome. The reality is we keep digging in the same holes. What we want to do is create new data, look at it in new ways, so that we can actually break outside of those holes that we've been digging over and over again and really find new paths. That's why the biology side is really exciting to us. I think we can also talk about the chemistry side. We've had many milestones with Sanofi and other partners advancing through, as well as being able to demonstrate how our chemistry is actually achieving things that other chemistries have not. I think we'll hopefully be turning over cards on the clinical side on that soon. It's been exciting to watch both of those come together. Yeah. No, that's a great way to sort of sum things up, I think, for the state of the industry. We hear more and more the actual generative AI models are not necessarily the big differentiator. There's still maybe an edge to be gained on compute, and you guys have your supercomputer in-house that you've already built out. But it's really more and more, it feels like the data side, and then how you pump that through the funnel internally and then spin the flywheel based on that data and generate new data from quality foundational data sets. That seems how you build a good platform. Maybe you can just talk about that piece, sort of how Recursion's built that from day one, and sort of where you see that unique data set being leveraged, either internally or through your partnerships. Yeah. Well, it's a really great point because if you think about why haven't we gone into more of that 87% where there's literally not even a drug candidate out there, much less something approved. A big part of that is because you can only go after what you have some sort of diagnostic or assay system or some way of understanding what good looks like. This is where coming in and being able to create and look at data in new ways makes such a difference. Because if you're just focused on the algorithm and you're just trying to look at the data that's already in existence, you're going to get a lot of the same answers. You're certainly not going to have differentiation from other people who are doing the same work. You need to really be creating novel data, to be able to look at it in different ways and analyze it. Most of the data that's created across the biopharma industry is not good for machine learning because it's been created in a format that's usually very local. It's usually done for a single project. The annotations are quite different from project to project, and you can't use it as fluidly. That's where starting from the beginning, more than a decade ago, we have been creating novel data in sort of a machine learning annotated format that we can then consolidate down. That's where we get to over 50 petabytes of having that data in existence that's very differentiated from what is in existence in other places. If Najat was on, Najat, our CEO, who used to head up AI at J&J, looked at it and said, "In all of the data inside of J&J, probably only about 30% of it was really something that could be used for machine learning, and even that had a lot of data wrangling." This is where trying to build up unique ways of looking at the data can get you down into really new spaces. I think that's where a lot of the excitement on our side, as well as across the industry is. How can I analyze this biology signal in a different way? Whether it's phenotypically or applying overlaying cellular imaging with transcriptomics and proteomics and other sort of sets, you get this much richer set of data. I think, last point, this is also a place where we've really felt the differentiation value of just having a lot of the breadth that we have because we started in the phenotypic cellular imaging. That was the original base, and it is sort of a new language for being able to look at biology, which is incredibly powerful. But then being able to overlay that with orthogonal data sets, like the transcriptomics I was talking about, or proteomics, or looking at real-world patient data and figuring out genomic signatures. What that allows you to do is actually synthetically or virtually compare the data output and the analysis. You can actually dramatically improve your ability to find real signals because no matter what source you are using, there is always going to be a lot of noise in it. We published a paper in Nature Biotechnology just recently that showed we were able to actually outperform models that were even up to 100 times the data set size that we were using. It was because our data was better annotated, and we were able to use multimodal sources to basically enhance the signal that we were seeing and make far better predictions on it. That is sort of how you get to more composite biology and systems biology and where we hope to go in the future. Great. Yeah. The saying, "Garbage in, Garbage out." That still holds true, maybe even more so today. Maybe to ground then some of the conversation that we have been having so far, you mentioned your Roche Genentech partnership. You recently selected the first neuroscience target here from your collaboration for further drug development. I guess what did that target need to demonstrate for you and Roche to be convinced that this is something worthwhile taking forward and why should investors maybe view this decision as important validation for the platform? Yeah, absolutely. Just to take the step back on the partnership as a whole. Roche had originally given us $150 million to go out and basically build maps. Most of that was to build maps in two different neurology cell lines, so neuronal cells and microglial cells. Those maps were basically whole genome knockouts and other perturbations of those cell lines, and then looking at how did it morphologically change, and then we overlay on some of the transcriptomics signals to understand what might be happening within those cells. Then they optioned in both of those maps. They do not get any of the data, but they are able to query the maps, basically. That was two $30 million payments to be able to do that. That is already all happened. Now, basically, the milestone that we just got is from those queries, we get a list of potential targets. From those targets, we selected a few that we then wanted to go through and do validation on. Some of that validation is still AI-based and really how we selected the targets that we wanted to do. Most of it, we went to experimental systems, and we said, "Let's test this in a disease model we know. Let's look for how this is disease modifying to different cells of interest or diseases of interest." In a very classical way, demonstrate that this novel virtual finding is having real experimental impact. We went through a process with Roche. They are obviously a world leader in neuroscience, and that was what triggered the validation. Now we're taking that target, and Recursion is designing the molecule to be able to drug it. Okay. Got it. I think one point that we shouldn't gloss over is that this was a target in neuroscience that is novel, right? There's been decades and decades of research in CNS diseases and your platform and your neural maps were able to uncover something new. So talking about digging holes in fresh soil. Yeah, absolutely. Well, by definition, everything that we do out of that partnership, this isn't going to be something that you can find in the literature or that someone's doing an alternative program for. This is wholly new work in neuroscience, so really exciting to see that progress, and hopefully a lot more to come. Okay. Maybe you could just remind us of the structure of the partnership with Roche. How many candidates could you bring forward? What are the economics around bringing those forward? Then maybe we can talk about how that structure is similar or different from your Sanofi partnership as well. Yeah, absolutely. The Roche partnership, originally a 10-year term, which we can extend, and we are about halfway through it right now. I almost feel foolish stating the number, but it is up to 40 design programs that we can advance. Actually, part of the rationale for the Recursion-Exscientia merger, which is now almost two years ago, was to bring the design capabilities in-house with Recursion's target ID. It is actually really exciting that Roche wants Recursion to do all of the chemistry work and drug design work because, prior to the merger, I do not know if that would have been true because we really have built out the capabilities. It is great to see that coming together. Now what we hope to do is basically create a pipeline of additional targets coming out in neuroscience. We also have work ongoing in some GI oncology as well, and to be advancing that as a long-term partnership with Roche. Sanofi is a little bit different in that there was not a target ID portion to it. This was a legacy Exscientia deal, so it was really focused on, "Hey, there is a target that we mutually are interested in. No one has ever been able to drug it before. Can we advance a candidate in it?" Now we have already seen five programs hit their first discovery milestone there, and so the next milestone for all of those would be basically opting in for Sanofi to take it forward into clinical trials. That is really exciting because that not only marks that we will be advancing, again, really exciting new potential blockbuster programs, but also that it ends our operational obligations. All of the payments from development candidate onwards are basically profit for us. If you look at the Sanofi collaboration, each program has the potential for up to $343 million in milestones, $193 million of that is pre-commercial. This is not some massively back-end loaded deal. Our royalties on it are average in the low double digits. We actually capture a pretty substantial part of the NPV. Both of those programs and the Roche design elements are actually similar to Sanofi, not quite as high on the economics, but close. For both of those, they are really designed to be more of collaborations, but ones where we are always at breakeven or profit on a direct cost basis. Sanofi and Roche and our other partners pay us ahead of time for our expenses. It is a really capital efficient way for us to grow value. Yeah. That was an important point that I think you also mentioned on your 1Q call about the cost to service these partnerships, and that is a question I have gotten. So, it is good that they are designed to not be a burden on that. Yeah. We have seen, in the space, a lot of different approaches to monetizing these AI drug discovery platforms. You have got the Schrödinger in that world that are more like a SaaS type- Yeah Revenue model, and then you have got Insilico, which are maybe kind of a mix. Then you guys are more of a hands-on, let us do interesting science together- Yeah And leverage the platform to push those forward. It's a little bit more hands-on, but you also get larger chunkier deals out of that as well. Maybe you could just talk about the philosophy around the partnership model and how you balance that with in-house development. Well, and it's interesting. If you go back to our original mandate as a company, there was really two parts to it. One was how do you change the probability of success using technology, right? We're in a 95% failure environment. No one's making data-informed decisions, not because they don't want to, but because they can't. The data's not good enough, the models aren't good enough. You can't make good predictions. That's how you get to a 95% failure environment. So please improve on that. The other part of it was how do you make this into an actual business model rather than just a binary risk bet? That's where our ability to do things at scale more efficiently really comes into play. You can almost think of the partnership business as an outgrowth of that. From early days, we decided having our own therapeutics was really important because it allowed us to demonstrate that the platform was working and also we were creating a massive amount of value, so being able to capture it as we get into those points. That's where all of the upcoming clinical data is so exciting because those are obviously massive potential transitional points for us. The partnership business, though, is a beautiful part of being able to fund the company, build the platform, and grow the long-term NPV really well. It makes sense because we do do things at, or we can do things at scale. We would actually be leaving some of our capability dormant if we didn't have the partnership build. So the fact that we can get paid early on, use that to actually do a lot of applied development. One thing that most people don't know, about 65% plus of our budget is actually applied. Even when I'm talking about platform and technology development, we're doing it on real programs. That's part of our edge. We know if our models work, because if they don't, the drug doesn't get made, right? Or the biology doesn't work out. The partnerships is a great applied platform for us, where we build out our platform, test our models, and be able to add that in as a part of the overall product engine. We've always loved it. We don't want to become a service company. That's a different set of economics. It's a different business model. There's a lot of infrastructure. In fact, how we run our partnerships is basically exactly how we run our internal programs. We just have a partner that we're strategically working with and talking about what good looks like. That's where we differentiate in our model from the more service-oriented side. Yeah. I'd say your in-house pipeline is also a differentiator for you guys, and you're doing quite a lot across oncology and I&I, other areas. Maybe we can talk maybe for the next 10 minutes or so on the internal pipeline. Yeah. Maybe starting with FAP, Ben, just because that's lead asset, you can- Absolutely bounce around, but it's the furthest in development, and we've got maybe more clarity here. Maybe talk about that program, sort of the origin story, and then the disease, if people aren't familiar. Yeah, absolutely. FAP, if you are not familiar, just a quick background. More than 50,000 patients, U.S. and EU5. That is probably under-reported because that number is about 70% hereditary, but it is possible to get this through somatic mutations as well. A number of those patients may just get caught at the colorectal cancer state rather than when they actually had the FAP. But 50,000, very large for an orphan indication, starts with an APC mutation that basically leads to chronic cancerous or malignant, however you want to say it, polyp growth. None of those polyps that are growing are benign. They will develop into cancer eventually. This typically starts in the colon, but it actually spreads throughout the GI system, so down into the rectum and up. It can even reach up into the stomach. As this patient progresses through decades, they will typically have a colectomy or a colectomy in their mid-20s, and then will be going through basically surgeries, excisions to be removing those polyps throughout their entire life. They can come in and see their doctors several times a year, if it is a serious case, to be investigating how the polyps are growing and have them removed. There is, on average, about 10 major surgeries for these patients throughout their lifetime, and about 70 treatments with excisions. It is just a massive surgical burden and quality of life burden on these patients. What we were able to demonstrate is that we were able to bring down within three months, a little over a 40% reduction in the polyp burden, and so that is a combination of size and number of polyps. Some of these patients can have hundreds or even thousands of polyps spread throughout their GI tract. That is a massive change in a short period of time. What was also really exciting, we are the first drug to ever show that you could take the patient off drug and maintain that response. We actually had a slight deepening of response in the data over the three months that the patients were off drug. But that is very exciting. More to come on that. We are currently in discussions with the FDA on the pivotal trial design, and we will give an update on that later on this year. We are also presenting at a conference most people probably have never heard of, but it is the one where all of the docs that care about FAP go to, and we are presenting at the presidential plenary with the data. Can give you some more detail on why they are so excited about it. I know you are going to ask about FDA trial design because everyone does. The short story is going to be we cannot get in front of the FDA, and we are going to let those talk through. But we know from the existing trials that are out there, which is one, there is a baseline that would be acceptable that we can move forward with. We think even if we do not change the design endpoints at all from that, we would still be able to drive better enrollment through a ClinTech platform, which we can talk about as well. Okay. Just to summarize, it is a large indication, no approved medicines. It could result in cancer if it is not treated, and even if it is treated with repeat surgery, oftentimes patients still get cancer. If you have thousands of polyps, how are you going to surgically excise those, right? It really makes sense for a drug to come in with a systemic mechanism of action to come in and reduce those. You have shown, I think 43% after 12 weeks of treatment, and then 53% speaking to that deepening. I guess thinking about the conference presentation on November 2nd, what do you think we will learn there in terms of is it longer follow-up, maybe individual patient data? It sounds like a great platform for you guys to garner excitement for enrolling a potentially registrational study, right? Activating- Yeah site with new investigators. Yeah. A couple of different pieces. At that conference, we will definitely be providing longer follow-up, which is exciting as well. The first cut we did was three months of treatment, and so being able to look at these patients over the three months treatment on and three months off. Most of the drugs that have been attempted in this area, and there is not a lot, were over a 12-month period and showed much less of a response than we did. Being able to talk about the durability of it, I think is important. Also with looking at different ways to be able to manage some of the known side effects of MEK1/2 inhibitors, which appear to be very manageable. We will give you additional data on that at the upcoming conference. A really important point, we have effect both in the upper and lower GI. The one other drug that is currently, it is called eRapa, it is basically an encapsulated rapamycin, showed a little under a 20% response after 12 months. Importantly, it was all in the lower GI, and there was no effect on the upper GI. The upper GI is actually where you get a lot of the more serious polyps as the disease progresses because you cannot do that with a normal colonoscopy. You actually have to go through a more invasive endoscopic procedure to be able to monitor and excise the polyps in there. That is also what can end up leading to things like a Whipple procedure, where they are going in and obviously taking out a really damaging amount of your internal organs. We were really excited, and we will have more detail on it to show a very similar response in both upper and lower GI. This is based on some of the PK properties that we had selected the drug for, which gives it a real advantage. Okay. When you think about obviously you need to iron out the details with the FDA, what a registrational study could look like. What are sort of the main takeaways you expect to get from those conversations? Is it sort of around whether polyp reduction could support an approval, kind of the length of follow-up? Do you have a sense of what a reasonable comparator would be? Is natural history sort of used in this indication, given there really are not any approved medicines? Yeah. What I can say is we have had productive discussions with the FDA. No one has ever gone in with either the depth of response or sort of the natural history data, to your point, that we have been able to show. We actually, this is another place where we can put our resources to use to be able to look at data in a new way. In a very short period of time, we were able to create a LLM model using over 250,000 patient records to look at clinical practice. What is standard clinical practice of patients with FAP? What are doctors actually doing? This includes all the physician notes and being able to query that and saying, "Okay, in this situation or with how do patients progress or when do they get surgeries?" Those sort of things. This is new information that has never been seen before. We also created a natural history database with one of the universities in the Netherlands that tracked FAP patients over 20 years, and we are able to show that these patients do have spontaneous annual, or they do have continuous annual polyp growth. That averages in the north of 50%. There is real polyp growth that continues to happen for these patients. We are going to go in there, but I do want to set the base. Even if it ended up the exact same trial design as the one that had been approved, we would be fine. What we are able to do with our ClinTech platform is we have shown increases of 30%-60% in baseline enrollments, and we have put out some of the literally site-level statistics on what we were able to achieve. We do this by starting with the data and finding out where the patients are and driving the CROs rather than having the CROs drive us. We have just been able to achieve different things in our clinical trials. We would focus on driving enrollment and getting to the right patients and understanding the right patient populations, to get that trial done in the best way possible. We will see. Maybe there are endpoint changes, maybe there aren't, but we feel good about it either way. Okay. Yeah, I am glad you brought up the ClinTech platform because as we know, the bottleneck, even with all the AI tools we have, continues to be clinical trials in terms of getting drugs to approval. If there are ways to speed enrollment or even recalculate your powering assumptions to decrease the sample size that you need, I think these are all things that you guys are doing in applying to your studies, right? Yeah. We use real-world data literally on every single one of our programs, and we build AI models around it. We run simulated trials. We are doing all of the data science around the clinical aspects as well, and it does really make a difference. We understand our patients much better and which of our inclusion/exclusion criteria really matter and how it could change. What sort of drug interactions we have to be most focused on. All of those sort of things, we are able to just get a much higher level of understanding for our patient populations before even starting the trial. Okay. Yep. That makes sense. Maybe turning, Ben, to a couple of questions around your oncology portfolio. You are bringing PI3K alpha towards the clinic. I think go/no-go is targeted for second half of this year. You also have RBM39, which is another asset for solid tumors. I think we will get second-half dose escalation data as well from there. I guess maybe just talk about your efforts in oncology and sort of what is getting you most excited from the emerging data. Yeah. Let's start with the RBM39 program because it's a really exciting one. This is one where it's basically we looked at a transcriptional target, CDK12, that the industry has always wanted to drug, but it is actually a really poor pocket to be able to drug. We just took a more phenotypic approach to understanding the biology, and what we found is RBM39 actually results in a very similar biological change to inhibiting CDK7. That sort of led us down a path, and we ended up creating a novel degrader to be able to go after it, and this is a first in class that we brought into the clinic. Really exciting early PK/PD profile coming through. Very large potential patient population. This is one where you can think of it sort of like a next-gen DDR drug. Certainly if you've got a genomically unstable cancer, this has a lot of potential from a mechanistic point alone. Or if it's combined with drugs that cause that genomic instability, it could be really exciting. More data on that soon. Second half of this year, we'll give an update. That's a program we really like, both for its novelty, but also for its potential in where it could go. As far as the PI3K inhibitor, it's funny, we get asked the question a lot like, "Why in the world would you do another PI3K?" There's a lot of them out there. This is interesting. If you look at a number of programs that were in the pipeline coming over from the more design-oriented side with the Exscientia. PI3K, we looked at it and saw things like hyperglycemia being a dose-limiting side effect for a lot of the PI3K inhibitors. We wanted to create something that had far more selectivity so that you could go far deeper because that hyperglycemia is caused by the wild type inhibition. It has two negative effects. One, obviously hyperglycemia, and the patients can be very limiting on its own. But second, there's a lot of theory that that actually causes the tumor to grow more quickly. What we wanted to be able to do is really knock out that signal so that you could drive maximum potency on the H1047R mutants and really get deep into it. So it's a good example to like LSD1, like MUL1, like CDK7. CDK7's a little bit different, but like those other two, where there's a single really clear side effect that if you can remove that side effect through chemistry, then you have a potential of really unlocking that target class in a new way. That's the goal. Even though the second-gen PI3Ks have shown less hyperglycemia, you still do see the hyperglycemia. In fact, those hyperglycemia rates are significantly higher if you're pre-diabetic, and they exclude the diabetic patients. So there's a diabetic and pre-diabetic orphan population, but even in the non-diabetic patients, I think that you're getting into some limitations in being able to push dose because of the side effects like hyperglycemia. That's not the only one, but that is obviously one that'll cause a lot of focus. Yeah. Even the next-gen, even if they have less hyperglycemia, they have other A's. I think stomatitis is a big one. Yeah Does your molecule avoid that? Yeah. We haven't seen any of the signals so far, but obviously we're IND cleared and should be starting the phase I soon, and we'll get a signal. But it looked very clean in the pre-clinicals that we ran and we didn't see. So we're getting more than an order of magnitude more selectivity over the other compounds in the space. I think it's also a nice example. There's a lot of me-too chemistry in the space. You just don't see a lot of differentiation. It's a little bit better. But if you want to get to an order of magnitude plus differentiation, you really have to be able to look at novel chemistry and go down different routes and explore the space in a different way. So that's another reason why it's a nice highlight too. We can actually go to places the rest of the industry can't because of what we're doing with AI. Yeah. Right. I want to pause for a second and see if there's any questions from the line. I'm not seeing any raised hands, but, operator, if you could maybe read the instructions for asking a question, and we can see if anyone has one here in the last five minutes or so. I think operator is maybe on mute, but I'll just say that if you do have a question, feel free to email me or I think there's a raise hand feature as well, and I'll be tracking that. Ben, I do have one question, and I think I would be remiss if I didn't ask the CFO a cash runway and capital allocation question. I was about to say, we got to talk about cash. I know. You're probably like, "Why are we saving these questions for the end? No, I am very happy we are saving these questions. This has been a core focus for Najat and I and the whole organization. If you look at our pro forma pre-merger expense to today, we're looking at about a 40% reduction. Honestly, we're not really doing less. We have eliminated a few of the programs that just looked like they weren't going to have impact, and that's really a sign of how we've gone through the entire budget. It's what can we see a clear line of sight that this is going to make a difference? We do that for everything. We even do that for G&A. This is across the entire company. If we can't see that clear line of sight, then we don't do it. If we can, then we say, "How can we optimize this work process? How can we ask the hard questions first? How can we get the data that we need and do this in the best way possible?" We're just doing a lot more with a lot less. That's been really exciting to see. We ended the quarter with $556 million in cash. We expect that gives us a runway at least to early 2028 without any additional financing. We're in a good place to turn over some of these data cards that are coming up. REC-4881, RBM39. We've got good partnership milestones that hopefully add cash as well. Hopefully we get across the line with our first development candidate with Sanofi, which I think would be really, really exciting, both in the program, but also in that partnership. I think we've got a lot going on. Yeah. But trying to do it as efficiently as possible. I think that's the name of the game, and certainly something that AIs, and Recursion specifically, has been built from day one around. Maybe just a final point to end on here, Ben. Just to focus investor attention over the next 12 to 18 months, what do you think would represent the strongest proof that the Recursion platform is working? Is it kind of the clinical efficacy on the in-house molecules? Is it progress with your current partners or new partnerships? Where do you think is the greatest opportunity for validation over the next year or so? I'm going to put it into the investor context because we're publishing papers that benchmark top of industry across multiple different areas in AI drug discovery and development. So I feel like that's well-validated. We've hit a whole bunch of milestones, so I feel like our partners are saying our technology is doing things that they couldn't do and that add value. So you really come back to people looking for clinical data. I think a majority of people who follow the therapeutics world want to see more of that data come through. I think 4881 was a wake-up call for a lot of people because it was pretty compelling and biologically unexpected. So hoping to surprise more people in those sort of ways. I think people are paying a lot more attention now than they were before, which is great. The other piece I'd say is just looking at the accumulation of data. We don't expect all of our clinical programs to work. That would be crazy. But if you start to see more programs reading out positive data beyond REC-4881, we've got RBM39, we have CDK7 and BOL-1, LSD1, PI3K coming into the clinic, some other programs that are coming up. They all start to point towards the same thing. Is the platform creating differentiated medicines? Because that's what makes a difference. I think that'll be a meaningful turning point in perception. Yeah. Very good. Well, I think with that, we're over time, so we'll have to end it there. But Ben, really want to thank you for the great discussion, for participating in the conference, and for Jill for hosting it. Looking forward to all the updates over the coming months. Sounds great. Really appreciate you having me. Thank you. Thanks, all. All right. Let me admit her. If Paul is on. Hi everyone. I'm Trey Brown. I work together with Jill Hall on our U.S. SMID Cap strategy team within BofA Global Research. We have a few sessions going on concurrently, but Jill and I just wanted to welcome everyone to our two-day annual Mid-Cap Executive Insights event, which provides opportunities to hear from corporates across the small and mid-cap space, where BofA has great breadth of coverage. Our analysts cover nearly 1,000 small and mid-caps in the U.S., and Jill and I have also been expecting continued leadership from mid-caps in the back half of this year. Please feel free to reach out if you need the schedule or want to sign up for any additional sessions today or tomorrow. We have nearly 20 companies joining, or if we can help sign you up for either our small and mid-cap strategy research or our daily compilation of mid-cap fundamental research. With that, I'd like to pass it over to Mihir Bhatia, our consumer finance analyst, to introduce Upstart Holdings. Thanks, Trey. Thanks everyone for joining, and especially to the Upstart team and Paul. Really appreciate you guys joining us today. Before we get started, one quick disclosure statement that I have been asked to read. Today's discussion may contain forward-looking statements that relate to future results and events, which are based on Upstart's information available as of today and are subject to risks and uncertainties. Actual results may differ materially from these forward-looking statements. The discussion may also include non-GAAP financial measures, which are not a substitute for GAAP results. Please refer to the company's filings with the SEC and its IR website for additional information, including GAAP to non-GAAP reconciliations along with other disclosures. Okay, with that out of the way, the lawyers should be happy, so we can get started. Again, like I said, a lot of you already know Upstart. I think I recognize most of the names of the folks joining. We will go through, I think, right at the start, we will ask Paul to give us a quick overview of the company. Before that, I again, just want to say thank you to Paul and the Upstart team for doing this conference with us today and for the opportunity to host you all today. So thank you, Paul, for joining. Let's get started. Maybe I will just kick it off with that, Paul. You have been with Upstart right from the start, 14 years now, I think. You recently took over as CEO. Maybe for the benefit of anyone who is newer to the story, just give us the quick 90-second version of what Upstart is today, and then I think the part that matters for investors, that excites investors. Why does this business compound for 35% over the next few years? Yeah. Really short tagline would be AI for consumer lending. We operate a marketplace business where consumers can go and shop for offers of credit. We do personal loans, auto loans, HELOCs. Really, over time, we will offer the entire suite of consumer credit products. Our real strategy from the beginning has been to say, "Hey, there is a whole bunch of transformative technology innovation happening around how we use models to make better predictions and understand patterns and data." That innovation has largely not made its way into consumer lending, which is arguably the most important place for it to go because if you think about the history of consumer lending, this is the world's oldest industry. Almost everybody borrows at some point in time. Actually, surprisingly, a large number of people, depending on your exact metric of preference, something like 50% or more of people in the U.S. we think are underserved in how they access credit. Either it costs too much, they cannot get approved, it takes too long, and it is all fundamentally because the ability of the models and the lending companies to understand their risk is too limited. What we have done over time is we have built models that can both better understand the risk, which we call better risk separation, and automate away a bunch of the process. As a result, we are able to radically reduce the cost and complexity of credit for everyday Americans who are looking to borrow, and that is just about everybody. Great. Maybe just on the second part of that question, though. Why is this going to be growing at a very high rate for the next few years? Yeah. It is exactly those two things I said. One, this is an industry that is relevant to almost everybody. The addressable market here is enormous. It is almost laughably large if you try to do any kind of math against it. At the same time, usually industries that are that big are really saturated, really well addressed, all the sort of innovations have already gotten plugged in. That is just not the case in the consumer lending world, which I think for various reasons historically has tended to move slowly in adopting new technologies. We really have been, for a number of years here, the first, and we think we have quite a large lead in taking a lot of the innovation happening in AI and applying it to this space. We think it is one of the best possible applications there is for AI to do good and serve the consumer. I think you put those things together, disruptive technology transformation against a huge market that has not really fully ingested that disruption, and I think naturally you are going to get a business that has the ability to compound for a very long time. Got it. I do want to dig in on that advantage that you all have, but we will get to that in a few minutes. Before that, I did want to just highlight that while you have been with the company a long time, you just recently took over as CEO in the last few months. I think when you took over, one of the last few weeks, I think I was reading, you have called it the second leg of the race, right? Taking over as CEO. Maybe just talk about that a little bit. Specifically, I think one question we get from investors is, "Well, Paul has been there a while. What is really changing?" So maybe talk a little bit about what investors can expect in terms of changes. What is going to be different under Paul than maybe under Dave? Yeah. Yeah. Obviously, Dave and I worked together in close partnership for a long time. We started this business, as you said, 14 years ago. The thing about this business is that it surprised us in how long it took to build. Well, at first, I think when we started, we thought, "Oh, we will just do this, and within a few years, everybody is going to be chasing us on this same race of applying AI to credit, realizing how large the opportunity is." I think we have just been consistently surprised how slowly that has taken to happen. I think it is actually for a lot of the reasons that it took us so long to build what we call the first leg of the race, which has a lot to do with the fact that consumer credit, maybe credit in general, but certainly consumer credit, obviously it is highly regulated. It is an industry that has a lot of entrenched ways of doing things. If you think about what does it take if you are going to say, "Hey, we have this completely new way of understanding credit risk," what do you need to do to actually make that a market reality and do that at scale? Well, it turns out you need a whole bunch of different kinds of very old institutions to buy in, right? So you need rating agencies to understand the risk of this stuff so they can rate these things in a risk-appropriate way. That unlocks financing for capital partners. The capital partners themselves, of course, have to buy in. Regulators have to understand it, banks, et cetera. You have all of these different kinds of institutions that really form this network around maybe the traditional way of understanding consumer risk. We came in and said, "Hey, we have this completely different way of doing things. Never mind that this person's FICO score may look like this or that." It took actually a bunch of years to overcome that because then once these people buy in, then you are able to start making the loans. Then the loans themselves take two, three, four years to prove themselves as actually good performing loans. Guess what? The reality is, the first time you build a model, it is not going to be very good. So the first model is going to have a few things it gets wrong, and then you are going to have to tune that model, and then you have to start that two to four year clock again. It ended up taking us really the better part of a decade to really get the first leg of the race done, where we built the foundation for the company, which was we acquired really valuable proprietary data. We talk about this in terms of rows and columns of data that the company uniquely has, where we gathered thousands of columns of data about people's characteristics at time of taking out a loan. Then we had millions of rows of repayment data, and it's that matrix that actually allows you to train the types of models that we developed. Of course, we had many years where we actually developed proprietary algorithms to understand the patterns in that data. Those two things work together, high volume of data, high complexity of algorithms. You really can't have one without the other. That was on the technology side. At the same time, we had to build up the credibility with institutional capital markets, the rating agencies, the regulators, to really believe that this is something that can work and to see the evidence and prove it out. Those two things took a really long time, and we finally got to this place where we said, "Okay, it's indisputable at this point that this is just a better way to lend. That if you do this, then you get a tremendous accuracy advantage that unlocks either much higher approval rates at the same loss rates or equivalently much lower loss rates at the same approval rate." That's something that we think we proved in phase one of the company. Of course, we took the company public. We built an incredible team that's able to operate this business with leverage and scale. We started taking the same idea and going, "Hey, we're not just going to do this in unsecured personal loans," which was our first product. "We're going to do this in every category of credit that's relevant to the consumer." We rolled it out across auto and home and short-term lending. That's where the company sat at the moment of transition, and that's why we called it the leap from the first leg of the race to the second leg of the race. In the second leg of the race, I think we have exactly the foundation that we need for the company, and now it's about really tactically and strategically applying it to the right places at the right time. The very first thing that I prioritized in 2026 was I said, "Well, if you look at the state of the business today, you look at the stock price today, one thing that's very clear is that the business is operating with an inordinately high cost of capital implicit in the stock price where investors and the market just is skeptical that this is a business that can sustainably deliver high rate of profit growth." We said, "Well, actually, the market is wrong about that, and we know just the way to fix that." Instead of saying we're going to invest in every possible thing we could do at once, we streamlined the company priorities down to a very short list, very centered around what we call contribution profit, which is our single best measure of the operating progress of the business. We delivered exactly that in Q2, you can see it in our results. We said, "Well, first thing we are going to show you is that in our core personal loans business, we do not have any kind of intensifying competitive threat or declining margins or any kind of structural compression there." Actually, this has just not been our number one focus for a little bit, now it is. We did 3.5x the growth of the prior three quarters put together in a single quarter. That drove then a massive increase in our contribution profits. We got to record contribution profits in Q2 that surpassed Q4 of 2021 back when macro conditions were much more generous in 2021, of course, than now in 2026, whether you are looking at interest rates or consumer default levels, just night and day marked difference. We still did record contribution profits in Q2 of 2026. Second thing we said was, "Well, we have a few new products that we are really excited about, in the last few quarters, we have proven that these products have real borrower demand, they have investor demand, but we have not really shown that these products are going to be profitable good businesses for Upstart. We said, "Let's make the number one priority of these teams to get to contribution profitable in these segments." These are home and auto. In Q2, we expanded their contribution margins by 61 percentage points. They are not all the way there, but they are well on their way to getting towards contribution profitable, we have said that we are going to be there by the end of this year. That has been really, really strong. Then I think the last big question has been, is this going to be a business that just structurally, constantly needs more equity capital in order to support the balance sheet or make the whole business work? Q2, I think, was also a really good proof point on that question, which is we did 23% sequential growth. 23% growth in originations in a single quarter is like $760 million. It is way higher than the normalized rate of growth that I think most people would be ecstatic about in the business. We did that while predominantly funding that with third-party funding. Our balance sheet loans declined to almost a two-year low in terms of the percent of total outstanding loans. Really, I think just a quarter where we were really hyper-focused on proving to the market that this is a business where you can expect all of the underlying businesses to have really strong margins because their underlying tech differentiation is very strong for that to be able to grow at a nice rate for us to do so in a very capital-efficient manner. Got it. No, that's interesting, and particularly on 2Q results. One thing that struck us on 2Q results, there's a few, I would say, highlights in the results. You mentioned the unsecured margin improving so much. What was also interesting was the growth in core personal loans. I think you are up 27% sequentially. The growth was probably, I think it was faster than the prior three quarters combined. To be fair, you had called it out in 1Q, but I did want to dig in a little bit on that. Maybe just bridge for us, how much of it was the model and the funnel improving versus you're just much more focused on driving that product and marketing investments in that product. Some of it was, of course, a little bit easier comp, but what drove that big improvement in 2Q, and what should we expect for 3Q with core personal loans? Yeah. I do think ultimately it's all downstream of management focus. I think the different categories of why it grew are really not so different in our mind. As a reminder, the number one way that our business grows is we invest in better funnel. That can be better models, better user experience, more automation. Those are, to us, all ways of doing roughly the same thing, which is increasing the percentage of people that convert, holding constant the applicant pool. At the same time, of course, the applicant pool is not fixed. We're always putting effort into growing the number of consumers that we have relationships with, the number of consumers that know about Upstart, the number that are coming to us. If you think about, you always every single quarter have these trade-offs you're making and which pockets are you going to focus on. I think in the preceding year, we had been a little more focused in some other segments. We had been a little more focused maybe on going kind of broader. In Q2, we were extremely focused on this core segment because it's a segment that we have such strong margins in, we have such a strong level of differentiation in, and we just wanted to prove and make clear that, with a little bit of focus on this, there wasn't any kind of fundamental change in the size of our advantage or compression in the margins. It was just actually a thing of, like, if we focus on this, then it's going to grow a lot, and that the sort of untapped opportunity here is just very large. I think that was disproportionate maybe in Q2, but we certainly will continue to focus on this as a strategy because it's just such a good way for the business to generate contribution profits, and contribution profits kind of pay for everything else we want to do. Right. I guess maybe if I'll push you a little bit on that, exactly where you ended, right? If you continue to focus on it and it has such a large market, you can drive a lot of contribution margin, which lets you invest in the rest of the business. Why not push even harder on this and really generate the cash flow, if you will, to help you invest in the rest of the business? I guess where's the choice between investing in growth of other products versus investing in more of your time in just driving personal loans? Take us through that decision. When Paul sits there and looks at it, how does he say, "Well, that's enough for that. We need to feed or water," whatever analogy you want to use, "some of the other products, too? Yeah. It's an evergreen debate at Upstart. We probably could go even harder in core personal loans, and I think that would start to come at a pretty significant expense to our new product growth and our ability to invest there. Ultimately, we're like, well, you don't want to just solve for the very long term. Maybe if you just solve for the very long term and you don't worry about anything in the short term, then you're just going to invest a ton and it's going to take a long time to show profitability. If you only invest in the very short term, then I think you're just never going to reach your potential as a business. To me, you have to land somewhere in between and you have to do it in a way that's mindful of what your implied cost of capital is and how much sort of credibility with investors and markets, and we hope to earn more and more of that over time. I think certainly, there are businesses that have earned the right to invest more aggressively and invest for a longer duration, and we hope to be there one day. But I think I'm just cognizant of the fact that today, the reality of our implied cost of capital is very high. I always say, like, well, there are these really great investments. Maybe this will pay off in five years or something, and that's really nice. If you do the math on that and you put it in a model that says that your IRR on that is really good. But then I look at the stock price, I am like, "Oh, well, what I think the implied IRR on the stock price is even higher than that." Those are, I think, some of the questions that go through my head in terms of thinking about how far out we should be investing. We want to be really smart, really rational capital allocators in how we think about those decisions. I do think we are landing in a place that is a good in between, where we are, I think, going to do a really nice job of continuing to grow our profitability as a business. At the same time, we are keeping open sort of the entire addressable market of this business over our relevant lifetime. I am not super old, but I also like not going to wait around for the rest of my life for us to achieve the whole market opportunity here. I think that is kind of the balance we are trying to strike. Sure. I think one thing that really impressed or struck a lot of investors in 2Q was the sharp improvement in contribution margin for some of the secured products. I think you mentioned it earlier, also 61 percentage points of court improvement right there in contribution margin. Can you talk a little bit about that? What clicked? Why is this suddenly seeing this hockey stick almost, if you will, growth in contribution margin? Is it just a matter of these products are now at scale and have found their product market fit, if you will? What should we expect from here beyond, go beyond 4Q where you are getting to break even. What do the contribution margins of this product looks like at scale? Yeah. So both Home and Auto in the preceding quarters had really achieved kind of what I think of as the first steps in building out a new product, which is proving that borrowers want this thing, proving that investors want this thing, a real marketplace that works. It was time for these products to prove the next thing, which is that these are things that can actually make money and be good businesses for us. So we really fairly sharply turned the focus of these teams over the last few months to be, your number one goal is to get to contribution profitable. We do not care how much you grow the business. Your number one goal is no longer proving demand. It is now sort of proving unit economics. That just reshuffles the sort of order of things that you're going to be doing. There are a bunch, of course, a bunch of things help both. If you think about what has to happen for these businesses to have good unit economics, well, you've got to do some optimization of costs. These really matter in secured products. If you look at HELOC has pretty significant costs when it comes to verifying an applicant for a loan, much more than personal loans. Levels of automation are much lower, and the just number of things that you could possibly automate is just so much more, when you have to deal with liens and all of that. A lot more focus on the kind of cost side of it, then also a lot more focus on optimizing our take rates, which in new products tend to be totally unoptimized. Where the starting point is just like you just pick some kind of slightly arbitrary flat fee and just charge it equally on everything. Really, if you compare that to what we do in, say, personal loans, there's a big difference in personal loans. We have intelligence when it comes to knowing which offers we are adding a lot of value to the consumer in, which ones we're only adding a little bit of value, trying to set our take rates in proportion to how much value we're creating. There's a very similar thing going on now in some of these businesses where if you take auto as an example, there's just a huge amount of variation in the dealership. Some deals we're uniquely the only offer. The only way you're going to buy this car is with an Upstart loan because no one else can understand that risk right. Then there are other deals where it's a fiercely competitive free market, and in that case, it just doesn't make sense to try to take the same amount of economics in each case. We're just starting to get smart to that and optimize around our take rates. Those will continue to be dimensions that we optimize along. I've said a few times that is going to be both a fast story and a slow one in the sense that I think there's going to be this very fast ramp to contribution profitable. Just because we're very focused on it. It's our number one priority. Then it's not like we're going to be done. If you look at our personal loan margins, they continue to expand for years and years after that product was mature, and that should just come as a function of how much value we create. That's ultimately what I care about doing in this business is I want consumers to get a delightful product, something that saves them a ton of money compared to the next best option. I want us to make more money as we save people more money. I do not want to try to take too much all at once up front. I think that can be a pretty bad trap for these sorts of businesses where you say, "Well, I am your best offer, so I am just going to take every last dollar on the table." I want to always leave plenty of money for the consumer on the table. Then as the amount of money that we bring to the table grows, we can keep an increasing amount of it. That is something that I think will play out more slowly over the years. Got it. We are about halfway through, so just wanted to flag for any of the listeners, if you have any questions, you can raise your hand or shoot me a Bloomberg or email, and I am happy to ask it on your behalf. Or if you want to just raise your hand, we can call on you. But one question that I know we will get, just because I have gotten it a ton, is around UMI, the increase in UMI to the top end of your. I think for this year, the guidance is 1.4 to. You had assumed 1.4 to 1.5. UMI is probably close to the top end of that range, but you have held your guidance. I guess what is the offset? What is helping you come in within your full year guidance despite UMI being higher? At what level of UMI does the math stop working and force a revisit either to medium-term or short-term guide? Yeah. We have shared that every 5 points of change in UMI is worth 5%-10% in relative size of originations, which tends to be pretty proportional to everything else, kind of revenue and contribution profits. If you do the math on that, it is a pretty sizable effect. I think if we were in the low or mid part of the UMI range for the year that we thought we might be in, probably we would have been raising guidance. I think that is because we had a pretty extraordinary Q2 in terms of execution on what is in our control. We control our ability to build better models. We control our ability to build better user experiences, more automation, reach more customers, and all of those things, I think we have done exactly the right things on, prioritized exactly the right areas and businesses, and I think the results show that. I think, if not for the rise in UMI, I think we probably would have landed in a pretty different spot on guidance. As it is, I think we look at those two things and say, well, there's kind of a great execution on the one hand, and sailing against a bit of a macro headwind on the other, and those kind of net out. So that's how we landed where we did on guidance. Of course, we also want to make sure that the bar for making any changes to guidance is just high. We always want to make sure that people understand that if we're going to affirmatively come out and say something, we really mean it, and we have a lot of confidence behind that statement. Got it. I actually see a couple of hands up already. Luke, why don't we go to you first, and then Jeremy, we'll come to you next. Luke, if you want to go ahead. Yeah. Great. Thanks. Can you hear me okay? Yes. Okay, great. Thank you guys for doing this. I have a couple quick questions. The first is just on kind of the commentary on the focus on the core in 2Q. You put a lot of emphasis on it, and it was really strong growth year-over-year. It sounds like that was a big push for you guys through the first half, and then the second half is kind of getting to contribution profit neutral. I think it was breakeven by 4Q in the secured product. Should we basically say that the growth rate that we saw in the core personal should flatten out into the back half of the year? Or as we look into next year and obviously beyond, how should we think about the sustainability of the growth rate on the unsecured versus the secured? Because obviously, getting to contribution profit margin breakeven on the secured side will, rising tide lifts all boats on the overall contribution profit dollars. Yeah. Our strategic focuses and priorities aren't really changing. Growing in core personal loans is still very high priority for us. We've also shared that getting our secured products contribution profitable is a top priority for us. You're certainly right, those are some of our top priorities. I would not say that we are significantly de-emphasizing core personal loans compared to before or anything like that. It's still right up there at the top. Now, having said that, our guidance is our guidance. We have two pieces of guidance out there. One is around this year, and one is around our growth rate that you can expect over the next few years. We've been guiding to a 35% compounded growth rate. Of course, because the core business is such an important part of that, those are going to be somewhat tightly related to each other. I would just point you back to that if you're looking to model something in terms of what kind of growth rate we think is sustainable for this business. Then, of course, our job is to execute as best we can and do the very best possible job against that guidance we possibly can. It is, of course, in context of what's going on in the macro, because our very first, maybe the meta priority of the business is always do credit right. If you get a bit of credit tailwind, that's going to probably push you a fair bit ahead. If you get some of a credit headwind, then maybe the opposite direction. TLDR is look at our long-term guidance, and I think that probably is going to be closely related to what's going on in the core business. From an execution perspective, on things that are within our control, AKA things that are not macro, the core continues to be very important to us, and we're going to do everything in our power to grow that business. Okay. I only have one other one, and then I'll let Jeremy go. Just as we think about the adjusted net income, obviously, it sounds like you think that the stock price is undervalued. Can you give us some benchmarks? As a traditional financial investor, a lot of my focus is on price to earnings, price to book, but then also understanding the capital contribution. I know you guys highlight that I think it's 5.9% of your loans outstanding are held on balance sheet. That doesn't actually include the co-invests, which also include I think another 5% plus of capital required on the balance sheet relative to overall. Can you give us guide rails as to I was surprised to see the stock buyback in the first quarter just because if we think about your overall call it return on capital from maybe the co-invest, I think you guys have said it's high teens over time. If we're looking at a stock that's 3 x tangible book value and 12x, 15 x adjusted earnings, but really 30 x earnings if you back out SBC. How should we think about the guardrails around capital allocated towards the stock price versus capital allocated towards the core business? Yeah. A lot to unpack there. Let's see. Probably the first thing I would say is that we don't see the business as primarily being funded by our own equity capital. It is certainly true that we have some amount of equity capital that's required to operate the business. We think that over time, there's going to be increasingly efficient ways to do that. That's why we've talked about this 5.9% number, which starts to nudge people towards thinking about this in terms of as a proportion of the total size of the pie of originations that, of course, has been growing very quickly and we expect to continue to grow. And so maybe that's the first thing, which is we're going to be really efficient about what fraction of all the originations requires Upstart equity capital and how much Upstart equity capital is required, whether that's in the form of directly on balance sheet or to your point, the risk capital co-invest, which is a nice thing for us. It works out because it secures these long-term capital commitments from partners that's fairly unique in the market, gives the business a lot of resilience, is a really good thing. But then, really the value of the business doesn't come from the ROE of the money that goes onto the balance sheet or the risk capital. I think of those almost as this just a necessary part of the supply chain to make it all work. The real value is coming from the growth in the contribution profits of the business. That's mostly fee revenue that's getting earned on transaction. And the growth rate in that I think is actually, that's probably the one place I would look and say, "Well, that's actually the thing that's extremely uncommon for I think your typical comp that you might look at and say, oh, what's the typical price to book or price to earnings?" I think any kind of multiple is fine over a sufficiently long timeframe if you're considering the growth rate of the underlying business. I think it's just really hard to be like, well, I'm going to compute at an average financial services business on a price-to-book basis in 2027 when that business is probably growing at a rate that is a fraction of the rate that our business is growing in. That's, of course, what it ultimately comes down to, is the growth rate in the business. In our case, I would say the growth rate and the contribution profit, which is of course real fee revenue. It is true that it's powered and made possible by the amount of equity capital that's either supporting balance sheet or risk capital. That's why I've laid out my framework of priorities as we want to grow contribution profits, and we want to do it in a way that's really, really efficient with equity capital. I think if you believe those two things about the business, then you get to a pretty different view of the stock. Okay. My only pushback would be, obviously contribution profits have grown significantly over the last year, but on the fixed cost basis, those are also up 30% plus, which doesn't necessarily lean into the operating leverage that I think you guys are trying to. Obviously over time, that should manufacture the operating leverage, just given the growth and the TAMs of the businesses. But if you're growing contribution profits by a certain amount, and you have in the last year, you would think more of that would fall to the bottom line, when in fact 30% plus in the fixed expense base year-over-year hasn't really let that come to the investors. Yeah, you're absolutely right. The operating costs have been growing. Some of that is intentional investment in some new areas that we think are going to pay off nicely. But in any event, we've, I think, made it pretty clear at this point that I think the lion's share of that growth and the rate of it has happened, and that looking forward for the rest of this year, the growth rate is going to be much, much slower. I think the operating leverage will start to get a little clearer as time goes on. Sure thing. Why don't we go to Jeremy and then I actually had a couple come in over email too. Oren, we'll come to you after I go through the email ones, but let's go to Jeremy next. Great. Thanks so much. I appreciate it, Paul. Appreciate you doing this. I guess in terms of what happened on the quarter and the messaging, that was all quite clear to me. From our vantage point, you guys did exactly what you said you'd do, exactly what the investment community, I think, wanted you to do. The profitability is inflecting. You executed well despite the UMI ticking up a bit. You have a much stickier capital base now. You're buying back stock, both personally and as a company. You and Andrew are both incredibly incentivized for the stock to appreciate and be a multi-bagger from here. My question is what has been the investor feedback? Because I'm trying to understand why the stock is where it is and sort of what the feedback from the investment community has been. Because objectively, just the reaction to the last quarter and the stock sort of not getting more attention given what has changed over the last six months for the company, which seems like a real inflection point to me across a number of dimensions. What do you think has been from, I'm sure you've had a ton of investor conversations, has been underappreciated or misunderstood? Yeah. We've certainly spent a lot of time with investors since the meeting. I do think that across the board, there's been some pretty strong appreciation of the quarter. I think us doing the things that we said we would do this quarter, I think everybody's appreciated that. I think the sort of outstanding questions, I never felt like it would just be one quarter and everybody would be sold on the business or its outlook. I think there are some outstanding questions. I think you just heard one around OpEx, which is people are confused why our OpEx keeps going up. I think it makes it a little harder to model how much operating leverage you should believe this business has over the next few years. I think that there is some concerning questions around how much overtime, how much equity capital this business will need. There's this bank thing coming up and what is that and what are the implications of that. I think that's maybe just a bit of a new thing that not everybody's really familiar with yet. On that particular point, we've stated, I think pretty unambiguously that we think we have sufficient capitalization to open Upstart Bank early next year. It's going to be pretty accretive to us operationally because it's both sort of economically more efficient and operationally more efficient, lets us sort of get to more states and make more offers. So it's a really good thing for us. But I think there's just a little bit of consternation around that. Then I think probably most fundamentally is just one quarter is probably just not enough if you're just fearful that actually if you look at the past N quarters of this business and you say, well, okay, there's this one quarter where it seems like you're able to grow profitably and then there's all these other quarters where you weren't. Maybe this quarter was just like a lucky fluke. Maybe it was only because the macro was supportive or something and that's about to change. You look at that and you just have, I think, reason to worry that one quarter is not enough to prove the thesis. I can't blame anybody for feeling like they want to see more proof points. So I think what's in our control is we're going to keep doing is just like this short list of priorities. We're going to continue proving that our core personal loan business is really strong, really differentiated, and something that can grow for a long time to come. That we've got these new horizon businesses in home and auto that have enormous TAMs that are going to become real businesses that are unit economic positive and something that you can believe in to give us a runway to grow for even more years to come. That we're going to do all this while still operating in a really capital disciplined, capital efficient way. I think if we can show those things, then whether it's one quarter or two quarters or five quarters, I think that eventually lots of investors are going to have to change their minds. So maybe I am going to preempt it a little bit here and just jump in. I was going to ask this question at the end, but Paul, wondering since you became CEO, you have obviously increased your engagement with investors, analysts. What is the one thing that you think people get most wrong about Upstart or struggle to understand about Upstart? Is there something that you just, when you meet with them compared to your understanding and from your seat, is just baffling, something that people are just not getting or not understanding? Well, I said this on the original earnings call. I think the answer is everything. I think when I checked to see earlier this year, there just really had never been a bigger gap between how we saw ourselves at Upstart and how the rest of the market saw us. It is like our business was stronger than ever. It was like we had the best tech we had ever had, the most committed capital we had ever had. We had more borrower relationships, more customers with traction in home and auto. We used to just be a one-product personal loans business. All of these are things that are real wins that are kind of durable no matter what the macro environment is. The opportunity is just so big, given this kind of intersection of the AI disruption meets one of the largest, oldest industries in the world. I think just like the market was like every element of the business was an area of concern. It was just like, "Hey, your margins seem to be compressing. You have been growing for multiple quarters but not actually growing your contribution profits. You have just dipped back into GAAP unprofitability. How are you going to fund all of these loans? Are these new businesses ever going to be real businesses, or are they just kind of things where you can grow originations but again, not make profits out of it?" I think that it is actually a lot of different concerns, each of which is a little related to the others, but sort of an independent thing that you have to get right. I think we are just going to go and take all of these questions head on because we have a lot of confidence that actually the trajectory the business is on is going to naturally show that you can get a ton of profitable growth for a very long time in a really capital efficient way. Of course, the great news is if we do that, kind of no matter what I think what investors or the market wants to do with us in terms of what kind of multiple they want to use or valuation framework. At the end of the day, I think all of them are going to go up with profit. Of course, if you have the profits, then you get to bet on your own future, and that is what we are planning to do. Got it. I think we have about five, seven minutes left. I am going to go back to investor questions now. One question that we did get was about the full Q4 2024 vintage. It looks like it has been, I think, underperforming a little bit targets. What is driving that there? I think this quarter it moved towards underperforming targets. What is driving that? Is there something specific in that vintage that we need to be aware of or that you are watching? No, nothing terribly specific. Overall, credit performance has been really strong. We have been really happy with the returns that have been getting delivered to investors. We always talk about how the returns have been consistently many hundreds of basis points above the spread of treasuries averaging, I think something like 600. They are really strong returns. Just kind of naturally, you are going to have some vintages that are a little over, some vintages that are a little under. I think that is pretty par for the course and not something that worries us terribly much. Some of that variation which vintages do well and not well is just going to be correlated to the changes in UMI because we, as a first-order approximation, think of us as underwriting to the UMIs at the time of the underwriting. If UMI subsequently goes up a lot or goes down a lot, that is going to give either a tailwind or a headwind to the credit performance of that particular vintage, of course, because UMI mathematically is just going to be something that is linearly correlated to the rate of defaults. That is part of it. Then there is sometimes some idiosyncratic factors that matter a little bit to each vintage. No, when we talk about underperformance of these vintages, I would say our effects overall tend to be pretty modest and except for that period of time right after the stimulus ended when there was this huge upswing in UMI. Obviously, that was a larger effect. Really since then, we have been very happy with the credit performance, and it is just kind of some natural variation and then some UMI-driven variation. Got it. Oren, I know you have had your hand up for a while. Why don't we go to you? No problem. Thanks, Mihir. Thanks, Paul. Thanks for the time. I will be brief. I think maybe a couple other things is cash flow generation is probably another focus of the market. But in terms of balance sheet growth, that is probably another thing people are focusing on, which I think you are doing a great job of pulling that back. But related to that on the balance sheet, just curious on your convertible bonds, since you are buying back shares why not go after those since those maturities are trading in the 70s? Perhaps just thinking if that is a good use of capital. Thanks. It is certainly something we looked at when we last did some share repurchases. But we mostly just looked at what we expected the sort of IRRs on each to be, and we just thought that the stock's IRR was just higher, and it was so much higher that it was like, oh, even though the other thing is more debt-like and there is some value in getting that down, whatever, they are risk-adjusted not the same. But nonetheless, the delta, we just thought that the stock IRR was just so high. So, that is how we think about these things, is just we are trying to maximize the IRR of where we deploy capital, and that could be internal uses or stock uses or various kinds of buybacks, and that is how we will generally compare them. Yeah, it makes sense. I am always wondering what you think that IRR is and where you think the stock goes. But just maybe lastly on just the cash flow generation how low, in terms of cash flow generation, you do not really have much at this point, but just curious to know how low you are willing to take cash in terms of those share repurchases, and when do you expect to kind of ramp up on free cash flow? Well, the business is growing a lot, and we expect it to continue growing a lot. And I have said that it is really important to us that we keep a sufficient level of investment that the sort of long-term kind of addressable market here is on the table. And so that basically means that I think we need to keep adequate amount of cash for all our various initiatives. Of course, like capitalizing this bank is something that is going to use cash. So that is why we are being mindful around that and we want to be mindful around the amount of cash. And so I would like to generate more cash, and when we generate more cash, then we get more options of what we do. And until we do that, options are more limited. Got it. Thank you. Appreciate your time. Maybe just turning very quickly to the cash line product. It's one of your newer products. Talk a little bit about that product, just in terms of how you work, how you're funding it today, what the end state looks like. Also maybe just spend a little bit of time on the modeling aspect of it, because it is a little bit of a different product than your personal loan or even the always-on credit, if you will. So maybe just spend a few minutes on that and just help us understand how big of a lift was it to put that into practice. Yeah. The cash line product is a great product for us. It's one that I wish we had launched sooner and earlier in our history. It's a product where you have to be really good at underwriting because you're serving a consumer that's fairly financially stressed most of the time. So it's really important to underwrite that consumer well. But it's also a product that can be extremely high demand. It's a product that people kind of search for proactively. Most credit products, you have to find your borrower, and this is one where the borrower finds you. So that's a pretty amazing fact about that as a business. As a result, it's a business that if you do right, I think can be a very profitable business. So we're really early stages on that, kind of in my steps one, two, three, four of building a new product. It's just kind of cleared step one, which is it's proven, wow, people really want this thing, and then you've got to prove everything else about it and figure out the right kind of funding rails for it, the right sort of lock-in kind of credit calibration on it, and then get unit economics. So there's still a bunch of work to do on this product, but it's extremely high potential. It's a really good match for the core competencies we have as a business. So I wish we started sooner, but next best time is now. Got it. I am being flagged that we are now past time, so I will have to stop it there, though there are a few more questions we'd have loved to get to. But I think it was a really good discussion. Appreciate all the investors being super engaged and asking questions, too. Thank you again, Paul, Sonia, and team. Thank you so much for taking the time and joining us today. Thank you.