This article is part of Fintech Leaders, a newsletter with 80,000+ builders, entrepreneurs, investors, regulators, and students of financial services. I invite you to share and sign up. If you enjoy this conversation, please consider leaving a review on Apple, Spotify, or Youtube.
Today I sit down with , the AI-native infrastructure platform transforming mortgage servicing in America. They have quietly built a top-10 national mortgage servicer with over $100 million in revenue and more than one million loans under management, all while achieving profitability. Valon is now selling their battle-tested platform to the largest servicers in the country. They are backed by a16z, Westcap, Jefferies, and Alleycorp, amongst others.
What makes their story remarkable is how they built it: $100 million in revenue sold entirely through founder-led sales, no marketing team, and a seven-year detour of becoming a servicer themselves just to prove their software works. Their journey from burning $4 million per month to 70% operating margins offers a masterclass in operational discipline and unconventional company building.
Below are four of the most impactful and actionable insights from our conversation. If you’re eager for more, the full discussion awaits:
Why building a company through pain and suffering creates unbreakable moats
Andrew and Linda share the unglamorous reality of Valon’s early days: creating $50 loans to employees and friends just to hit TransUnion’s 50-loan reporting threshold, camping out in the NY DFS lobby for two hours to rescue a license application, and manually reconciling thousands of transactions from 10pm to 2am while appearing professional to the outside world. They serviced hundreds, then thousands of loans completely by hand before the software could handle the load. Linda recalls spending three hours with Andrew just to onboard their very first loan, mapping every field and document in a spreadsheet so they could manually load it into the system.
Mortgage servicing is filled with regulatory catch-22s that make innovation nearly impossible. You need to report to credit agencies, but credit agencies won’t work with you until you’ve serviced loans. You need licenses, but getting licensed requires experienced operators on staff that startups don’t have. Fannie Mae and Freddie Mac typically require 24 months of profitable operating history, a death sentence for tech startups. Valon only broke through because of a COVID-era expedited approval program for tech companies, one of very few to get in. By living through every edge case themselves, they built both the credibility and the knowledge that no competitor could easily replicate. As Linda accurately puts it: “Anyone who wants to do this after us would have to go through the exact same pain and suffering.” And Andrew explains it well: “I can sign the NDA, but I honestly don’t mind that much, because I know there’s no way you’re going to be able to do this. It’s pain and suffering. It’s just time, pain and suffering.”
Why they’re building two companies at once: an operating business and enterprise software
Valon started with a vision to sell mortgage servicing software. The original company name wasn’t even Valon, it was “Go Service One,” a joke on Ready Player One, reflecting their goal of building a system for servicing. The problem became apparent within three to six months: the industry had zero trust in outsiders. These were companies that had spent hundreds of millions of dollars trying to improve their own systems and failed spectacularly. Interestingly, they weren’t just skeptical of a couple of smart founders claiming they could build better software, they were skeptical of their own ability to evaluate technology after so many expensive failures and millions of dollars wasted.
The Valon team made a radical decision: they would become a mortgage servicer themselves, not as the end goal, but as proof that their software worked. This meant eventually building two companies under one roof with completely different skill sets and business models. One as an operating company focused on processing, performance management, compliance, and customer service. The other as a pure software company focused on infrastructure, APIs, and scalability. Over time, they have had to become experts in both domains. As the servicing business grew to new heights, every year the team had to hold serious conversations about whether to abandon the software dream and just run the highly profitable servicer. But they kept returning to their original mission: solve the underlying problem for the entire industry. Today, Valon operates over one million loans as a top-10 servicer with materially better economics than incumbents and they have started deploying the Valon Operating System to major industry players who’ve watched them transform a typically low single-digit margin business into something with 70% profitability.
Selling $100 Million in contracts with no sales team
Valon built $100 million in recurring revenue for their operating business and now has a huge software pipeline without a sales team, marketing team, or even a product name until their board member Brian McGrath insisted they needed one. When they finally named it, they went with “VOS” (Valon Operating System). Andrew sold the original operating business almost entirely himself, flying around the country looking for loans to service. Their realization when pivoting to software sales was stark: nobody at the company, besides the founders, had ever thought about go-to-market. The entire organization had scaled without developing those muscles.
Their current software approach is equally unconventional. Andrew, Linda, and Brian fly around the country meeting with the largest mortgage companies and present a simple case: here are the results we’ve achieved with our own operating business, here’s the value we think we can deliver to you, do you want to do this with us or not? Then they focus on deployment rather than sales. The mortgage industry is practical and functional in ways that consumer tech is not. These aren’t companies that care about branding, beautiful interfaces, or how innovative your pitch sounds. They care exclusively about economic value and operational proof. When Valon walks in and shows they’ve taken something that’s usually breakeven or losing money and turned it into 70% operating margins, the conversation changes immediately. Servicers look at those numbers and realize existentially that they need the same technology.
How CTOs can succeed in an AI-first world
When OpenAI released O3 on Christmas Day 2024, Andrew immediately recognized the implications for Valon’s entire strategy. He’d already been experimenting with LLMs for tasks like optical character recognition, but O3’s capabilities made the trajectory undeniable. Whether AI reaches true AGI or just becomes extraordinarily powerful, two possible futures emerged, and both pointed to the same conclusion: Valon needed to be the operating system layer for mortgages, the infrastructure that LLMs call, not a business that gets automated away or squeezed between foundation models and OS platforms. The first thing Valon did after the New Year’s break was hold an all-hands meeting with a presentation titled “AGI is Coming.” The message was unambiguous: we need to accelerate our transformation from servicer to software company within one to two years, not five to ten, because waiting means either obsolescence or getting crushed by companies that move faster.
Valon got somewhat lucky in their positioning. Because they’d always focused on core infrastructure: money movement, ledgers, system of record representation. Rather than flashy applications, they were naturally structured for an AI-native world. Their infrastructure was designed so that deterministic API calls would handle critical functions like moving money and recording transactions. LLMs would orchestrate workflows and make decisions, but they’d call Valon’s infrastructure to execute. All of their recent development, built in the last ten months, included evaluation frameworks, monitoring systems, and controls for adjusting how much work humans do versus how much LLMs handle.
Andrew’s guidance for industry CTOs focuses on three areas:
Understand what part of your infrastructure and product is truly defensible and won’t be worked around
Actively push your engineering teams to use LLM coding tools like Cursor and Claude, because adoption requires deliberate effort and changes how people work;
Recognize that organizational design is fundamentally shifting. High-agency people with strong business context and AI skills will be 10 to 20 times more productive than before. Companies will need dramatically fewer people but must compensate them significantly more. Valon has even codified AI usage and skill as part of their performance review process.
The Unfiltered Q&A: Andrew and Linda on Building Valon
The four principles above shaped Valon’s story. But principles without context are just platitudes. What follows is the unfiltered account of how it all happened: regulatory moats, founder-led sales, and why they’re racing to become the operating system for mortgages.
Miguel Armaza: Linda, tell us how Andrew convinced you to leave a bunch of really good places—Google, Goldman Sachs, Two Sigma?
Linda Du: In our group chats, we actually call Andrew “Nostradamus” because he’s always ten steps ahead of everyone else. We were friends in college—we both studied computer science, and Andrew was one year ahead of me. If you look at my career, it looks like I spent my entire career following this man around.
He interned at Google as a software engineer, next summer I interned at Google. He went to Goldman Sachs, next year I went to Goldman Sachs. He went to a fund, next year I went to a hedge fund. It was always the same pattern. He would call me out of the blue and say a bunch of words that were very convincing about why this was obviously the only next career move I could make. And then I would go and make it.
Finally, when I was at my last fund, he called and said, “Hey, I have this great idea. It’s time for us to build a company.” Back to the ten steps ahead: I suspect that ten years ago, he knew he wanted to start a company someday, and he was making sure I had exactly the right skill set to partner with him on building it.
Andrew Wang: When you think about who you want by your side, the most important thing is what David Haber calls “safe hands”—someone you can trust with whatever you entrust them with. But the hardest thing to find is someone who you can entrust anything and everything to, and someone who can continue to grow alongside you.
If you’re thinking the way we think—ten years, fifteen years, twenty years—you have to find people who are able to go along with you in that journey. Being able to have that person alongside so you can talk about what happened in the past and how you’re thinking about the future, those things are so important for the development of a company. And even more so for the type of long-term thinking company that we run today.
Miguel Armaza: You hear stories of founders doing things that don’t scale. Share those initial days—the stories of doing things that don’t scale.
Andrew Wang: When we first started, we needed licensing and approvals. The only way to get California approval was having an existing license with the Federal Home Loan Association. That’s quite difficult when you have no loans, no history, no track record. There was a carve-out by getting a Department of Real Estate Broker license. We scrambled around looking for who we knew who’s a family friend who’s a real estate broker. It turned out one of my co-founders—his long-lost cousin who was estranged—was a real estate broker. My parents were friends with them, but his family was no longer really close. We reunited them and said, “Do a solid for your cousin, please be our registered individual.”
Another example: you’re required when you report to credit agencies to have a track record. They want at least 50 to 100 loans. We’re in the business of making mortgage loans—the average mortgage loan is a couple hundred thousand dollars. We’re not going to make 100 mortgage loans before we can report to credit agencies.
Linda Du: And nobody will give you 100 loans to service if you can’t report to the credit agencies.
Andrew Wang: Exactly. It’s all catch-22s. We found which states had exemptions for small loans, then went out and gave loans to a bunch of friends in order to be able to report 100 loans and show we had them. We went through all this exercise and effort just to be able to report to credit reporting agencies, which would then allow us to actually get licensed and work with different large counterparties.
Linda Du: All of these catch-22s are just why mortgage hasn’t been disrupted or innovated in the last couple of decades. Even just getting licensed was such a journey. Andrew literally had to rescue our DFS application. The New York DFS is known to take probably two to three years to approve an application. When we applied for our license, it was just before COVID happened. I realized if I don’t make sure they have all the materials, nobody’s going to accept materials anymore. So I ran down to the mail room where the DFS was located on One State Street down in the financial district. They told me they couldn’t find our package. I literally sat for two hours in the lobby, constantly asking them to look for it. Eventually they sent someone back and were like, “Oh, we found it. We just misplaced it.” I was like, “Thank God, but also, what the heck.”
When we boarded our very first loan onto the system, Andrew and I sat there for three hours and literally mapped every loan field, every document, in a spreadsheet so that we could manually load it into the system. The only way to build something like this is if you do it yourself for a while, because there’s so much detail, so much nuance, so many edge cases. Andrew and I have manually, personally, by hand, serviced thousands of loans.
Andrew Wang: Today, that’s actually a selling point. When we tell our clients they should use our software, we can tell them if it goes wrong, we have serviced thousands of loans by hand and we know exactly how to do it. So we’re going to be able to get this done.
Linda Du: What we always tell the team, and honestly what I tell myself at midnight when this is happening, is you’re creating moats for yourself. Anyone who wants to do this after us would have to go through the exact same pain and suffering.
Miguel Armaza: You started wanting to sell software but ended up becoming a servicer first. Did you ever consider just focusing on being the servicer and forgetting about the software?
Andrew Wang: Our original company name was actually “Go Service One”—a joke on Ready Player One. The idea was building a system for servicing. Within three to six months, we realized nobody’s going to use your system. You have this super regulated, highly complex industry that just came out of the Great Financial Crisis. They look at guys trying to build this system and say, “Not only do I have zero trust in you, but we’ve also spent hundreds of millions of dollars trying to improve these systems and we’ve fantastically failed. So we don’t have trust in you, and we don’t have trust in ourselves.”
We needed to build our own operating servicer alongside the software to demonstrate that the software system not only works but is so much better than everything else that they have to pay attention to us. When you think about this for a moment, you’ve decided to sell a software system, but you now are deciding to pivot into building an entirely separate company, and you’re saying this company’s sole purpose is to show that my software system works so I can do my other company. We somehow decided to build two companies instead of one. Building one company is hard enough. Two companies in very different domains—one’s an operating company about processing and performance management, the other’s a true software company. We needed to somehow be good at both.
Every year we would revisit the conversation about whether to just focus on the servicer. As we became much more successful and efficient, it became a real question. Today we’re industry-leading, best in class, top ten in the country serving around a million homeowners. But when we look at that, we ask ourselves: what did we set out to do? We have to be true to ourselves. We set out to build servicing software to solve this problem for the industry. If we just do a servicer, we’ll make a very profitable and interesting business, but we won’t actually solve the underlying problem.
If you’re going to do it and it’s going to be that hard, you might as well shoot for the moon. You might as well do the biggest enterprise value creation thing possible and the greatest impact possible. And if you build a company and you’re aiming for not such a great goal, people don’t really want to work with you. People are not as interested. It’s just as hard, but it’s easier to convince and galvanize your company going for the hardest things versus an answer in between.
Linda Du: Mortgages impact 80 million plus homes in the US. They underpin 13 trillion dollars of consumer debt, and today they still run on technology that was invented before the internet was created. That is a crazy thought. Something we always tell the company is, if we weren’t working on this problem, literally nobody else is. There’s actually nobody else out there trying to do this.
Miguel Armaza: You’re operating a servicing company, then you start selling software to your competitors. Are you now trying to grow both, or are you mostly focused on the software?
Andrew Wang: We are definitely today focused on the software side, and that is by far and large 100% of both Linda’s and my focus in terms of what we do day to day, as well as the majority of the company. We do have a top-10 large servicer in the country. We can’t just drop the ball on the homeowner. So there is a team that continues to run it and operate it, but it’s not something we’re looking to go through the hyper growth that we did in the past. The software part is definitely the biggest focus.
Miguel Armaza: How can you share a little bit about your go-to-market motion? What did you learn in the process? How does that team look like?
Linda Du: When we finished building the operating business and started figuring out how to go to market as a software company, there was a moment where Andrew and I looked at each other and realized we have no go-to-market motion at this company. The entire way that we scaled the operating company was Andrew and I would fly around the country looking for loans to get. Nobody, not a single other person at the company, had to think about go-to-market.
When we started on the software side, we took the same approach. Andrew, I, and now Brian fly around the country, and we talk to the biggest mortgage companies. We basically tell them, “Hey, these are the results that we’ve achieved with our own operating business. This is the value that we think we can deliver to you guys. Do you want to do this with us or not?” Then we’ve started building really a deployment organization. It’s about delivering the software more than it is about selling the software.
The mortgage industry is a very practical, functional industry. It’s not an industry where they really care about how cool the product is. It’s really about economic value that you can generate. The best marketing for our software is actually our operating business, because we can show that we basically took something that’s usually a low single digit, break even, sometimes net losing money business, and turned it into something that is 70% plus operating margins. That is the number one thing where a servicer looks at that and says, “Hey, whatever they have, I think existentially I need to have as well.”
Brian was on our board from the very beginning and recently joined us full time. He has been astonished that we have no sales team, no marketing, no branding. We actually did not have a name for the product until he mandated that we needed to name it. With all of this right now, if you look at our software sales pipeline, it’s millions in ACV of people who want to sign up for the software, and all of that with no marketing, no sales, no branding.
Andrew Wang: Another fun stat that to this day the board still makes fun of us for: we sold over 100 million dollars of recurring revenue and contracts for the original operating business with basically just me as the key salesperson. We did founder-led sales from the beginning to the end for the operating business. We’re still doing it right now, and we’re going to build a billion dollar revenue, annual revenue business doing founder-led sales.
Miguel Armaza: Something that has become very important the last three years is not just profitability, but efficiency. You had a turnaround—you went from high burn to a very different story. Was there a wake up call?
Linda Du: In 2022, our operating business was growing a lot. I think it was 100x-ing, 10x-ing year over year. We had gotten all of this advice from really great people saying, “You’ve now hit growth stage. In growth stage, you hire people, you hire senior people who have scaled companies before. You just hire, hire, hire, and then growth follows.” This is the classic venture playbook. We raised around mid-2022, and then by end of 2022 it was basically VC winter. No more funding. It was way harder to raise money. The bar had been raised, and all of a sudden, cash was king.
All the really great senior people that we had hired into the business were like, “This is fine, we’re in growth mode. This burn is okay. If you’re not burning money, it means you’re not investing in growth.” Andrew and I looked at each other, we looked at the numbers, we looked at the funding markets, and we were like, “No way. This is actually very not okay. We will die as a business if we keep going down this path.” It’s one of those moments where our job is to make the hard decisions. Number one focus at that point became profitability and efficiency. How do we take this operating business that we’ve scaled and actually turn it into a cash printing machine?
We looked at basically every facet of the business. We’re moving billions of dollars a day. People have made money moving money. We figured out how to monetize float. We figured out how to monetize all of the disbursements and transactions that we were doing. We looked at our lending business and we knew an originations business makes money in its simplest form. Our team was working on all sorts of fancy projects that were great for the long term. We basically looked at them and said, “No, your number one job is to turn this into something that prints money. Everything else we don’t care about—just get more efficient at targeting, just get more efficient at conversions.”
And then when LLMs started becoming a thing, we looked at all of the labor costs that we were running, and we basically, one by one, just started LLM-ing them away.
Andrew Wang: As we have been growing as a company, we’ve grown our engineering team. The number of hires we made this year equals the last two to three years combined. We’re much bigger, but when it comes to the actual expenditure versus the revenues that we’re bringing in, it basically hasn’t moved. That’s just because the operating company grows and makes so much money that it covers all these costs and lets us pursue relentlessly the goal of building the software company in this space.
Miguel Armaza: It’s easy to retrofit if you’re getting started now because you’re AI first, right? But even if you have three to five years of infrastructure there, inevitably there’s going to be some retrofitting. How did you approach this?
Andrew Wang: We got somewhat lucky in this exercise. Part of it was because our goal was always to build a software company. When you’re trying to build a software company in this space, the strategy we decided to pursue from the beginning was being very, very focused on the core infrastructure—money movements, ledgers, underlying system of record representations. These are things that will not get out-lapped.
Stripe is today working with ChatGPT and powering different applications, but the fact that they’re moving money—the code that literally calls the money movement, the ledger transactions—that is not going to be an AI piece of code. That’s not going to be an agentic piece of software. That is going to be a deterministic call based off of an LLM calling it. Because we were so focused this entire time on building this hardcore infrastructure to represent the world correctly and get the single source of truth representation done in the right way, when we started focusing on software, we had to go build all the infrastructure for how these APIs would be called—the way that LLMs would call this API.
All of our infrastructure when it comes to actually powering how all this work gets stitched together—calling the APIs—all that was done with AI in mind, really thinking about what would an LLM-native operating system look like. For us, it’s actually quite easy today. We actually do this internally today, of saying, “We’re going to build all of our infrastructure in a way where it’s very easy to slot in not just the usage of LLMs, but the eval set creation, the monitoring of it, the changing of how much you want humans to do versus how much you want LLMs to do.” All these things are things that we built honestly in the last ten months. So it’s very easy for us to say that’s something we built with AI in mind.
Miguel Armaza: Do you think AI is also going to change how you think about organizational design, or are a lot of the tried and tested models still relevant?
Andrew Wang: There’s a constant debate about what’s going to really change when it comes to more powerful LLMs that can do more work. I think people are generally aligned with the fact that, at the minimum, people who have high agency, have a lot of context, and have skill with LLMs are going to be 10 or 20 times more productive than before. That changes the model for companies when it comes to who you’re really prioritizing.
Historically, companies have focused not just on the 10x engineer but also been actively aware they needed the 2x engineer and the 1x engineer because there’s a lot of work to get done. What happens when you scale all those different things under this new paradigm? The number of people you need drastically reduces because they can do more. Obviously you can continue to grow that organization, but the rate at which you need to grow the number of people decreases dramatically. That also means your philosophy around the type of people you want, the longevity you need, all those different things change as well.
We’re talking about coding in this specific example, but that’s not just coding. That’s legal. We historically have had a number of lawyers in the company. We basically have three today, and we’ve stayed with three because as we have increased the workload dramatically, tools and legal AI have gone dramatically better. You’re just going to see more and more of this, which means you’re just paying more and really compensating people to be very good at this skill set in order to get the same amount of work done.
One thing we had to do, and really what I spent a bunch of time initially doing earlier this year, is sitting down with all of our engineers and showing them how you could use LLM coding, whether it be Cursor initially, and then it became other tools. It’s a very active effort, and it’s a question of how fast you need to move. You need to spend that active time and effort really pushing the product and usage to your team and coming up with an ultimate long-term strategy of how you’re going to get everybody to use it and understand it in the right way.
Linda Du: When you have a large organization already—we didn’t start the company this year, we started with a team of already 100 engineers—when you need to drive change and adoption, you really have to figure out how to systematize it at scale. As an example, we’ve actually codified AI usage and skill as part of our performance reviews. This is just something where you constantly have to tweak the machine and how your company is operating.
Want more podcast episodes? Join me and follow Fintech Leaders today on Apple, Spotify, Youtube or your favorite podcast app for weekly conversations with today’s global leaders that will dominate the 21st century in fintech, business, and beyond.
Don’t forget to subscribe so you don’t miss big conversations and insights with the giants of Fintech.
Previous Episodes You May Enjoy:
