On this episode, I spoke to Andrew Wang, CEO and cofounder of Valon. Valon is a Series C vertically integrated mortgage servicing startup. Founded in 2019, it services over 500k loans today and serves roughly 1% of all homeowners in the United States.
This was a special episode because Andrew was my former boss. I worked at Valon for two years, joining when it had 50 loans, and had a blast growing the company. I’m biased, but I think Valon is a startup with a lot to learn from.
We discussed Valon’s origin story, how regulated industries are adopting AI differently, what makes for a successful AI rollup, and why people still underestimate the fact that growth solves all problems in startups.
Watch on YouTube, listen on Spotify.
Andrew’s Key Takeaways
Andrew saw Valon as a “high alpha, low beta” business. Valon isn’t betting on a big macro trend, other than people wanting to continue buying homes (safe bet!). Instead, Valon is a bet on the company’s ability to build technology that reduces the cost of mortgage servicing. Doing that is sufficient for success in this space.
Mortgage servicing is the poster child for the “Spaghetti Code” problem that plagues many regulated industries. Companies in healthcare, banking, and mortgage build hundreds of wrapper systems on top of their core 30 or 40 year old system of record software and struggle to effectively integrate everything. That’s why, despite the complexity, Valon is building its own system of record software to cover all aspects of mortgage servicing.
The “Congress test” makes it challenging to apply LLMs to certain regulated use cases. In regulated spaces, it’s even more important to be able to explain why your software made the decision it did if Congress/regulators ask. The fact that models hallucinate and are not deterministic poses challenges for many use cases.
Most AI rollups will fail for similar reasons to previous tech-enabled services businesses. Andrew is bearish on most AI rollups, but thinks they’ll succeed when two things are true:
AI fundamentally changes the unit economics of the business: If your biggest cost is marketing or sales and AI doesn’t have a big impact on that, it’s probably not a good candidate.
There are network effects or other first mover advantages: Most rollups have no true technology edge, so you need some reason why you can have a durable moat long term.
Startups are about making 15 decisions with 75% accuracy and not 5 decisions with 90% accuracy: Everyone knows you need to move quickly with startups, but people still underestimate how important it is to just keep putting “shots on goal” that are good enough and not obsess over perfection.
You can be overly first-principled when building products. Building from first principles is generally a good idea. But if you lack sufficient context about a problem, you actually will build the wrong thing if you are just thinking from “first principles.”
Alcaraz > Sinner: Andrew (a big tennis fan) is predicting Alcaraz will win more grand slams than Sinner over the next 10 years (I’m skeptical!) and that João Fonseca is the new phenom to watch.
Timestamps:
00:00 Intros
01:21 The Origin Story of Valon
05:03 Why Valon built a services business, not just software
07:24 The Spaghetti Code problem in every regulated industry
12:02 How Valon uses AI and the “Congress Test”
20:21 AI Roll-ups
28:23 Fundraising lessons and why growth solves all
33:43 Managing a large operations organization
36:30 Hiring lessons
41:19 Balancing first principles with industry wisdom
43:58 The importance of quick decision-making in startups
48:03 Investing vs operating
50:43 Andrew’s thoughts on regulations
55:09 Alcaraz vs Sinner
Share this post