Stepping into HVF, it seems like any other startup office in San Francisco — though this one is near Jackson Square, not SOMA, and it houses not one company, but many. At its helm is Max Levchin, who created HVF — which stands for Hard, Valuable, Fun — as a R&D lab and incubator hybrid for data-driven startups.
HVF is “his intellectual retreat,” one he deserved after being one of the original founders of PayPal and then creating social apps Slide, which sold to Google in 2010. Levchin then stepped out of the limelight and the startup rat race to focus on his young kids and his cycling passions. The period was short-lived; after he founded HVF in 2011, the company has spun out three startups: Affirm, Glow and most recently, Homer.
While HVF is still his retreat, he’s also slid back into the life of an entrepreneur, serving as CEO of Affirm and chairman of Glow, while balancing life as a dad. I spoke to Levchin in July, two days before the release of Homer, about Silicon Valley, his love of data and why he’s betting big on Affirm.
Biz Carson: When you last spoke with Om, you were talking about the startup cycle being four years and you were looking more long-term at 10 or 11-year cycles. But since then you’ve spun out two startups from HVF and become CEO of one. So, has your view of that cycle changed? You seem to have jumped back into the rat race of it all.
Max Levchin: It has not changed at all. Both of the startups that we’ve spun out either fail on their own timeframe, or they go for a very long time to be impactful. If anything, I’m practicing exactly what I preached, which is not strange because we talked about my commitment to solving problems that I personally find compelling and critical, especially as I age and witness the human condition a little longer. Affirm, if it keeps doing as well as it is, is hopefully a multi-hundred year company.
There’s a whole sort of a constellation of aspiring bits that drove me to Affirm and drove Affirm to be created — New York has to do with multiple. I had dinner with former Senator Bradley in New York, who sort of inspired me to start thinking about the impact of the financial crisis on Americans and how lending has really seized up even though federal interest rates are real low. And, every time I go to New York, I always walk by the private banking building for J.P. Morgan. It’s a really good reminder that basically the most durable companies in America have generally been financial services. If we are going to line up ourselves against meaningful and inspirational competitors, we might as well be the guys that run the distance.
So how do you think Affirm is going to go past that four-year life cycle?
I will gladly give you extraordinary odds in the bet, so long as it is around in the sense that we could fail. Financial services company fail more often than they succeed, and they probably fail more often than the average startup because you’re playing with real money and you can’t easily pivot out of financial commitment. Trust can be destroyed a lot faster than the main stuff. So long as we are growing and succeeding and signing up users and getting merchants and everything we have plans for, I would say I more or less guarantee many, many more years of Affirm.
How much of that guarantee do you think is based on Affirm’s foundation and your trust in data?
That’s what the company is all about. I think the next great financial institution is going to be fundamentally about the dynamic relation with a bank. Today’s banking services have an entirely static relationship with their customers, both on a merchant side and a consumer side, and data is what allows for that relationship to become dynamic.
But Affirm’s looking at things like whether or not I have a Facebook profile when they’re judging my creditworthiness.
On a giant scale of things we look at, that’s a tiny, tiny bit. We actually found, not to belittle any form of data, social media profiles are far less information rich than we had hoped for. I don’t mean information rich, but predictive or information value rich in relating to or correlating to a payment.
Why were they less useful?
That’s not mine to ask or answer. Seriously, a big part of being honest with yourself about data science and data analysis is that the second you start explaining it to yourself is that you are just over fitting data. That’s exactly how we get into stereotypes. You can’t really allow yourself to make a narrative out of something you just figured out numerically. You have to think through the explanation because that’s where you find the most interesting opportunities.
A classic example instead of being so abstract: The FICO score is pretty coarse. It looks at very few variables and it says “Here, between 700-750 you’re pretty great. But 600-650, you’re damaged goods.” It’s actually really predictive, and you’d be a fool not look at a FICO score if you are a financial services product because it is in fact extremely effective. The problem with the FICO score, for example, is that you gain jobs and you lose jobs. And in economic crises, it correlates really well with people’s ability or non-ability to repay their debts.
But personal economic crises are not represented in FICO at all. FICO says, “Oh look, you missed a bill. Bummer, lower FICO for you.” But the reality is, if your bank had a dynamic relationship with you, it would say you lost your job, so here are two ways to go about this: we can wait for the FICO to update and we can up your rates so the risk can be offset by price and oh, crap for you, you’re going to have to go to a 29.99 (percent) APR.
Or they can say, wait a second, given where you are and where you’re trending and the fact that you lost your source of income, it would be a lot easier if we offered you a service or changed the terms so that we actually lowered the burden so that you wouldn’t be constantly stressed out and facing immediate financial ruin. You could paint a terrible story and say they’re too big to fail, these guys are just out to get the little guys, but it’s not necessarily true. They’re just too coarse and slow.
You said that five years ago this would have been a crazy idea, but why had people not thought of this before?
So people were trying to do stuff like this before and they inevitably run into an “oh crap moment” that prevented them from moving forward. Bill Me Later, which we think as a spiritual predecessor for Affirm, was a good product and service, but during the 2008 financial crisis, they got themselves into enough trouble that they had to sell to eBay. In some ways I feel like we’re carrying the torch forward.
Every 10-15 years, somebody says there has to be a better way than FICO. Sometimes it works, sometimes it doesn’t. At this point I’m hypothesizing, but one easy truism is Silicon Valley does live on a four-year cycle. If your company is a financial service startup and you go through a financial crisis and you survive and your customers still like you, that’s when you know you’re for real. PayPal did that, a couple times at this point, Bill Me Later did not.
So why are you betting on Affirm to weather those storm that have taken out the competition?
I think we’re in a better position to do that because we built the business around its dynamism. A deer in headlights is hopefully a little less likely for us because we’re constantly rebuilding our models, constantly figuring out the relationships with consumers that we have, constantly figuring out what is happening to someone’s credit worthiness and how we want to respond. But, time will tell. There’s no guarantee of success in Silicon Valley.
All your startups are based around data, and as more things come online they’re generating a ton of data. Do you think there’s almost an overabundance of data or data-driven startups?
It’s a little idealistic, so I have to maintain a skeptical look when I say this, but I think it’s very similar to the discovery of oil. We figured out that oil was this an amazing thing that’s in Earth and all we have to do is find the right place to tap it. This liquid would come out and we could do things with it from rubber to gasoline. It sort of fuels us as a species except we know we will run out one day. We’re not making any more oil, and we’re not making as much as we’re consuming. And data is this endless commodity that we’re generating ourselves. It doesn’t cost much to generate at all.
And as we’re consuming data, we generate more, so it’s this self-renewing source. Some of it isn’t as useful as others so it’s a bit naïve to paint it all with one happy brush, but we’re getting more and more sensors, we’re getting more and more bandwidth, we’re storing more data. The opportunity there is essentially limitless.
The downside of course is that we have lots and lots of it and in many cases we’re overwhelmed and don’t even know where to start. The human wearable devices are still in the zero-eth inning. The first inning will involve more than one percent of the population. Today’s quantified-self users are generally curiosity seekers and fitness junkies and people that are already fine. Data is something that gives them interesting conversations basically.
The problem is the sensors have to become passive. You can’t really convince someone to log their food all the time unless they’re extraordinarily disciplined. Even the most disciplined people have other occupations. I’m really good at tracking pretty much anything I want to track, but I have two kids. If one of my kids comes up and wants to do something, then the data record will be truncated or may not even happen. Most people have priorities that trump data gathering.
We’re not there yet with sensors and preventative medicine, but yet you’re trying to do that with money and data. Is Affirm on the early side of things, or is the sensor part late?
I think we’re just on time. I really think we’re living in the single most interesting time that I could have predicted as a computer scientist.
When I was doing computer science at Illinois, my kind of defining moment, which was weird because I remember it very vividly, was in a bathroom. I was walking out of it and I had been reading up on this thing called SQL, which was invented in the 70s or even earlier, and I was trying to put together: Why do I want to learn another programming language? Why does this even make sense? I like my code written in C and this SQL thing is kinda like a human trying to talk to a computer, it just seems unnecessary. The interface is stupid, but the idea of having a set math available on an arbitrary scale is profound. Being able to do set operations on giant chunks of data independent of the length of data is mind-blowingly important.
I remember walking across rooms thinking I’m going to spend my entire life perfecting this notion of set math on datasets as big as possible. Whatever the biggest data set I can get my hands on, that’s the one I’m going to work on that day. And it literally played out that way so far. I’m very excited to be living in a world where sensors are just pooping out tons of data all the time.
Here are a few video interviews of Levchin from our archives: