Pod Insights a16z Podcast 3 Ways Startups Are Coming for Established Fintech Companies -- And What To Do About It About this episode In the world of finance and financial technology, established companies are starting to look at the rearview mirror at fast-approaching startups, wondering just how true the phrase “objects in mirror are closer than they appear” might be. What is causing this acceleration in fintech startup traction? What should the incumbents be nervous about? And, more importantly, what should they do about it?
Alex Rampell, a16z general partner (and former fintech entrepreneur as the CEO and co-founder of TrialPay), joins a16z operating partner Frank Chen to discuss the changing fintech landscape. They discuss both lessons learned from the perspective of the the hungry and fast fintech startup, as well as concrete advice for execs in the hot seat at established companies.
Transcript 10,515 words · 2 speakers
0:00 Speaker 1 Well, hi, welcome to the A16Z YouTube channel. I'm Frank Chen. And today I am here with one of our general partners, Alex Rampello.
0:08 Speaker 0 I'm super excited that Alex
0:10 Speaker 1 is here. So first fact, we both have sons named Cameron, so affinity there. And then two, one of the things that I really appreciate about Alex, and you can sort of see this from his young chess playing days is he understands fintech and incentives and pricing backwards and forwards. And so FinTech has this hidden infrastructure on how do credit card transactions work. How do bonds get sold? How are insurance policies priced? And there's deep economic theory behind all of these, and Alex understands them all. So you're going to have a fun time as Alex takes you through his encyclopedic knowledge of how these things are put together. And so excited to have you. Yeah, it's great to be here. So what I wanted to talk to you about is I'm going to pretend to be in the seat of a let's call it an incumbent fintech company.
1:02 Speaker 1 So I'm a product manager at Visa or a Geico. And I am looking in my rear view mirror, and there are startups in the rear view mirror. And I'm very nervous that the startup in the rear view mirror, exactly as the mirror says, objects in mirror, maybe closer than they appear, is like, wow, they are catching up to me faster than I really want. And so I want to understand like what are startups doing? Like, how would they mount an attack on me, the incumbent? And we're going to talk about sort of wedges they they can use. Right. And then that's sort of the first half, like, how are they coming after me? And then the second half, let's talk about like, and what should I do about it? So that's sort of the premise for R.
1:48 Speaker 1 So why don't we start with the the attacks? Like, how would a startup come for me? And one way they come for me is they come after my best customers. Right.
1:58 Speaker 0 So this is the interesting thing about financial services in general, because you know, there's a sharp television hanging on the roll wall. And Sharp knows that they make more money every time they sell an incremental television. So more customers equals more money, cause effect. And the interesting thing is that for many kinds of financial services, that is not true. Because what you're really trying to do is assemble a risk pool.
2:22 Speaker 1 And
2:22 Speaker 0 the best example of this is insurance. So what is car insurance? Car insurance has good drivers, okay drivers, and bad drivers. And effectively, your good drivers and your okay drivers are paying you every month to subsidize the bad drivers. The same thing goes for health insurance. You have people that are always sick, you have people that are always healthy. And if you are an insurance company that only provided insurance for very, very sick people, or if you're a car insurance company that only insures people that get into accidents every day, there's no economic model to sustain that. You actually have to accumulate the good customers and use them to pay for the bad customers. And the interesting thing about this is that from the perspective of the good customer, it's not fair. And I'm not, I'm not talking morally or philosophically, but just from a from a capitalist or economic viewpoint, it's like, okay, I want life insurance, and um I eat five donuts a day.
3:16 Speaker 0 I just had a donut today. I don't need five a day, but I have one donut every Friday as you can testify. Right. Um and then I have a friend who goes to the gym five times a day, never eats a donut. That guy's probably gonna live longer than me. Hopefully not, but probably uh probabilistically, he's probably going to have a better time than I am in terms of life expectancy. So why is it that we both pay the same rate? And that just seems unfair to him. It seems great to me. Right. Because he's subsidizing me.
3:44 Speaker 1 Yep.
3:44 Speaker 0 Gym guy, subsidizing donut guy. Exactly. Exactly. And that seems unfair. And then the startups can sometimes exploit that psychological uh uh unfairness, like that that feeling of unfairness. So and it kind of does two things, because from the big company perspective, if you were to take away, think of it as a normal distribution. So most people are in the middle and they're just gonna live whatever to the to the average of 79.6 years or whatever it is right now. Some people are gonna live forever. They're the ones that you know have the olive oil go to the gym and do whatever it is that they do that uh that makes them live a long time, great genes, and then some people are gonna die early. And from the perspective of the startup, if you can get all of the people that are going to live much, much longer, you're going to be more profitable. Or the same thing for car insurance.
4:30 Speaker 0 If you can get all the people on the good end of that distribution curve, you're going to make money. And then the nice thing is that if you're starting a brand new company and saying, hey, I give you a loan if you can't get a loan, who's going to sign up for that? People who might be bad.
4:45 Speaker 1 If I
4:46 Speaker 0 say, I'm going to give you insurance if you can't get insurance, who's going to sign up for that? The people that are eating all the donuts. And that might not be very good. So it actually has this nice kind of uh symbiosis between if you do it correctly, you get positive selection bias, and that you you establish a new criteria. Part of that new criteria is based on data, but part of it is based on psychology.
5:07 Speaker 1 And
5:07 Speaker 0 the psychology is I'm treated unfairly, I want to be treated more fairly. That yields a lower price for people for pretty demand-elastic product. So I say I can get life insurance at half the rate because I'm going to the gym. That sounds great, that sounds fair. But to answer your question, what the incumbent might be left with is not uh half of the number of customers. Like that could be the case. It could be half the number of customers, but it could be half the customers, and all of them are like entirely unprofitable.
5:35 Speaker 1 Right.
5:35 Speaker 0 They took up all the profits. They didn't have to take all your customers. They just had to take the good ones. Right. So actually, and if you just take in the the funny thing is that because not it's not like I want to get, oh, Geico has X million customers, I want X plus one million customers, you actually might want one-tenth as many customers as Geico.
5:51 Speaker 1 Yeah.
5:52 Speaker 0 Because if you can just get the good ones, I mean, what if you give people a 50% discount, not a 15% discount, like Geico always advertises about, but a 50% discount on their car insurance, and these are the absolute best drivers in the country. How many claims do you have to pay out on the best drivers? You might have to pay out nothing. Literally nothing. And if you have to pay out nothing, and there are these uh mandatory loss ratios for different insurance industries, so I don't want to get into that. But yeah, imagine that unregulated, you can pay out nothing. Consumers feel like they're treated very fairly, they're rewarded for better behavior. This begets positive selection and not adverse selection, then you're going to have the most profitable lending company or insurance company in the world, because it really is a unique industry where more customers is actually worse than less but more profitable customers because each incremental uh customer is like a coin flip of profit or loss.
6:45 Speaker 0 Yeah. Might generate profit, might generate loss. And that's not true for the vast majority of industries. Like Ford never sells a car saying maybe we'll lose money on this customer. Right.
6:55 Speaker 1 Right. They just like, I I need everybody to buy a Ford F150. If you don't buy an F150, I need you to
7:00 Speaker 0 buy this other thing that I said, the expedition or whatever. They might lose money on the marginal customer until they hit their fixed costs. But they're never going to have a coin flip of when they sell the car, hmm, we might maybe we shouldn't have sold that car. But that's what every insurance company has when they underwrite a policy. And that's what every bank has when they underwrite a loan.
7:18 Speaker 1 Yeah. So auto insurance companies need to find people like me. I have this old Prius, right? First, it's you know hugely reliable car, and you know, and then I drive like a grandma because I'm optimizing for fuel efficiency. So like, you know, I rarely go above 65, and so like rarely safe. I've never filed a claim. They need more customers like me. Right. And that's what drives the profits. Yes. Right. Because there's no payouts.
7:41 Speaker 0 Well, not only does it drive the profits, it actually subsidizes the losses because there are a lot of people who are the inverse of you, and you're paying for those people, and the transfer mechanism is through geico. Yeah.
7:54 Speaker 1 Um I saw an ad in my Facebook feed recently for HealthIQ, and I think they're doing something like this too, right? So the I think the proposition was, hey, can you run a mile in less than nine minutes? Can you bench press your own weight or something like that? There's all these like, ooh, healthy people. Uh and is that the mechanism they're exploiting? It's exactly that. Um
8:11 Speaker 0 I would say the first company to probably do this on a widespread basis in FinTech land was SoFi.
8:17 Speaker 1 And
8:17 Speaker 0 SoFi said, hey, you're really smart. They actually coined this term, they called it the Henry. High earning not rich yet. Because if you look at how student loans work, it's like everybody gets the same price on their student loan. Right. And it doesn't matter what your major is, it doesn't matter what your employment you know uh prospects thinking, what your employment prospects are, everybody gets the same rate. You get a look, you get this rate, you get this rate, you get this rate, because a lot of it is effectively underwritten by the US government. And that's not so think about it again from the twin pillars of psychology, where I mean psychology of the borrower. Like, how come I'm paying the same rate as that person who's going to default? That's just not fair. I'm never going to default. In fact, I'm going to pay back my student loans early, right? So that helped. And then again, positive selection versus adverse selection, because refinance has this concept in general, because I would say if you're planning on declaring bankruptcy, or if you're saying I'm going to uh I'm going to join Occupy Wall Street and never pay back my loans, and I hate capitalism, why would you go refinance?
9:18 Speaker 0 It just doesn't make sense. Right. Um, because you're just going to default. Right. So if you raise your hand, and actually it's interesting, even on the other side, there are a lot of companies in what I would call the debt settlement space. And this is something that most people don't know about. But if you listen to like some interesting talk radio, you'll hear all these ads for debt settlement. And what is debt settlement? It's saying, hey, do you have too much debt? Um if you call us, we will negotiate on your behalf and pay off your debts, and then you just owe us. And you kind of need this intermediary layer because imagine that you owe 10,000 to Capital One and you can't pay it back. And you call Capital One. It says press one for your balance. Press To get a new card mail to you. Press three if you don't want to pay us the full amount and want to pay us less. Everybody's going to put it.
10:04 Speaker 0 Everybody's going to be able to do that. Everybody's doing it.
10:07 Speaker 1 This is why they don't offer that option. They don't offer that option, nor will they ever.
10:10 Speaker 0 However, on talk radio, and these this is very big in the Midwest, like you'll hear, you know, freedom financial. Go call Freedom Financial and we will settle your debts for you. So they call Capital One and say, look, Alex can't pay you back. We'll pay you $2,000 right now, and then you're going to get rid of the loan. And you're like, well, we're not happy taking 20 cents on the dollar, but it's better than zero cents on the dollar. Fine. We'll take it. And then you owe Freedom Financial the 20 cents. But why do they feel comfortable underwriting that? Because you you rose your hand, you know, said, I want to get out of debt. And that that's positive selection bias right there. Because people who are just deadbeats, because you know, behind every credit score, if you think about how that works, it's willingness and ability to repay.
10:55 Speaker 0 And the psychological trait of the willingness is in many cases as important as the financial constraint of the ability. Because if I if I owe a million dollars to somebody and I only make a hundred dollars a year, it doesn't matter how honest I am, I can never pay that back. Right. It doesn't matter how long, I mean I could live 10,000 years and I guess I could pay it back. But otherwise I can't pay that back. But the willingness to repay is interesting. And that's that's very important. And that's again this kind of psychological trait that's captured in this idea of positive selection. So what did SoFi do? They kind of again hit this this twin pillar, which is I want to only get the good customers, I'm going to reprice them and steal them from the giant pool that again, normal distribution, these are the losers, these are the whatevers, and these are the people that you have no risk on whatsoever.
11:44 Speaker 0 Let's steal all of these people over here.
11:46 Speaker 1 Yeah.
11:47 Speaker 0 And it makes them feel good. It's a better marketing message because it's differentiated. Like how do you compete with everybody? It's like, hey, we're just like Chase, but smaller and a startup and not profitable and you probably shouldn't trust us. Bad marketing message. Right. Good marketing message is you're getting ripped off. We're going to price you fairly, come to us. Yeah. So SoFi did this for lending. Yeah. And what did health
12:11 Speaker 1 IQ do? So
12:11 Speaker 0 health IQ did this for health, or really for life insurance. So they started off with a health quiz. Because they, I mean, it seems almost self-evident that uh healthy people are healthy. I mean, it's it's a taut, it's a tautology. Like healthy people are healthier than not healthy people, but can you actually prove this
12:28 Speaker 1 from
12:28 Speaker 0 a life expectancy perspective? So they started off with just recording data and then building a mortality table. And it turned out that this uh that you know that what I would assume is a prima facie case turned out to actually be correct, which is these healthier people do live longer than not healthy people, and then they turn that into both a positive selection advertising campaign with which differentiated them from a brand perspective, but also left them more profitable. So what they do is they say, yeah, can you run a nine or an eight minute mile? Can you do these things to prove that you're better than everybody else? And why is that important? Well, from their own balance sheet or profitability perspective, they want to get these good customers versus you know a brand new life insurance company that said, hey, life insurance takes too long to get, it's a big pain, and it's expensive, we'll underwrite you on the spot in one minute, no blood test.
13:18 Speaker 0 That's gonna be adverse selection. Right. That's like, ooh, I think I'm gonna die soon. Right. I want to get everybody rejected me for life insurance, I'm going to that company, as opposed to here, they're only getting the customers that kind of hit um that think they're gonna hit the underwriting standard, which is great. Um they think it's fair, so it's a differentiator from a brand perspective, and then it turns out that again, each marginal customer in insurance is kind of a coin flip, they're getting a weighted coin because they're only getting people on the far right side of this normal distribution.
13:51 Speaker 1 So wedge number one is exploit psychology, right? Uh positive selection rather than negative selection, and what you'll end up with because of this sort of unique dynamic of the fintech industry is you end up with the most profitable customers. What's wedge number two? We're gonna talk about sort of new data sources and what startups can do to sort of price their products smarter than encumbrance.
14:12 Speaker 0 Right. So imagine that you have a group of a hundred people, and of the hundred people, half of them are not going to pay you back. So think of this as the old combinatorics problem of you know, bins and balls. So you've got this giant ball pit, you scoop up a hundred balls in your bin, and uh half of them are going to be bad, half of them are going to be good. So what's a fair rate of interest if you're a lender that you you have to charge this whole bin if half of them are going to default and you assume that you can't lose money? Well, the answer is going to be 100%.
14:46 Speaker 1 Oh, right. Because half of them you you have to make up for all the things. So half of
14:50 Speaker 0 them, you know, you lose all of your money, half of them you double your money, so you're you're back to square one. Now you're
14:55 Speaker 1 even.
14:55 Speaker 0 Now you're even. Um so the problem is that that's that's not good. Yeah. Because, well, in the United States, you can't charge 100% interest. Right. It's called usury. Right. Um, there are other parts of the world, again, illegal, step one. You're up illegal. So that's a problem. But what if you can use different data sources to, again, it's not positive versus um adverse selection, as in for some of the insurance companies, but it's saying, can I collect more forms of data so that instead of saying the only way that I can make my operation work is to charge an interest rate which actually turns out to be illegal, can I come up with more data sources that effectively, even though discrimination sounds like a terrible word and it's normally used in that construct, if you discriminate against criminals, that's fine. I mean, some of the people that try to take advantage of lenders are actual like organized crime.
15:41 Speaker 0 You don't want them in your bin, right? You want to throw them out. How do you how do you take more data sources and actually start measuring this? And the interesting thing here, and it's somewhat unfortunate, but um you have a giant market failure happening in many different regions of the world because in the United States, like the top interest rate that you can charge, it's regulated on a state-by-state basis, but Utah has a 36% usury cap. So a lot of people export that cap. Um that's a lot less than the 100% that I was mentioning. And there are lots of ways of kind of gaming that system, and you charge late fees and you charge this fee, so it actually might end up looking more like 100 or 200%. But um, so you can't charge more than 36%, and then you actually can't use certain types of data if they are prone to having an adverse impact. So if you think about how machine learning works, I always kind of describe it somewhat over simplistically as linear algebra, where I have here's every user that I've ever seen, here's every attribute that I've ever measured, and what I'm looking for is like strange correlations that I can't even explain.
16:40 Speaker 1 So
16:40 Speaker 0 I'm gonna ask you, uh well, I'm not even gonna ask you a lot of these things. It's like how long did you fill out this field for on my loan application? Right. Did you enter all caps or not all caps? Like just all of these different things. You take
16:50 Speaker 1 the slider on how much do you want and jam it all the way to the right, right? All
16:54 Speaker 0 of
16:54 Speaker 1 these things.
16:54 Speaker 0 I can ask you, do you have a pet or not?
16:57 Speaker 1 That
16:57 Speaker 0 might be interesting. I don't know if that's a leading indicator of default or not, but I want to collect all these different variables, and then at the end of the day, I'm going to see default or not default. That's the output, and then I'm going to see what's correlated with that. And it's a little bit of this, it's a little bit of that. I can't explain it, but the computer can. Now the problem is that in the United States, you actually can't do this because it might have an adverse impact. And what is an adverse impact mean? Uh there actually was outright in terrible discrimination in lending in the United States, but there's unfortunately terrible discrimination in many things in the United States, but lending was one of one of several or one of many. Uh so imagine that I said, are you married or not? Oh, you're not married, I'm not gonna make you a loan. Well, that's illegal now.
17:42 Speaker 0 Um are you this race? Oh, I'm not going to make you a loan. Well, that's illegal now. So what did people do to get around the people that were actual racists or actual, like maybe they weren't racist or discriminatory at heart, but they were picking up on cues, they'd say, Oh, what part of town do you live in? Oh, you live on that part of town. Well, that's like 100% correlated uh with this race or this gender or this, that I'm not going to make you the loan. So the law was strengthened, so there's a law called fair lending in the United States, and then one of the components of it is this idea of called adverse impact. And it's different than adverse selection, it's saying, I don't care what you said you did for why you rejected Frank for a loan. Um, if it turns out that everybody in your reject pile has a disproportionate, you know, gender ratio, race ratio, something like that, I'm going to assume that you're underwriting standards or having an adverse impact.
18:32 Speaker 0 So
18:33 Speaker 1 you as a bank couldn't say, hey, look, I asked him if he had cats, and I'm using that to make the loan decision. If it turned out that having cats was correlated with being in particular race, um they couldn't use the cats answer to deny you a loan.
18:50 Speaker 0 Correct, because that was, and in all fairness to the law, this is what people used with your geography. What zip code do you live in? Oh, you live in that zip code? Right. 100%, you are a member of this particular race, and the intent all along was to discriminate against people of that particular race. Um but now instead of using loan officers that use, you know, God knows what to decide, do I want to make you the loan or not? You're using a computer, you can look at the code. So I think there is a lot of uh there there are some anachronistic laws that have to catch up here, but let's take an area outside of the US to answer your question where perhaps you don't have interest rate caps. Um because you know, the the thing that a lot of people say, oh, you know, 200% interest is terrible. 500%, that sounds that sounds awful, you should go to jail for that.
19:37 Speaker 0 But what does APR mean? APR stands for annual percentage rate. And uh, what if I'm giving you a four-day loan? So I say, okay, I'm gonna loan you nine dollars right now, you don't look very trustworthy. Uh I want you to pay me back ten dollars on Monday. Yeah.
19:51 Speaker 1 That doesn't sound so bad. Yeah, it's like
19:53 Speaker 0 you're gonna pay me a dollar. But what is that on an APR basis? That's like 9,000 percent.
19:59 Speaker 1 Right.
20:00 Speaker 0 Made that up. But it's it's probably about that, right? Because it's 10% every four days or every three days, 10% every three days, and that accumulates, like that's a lot of money or a lot of interest on an APR basis, but it's the wrong metric because effectively it's like trying to figure out what your marathon time is based on your hundred meter dash. Like the winning marathon time would be an hour. And that's not true. We know that nobody can run a two-hour marathon right now yet. So maybe Angela can. Maybe Angela. So uh so there's a company that we invested in called branch. And what they're doing is they just collect every form of data possible and they look for these strange correlations. And the interest rates on an APR basis might be high, but they're really charging like a dollar. Uh these are small loans. They're very, very small loans.
20:45 Speaker 0 So I loan you, and actually the other interesting, like one of the nice data points that they're accumulating over time that is a really interesting idea, I think. It's not new. In fact, it's almost back to the future old, where they loan you a dollar, if you pay it back, they loan you two dollars. If you pay it back, they loan you four dollars. If you pay it back, they loan you ten dollars. And they they ladder up your credit and they keep that information proprietary to
21:10 Speaker 1 them.
21:10 Speaker 0 Um because induction turns out to be a pretty good formula for figuring out not so much the ability to repay, but the willingness to repay. You've established a pattern of willingness to repay, but they also look at where were you today? Um again, you provide all this information in order for them to crunch this in order for them to give you a loan at ideally a lower rate, because the more information because it's kind of twin pillars, right? Yeah. The less information we have, the higher the rate that we have to charge. Not because we're evil, but because otherwise you're gonna have a market failure, like you have in lots of. It's a big ball problem, right? Exactly. Because
21:44 Speaker 1 you have no idea how many deadbeats.
21:46 Speaker 0 Exactly. And if I don't have any idea, I either have to charge a high rate or not charge anything at all, and not charge anything at all doesn't mean like everybody gets a zero percent loan. It means I don't make any loans. Right. And like both of those are bad uh outcomes. The better outcome is you accumulate more data and you figure out here are the good people, let me not accept the bad people, because again, the way that the good people end up paying more money is if the company starts accepting more bad people, because it goes back to what I said at the beginning, which is more customers in this unique industry often is bad if you don't understand how to select them correctly, and for many of these newfangled lending and insurance companies, the default customer is going to be adversely selected because if you're a new lender and you have no underwriting standards, you're basically your advertising free money never pay us back.
22:33 Speaker 1 Yeah.
22:34 Speaker 0 And those are the people that will be attracted to you, both the criminals and the non-criminals in droves. Yeah.
22:39 Speaker 1 So this is sort of startup attack wedge number two, which is I'm gonna generate a new data source that allows me to price my product in a way or reach a customer that a traditional company would never even try, or they don't have the data source, so they have the bin and ball problem. So what are the types of data that branch went to go get to try to figure out should I give you a loan of a dollar or two? Well,
23:01 Speaker 0 the the other type of data. So branch was somewhat unique in that they said we're going to get data from your phone. Um it seems odd. Because like most lenders in the developed world, uh, or not developed versus undeveloped, it's really like with developed credit infrastructure. Yeah, if there's a credit
23:20 Speaker 1 bureau.
23:20 Speaker 0 They look up your credit report. If it's good, they make you a loan, if it's bad, they don't make you a loan. It's actually not that hard. Um there are all sorts of nuances that you can layer on top, but this is how it's been working for a long time in the United States as an example. Whereas there it was like, okay, uh, where did you work today? Did it look like you work today? Right. Um so it it was stuff like that, and even like how many apps do you have on your phone? Like weird stuff that you would never assume actually has any kind of indication of willingness or ability to repay, but in many cases it does. Like, are you gambling? Um, well, if you have a gambling app on your phone, you're probably gambling. Maybe that's good. Yeah, maybe it's bad. It's actually not making human judgments. It's and it's also not looking at any one of these unique variables as a unique variable.
24:05 Speaker 0 It's looking at them in concert and then correlating them with these outcomes, or really observing the outcomes and then linking them back to all of these different inputs.
24:15 Speaker 1 Yeah, I remember talking to the team when I was researching my last machine learning presentation, and the fascinating things that I found were if you got more texts than you sent, you are more credit worthy. If you had the gambling app, you were more credit worthy rather than less, which is not kind of what you would expect. Uh if you burn through your battery, you were more likely to default, right? Uh so like all of these things where human alone officers would never really guess. Right. Um they probably would guess the wrong way. Right. Because many of them are
24:45 Speaker 0 counterintuitive. Right. And then many of them are not, they're not uh unilateral. Like so it's not just I mean, I don't know, but it's not just the battery thing, it's the battery thing with this, with that with that. And it's like, you know, humans can only really observe three dimensions plus time, so I guess four. Right. And these are you know, 9,000 dimensional problems. So it's just it's much, much more challenging for humans to really grok.
25:08 Speaker 1 Yeah. Got it. So that's sort of the second category of attack, which is you generate a new data source, and then that allows you to price or find customers in sort of a more cost-effective way. Um let's talk about the third, which is around um sort of fundamentally changing behavior. Right. So why don't you talk about maybe earnin' is a good example of this?
25:30 Speaker 0 Yeah, so if you assume that humans are static, so they're they're they're born, both of our cameras were born, and their DNA is set upon birth. Uh maybe it changes a little bit with some mutations from some gamma rays here and there, but it's it's set upon birth, and then human behavior never changes. And that's one way of looking at things. And then you can think about adverse selection versus positive selection. Good drivers are always good drivers, bad drivers are always bad drivers, let's just get the good drivers. So the other the other category, and it's not to say that these other two groups don't do this, but um if I look at a company like Earnin, um, most payday lenders are reviled
26:05 Speaker 1 because they
26:05 Speaker 0 charge high fees, they don't educate their uh their borrower very well. Um, it actually provides a valuable service because if I'm getting paid next Friday, but my rent is due today and I don't have money, do I want to get evicted? No. Right. Uh I want to get paid right now, and the only person that does this is the payday lender, but the payday lender is competing with other payday lenders for advertising in the local newspaper or something. Right. And if they're able to rip me off more, not because they're evil, but because they have to you know afford the advertising spot, they're now incentive to do so. So it's just it's a it's an it's a vicious cycle. Um so let's talk about Earnin. So what Ernin does is they say, okay, uh, we know that you've worked this long. So again, new data source, because the phone's in your pocket and you work at Starbucks and you're getting paid hourly, and we've seen the phone in your pocket or in your locker in the Starbucks office and you know, nearby the barista counter uh for eight hours.
27:00 Speaker 0 So you worked, we saw your last paycheck hit your bank account, we know that that's where you work, we're not taking your word for it. We have real-time streaming information about this. Um we will give you your money whenever you want. Not money that you haven't earned yet, but money that you have earned, but you actually haven't gotten paid for yet. And then you can tip us. There's no cost. If you want, you can do it. No
27:23 Speaker 1 interest, no fee, nothing.
27:25 Speaker 0 If you want to pay us nothing, that's fine. I mean, we would appreciate it if you pay us something because obviously we're providing a valuable service for you. And then you can even give tips for your friends. There's this community that's really emerged of people on Earn in. And actually, if you look back, a different business model, but uh this idea of microfinance in general. So if you think about Mohammed Yunus and what what he did, um, this idea of can you encourage people to pay back loans using social pressure? So again, not adverse selection versus positive selection, but actually trying to force everybody down positive behavior. Let's get
27:59 Speaker 1 the community to encourage repayment.
28:02 Speaker 0 Right. Um because then uh saying, or like let's get the community to encourage people actually driving safely. Because there's underwriting at the time of uh admission. There's underwriting based on ongoing behaviors. So like many of the car insurance companies that are brand new are saying we will re-underwrite you, but like, yeah, if you drive like Frank when you signed up, great, but now you switched into like race car driver mode and you were trying to hack us, but we're actually monitoring your speedometer at all times. So guess what? You got a higher rate now, so that might encourage you to drive safely. If I'm if I'm Frank and I drive safely in my Prius, um, but then I decide, and then I got a really good rate on my car insurance as a result. Right. And now I'm like, aha, I game the system.
28:48 Speaker 0 Now I'm going to drive like a maniac.
28:50 Speaker 1 Right.
28:51 Speaker 0 Uh well, the nice thing is that you can make underwriting dynamic and you can say, all right, it's we're actually going to re-underwrite you every day. Right. So we had the positive selection to try to attract the Franks. We have the continuous evaluation to try to uh encourage the right behavior post-Frank sign up, and also to you know stop the gamification of it's like I'm gonna pretend to be safe and then be like a maniac. Um but then how do you actually get what if Frank was a bad driver initially? Uh doesn't fall into my positive selection loop, but I still want to try to make Frank a better driver. Yeah. If I could turn him into a good driver,
29:30 Speaker 1 he'd be profitable. Right. So
29:33 Speaker 0 that that's the flaw with kind of wedge one and wedge two of like creaming the crop, or really wedge one, which is we're gonna cream the crop, we're gonna do what SOFI did, we're gonna do it with health IQ. I mean, it's a great strategy. But the rest, again, if you assume that uh it's it's all nature and there's no nurture, then perhaps there's nothing you can do. But if you can actually try to nurture better behavior, you actually see better, you do see better behavior, and then the profitability goes up. Up. So the interesting thing there is that you're still finding mispriced customers, but you're actually helping turn them into correctly priced customers. So, you know, somebody like a bank would turn away that customer and say we don't want them because they have a 500 FICO, which is really bad. And then you have to figure out, and as would all of the new startups that are saying we only want the best customers, we want to leave the banks with the bad customers.
30:26 Speaker 0 But it's kind of the twin pillars of can you identify something that's below that credit score or below that driving score or something, and then can you encourage positive change? And if you can, then you can start actually creaming the crop of the bottom half of the customers. Right. Right, not even the bottom half. It's the customers that are just neglected because nobody wants to underwrite them. And then you do that, you take them on because you have a secret to change their behavior. Right.
30:54 Speaker 1 You're seeing a lot of companies that sort of are using behavioral economics research to figure out how do I nudge people into better behavior, right? And so this would be an example of how you're trying to change behavior to get the profitable customer.
31:09 Speaker 0 Right. So, you know, uh there is one company in the lending space a while ago called vouch. I think ultimately it didn't work, but um when you apply for a loan, it actually kind of taps your social network and it requires that they do a reference for you. Either a reference in terms of like, yes, Frank is a good customer, you can trust him. Right. And even kind of a co-commit. So I'm getting a loan for a thousand dollars, and you say, yeah, Alex is okay, or I'm I'm saying Frank is okay, and uh if he doesn't pay you back, I will I will put a hundred dollars in. Right. Because that's how confident I am. And it's not all thousand, but it's a hundred, and then you're my friend. I go bowling with you. Right. We go take our cameras out together, right? And if you don't pay back this thousand dollars to this kind of faceless, large, evil corporate entity, not really, but if you don't pay that back, I'm on the hook for a hundred bucks.
32:02 Speaker 0 Yeah. I'm not going bowling with you anymore. Right. So there are other things that are really interesting to try to encourage the correct form of behavior. Yeah. Um when, and actually, part of it is just making it personal. Like that this was the whole unus theory, which is if you are kind of held accountable by your peers, that is so much more powerful than getting a collections call from Citibank. Yeah. Like you're like, oh, that's the collections number, iPhone block.
32:28 Speaker 1 Right.
32:29 Speaker 0 Done. Right. But how am I going to block my friends out? Right. If Alex calls me
32:32 Speaker 1 and says,
32:33 Speaker 0 You
32:33 Speaker 1 really got to
32:34 Speaker 0 pay that loan back,
32:34 Speaker 1 otherwise I'm out
32:35 Speaker 0 a hundred
32:35 Speaker 1 bucks, right? That's much more powerful. I mean, this has worked great for a lot of health in a different domain, right? Which is the if you are trying to get a pre-diabetic patient not to get diabetes, the most effective thing to do is lose something like six or seven percent of your body mass, and the way they do it is they get you into a group. They mail everybody a scale. Right. Everybody sees your weight in the morning, right?
32:57 Speaker 0 Like that's a powerful motivator. Yeah, I mean, that this stuff, psychology is very powerful. So there are a lot of tricks that you can use here. And if you understand the impact of them, you actually have to reassess your entire branding and customer acquisition strategy. Right. Right.
33:13 Speaker 1 All right. So remember I opened up pretending to be the product manager at Visa. And now we've gone through all of these three categories of how the startups are coming for me. And like I'm starting to sweat here, right? They can come get my best customers, they can generate new data sources that like I would have a hard time doing. They can actually even go after sort of worse customers, change their behavior, turn them into profitable customers. I'm scared now. Like, what in the world should I do? Like you're in my seat, you're the head of innovation or head of strategy or head of digital at one of these big thin tech companies. What should I do with with respect to startups?
33:51 Speaker 0 Well, I think it's actually very hard for a company that's trying to be all things to all customers. Um because if you look at what SoFi is, look at SoFi's brand. Brand is, you know, we are the high, like if you're great, you're good enough for us. If you're Henry,
34:03 Speaker 1 right? If you're a Henry, you're
34:04 Speaker 0 good enough for us. Health IQ, if you're healthy, you're good enough for us. So on that sector of the curve, you know, how does Geico say, hey, if you're a good driver, go to this special part of Geico? If you're a regular driver, you still save 15%. If you're a bad driver and you had a DUI, well, we can we can cover you over here. It's just it's a vi it's lost in this kind of giant Geico, you know, gecko marketing message. So in many cases, it actually helps to have sub-brands and divide this up, which is is somewhat anathema to a lot of companies that want to say, how do we get as much efficiency and synergy as possible? We're gonna have one overarching brand. And you know, one of one of my favorite examples of this kind of different industry, but the highest end of the highest end of jewelry is Tiffany and Co. or one of the highest ends at the highest end.
34:51 Speaker 0 And for a long time, it was owned by Avon.
34:54 Speaker 1 No, really?
34:55 Speaker 0 You know, the Avon lady Avon. So uh, and if if Avon bought Tiffany, which they did,
35:00 Speaker 1 and
35:01 Speaker 0 they said, okay, we're gonna rebrand Tiffany and Co. as Avon, yeah, like that doesn't work. Yeah, yeah. Like you're not gonna get, you know, 80% gross margins on whatever they sell at Tiffany and Co.
35:11 Speaker 1 Right. Uh for breakfast at Avon's just doesn't have quite the right ring. It doesn't work.
35:15 Speaker 0 And then for Avon to say, okay, you know, the the door-to-door salesperson or sales lady with the pink Cadillac that's going around like we're now going to have her push, you know, $2,000 bracelets as opposed to the normal $10 fare, like that's not going to work either. Yeah. But it actually can make sense if you want to just appeal to more customers, you have different brands, and you don't want to all suck them together. So you can imagine instead of having, you know, Geico could be your generic brand, but then you could have uh like I I think I mentioned this to you once before, a friend of mine is Mormon, doesn't drink alcohol and says we should have Mormon insurance for cars, because it's just totally unfair. Again, going back to the psychology point, like why is it that I'm paying for the drunk idiot that goes through the stop sign?
36:01 Speaker 0 I don't drink, I can prove that, I will never drink, I have a million friends just like me that will never drink. We should all get car insurance, we should all get a 40% lower rate. Do they think of geico when they go there? Maybe they could, but it could be like Mormon current, or sorry, I'm not good at branding. But you can have a separate brand for all of these separate subgroups and have the same underlying infrastructure behind all of them. But again, part of this is just how do you brand and how do you market effectively? Because if you look at the efficacy of health IQ ads or the efficacy of SoFi ads, they're so much higher, because again, you have this large group of people, or many cases small but valuable groups of people that feel like they're being treated unfairly. So yeah, Geico is save 15% on auto insurance, click here.
36:48 Speaker 0 Mormon car insurance uh advertised to LDS members in Utah, shooting fish in a barrel, that's gonna have a dramatically higher click rate. And then many of these products are also very demand elastic. So I'm not saying save 15% on car insurance, I'm saying save 80% on car insurance. It's very easy to do, click here, positive selection bias. Uh that's gonna work better than like Geico, but we also have something for Mormons too. Right. Yeah.
37:12 Speaker 1 The goal is to find the LDSers and the hyper milers who are really safe, et cetera, et cetera. Right. And so it's very counterintuitive because if you're at a big company, you're thinking scale, how do I get the next increment of revenue growth or profit, and you're saying, actually go the other way. Don't try to make your single brand bigger. Try to think about a dozen sub brands, each going after sort of the perfect market for them. Right. How do
37:36 Speaker 0 you positively select into a submarket? Well, the uh the other uh side effect of this is that you know, part of the uh the asymmetric warfare that some of the startups have is that if you wanted to kill Geico, you wouldn't you wouldn't steal 100% of their customers. Because if you did that, that would almost be too obvious. You'd steal 20% of their customers, but only the good ones. So imagine that Geico could actually devolve or evolve, depending on your point of view, into 10 sub brands. There's no more geico, but it's just like the 10 sub brands basically select for the right types of customers or even help uh judge and improve behavior from other subsets of customers, and then expel the 30% that are just bad news. Right. And if you can expel the 30% that are bad news, you might say, okay, well, all of this dissynergy of going from one brand into 10 sub brands, well, that was idiotic.
38:25 Speaker 0 Because now I have fewer customers, but actually, no, it isn't. Because you might have fewer customers, but it's not like selling widgets. You're selling probabilistic widgets where in many cases you have negative gross margin when you sell a widget. So it's important to figure out how do I get the good ones, uh, keep the good ones, and then get rid of the bad ones. Yeah.
38:46 Speaker 1 So that's one strategy, which is sort of sub-brands and sort of customer segmentation. What if I've been told by main management team, go find a bunch of startups to work with, right? Sort of somehow figure out a marketing or um co-selling relationship so that we can start experimenting with some of these new models and we can keep an eye on the startup community so that maybe you know we can put ourselves in the best place to buy them if it turns out working. Is there a
39:13 Speaker 0 a way to do that? Well, there are many ways to do that. The probably the easiest way that is often counterintuitive for a lot of big companies is I I call this the turn-down traffic strategy. So Chase turns down a lot of people for loans, either because, again, it's the bin and ball problem where it's like, well, you might be good, you might be bad. Sometimes it's not even that. It's like we think you're good, but we just can't profitably underwrite a $400 loan. But Chase has all the traffic. So what is turn down traffic? It's saying, okay, we rejected you. Hey, here's a friend that you might like.
39:45 Speaker 1 So
39:45 Speaker 0 this is not cream of the crop. This is the bottom tier on the ingestion point for a big financial institution saying we don't want you, which kind of is kind of a mean thing to say, uh, a way to ameliorate that potentially is saying we don't want you uh Because we're not smart enough to, hey, it's sorry, we're working on it, all our systems are down, but here's a great startup that does. Now, why would you send customers to a startup? Well, the number one thing, like Geico spends 1.2 billion dollars a year on advertising.
40:13 Speaker 1 Yep.
40:13 Speaker 0 It's really hard to compete with that from a So if I could not spend a dollar of advertising but give 90% of my net income to Geico as a startup, I still might make that trade. I mean, we don't always like this because we want to see do you have your own acquisition strategies, your own acquisition channels, you're not dependent on the big company. But from the big company's perspective, turn down traffic is often brilliant. Um because it's saying here's somebody that knows how to underwrite better than we do, or more profitably than we do. We're going to send our customers, you know, otherwise, what happens? Um, and this is what I think Amazon got right in an era where everybody else got this wrong. Amazon said, okay, you're on Amazon's website and you're looking at the Harry Potter book. And then right next to our Harry Potter book is an ad for Barnes and Noble for the Harry Potter book.
40:58 Speaker 0 Barnes and Noble is like, this is amazing. We can buy ads on Amazon's website. They're so stupid. We're buying ads and stealing their customers. But every time you click on that Barnes and Noble ad, Amazon made a dollar, and it's a hundred percent gross margin. They share that with nobody. There's no cogs on that. And then they can use that dollar of pure profit to lower the cost of their Harry Potter book. Which actually made more people want to go to Harry Potter. Go to Amazon to look for Harry Potter than go to Barnes and Noble that said, we're s we're locking you within our walls. It's like a casino with no clocks, and we're gonna pump oxygen in. So what a lot of big companies don't get is that Google is just one click away. Like, why give all the excess profits to Google when I go to Chase, I get turned down for a loan, and then I go back to Google and I say, where else can I get a loan?
41:43 Speaker 0 Well, Chase should be sending you there.
41:46 Speaker 1 Um
41:46 Speaker 0 actually they're starting to do this. So that that's one strategy that I think has a lot of legs.
41:49 Speaker 1 Yeah, so turn down traffic. That's super interesting. Look, you sp you spend all the money to bring them to your site, right? And otherwise you would have just lost them, right? And that's sort of sunk sunk costs. You get something out of it. That's fantastic. Well, why don't we finish this segment out? And I want to do a lightning round with you, which is I want sort of you know, instant advice for somebody in this seat. I'm an exec visa or a geico. Um, and so I'm gonna name a category and you sort of just uh of how to deal with startups and you can react to it. All right, so category one is you should always invest super early, uh as early as you can into a startup.
42:24 Speaker 0 Uh so again, remember adverse selection versus positive selection. So I would say the company so this is what you have to get right, which is if you take nine weeks to make a decision and like you know, we'll decide within a day, or Sequoia or Benchmark or some other great venture capital firm will decide within a day, like you're not going to get good deals if you take nine weeks. So it can be very, very important to invest really, but like the best things always seem overpriced. Like this is something that we've learned, and it's the same thing with underwriting your own customers, which is like if something's too good to be true, it probably is. So some of the best things are actually very expensive.
42:58 Speaker 1 Yeah. All right. Just given those dynamics, just wait for the later rounds. Let all the venture guys take all the risk, and then like you plow in late. That should be my strategy.
43:09 Speaker 0 I I think in general, that's probably a better strategy. But again, saying like, ooh, we're getting a great deal on this one, that's probably then you know that you're the adverse selection source of capital as opposed to okay, here's something I can't believe we're paying this much money for it. We have to fight our way in. There are 10 other people that want it. You probably know you're on to a good customer, if you will, uh, or a good investment. Yeah.
43:33 Speaker 1 All right. Partner with as many possible startups as you can, because you don't know who's going to win. So let's open up a marketplace. Let's a hundred startups that I have either turned on traffic relationships or something.
43:44 Speaker 0 I think that I think actually that does make sense. I mean, there should be some kind of gating item to make sure, like, maybe not a hundred, but um, how do we how do we stay close to different models that are working well? Because the main advantage that the incumbents have, again, depends on like lending or under or or insurance, but it's typically something around cost of capital and something around distribution. So if you have both of those and you're not using it to the fullest extent, like you turn down a lot of customers, like you should try to find an intelligent way of using this and using that's your unique, that's your unique thing. Like venture capital firms don't have that. I can't fund somebody and send them a million customers tomorrow. Right. But Geico could. Yep. So uh but you can't do that a hundred times, you can probably do that some some sub segment of times according to how much you know additional traffic or whatever it is that the unique advantage that you want to bring to bear.
44:36 Speaker 0 Yeah. All right. Now
44:37 Speaker 1 on MA strategy. Um MA strategy one. In um buy super early, four it's proven to work because presumably the prices are lower. So MA strategy early. Focus on early stage companies. Uh I'm a big
44:54 Speaker 0 fan of what Facebook's done with MA, and I encourage everybody in pretty much every other industry to do this. So Facebook has two formats for MA. One is we buy the existential threat that could kill us, and we price it probabilistically. So surrender one percent of our market cap to buy Instagram. That was way overpriced. But everybody said now. But like there's a one in a hundred chance that this is going to be bigger than Facebook. We should probably surrender one percent of our market cap. WhatsApp, 7% chance, or whatever it was, I think it was 7% of Facebook's fully diluted market cap was spent on WhatsApp. These were brilliant acquisitions. Oculus, I mean, Oculus hasn't turned out the same way that WhatsApp has, perhaps, but like same idea. It's like here this could be the new platform. If we don't buy this, and Apple does, we are subject to their random whims and fancies.
45:40 Speaker 0 So that's that's category one. Category two, and this is super counterintuitive for a lot of companies, buy the guys that failed trying. Because they had the courage and the tenacity to try to go and build something new, and that's what you want in your company as well. And then this is the most counterintuitive part is like take the person that failed and put them in charge of the person that was successful.
46:02 Speaker 1 Yeah.
46:02 Speaker 0 And that's the part that that's breaking glass. Or a big company, that's so hard.
46:06 Speaker 1 But but you reward
46:07 Speaker 0 your execs on success, not on failure. Right. But in many cases, it's like you have a big company that's been trying to build this thing for 10 years. And if they build it, they will get one billion customers. Yeah. Because they have or making that up. They have the distribution. Then you have the startup that actually built the thing in like a week. Right. And they built it for a million dollars. And that would take the big company like a billion dollars in 10 years to do. But like, oh, the company failed. Oh, that's a bad company. These are bad managers. But actually, you want to take them and put them in charge. And the the joke that I would always make is like if Amtrak buys Tesla, the worst thing that Amtrak could do, because Amtrak is probably more profitable than Tesla at this point. But if Amtrak were to buy Tesla, the worst thing they could do is say, okay, all of you Tesla Bozos, you work for us. But the whole point of a lot of this other form of MA is you're really trying to buy products that you can push into your distribution.
46:57 Speaker 0 And you're trying to buy talent that wrote the products, that built the products that understand that. And the only thing that they needed, the only gap between them and actual huge success is distribution, which these big companies have in droves.
47:09 Speaker 1 Yeah.
47:10 Speaker 0 So that makes
47:10 Speaker 1 perfect sense. Maybe just a piece of advice on how to actually make that happen. Because you have this dynamic where you're a big company, you just bought a failing startup, right? You have all of the execs inside that have earned bonuses consistently over year for awesome performance, right? You've rewarded success. And now you're gonna say, I'm gonna take this guy that kind of failed, and like
47:32 Speaker 0 you work for them. Right. Like, well that's hard to do inside a big company. It's very hard, but I mean, in some cases, you just want to do it early. I mean, I think it actually where it works best is where you say, we need this product. Yeah. We need this product to exist. We don't have it right now. We haven't spent eight years trying. Uh rather than saying, let's go assemble a team and I'm gonna rely on like something that's just not in our core DNA, here's how we're gonna go shopping. We're not gonna go shopping and value this. And again, we don't, this is not a self-serving comment because if somebody buys one of our failing companies for 10 million dollars and we have a billion dollar fund, it doesn't matter. Like we want the companies that actually beat the incumbents. But the incumbents, the way that they can actually do great is to adopt more of this Facebook mentality. And like the key thing is that uh many of these acquisitions, these kind of aqua hire, that's the portmanteau of acquire and hire, uh, these aqua hire acquisitions that Facebook made, these people now run big swaths of Facebook.
48:27 Speaker 0 So I agree, it's hard to do if you already have a leader in place. In that case, it just requires a very strong-wheeled leadership team and an actual overt strategy that this is what we do. It becomes easier if it's like, okay, we're trying to do this new thing, rather than assemble our own team and they don't know what they're doing, but they're well intentioned. Uh let's go buy a company, but let's buy a company that hasn't already done the thing, but a company that tried and failed to do the thing, but we're pretty sure that these are the best triers and failures in the business. That's the hard thing to uh to to really measure because most people are used to measuring outcomes and not process.
49:01 Speaker 1 Exactly.
49:02 Speaker 0 And the key thing to make this strategy work is you actually want to over-allocate on process and you want to weight outcome to almost zero because you're buying the outcomes that were in fact zero. Yep.
49:12 Speaker 1 The uh Mark is about to uh interview Andy Dub, uh, the thinking and bets break, right? And this is sort of the essential thinking and bets motion, right? Right, which is don't confuse a bad outcome with sort of uh a bad bet. Right, right, right. Exactly. Awesome. Well, thank you so much, Alex, for coming in and sharing your thoughts. For those of you in YouTube LAN, please like and subscribe. And for the comments thread on this, I'd love to get your
49:37 Speaker 0 input on what you thought of Alex's idea that
49:41 Speaker 1 what you really should do is not go after more customers, but instead go after only the best customers. So, what are examples that you've been trying in your own startup where you're trying to implement that idea? So, see you next time. Go ahead and subscribe to the channel if you like. Like it and see you next episode.
Previous What’s with All the Bio M&A in 2019?: A Quick Take Next a16z Podcast | Incenting Innovation Inside, Loonshots to Moonshots Page updated August 25, 2026 · transcript and summary are machine-generated
