为什么 AI 智能体终于可能重塑信用卡

The card-payment interface is, so far, the best user interface ever created. This is the largest market in the world—more than $100 trillion in size. Existing payments are enormous in aggregate, but when you look at the size of individual transactions, the numbers are small. Ironically, most transactions are small, while the biggest revenue opportunities in payments are small in total. There will always be opportunities to use another kind of payment device to meet a basic need, but as total payments grow, convenience is the main thing that changes. The credit card is still the best user interface we've created.

AI is already available, so maybe that will lead to another discussion. An agent can do it; I still believe that. No, that's it. One more thing: what is it? Surprisingly, there's also more failure.

1. Meet Max & Alex: 25 Years of Thinking About Fintech

I submitted a new idea in digital payments related to cryptography. I went onstage and was humiliatingly sent away because we weren't anonymous. PayPal's big innovation, for us, was that anonymity wasn't necessary. In that sense, this is very familiar—it's old.

2. What's Surprised Them: Apple Pay, Google Pay & Changing Consumer Behavior

Before we met, you had already helped establish the field. You've guided fintech for 20—maybe 25—years, analyzing the sector and thinking about where it's going. Since you entered this field, which of the things that happened—or didn't happen—surprised you most? At the beginning of the 2000s, did the sector look the way you expected? And what about the growth of Apple Pay and Google Pay and the way they're changing consumer behavior? How successful have they been?

Maybe I got it right at the very beginning by accident. Surprisingly, yes, because consumer habits are incredibly difficult to change. One strange, random incident changed things: copying a magstripe was so easy that merchants were forced to change. You may remember that the new cards had a small chip, and you had to insert the chip rather than swipe it. The terminal would say, “Don't swipe it; insert it,” right? That chip is much safer than a magstripe.

3. The Allen & Company 1-800-Flowers Demo

Visa and Mastercard, along with Europay—that's why there's an E in EMV—created the standard. The E is for Europay, not Eurocard, as I used to think. Eventually, it became the standard in all the machines. If you don't want to accept it, for example, if I go to Best Buy and buy a TV, take it home, use it, and then say, “No, I didn't buy the TV,” a magstripe transaction would leave Best Buy out the money, and I would have received a free TV. I'm exaggerating slightly, but every merchant had to get new machines.

Those new machines also had contactless capability. You no longer had to dip the chip or swipe the card; you could just tap. When those machines first arrived, nobody tapped. Now it's the most common thing. Consumer behavior is usually difficult to change, but COVID forced every merchant to change its machine independently, and that changed everything.

No, but you clearly predicted this because Palm Pilot payments were already there. You had a PDA and a wave, but you guessed these three topics would change behavior.

No. Consumer behavior changed everywhere: new merchant payment terminals and mobile telephony changed the world.

4. Max on the 60-Year-Old Rules Still Running Payments

And you, Max? You've been thinking about the future of money for a long time. What did you dream would happen but didn't? What has surprised you? For the past 30 years, you've been analyzing this sector, trying to understand what it means and where it's going.

I think Alex's point is worth explaining further. Apple Pay's and Google Pay's interaction with the Visa and Mastercard networks is really excellent and subtle. The networks impose a hard limit of 2.5 seconds for a transaction between the issuing bank, the merchant, and the acquiring bank. The entire process has to happen in 2.5 seconds; otherwise, the transaction is retried or canceled. That leaves very little opportunity to discover anything intelligent.

Online, you can play games. If you do e-commerce, you can tell Visa, “We'll submit your card, but first we'll perform some antifraud checks.” You may think you're reducing your responsibility and adding a few more tasks. But offline, once your card has been presented, you have only those 2.5 seconds.

Google Pay and Apple Pay create secure elements in their chips. They can say, “I already know your card,” and that changes the total time. Before Visa and Mastercard even speak to the card, they can perform all kinds of tasks. What's surprising is that this 2.5-second constraint doesn't need to exist. Visa and Mastercard still haven't introduced a new standard.

They could ask issuers to bid for better credit quality or allow companies 15 seconds to perform other innovations, but the rules from those early days are almost the same. They've been in place for about 60 years, not 30. That's a critical point.

In payments, the surprising thing is that this is the world's largest market, yet almost everything is small. Small niches lead to innovations and change a market worth more than $100 trillion. There probably aren't $100 billion niches in payments; they're probably much smaller. That's the thing I find most surprising.

Even when you go to the very biggest numbers, the transactions are strangely small. If I sent a $1 trillion wire transfer, America would announce, “We have $40 trillion in debt.” If Elon Musk—or, say, Rihanna—were worth $40 trillion and decided to impose a $40 trillion tax, it would be a wire transfer, but it wouldn't be easy. There's a lot of volume, but the more profitable opportunities in payments are low-dollar and small in total. That's surprising, right?

How did the two of you meet?

I think we have slightly different stories. In the end, I tried to find the original email, but it had disappeared into the mists of time. We met in April 2011. I wrote to you, “Tell me more about Bill Me Later.” Because PayPal had acquired Bill Me Later, and I was doing something different at work, I wanted to understand it. You sent a long, detailed answer, and we discussed the subject over email. We decided there wasn't a good business there—not a big one, anyway.

You mentioned it in a very strange way.

I said, “Pay me sooner.” I had bought the PayMeSooner domain.

Yes, that's me. I remember.

As a result, PayMeSooner seemed like a good idea. It still seems like a good idea. The idea itself isn't bad; it's just not really about payments. It's about loans.

Think about it this way: big companies get good payment terms, while small companies are constantly struggling. Imagine I'm GE and you're Erik Torenberg, a small merchant. I tell you, “I'll pay you in 90 days.” You send me an invoice, and I'll pay you 90 days later. Meanwhile, you have to pay your employees' salaries.

So you go to a bank or someone else and say, “Hello, I'm Erik. I'm running this business. Most of the money coming to me comes from GE, a very great company, and they will pay me in 90 days.” The bank says, “Yes, we'll give you the money, but we'll charge you 15%.”

Meanwhile, GE can issue bonds at SOFR plus 10 basis points. If GE is paying 5% for credit and you're paying 15%, that seems a little meaningless, because you're borrowing against GE's credit—something like that.

There's a method called factoring, through which you can sell your invoice. You can sell the receivable that's coming to you. That's somewhat like an interest-bearing business, but it's extremely expensive. In fact, it's not technically an interest-bearing business, because usury laws don't apply; you're selling future money.

I never thought so, because the entire financial system runs on businesses waiting for payments. I get into debt because you're not paying me quickly.

Or, conversely—sorry—I won’t pay you quickly. Yes, so we do it. These days, there are some good businesses in both accounts-payable financing and accounts-receivable financing. So this isn’t a bad idea, of course. But in a credit sector with no control, the income opportunity for consumers is lower than for the sector, because it’s a convenient item that’s understandable to everyone, and there are many organizations offering it at low charges. What do you do with a $40 trillion wire transfer? Why transfer it? Is it profitable or not? That’s the source of the opportunity.

5. What Idea Should Exist But Doesn't Yet?

Yes. The world is like this: you think it should work anytime, but there’s still another one that doesn’t exist. Any idea? Is there?

Here, clearly, there’s a possibility. And for some reason, that’s it—it didn’t happen. Or, as Max said, bad thoughts? No, just the first thoughts that arrived.

Anything else on the subject? You said it didn’t happen. Are you surprised?

Biometric payment is a great way for us. We’re never searching for that. In movies, it mostly appears as authentication for people: fingerprints and eyeballs. However, we still pay only with chips. The card-payment interface is, so far, the best user interface we’ve created, and people haven’t tried to improve it.

Remember, before PayPal arrived, MasterCard and a gas-station network made a wand-like device together. With it, at the station, you just waved. Payment was done, and you could fill the car with fuel. When I saw that, I said, “Oh, this definitely does replace credit cards,” I thought. But it didn’t happen.

There’s an interesting lesson related to innovation right here in payments. Critical mass is unknown at first, but afterward, one thing is obvious. If you can’t reach that level, you fail. In payments, “It’s okay” will not produce results. To be successful, everyone must use your widget or network; otherwise, it will disappear into the mists of time.

Actually, this is a very meaningful example. I like the idea of a small wand for car keys that lets anyone fill up with fuel at a glance. But it works only a little faster than a credit card in your pocket, and by and large, there’s no difference. That’s also fine. It works. That’s why I’m waiting and looking for a completely different way to identify the person making a payment. Apart from mobile phones, so far we haven’t had anything like that.

Amazon stopped that palm-payment program.

That’s too much for me; I actually liked it. It’s not fingerprint payment—that’s for the palm of your hand.

Yes, it’s related to the palm of your hand.

Yes, it’s palm-related. I liked it. Which Whole Foods? Yes, next to it. The whole thing is there. Go to Whole Foods; I used to buy groceries there.

I know. Me too—I used to use it.

Actually, it won’t be that fast, either, but it will be fun.

Yes, actually, it may be even slower.

6. Pay With Your Identity: Social Credit & the General Store Model

How about, “Tell me my future and give me these grapes and fruits,” as you said? That means you’re giving me money. You’re the only one who gets the future. Okay, I thumb it. I called it “thumb” by mistake; maybe that’s because it’s close to the name of the company.

That’s true.

7. Did Crypto Feel Like the Realization of a Dream, or Orthogonal?

When the crypto industry became popular and some of the major projects started, did you feel excited about anything? Or was it your long-term dream? Or did you feel no particular connection to it?

I often observe trends late. What I saw, first of all, was that no one was receiving it. But in my case, PayPal seemed very far ahead, right? In a way, yes.

To speak completely about PayPal, I wanted to make something different: a special thing working on low-power chips. I came up with methods for using cryptography on a very low-energy chip, and eventually we made it work. Then we understood that all we were doing was encrypting a lot of data and quickly encrypting and decrypting it. Nothing else was happening. We took that away from the payments side, and PayPal was born.

But before that, I went to the DigiCash bankruptcy ceremony. It was the original source of digital payments in the world, and it’s a very painful subject. It happened in the courtyard of Stanford University in the 1990s. There were very sad cypherpunks saying that the time for digital payments still hadn’t come. As we talked and listened, it seemed like we had arrived very late.

Whose original idea was it? Blind signatures—David Chaum was behind them, but he wasn’t there. He had thought brilliantly, but the market didn’t receive it properly, and he was wandering desperately through the streets of Palo Alto. After that, we immediately started PayPal.

I went to a cryptography conference. If a new idea in digital payments was shown, I wasn’t safe—I was driven off the stage. That’s for sure. PayPal’s big innovation was: What if we didn’t care about anonymity? Actually, people didn’t need it; they just wanted to pay for coffee or online purchases. Even in that game, we entered very late. It seems to me that the game was already over, and all the enthusiasts had already left the sector.

Not to create confusion with cryptography, but when I talk about cryptocurrency, I mean the real thing. I read the Bitcoin white paper and said, “Wonderful.” Solving the Byzantine Generals Problem that way was very intelligent, I thought. The mathematics and cryptography were all known, but I was surprised by the method—and so was everyone else.

But as a currency or a payment method, I never thought it would change anything, not even for a moment. I still don’t believe it’s a payment method. It’s been a long time, but as a currency, an asset, and a store of value, it has proven very successful.

So, in my opinion, I saw cryptocurrencies expand and become more popular. Now stablecoins have arrived, and they have very clear uses. But in my view, “With this, I’ll buy a cup of coffee” hasn’t reached that level. For all kinds of payments, I think that’s the most important and authentic use. I would argue that, as currencies—as a way of storing value—this is an excellent tool.

One last point: with Bitcoin or Satoshi, what matters is when it’s spent. To me, that argument is wrong. If you put spending next to them, buying coffee is the most important measurement, because its size, frequency, and practicality are very important. To meet basic needs, there will never be an opportunity to use another payment method.

If you’re thinking about a $40 trillion transfer, surely more time will be allocated to finding the safest, fastest, cheapest way. But if you’re going next door to the bakery for coffee, and your cryptocurrency wallet password is very long, you’ll search your pockets for cash. If it isn’t there, you’ll take a debit or credit card. So, as the total payment decreases, user-interface priority takes over. Cost rules everything. Overall, as the total amount you send decreases, comfort is more important than the rest.

8. The Origin of Affirm: Solving the Pajama Problem

Yes. How did Affirm start? You had ideas about becoming a lender. How did you proceed? I want to take us down memory lane.

Come on, yes. Before you tell me—okay, our stories. That’s enough. Let’s hope so.

Yes, ask us different questions.

No, different rooms to be kept. This is a prisoner’s dilemma, right? If our stories are enough, we can go. If not, we can be here all day.

According to my memory, I ran the company TrialPay. It provided alternative payments for digital goods. You don’t want to pay money to throw sheep in a social game or do anything useless. Is it worth it or fun? No. Would someone do that?

But you know, right? You buy coins in FarmVille. You’re doing something. You play poker and buy chips—virtual poker chips, sure. You’re paying money. If you sign up for GEICO, you get them for free. There’s a lot of economic value in GEICO. Do you use Progressive or GEICO? Does it really work for you? It doesn’t really work for me. If a credit card is required, or if you sign up for Netflix, I will do it to get FarmVille coins. That’s what we did.

Except for Zynga, I got every social-gaming company signed up. So that was the slide.

Alex and I went to the Allen & Company conference. I didn’t want to go because my wife was full-term pregnant. This was in March 2009, around March 1. My son was born on March 28, 2009, but her due date was March 16. That’s why I said, “I’m not going to this conference.”

Then Alex said, “This is a very good conference. You have to go.” I said, “My wife is full-term pregnant.” He said, “Go to the conference. If she has pains, if labor starts, we’ll send you back on the plane.” I said, “Okay, that’s good. I’m going.”

So I went to the conference, and we met there. I think you succeeded in persuading me.

But later, you wrote a note in Russian, I think.

Yes. These days, I don’t do that work because now I use Gemini or ChatGPT to write a note. I used to do it myself; maybe there would even have been grammar errors. In fact, maybe it was ChatGPT, not GPT—informative? That pangram? No, but I learned Russian in high school and college. I worked there and lived there for a while. And you said, “You’re Russian.”

I asked you, “How can I help?” I remember. I think the answer was, “After that, we’re good.” We became friends.

You had sold Slide to Google, I think. That’s why Slide, as a TrialPay client, couldn’t change. It was painful. But then you did something: you tweeted at me, I remember, “I’m looking for something to do. Any thoughts?” That caused me to contact you.

We met at Google and drank coffee there. I think there would have been a sign saying that guests were unauthorized in the Google building, but you didn’t care at all. There were limitations—it was almost like a law—but I don’t remember you revealing all your secrets.

Exactly. You said, “How much is a week?” I liked it. So we drank coffee right there and talked about Bill Me Later.

My memory is that paying for anything on a mobile phone seemed very difficult. Mobile phones were becoming much more common. We were upstairs in our pajamas, and if you wanted to buy anything, you needed a credit card. There had to be a way to solve this problem: how would you pay?

Apart from mobile phones, social networks were also becoming much more common. Maybe go back to the general store in the 1800s. If you went to the general store, you didn’t have cash nearby—literally and metaphorically. You didn’t have it near you, or even at home, but you were at that store.

If you were in the store, someone would say, “Okay, Max, I’ll take care of it. Don’t worry.” But these days, all they have is a cookie and an IP address, so you’re someone unknown to me.

Or, if you go offline to Walmart, a very nice greeter welcomes you. Even if the greeter welcomes you, you’re still completely unknown to them. You can’t implement the general-store model that way.

But if you have 500 friends on Facebook, that’s different. You have 500 friends and thousands of uploaded photos, which means you’re a low credit risk. You are the credit.

That’s an important subject. Credit offers will be sent in advance. Your score is an 800 FICO score. You have a Capital One card with $14,000 on it, and I know that. I have a customized mailer for you from Bank of America saying, “Why shouldn’t you refinance?”

But you go to Google and say, “My credit is fine. I don’t need money. I need credit.” Google might say that you’re a bad credit risk. So, if you go back to the general-store concept, the people who don’t look at credit and don’t have your wallet are, as in the 1800s, paying based on recognition.

Yes, exactly. I think I can add some more detail to this. I had run a company unrelated to social media for more than 5 years. I struggled a lot and gave many people good financial results, but ultimately it didn’t fulfill what I expected as an entrepreneur.

That year, I spent a lot of time thinking about what should happen next in my life. I wanted to start a company, but I knew I had to think more carefully about starting the next one.

My wife said something like this: “When you were at PayPal working on antifraud, you struggled a lot, but you were also very happy. Then you worked to death. Even when you appeared tired, you were actually very happy during those days.”

She said, “You’ve had enough of financial services. I know you don’t want to go back to payments, but give this one opportunity a chance. You love payments companies because they work.”

The reason for my reluctance to deal with TrialPay was that I couldn’t bear the thought of starting another payments company. For that reason, I went into social media. I had no relationship with those things. So, when I slowly came back to working on payments, we started talking about this.

The short name we used for the idea was “Social BML.”

The comparison to a general store is interesting. I don’t remember which of us knew this, because you speak Japanese too, and you told me about it. In various financial systems, social credit is a recognized idea.

In Israel, at grocery stores, paying later—or paying the next time—is still very common. In small towns, when you pay your bill, the cashier looks at you and remembers you. If you forget your money, they’ll take it the next time.

It’s even more organized in Japan. A long time ago, even if they didn’t know you, you could give your business card to the store owner. If you had a card from a nearby business, they would write your purchases on the back of it, and that would become your total bill.

You would just put it on your account. You know, this is clearly a very American concept.

So that’s where we were going. At the time, I was speaking with a startup and learning about the industry. They were trying to create a social score based on a social credit score in China, and it was used for much more than credit.

We were moving toward the idea of payments. I had another metaphorical version of it, too. I was watching television, and there was a great advertisement. TV programs used to have advertisements in the middle, which I liked.

The advertisement promised an object at a price I was willing to pay. I could buy it on my phone, but I was in bed and didn’t want to get up. So, was there any chance of bringing those transactions close to us in real time? Improving conversion was a natural outcome.

That’s the somewhat confused origin story. For the most part, what’s fun in our case is that we had slightly different opinions.

You saw payments as something we could do as individuals. I saw social media as a lot of machine learning. After doing that and recovering as a person, even if it was good, I didn’t think I wanted to do it again.

With fraud prevention and machine learning, my purpose was to build a great credit score. After that, someone else could take care of lending and payments. I had already done payments; I didn’t need to do lending. But making a good score would be amazing, because PayPal had existing data available to us.

We could get some data. Maybe we could buy it. We could collect data from Facebook. I consulted Mark Zuckerberg. I wondered whether they would give me their data. I tried.

9. Agentic Payments vs Agentic Shopping: The Real Innovation Ahead

When you want to complete a transaction, all of those things are connected in my mind. If there was any difference between our motivations, I showed more interest in the merchant side, I think.

We had similar inspiration. We were like, “Let’s test this.” We were in the process of selling to PayPal, I think, but at the last minute they left us. That wasn’t your PayPal; that was many generations later than your PayPal.

I thought I would never start another company. It was a very painful thing, and you faced that situation too.

So, anyway, we thought, “Let’s get started.” We thought we would hire other people. Then you called me, Nathan, and introduced me to Jeff, so we officially established the organization.

But I clearly remember one more thing, which you remember very negatively. You know, it was PHP.

I know.

We had a meeting. It happened in 2012 at the Allen & Company conference. That’s where this began.

Yes, that’s right. I don’t remember it negatively.

No, I know. But it’s funny.

We were looking at a list together. Your old chief of staff—remember? While looking at the list, we saw Jim McKelvey. We thought, “Oh, we both know Jim McKelvey.”

We remembered the 1-800-Flowers commercials, where there was a friendly person visible. You remember 1-800-Flowers? You can order flowers. So we sent him a message about breakfast.

There were about 200 attendees at the meeting, and anyone could meet anyone. We thought, “Hey, why shouldn’t we have breakfast with him?”

I stayed up all night and made this demo in PHP. I basically cloned the 1-800-Flowers site and built a whole system that let you pay with Facebook. Today, AI can do this in 5 seconds, if you want.

I made the checkout work. To correct for mistakes, I added a dozen roses to send to my wife. I clicked it and thought, “Oh, my payment card isn’t nearby.” Then I thought, “I can pay with Facebook.”

What does that do? Facebook Connect does that. I have more than 500 friends, so it confirms whether or not I’m a real person.

Facebook had very interesting data, as you said. There were internal signals for deciding whether an account was fraudulent—that is, whether it had been created as a real account. It wasn’t a real person because the account wasn’t the person, right? They’re the same?

Let’s hope so. But a single person can have multiple accounts, so it isn’t necessarily a 1-to-1 relationship.

That was our fun demo. Jim McKelvey said, “This is very good. Let’s do it,” as I remember.

Yes. There are 2 interesting topics that come from this. The first is the good breakfast. In the middle of it, he said, “Yes, the people who are with us in the service call me and say, ‘Send some flowers to my wife.’”

And, “Oh, I don’t have a credit card.” When they said that, we were like, “Oh, that’s okay.” We told them, “Thank you for using our service. Next time, we’ll take it from you.” He had enough confidence to think that this could be done as a post-pay product.

We basically put our hands up and said, “If you don’t get paid, we’re responsible. We’ll take it. It’s okay. We’ll bear it.” He immediately became fascinated by the product. He was a great supporter of ours and a successful professional in his own right. He was one of the people who believed in us from the beginning.

He said, “You’re great. You’re really doing well,” very excitedly. Then we handed over that responsibility. What is his name? He’s now running a successful startup. His name is Amit—yes, Amit Shah. Amit took on the responsibility, and he was very, very excited about it.

I remember doing this in Microsoft Word or Excel. We needed a price list, so I made our pricing sheet. I asked, “Why are we charging 7%?” In consumer finance, there are concepts called MDR and APR.

Merchant discount rate means that you give the merchant cash now, before the consumer pays, so the merchant gives you a discount. This is called a 7% merchant discount rate. But there was a table, and I think I made it up from time to time. Then there was an APR, but there was no actual APR here. This was just MDR.

I showed it to someone—I think it was Rob Fife, whom we had hired earlier as a real finance employee. I showed it to him, and he said, “Oh, free flowers.” He thought that, because we didn’t need to be paid back, the benefits for us were not great.

What was Rob like?

He was very smart but very sarcastic, so we didn’t take his irony seriously. It seemed like a great demo. I remember interviewing him for the organization. At that time, it was called Expedite, I think.

Yes, it was.

For a while, Expedite was the name. After that, we changed it.

What does that mean?

Expedite Software, Inc.—as in “Inc.” I think that’s true.

10. The 40 Years in the Desert & Finding Product Market Fit

But how did that idea develop? What happened? When did you really achieve product-market fit, and how did that sector go? Did it go well?

Every startup has to travel through 40 years in the desert. That’s a natural thing. Sometimes a lot of people leave quickly, or something happens and the company is destroyed, but I’ve never seen someone start a company, achieve product-market fit immediately, start sales within 24 hours, and have everything move smoothly.

So, going into the office, understandably, it can seem as though there will never be an end. In the end, we didn’t know whether it would end. For a while, we were wandering in that same kind of void.

With 1-800-Flowers, we had that experience of not getting along. He was never satisfied with the conversion rates or the user interface. For the most part, he was right. He was a little angry and demanding, but when it came to product and financial matters, he knew what he was talking about.

He simply wasn’t willing to pay exactly 7%. So we pushed the situation somehow. We were doing some business, but with other things, the flowers were selected. The option to pay us afterward was visible.

He used to say, “Look, my credit card transactions are hurting. If I don’t have to pay these charges, I’ll certainly send you more transactions, but you’re overcharging me for a card. This is the worst idea. I don’t want to pay anything. You’re the original one. You should be ashamed.”

We tried to convince other merchants, but one merchant said, “These are good, but this is a waste.” So that didn’t go anywhere. It didn’t work.

11. Beautylish, Mattresses & the 30% Conversion Breakthrough

At the time, we had another friend named Nils Johnson. He was the founder of a company called Beautylish, which still exists. They sell cosmetics tools and beauty products online. Beautylish was the only one that did change: they implemented it.

By then, its name was Affirm, I think. For their users, when they selected shampoos or perfumes, they were told that they could pay in 3 installments or 30 days later. As a result, the conversion rate grew by 30% immediately.

Then we got it: this is it. This isn’t a product problem; it’s a budget problem. If I can pay in installments, my purchasing power increases. We understood that immediately and changed it into a sales campaign.

Very quickly, small brands found it, especially direct-to-consumer brands. They would say, “We don’t care if you charge 1%, 5%, or 12%, because we’re looking to increase our sales. Our customers don’t have to pay cash upfront; they can pay in installments.” That brought a lot of change.

Within days, merchant CEOs were sending me love letters. They were the first recipients of our services. Many have since sold their businesses, closed them, or left for other fields. I still have friendships with many of them.

For example, Tracy from Tradesy—Tradesy.com. Tracy is a very good person. She supported our service from the very beginning. She emailed me screenshots of her dashboard and said, “This is the Affirm effect. Because of you, my sales have increased 35%. Please keep doing it this way.”

Okay, pajamas—that’s the problem and the solution. After that, mattress companies arrived too.

Oh, yes. That was a very large subject. I remember saying, “Wait, is there a company called Purple?” That surprised me. And Casper was a company. I didn’t know where they were coming from; they kept appearing one after another, and they brought a lot of change.

The original issue was how much more MDR could be charged. How do you push that? For the merchant, how do you show enough value?

The highest MDR charges are actually from for-profit education organizations. Some of them charge up to 50%. University of Phoenix, for example, is owned by Apollo, a private-equity company. There is an Apollo holding company—Apollo Education Group, or something with Apollo in the name—that owns it. University of Phoenix is, I think, a big one.

Many people consider online courses something they won’t pay for, so the benefits to the provider are mostly there. They’ll say, “Okay, you can take 50%,” but the deal is up to you.

Max, you said before that this leads to fraud prevention.

Mattresses have high gross margins. These mattresses don’t ship as full-size beds. They’re compressed into a cylinder, which is very small. Some of them come in boxes, and when you open them, they expand—like a sponge you put in water, right? Even the mattress does that.

So they had much higher gross margins. Therefore, there was a lot of flexibility around MDR.

Yes, very much.

No one particularly wants to be uncomfortable in bed. You don’t want to pay $1,200 for bedding, but when it’s divided into $30 payments, the conversion rate goes up dramatically.

There are some other technology topics to discuss in that story, but let’s take the next big step. Direct-to-consumer brands were great for expansion, and mattresses served as a foundation. We should definitely discuss that.

Before we got there, we tried Affirm as a payment provider for online education organizations, but we got out because they were willing to pay a higher MDR. The quality of the education was so bad.

People wouldn’t pay because they thought, “I thought I was getting a degree in basket weaving, and I’m not going to pay for that.” Then it turned out that their degree was worthless. They thought, “Why did I join this ridiculous course?”

What was that one called? General Assembly? Is that the name?

Oh, yes. We had classes called, “We’ll teach you coding.” Before AI, I thought coding was something you had to learn. But today, you can just ask ChatGPT to code.

Learning to code is important these days, but you don’t need to go to college for that. You only need to take a 6-week course. There are many University of Phoenix clones—all of them.

We looked at that trend and thought, “Oh my God, are these people ready to subsidize education?” That was very good and aligned with our goal.

And yes, there were lots of losses resulting from consumer dissatisfaction. We only continued in that field for half a year.

But mattresses—to the point, let’s go. Direct-to-consumer, or vertically integrated, businesses are powerful. There may be some factory in Brooklyn, or somewhere else—who knows? China. I thought Chinese mattress manufacturers would make them locally because I knew the industry, but they’re all in Utah.

12. What People Miss About Affirm Today: From Satisfying Demand to Creating It

Utah is basically mattress country, it seems. Casper may be in New York, but Purple is here, and the other mattress companies are there too. The strangest thing is that everything, of course, is concentrated in one place. So they make mattresses for very little money; they’re all basically the same.

I’m not trying to reveal industry secrets, but memory foam—and the basic value equation of memory foam—is very interesting. First of all, you buy this memory foam in a box, and it’s compressed. There are benefits, but it’s very heavy because you’re actually changing petroleum into foam.

Another thing is the replacement cycle. There was a Harvard Business Review article about Casper that came out around that time. Before Casper, this wasn’t an established idea, but all of these entrepreneurs read that single article, and that was it.

Basically, the article said that people change their mattresses every 7 years. If you’re a mattress-selling company, that’s the most important subject in the world. If you miss the opportunity, the next chance comes only after 7 years.

In mattresses, the margins on the foam and everything else were previously 80% or more at the retail level. That article suddenly gave birth to 4 or 5 companies. They were like, “If we can get into this mattress market, we can reduce the replacement time for your bed.”

Instead of buying a new mattress every 7 years, you could buy one every 6 years. Or you could buy the best mattress in a box and say, “This is it.” There’s a lot of profit in that.

Because of this, the mattress industry is full of opportunity. But more than that, you actually have to persuade people to buy a mattress. That’s what gave this idea importance: “You have no interest for 3 years. I’ll give you a loan.”

That was also the point of Affirm, right? There were similar events in Affirm’s history. There was an opportunity to collaborate significantly with the merchant through the MDR, and we understood that. Because of this, we were able to remove the APR—in other words, the consumer received a genuinely 0% loan.

The background here is very important. I’m seriously opposed to fake 0% loans, and I’ll keep fighting until they’re illegal or otherwise removed. If you go to your favorite department store, these days there will be a huge, invisible sign somewhere that says, “Get the store-branded credit card.”

Next to “0% APR,” there will be a small star. Its meaning will be something like: “For the next 12 or 24 months, when you make a purchase, you must pay at least 1 penny of principal on time. If you’re even 1 day late, interest is calculated from the beginning.”

If you take that card and swipe it for $1,000 today, 2 years later you’ll have to pay $3,000. This is called a deferred-interest credit card.

Out of our anger at that, we received a great opportunity. Our 0% really is zero. When you go to buy an inflatable mattress or a mattress that comes in a box, you have a 0% loan available. Even if you’re a month or a year late, we won’t change the price, because there’s no star next to the zero.

This is cheating. It’s ridiculous, and we’re fed up with it. Affirm has no star next to its zero—ever. We don’t charge late fees, we don’t defer interest, and we don’t use the industry’s gimmicks or frauds to squeeze people.

That was our origin story. The main reason we got away from all of that was simple: if you’re with us for 3 years, you’ll never get into trouble. You’ll never face unexpected disadvantages. That became another major growth point for us.

Every mattress company eventually came to us and asked, “Are you doing that?” We thought, “We have to work with them too.” If you fast-forward a little bit, today, when many people hear about Affirm, “Buy Now, Pay Later” immediately comes to mind.

What isn’t being said about the company today, or what are we missing?

I just reread an email thread related to the emergence of Affirm. One thing is that advertising and payments are getting closer. Your argument, with TrialPay and many others, is the existing theory behind these things.

Actually, on PayPal, there was a failed project called PayPal Shops. I personally just tried to buy something on Back Market through it. It gives you an estimate of what you might buy later, you know? And all of these things turn out to be true.

They’re just 15, 20, or 30 years ahead of you. But I’m not sure. Today, for businesses, satisfying demand is changing even more fundamentally. Someone comes in and says, “I want to buy this item, but this is my budget.”

They say, “I need a pair of shoes. I want the bag itself, but Affirm is ready to provide transparent financing terms. We can help you, in many cases, without interest.” You can take the bag and the shoes. It’s safe.

That’s Affirm’s brand promise, and it works amazingly well. It represents tens of millions of transactions and billions of dollars’ worth of business each year. Now merchants can tell buyers, “Hey, we’re getting started with a new product.”

More focus is being put on creating a platform. We’re selling an attractive new product at scale. In America alone, we’ve done transactions with more than 50 million people, and we’re now available in 4 countries. We’re growing rapidly.

So we changed from simply satisfying demand at the beginning to helping merchants create or generate demand. I expect that over the next 15 years, payments and advertising will compound. That opportunity is still ahead of us.

We’re very far ahead of Stripe, I’m sure, and very far ahead of PayPal. On the other hand, that reminds me of something. Maybe it’s in our discussion—I don’t know. I need to find the email.

For companies, customer-acquisition costs are going to be very high. Now, as a VC, the joke is that 90% of the consumer companies I see make me want to buy Google or Facebook stock, because all of their customers come from there. How do we get negative CAC?

One of the most excellent things about Affirm is that it has negative customer-acquisition cost. You’re paid money to get a customer. There are some thoughtful companies with similar features, but they’re not like that because they’re basically white-labeling.

If you come back to the mattress example, let’s say we did a payment delay. Casper is the friendly-ghost mattress company. Casper is a white ghost—that’s the character. When I was growing up, Casper was the name of a cartoon.

They don’t want to send you a notice saying, “You haven’t paid me back.” They want that work to be done by a third party. Compared with other companies, that’s B2B2C. It’s a very interesting category: business-to-business-to-consumer.

I tried this with TrialPay. Let’s say you’re Zynga and you’re playing a game. Through TrialPay, you could sign up for GEICO and get free coins. TrialPay was an intermediary—someone the user didn’t know.

We tried to send messages to users, but our email provider blacklisted us. I think it was because we had the right to send email to those customers, even though they didn’t know who we were. Having a relationship with millions of people at negative CAC is a gift. This is very powerful.

Doing this is very difficult. In venture, we never saw anything quite similar because there are many companies that are reachable at scale. Casper got to scale, but with a lot of CAC. A lot of these business models decay over time because the economic value all goes to Google and Facebook.

If you have negative CAC, you can start other products too, right? Because you actually have the customer. Above all, you want to have the customer relationship and the brand.

That’s what I got right with TrialPay—or, rather, what I didn’t get right. For example, I didn’t want to own Zynga’s customers, and I didn’t want to own Netflix’s customers. I just connected both sides.

But in Affirm’s case, merchants really want you to have the customer relationship. That’s a lot for them. They already have the burden of supporting the product, dealing with dissatisfied customers, and handling technical-support problems.

When you have a long-term loan—12 months, 39 months, or even 3½ years—someone has to deal with the customer if they’re late. The company has to say, “Hey, you’re late,” or, “You need to make your payment.” All of the communications coming from the lender create another kind of burden for the brand.

The merchant can say, “You know what? Affirm can do this. They can handle it.” That would be very nice. So the customer relationship is foundational to Affirm, and our ability to develop new products is really built on it.

Another important topic is credit scoring. Going back, we’re the only company in the industry really trying to do this. There are many competitors, and to some extent they’ve achieved success, but mostly there’s no one who goes long-term into debt. For a long time, that means a mortgage or a 15-year loan. That doesn’t mean 3½ years is average for Buy Now, Pay Later. Compared to a mortgage, 3½ years is a long time, but Buy Now, Pay Later is about 6 weeks.

To do that, you have to underwrite. You can’t just say, “I’ll look at your FICO score,” or, “I’ll look at your Facebook friends.” Shortcuts don’t work. You really need very advanced machine learning to do this work.

There are 2 excellent topics here, and they’re difficult to do. That’s why it’s so difficult to compete with us in this area. On the other hand, when billing notices are sent, when you communicate with customers, or when you tell them about payments, those interactions provide opportunities to offer users new services.

So, organizing long-term loans and keeping default rates and delinquencies under control is very difficult. But providing more services to users is the price you pay as a financial organization. In payments, most of these challenges are combined. However, this is our business: running through long-term products.

Yes, we’re discussing the history of Affirm. We’re doing a historical episode, and I want to ask a question about PayPal’s history as well. I’m going to be straightforward.

Recently, The Founders came out, and there have been a lot of discussions about PayPal: its leading effect in payments and fintech, as well as the people who came out of it and the excellent things they’ve done. The concentration of successful people is remarkable. Nothing today seems equal to it—not even the Thiel Fellowship, or anything else. There’s no company comparable to PayPal today.

Everyone there was very strong and highly capable, but at the same time, they had very strong personalities and some weaknesses, too. When you talk about the people who came out of that company and its effect on the industry, are there any stories that have been underestimated or left out—anything that was undone, if you know what I mean? At the time, when you were watching the company, is there anything you’d like to add?

Sure. This is an old question with a new answer. Maybe you’re hoping for both.

The standard answer I gave in the book is that PayPal’s history has already been written about a lot. The Founders is actually a really good book. The reason it’s so well written is that Jimmy Soni, the person who wrote it, thoroughly interviewed everyone and spent a lot of time talking to everyone involved in every part of the story. So, at least, it’s based on the facts.

For his research, I gave the same answer I give when we interview the basic team and the next team. One of the simple questions we asked was, “Okay, what are you going to do after PayPal?” The most common answer was, “This is the last one. After this, I’m going to start my own company.”

People would say, “You got your terminal degree in entrepreneurship here. If you come work with us, of course include us. You’re smart, you’re capable, and you’re prestigious. Please contact us and join.” That’s why so many people who came out of PayPal started companies. In a short period, we got YouTube, Yelp, Founders Fund, and LinkedIn.

That wasn’t accidental. We strategically attracted founders. Another thing we don’t discuss is that we learned very closely from the people in the room. We sweated together. We occupied rooms together, brainstormed, screamed at one another, spent time around whiteboards, and criticized other people’s thoughts as useless. That’s how we learned people’s nature.

Peter Thiel is the most philosophical, Elon Musk is never ready, and David Sacks is David Sacks. One thing all of them have in common is that, under pressure or when faced with problems, we know what they’re really like as individuals.

When you know everyone’s true nature, seeing Elon Musk carry humanity to Mars and accomplish such extraordinary tasks seems incredible. Even though he’s a human being, I’ve seen him in the company kitchen, sweaty, tired, and disgusted, a thousand times.

Peter, when he was in doubt while raising a fund, would call me from a trip and say, “Do we have money? Maybe.” They’re gods, but they’re also human. That’s actually inspiring. He’s also a simple, ordinary person. I spent a lot of time with him, so I know that. And he tries to pursue such big ideas. I should do the same. I should try, too.

Are there any other things left to discuss?

Agentic shopping? I’m not so optimistic.

And agentic payments?

I’m very optimistic about agentic payments.

Yes. Robots are us. On Friday night, I’ll buy clothes. I feel that’s the wrong way to think about it. How did we do home delivery before? We’d like to know.

But the counterpoint is that the best user interface we’ve created so far is the credit card, I feel. Agents are smarter than pieces of plastic—than pieces of rewritable, chipped plastic. So eventually, negotiations can reach a stage where agentic commerce becomes possible.

I feel the mistake in all the ongoing discussion around agentic commerce is believing that the correct object to buy is a robot. Can a robot convince you? “I like it.” “No.”

I have a bike, and I like to buy spare parts. I have 2 different bikes, and I spend a lot of time looking at different bike parts because I participate in that purchase. But when I say, “Okay, that’s it. I want it,” I think that’s the moment when you take out your wallet and start. For now, when you say, “I want it,” that’s when payments begin. So we’ll see some interesting innovations in the payments user interface.

I have a question about that. For the most part, I agree, but it depends on what I’m buying. What do I do when I’m buying something that needs to be researched?

AI is just another tool. Just as you ask your friends, I ask mine. This is a bike part—I’m asking you, and I’m also asking AI. I’m asking both of you.

But now I want a particular SKU, and it’s nearby. I have 19 different options in different places. What I’m getting, I don’t need to see the whole world. Some people will value time more than money. For them, time will be more important than money.

From a business perspective, none of this is happening now. So what happens in the future? Well, here it is: a UPC or, preferably, a SKU, at a low price for me to buy. That’s what I’m saying. For people who have more money than time, this is already the case.

That’s why I talk about CamelCamelCamel. It’s my favorite example.