Max Levchin's idea for the company Affirm was born from a painful personal experience. As a 16-year-old immigrant from the Soviet Union, he had little understanding of credit. He got his first credit card at 18, financed his first startup with it, and promptly fell into debt when the startup failed, leading to calls from collectors.

The real shock came five years later. After taking PayPal public and becoming wealthy overnight at 23, he tried to buy a fancy car but was declined for credit. His earlier mistakes had wrecked his credit rating, a fact that blindsided him.

Not only did I feel completely screwed by the moment where I found out that you're supposed to make the minimum payment and by the way, interest accrues into principal... It bit me again five years later where it's like, oh by the way, it wrecked your credit rating too, so you can't get a loan for anything.

This experience was so humiliating that it stayed with him for years. His friends would prank him by having restaurant staff say his card was declined, and he would turn beet red every time. In his late 30s, he sat down with a friend from his PayPal days and questioned why they hadn't tried to fix the negative aspects of credit cards for young people. They realized the AI technology they had built at PayPal to fight fraud could be used to create a better, fairer credit scoring system. That conversation was the seminal moment that led to Affirm.

The underwriting process for Buy Now, Pay Later (BNPL) services like Affirm is fundamentally different from traditional credit cards. Max Levchin explains that the credit card business model is built on the accrual of interest into the principal. The longer you take to pay, the more it costs due to compounding interest. They also benefit from late fees, which disproportionately affect smaller balances.

The credit card business model has a weird misalignment of interests where the lender tells you, 'please pay your bills on time,' but what they're saying sotto voce is, 'but not too on time and ideally take as long as possible,' because that's when we make the most money.

From its inception, Affirm was designed to correct this misalignment. The company committed to two core principles: creating a payment plan that never changes and never charging late fees. This structure means Affirm doesn't make more money if a customer is late or takes longer to pay. In fact, they lose money, as the capital they lend isn't free. This forces a 180-degree shift in perspective, where Affirm only wants to lend when there's a strong conviction the person will pay back on time.

This philosophy dictates their underwriting process. Instead of just a credit score, which might be gamed or inaccurate, Affirm focuses on a person's actual cash flow and ability to pay for each specific transaction. For a new user, Affirm might pull standard credit bureau data. If that's insufficient, they may ask to analyze the user's bank account cash flow to understand their financial situation. This entire process takes only seconds.

When it comes to the data used, Affirm avoids "creative" or behavioral data points, like measuring impulsiveness by how quickly a user interacts with the website. Max considers such methods unsettling. The underwriting field is highly regulated by laws like the Fair Credit Act, which prohibits using protected attributes like race, gender, or age as a basis for lending decisions. Affirm is careful not to use any data that even correlates with these prohibited bases.

Merchants share data on what customers purchase, providing insight into how people use Affirm. Contrary to some press reports, people don't typically finance small items like burritos. They might finance large grocery bills for a special event, but the average transaction is around $300. This data informs a key mental model for lending: if an item's useful life is much shorter than the time it takes to repay, it might be a poor financing decision. This prompts questions about an item's history and quality before approving a long-term loan for something that might not last.

When it comes to underwriting, new variables rarely have a major impact; most influence a decision by only 1-2%. The idea of a single magic variable that dramatically improves underwriting is a myth.

Anytime someone tells you, 'I found this magic variable, and if I just look at that, it's 30% better underwriting or 5% better defaults,' they're lying probably to themselves more than to you, but they're definitely lying.

Such 'magic bullets' are brittle and tend to fail as macroeconomic conditions change. The more robust approach is to use many subtle, compliant risk factors to build a comprehensive score that predicts a person's likelihood to repay.

Affirm has differentiated itself for nearly 15 years, not through advertising, but by building customer trust at the point of checkout. Retention is incredibly high, with customers who use the service a few times returning with 90% probability. The core of this loyalty comes from the company's transparent, no-late-fee policy. The moment of truth often comes when a customer stumbles and contacts the company expecting a penalty.

Our best customers are the ones that email us or call us and say, 'Hey, I was late to pay my bill, something happened, what's the late fee?' And we tell them there isn't one. That's the moment when they grasp how we're different from the rest of the industry.

This approach has led to a model where 95% of transactions come from repeat customers. The business operates in two primary ways. First, through interest-free loans, where a retailer pays Affirm a fee to cover the cost. This applies to both short-term 'pay-in-four' options and longer-term financing for larger purchases. The second model is an interest-bearing loan, where the consumer pays the interest directly to Affirm. In this scenario, all costs, rates, and payment schedules are calculated and shown upfront. The punishment for being late is not a fee but the inability to use the service again until the balance is paid.

This structure aligns Affirm's incentives with its customers. Since the company doesn't profit from late fees, it must underwrite loans carefully, resulting in delinquency rates about half those of the credit card industry. In the notoriously competitive payments space, where cheaper options for merchants exist, Affirm's value proposition is its honesty. Competing services that are cheaper for merchants often make up the difference with hidden fees for consumers. Customers recognize this and choose the service they trust not to penalize them if they make a mistake.

When merchants consider adding a buy now, pay later (BNPL) option, they evaluate it on three main criteria. The first, unsurprisingly, is the fee. These fees can range from below standard credit card rates, which are typically between 2.5% and 3%, to low single-digit percentages if the merchant is subsidizing consumer interest. Max Levchin clarifies that Affirm isn't positioned as a cheaper alternative to credit cards.

More important than the fee, however, are incremental sales. Merchants are focused on converting a marginal buyer who might otherwise walk away. If a 5% fee secures a sale for an item that would otherwise stay on the shelf, the merchant will gladly pay it. This directly ties to the second thing merchants care about: approval rates. A BNPL option is only useful if it approves a high percentage of customers. Max notes that Affirm's approval rates are typically higher than the industry average due to their advanced underwriting, which doesn't rely on crutches like late fees. They don't explain the complex underwriting to merchants, they simply demonstrate the value through higher approval rates.

If your next marginal buyer is either a no thanks or it will cost you 5%, if you have the margin of better than 5%, you don't want the item staying on your shelves. You would like to pay this 5% to get the marginal buyer to say yes.

The third factor is brand reputation. Merchants invest heavily in acquiring their first-time customers and want them to return. If the BNPL provider creates a negative experience for the consumer through harassment or unexpected fees, that bad experience reflects on the merchant's brand. This can deter repeat business. Therefore, merchants care about the consumer's post-purchase experience, as it's directly linked to their own brand integrity and customer loyalty.

Affirm's credit card strategy avoids the high advertising costs typical in the industry. Instead of broad marketing campaigns, the card is offered exclusively to existing Affirm users who are in good standing. It's presented as a next step for satisfied customers who appreciate Affirm's model and want to use it for more of their retail shopping.

Max Levchin describes the card as a "dual mode" product that a user can explicitly switch between debit and credit functions. For small, everyday purchases like a burrito, it functions as a simple debit card, drawing money directly from a linked account with no interest.

If you're buying a TV, you can buy it now with Affirm. In the app, you say, 'Hey, my next transaction, I want that to be a 12 months, maybe it's a 0% loan because the retailer wants you to have it interest free.' Maybe you're paying some interest, but you're setting up that transaction in the app and then your card is ready. When you tap it next time, it becomes a loan automatically.

This approach has been very successful. Without significant advertising, about 12% of Affirm users adopted the card quickly. It appeals directly to their core user base, which has already shifted away from traditional revolving credit and is looking for financial products without late fees or excessive interest.

The issue of consumers "stacking" loans by borrowing from multiple "Buy Now, Pay Later" (BNPL) lenders is a significant concern. Max Levchin explains that for Affirm, the user base overlap with other providers is currently minimal. Affirm users tend to be loyal, and their average transaction size is significantly higher than the industry average. However, looking five years into the future when the market is more saturated, the right way to address this problem is for all BNPL providers to report to the major credit bureaus.

When a lender taps into a credit bureau, they can see a consumer's current borrowing from traditional credit sources. Affirm is the only major BNPL provider that furnishes, or delivers, both positive and negative payment data to these bureaus. This has been a core part of their value proposition from the beginning. They believe that if consumers are paying on time, it should be reflected in their permanent record to help them build credit for future car or home loans. About 97% of Affirm's consumers are doing well and paying on time.

Affirm pioneered this process, working for a decade to persuade credit bureaus and score providers to correctly interpret and use the data they furnish. Max makes a strong appeal to others in the industry to follow suit.

If I can do one bit of advertising on your show, it won't be for a firm. It will be all of you out there. If you are in a buy now, pay later industry, furnish your damn data. It will help consumers and it will eventually accrete to your brand too.

Max suggests that the resistance from other BNPL lenders may be tied to their business models. If a company makes a lot of money from late fees, it can tell consumers that a missed payment won't go on their permanent record. Since Affirm charges no late fees, they are very pro-reporting. When asked about the financial health of the consumer, Max notes that Affirm's customer base is more financially responsible and would likely be the last to show signs of a macroeconomic downturn. As of now, he feels good about their loan book and the US consumer.

When considering the use of stablecoins in the payments industry, Max Levchin expresses a healthy dose of cynicism. He acknowledges a potential use case for simplifying cross-border commerce by reducing complexities around foreign exchange and compliance. If global transactions were all conducted in a dollar-pegged stablecoin, things could become easier.

If we just lived in dollar pegged stablecoins everywhere all at once, it would be a little bit easier.

However, he argues this scenario is not particularly relevant to e-commerce today. Most online shopping is not significantly cross-border. Consumers prioritize fast shipping, which means they are typically buying from a domestic subsidiary with warehouses nearby. The market for direct international purchases remains relatively small. He notes that if his company, Affirm, were to adopt stablecoins in the future, it would be because cross-border commerce had grown to a scale where it made a material difference.

When asked about using stablecoins for rewards programs, Max critiques the entire rewards ecosystem. He describes the credit card industry as a regressive wealth transfer system. Affluent "transactors" who pay their bills in full each month reap the rewards, while about half of Americans "revolve" a balance, carrying debt and focusing on minimum payments, not points or miles. For these consumers, the idea of a rewards scheme is preposterous.

For them, being told, 'Hey, your reward is a 0% loan' is incredibly powerful. And so before we get into using our hard earned revenue into stablecoins, we will just give it back to the consumers in a form of no interest.

For Affirm's customer base, the most valuable reward is a clear, low, or zero-interest loan, as the alternative is often high-interest revolving credit card debt.

When managing the cost of money, the key is not how to deal with rising credit costs, but how quickly those costs change. To handle interest rate volatility, contracts with lenders and loan buyers are structured to adapt gradually over a long period. This ensures stability and prevents sudden financial shocks.

These contracts adjust both up and down quite slowly and we're in no way unique in the industry. Everybody does that.

If costs do change, there are two primary levers. One is to pass the cost through to consumers. A small increase, such as 25 basis points, is not very significant and most people generally don't care that much. The other option is to slightly increase the cost for the merchant. Because merchants pay in real-time as a transaction occurs, the true cost increase to them is minimal. As long as rate movements are not overly violent, these mechanisms make cost changes a manageable part of the business.

AI is emphatically in use in a productive capacity. During major shopping events like Cyber Weekend, tens of thousands of consumer contacts are handled. A huge percentage of these interactions, which range from questions about payment due dates to refund requests, are managed entirely by AI.

This has not led to laying off customer service staff. Instead, it has allowed the human team to specialize. AI is very good at handling basic questions, like looking up an account to confirm if a payment is late. However, for more complex or unusual situations, such as a customer changing their name or address during a transaction, a human is better suited to handle the issue. This prevents potential AI 'hallucinations' from damaging a good customer relationship.

So we've been able to move our human helpers into a much more sophisticated, much more specific role. The theme in our customer service for last year has been specialization, specialization, specialization, where we train people now to serve very, very specific subset of problems because they can be effective and move faster. And so these tools are actually making them more efficient by letting them focus on just a very specific thing that they're good at.

Internally, the engineering group is not the single largest consumer of AI tools; the finance group is. Many finance roles at the company require coding skills, so their work often resembles software engineering, making them great adopters of AI tools. The legal team also uses AI extensively to manage hundreds of thousands of custom merchant contracts, finding errors and necessary corrections.

Another key use is in compliance. As a regulated business, the company is responsible for ensuring merchants do not advertise its services incorrectly. AI tools are amazing at scanning large volumes of advertising copy to flag inaccurate statements, such as a merchant claiming 'interest-free for everyone,' and automatically initiating a correction process.

When asked to design an ideal payment system from scratch, Max Levchin stated it would look a lot like Affirm. He explained that Affirm was built to be a system its creators could be proud of, which is why it has a clear policy of no late fees, no compounding interest, and no deferred interest. It is difficult to be proud of a business that makes half its money on late fees. He emphasized that the company maintains the moral integrity of its product, even while being intertwined with legacy systems.

Following the interview, the hosts reflected on the conversation, particularly the regressiveness of the credit card points system. People who rarely carry credit card debt receive rewards like frequent flyer miles, but these perks are funded by those who do carry a balance. Affirm's model targets people who need to extend their payments. If this model grows, it could cleave off a significant number of people who typically revolve their credit card debt.

But as this grows, you got to wonder, like if the maximo version, where you've cleaved off a significant number of people who roll, what happens to the points ecosystem for the people like us?

This shift could destabilize the business model of legacy credit card companies. If the number of revolvers declines, a key revenue source that funds the rewards ecosystem would shrink, potentially making generous rewards programs unsustainable.