Payments innovation has accelerated in recent years, including multiple fundamentally new payment modes. Most nations did not historically see a new mode in a typical decade. We've had at least five since iPhones hit pockets. One capturing a lot of attention and payment volume recently is Buy Now Pay Later (hereafter, BNPLs).
BNPLs are fascinating and unfortunately not understood very well, partly because (in an ongoing theme for Bits about Money) their core customer is not in the social class which spends most effort in understanding the financial system, and partly because they combine old mechanisms in a way which is very novel. Let’s dig in.
I work at Stripe, which builds infrastructure for the global Internet economy. We work extensively with BNPLs and with the businesses that use them, just like we aspire to work extensively with every way a customer and business would like to transact. I will note for hygiene purposes that there are three large global brands (Affirm, Afterpay, and Klarna) which are each publicly traded. My opinions are my own and are informed by my general professional expertise. Factual representations below are on the public record (or are my errors).
How BNPL presents to a user
The overwhelming use case for BNPL is in consumer transactions for discretionary medium-duration non-durables or medium-amount durables, for reasons which will be very apparent later. Think makeup, clothes, and exercise bikes, but not (generally) food or cars.
A new user for a BNPL is recruited either at the point-of-sale (POS) in an offline transaction or during the checkout flow during e-commerce. They’re presented with an endorsed third-party offer of credit: sign up for the named BNPL provider and, upon successful underwriting, get offered terms go here.
The bread and butter offering in the BNPL space is “pay in four installments”, though there are others (pay in 3, revolving credit, etc). To make this discussion succinct and comprehensible for non-specialists, we’ll talk mostly about the mechanics of “pay in 4.”
The timing of the installments is crucial to the offering: the first one is drawn immediately (typically from a debit card collected from the user during the instant underwriting flow), with the other ones being drawn automatically from the same card 2 weeks, 4 weeks, and 6 weeks after the transaction.
This is a credit offering; the user is getting use of someone’s funds for a few weeks. Another crucial point is that the credit is free to the user. The installments are equal and include no interest charge if paid as agreed.
What happens if the user doesn’t pay as agreed? That’s complicated, and highly sensitive to where the user is located and which BNPL they’re doing business with. It is dictated both by desired user experience, by how willing the BNPL firm (or their capital partners) is to subsidize credit losses, and by local regulation of consumer lending. Some BNPLs have no late fees; some have ones which are capped at low dollar amounts. Late fees are far less crucial to the BNPL economic model than they are for consumer lending generally, particularly subprime consumer lending, which BNPLs are sometimes (unjustly) grouped with.
BNPLs underwrite each user (and each transaction) individually. This is entirely done by a computer and happens in milliseconds.
Subsequent BNPL transactions follow the same model, except the underwriting process is faster since the user already has an account and doesn’t have to repeat data entry.
How BNPL presents to an accepting business
As we covered for credit cards, the payments industry charges partially to facilitate the orderly movement of money and data and partially for marketing. Businesses have three goals of marketing: increasing customer propensity to use them, increasing frequency at which they come back, and increasing amount spent (“basket size”, in retail lingo).
BNPLs pitch themselves to businesses as more marketing efforts and less simple payments rails. They’re more expensive than cards by about 300 bps. (Payments nerds always quote prices in basis points, hundredths of a percent, rather than in percentage terms. In addition to this being a shibboleth to get oneself mocked on Twitter, it reduces ambiguity. If I had said “3% more expensive than cards” you might not know whether I meant “so about 3.1% plus some fixed fee” or “about 6% plus some fixed fee.” It is, of course, the latter.)
Clearly BNPLs need something to justify costing twice as much (or more) as the default option. They claim they radically increase conversion rates (by X0%), basket sizes (by X0%), and repurchase frequency, without substantially cannibalizing card sales. (You can read the specific claims on their sites.) The math follows very quickly from there, particularly for retailers in segments like fashion and makeup who have high margins.
An important subtlety: installment financing is not a new thing in the world. Unlike retailer-provided financing, the transaction is over immediately from the retailer’s perspective; they get paid very quickly (settlement times are broadly in line with credit cards) and do not have to book a deferred revenue liability. This is extremely important for retailers, and is something they pay for both with cards and with BNPLs. Moving the credit risk to the financial sector insulates them from it. Moving the deferred revenue liability makes their businesses structurally more valuable.
Retailers depend on generating a high volume of transactions on an ongoing basis, and having a large percentage of them be financed by the retailer would tend to make the retailer look very highly leveraged, which stacks with other structural needs of retailers to leverage themselves (to e.g. finance inventory) and would tend to decrease their enterprise valuation. (For similar reasons, many large retailers that offer installments do so via arrangement with a financial partner who takes the risk in return for fees and, potentially, interest revenue.)
Who owns the risk?
Here’s where the financial engineering magic happens.
The BNPL is a software company brokering a very complex credit transaction in a fashion which makes it free to the user, by producing ex nihilo high interest debt, manufactured out of the interchange fee the business paid.
Most BNPL loans start with a bank, which is probably surprising to non-specialists. The dominant reason is that consumer lending is, in much of the world, a monopoly granted to the regulated banking sector. It was granted in return for that sector’s guarantee that it will not engage in the long and sordid list of anti-consumer practices possible in debtor/creditor relations.
The bank originates the loan but does not desire to economically own it. They retain a small piece (on the order of a few percent) and sell the rest to someone who has agreed in advance to purchase it.
In principle, that could be the BNPL provider itself. In practice, BNPL providers don’t want to own their originated loans either.
The dominant reason is capital efficiency: BNPL providers are software companies which code up consumer loan factories, but they do not themselves want to be in the business of borrowing and lending at a spread. They do not, unlike the large banks which dominate consumer lending, have gigantic franchises which attract deposits at extremely low interest rates that they could use to lend out. Instead, they’d have to lend from their own capital, and capital for software companies is extremely expensive relative to bank deposits.
So the BNPL provider looks for capital partners. There is a substantial amount of variation and nuance here, but broadly the counterparty is a large pot of money which seeks attractive yields at low but non-zero risk. That counterparty will offer the BNPL one of two things: a flow forward arrangement or warehouse financing.
Finance loves making simple things sound complicated because that ensures the continued employment of lot of finance professionals at high wages. Here’s the intuition you need for both of these instruments: “If you came to me tomorrow with a loan for $100 at 25% APR, since I trust you to underwrite good loans, I’d buy that loan from you. I have very few opportunities to buy low-risk 25% APR debt in the current interest rate environment. I don’t need to trust that loan, per se, I need to trust you, because I intend us both to be repeat players on this transaction and we both understand how statistics work. I am not buying the performance of that loan, I am buying the performance of random samples from your loan-creating machine.”
Multiply that intuition by a lot, and that’s BNPL financing in a nutshell. (We'll elide that, as large software companies, the software / marketing / administrative side of the operation also needs to be financed, could issue debt and equity, etc etc.)
The main difference between forward flow arrangements and warehouse financing is that there are two transactions in buying a loan and they don’t necessarily have to be coupled. One part of the transaction is providing capital. One part of the transaction is absorbing risk. In warehouse financing, the BNPL and the capital partner split the non-payment (or late payment) risk between themselves. In a forward flow arrangement, the capital partner agrees to bear substantially all of the risk. (From this follows some largely correct intuitions about which is more expensive, which banks are likely to engage in versus which is largely the province of hedge funds and similar, and the like.)
Where did that 25% come from? To save you from having to prove this to yourself in Excel, here’s the intuition. A $100 pay in four transaction needs $75 of capital backing it for six weeks (because the user immediately pays a quarter upfront). The average amount of capital required over the term of the loan is half of that, due to repayment. If the BNPL was willing to pay 2% of the original transaction (sliced from their fee) to the capital provider in lieu of interest, the capital provider would receive $77 in repayments for their committed capital for 6 weeks, representing about 2.67% yield on $37.50 over 6 weeks. This works out to 25% annualized, give or take.
Bam. You have just invented a relatively high-interest consumer loan out of nothing, without charging the customer a penny more than their purchase naturally costs.
Of course, it is not out of nothing; the business is paying for it, as a marketing expense.
Sephora wants to sell more Jelly Mint Glossy Lip Tint. They have a variety of options to incentivize this purchase, from giving you a discount to offering a free tote bag to underwriting use of credit (provided by Klarna). They are largely ambivalent which specific inducement it takes, as long as people buy more makeup, more frequently, from them.
BNPLs sometimes re-use existing rails
“Rails” are payment jargon for the technological and financial links between various entities which allow money movement. Visa provides rails, debit card networks provide rails, etc etc.
One interesting thing BNPLs are beginning to do is to pivot the model. Historically they worked only with customers sourced by partner businesses on transactions at those businesses. This is strategically speaking a rough place to be. Increasingly, they want to own the customer relationship and re-use it at other businesses, the same way the bank backing a store card can convince you to use that card everywhere else on the same network.
Card networks are the best network effects businesses ever, and to avoid having to re-build the merchant side of those networks, the BNPLs largely build this offering by riding on the existing card rails. They arrange for the issuance of a debit card (yes, of course, Durbin exempt in the U.S.) which you can use almost anywhere.
The debit card typically comes with two options. One is less interesting as a product: it works virtually the same as any other debit card, making a withdrawal against your bank account via whatever local mechanism there is for that sometime after purchase. In this case, the BNPL only earns interchange revenue.
The more interesting one is that they make it trivial to turn those debit transactions into credit transactions, generally after they’ve been made, within the linked app. As usual, these transactions are interest-free.
Who pays for the credit here? The BNPL firm themselves, partially out of the interchange revenue (using much the same logic as before), and partially out of their hope that this causes you to use their core offering more frequently at merchants they have a first-party relationship with.
As with all consumer lending products, controlling cost of customer acquisition dominates long-term profitability of the business. The best way to control it is to amortize it over long customer lifetimes. BNPLs don't want to acquire one transaction; they want to acquire a continuing stream of transactions from a customer, ideally changing their purchasing behavior for certain sectors and charging their partners for the change.
Finance as social commentary
The finance industry has not covered itself in glory over the centuries, and there is a reflexive distrust in many quarters that financial innovations could actually help consumers.
One argument made is that BNPLs, by extension of credit to people who don’t have traditional credit available, allow customers to overextend themselves. (This is, for social reasons, very rarely articulated as “I am better at math than women who shop at Sephora, think they will not make good choices, and accordingly think they should have less choices presented to them.” But it is that argument.)
One response to this argument is that BNPL users are less concentrated on the socioeconomic margins than they are believed to be. When credit cards were introduced, much sniffing happened over allowing people to borrow money to eat dinner, but credit and debit cards are used by the entire socioeconomic spectrum. Some BNPL users find it the only source of mainstream credit available. Some BNPL users are financially sophisticated, could trivially get (or already have) credit cards, and simply enjoy accrual accounting for capital purchases like e.g. clothes.
But to ease the concerns of benign paternalists, BNPLs have a structural advantage over traditional credit arrangements: their duration is known in advance and known to be very short. A credit card is a commitment which implicitly requires you to have some sense of your earning potential, career stability and growth, and outgoings over a multiyear time horizon. BNPL asks you to look six weeks into the future. Many people, even people less fortunate than steadily employed professionals, have fairly tight understandings of what their finances will look like six weeks out. “Pretty much like today” is a good heuristic.
BNPLs, due to their structurally high cost, tend to be used by businesses with structurally high margins. This means they’re less likely than credit cards to be used to purchase e.g. groceries. Stacking food bills for weeks might easily get someone at the margins to a point of overextension, but they’re far less likely to need to stack clothing or makeup purchases. If overextended, they can simply choose to stop or delay new discretionary purchases. Even defaulting on the BNPL is likely less impactful than defaulting on other debts; the chief consequence is not being able to use the same BNPL provider until one makes good on prior commitments. This is an impact, but it blocks the thing you use on discretionary purchases, not the thing which keeps the lights on.
Moving the cost of credit from the end user to businesses dealing with them also insulates consumers from self-digging interest / late fee holes.
A separate argument is that BNPLs increase risk to society itself, because society ultimately backstops consumer debt. How quickly we forget that the the global financial crisis happened, etc etc.
Because new sneakers are less critical to the capital stock of developed nations than housing is, the total risk engaged in by all capital providers to BNPLs is several orders of magnitude less than mortgages and always will be. There will never be socialized losses for BNPL capital providers; they have a profitable and contained niche. The parts of it within the regulated banking system are structurally low risk and in the regulated banking system.
The argument that BNPL and capital providers have to themselves, though, is even stronger than that sketch.
We have decades of financial and ethnographic experience of what happens when modern consumer debts go bad. At a micro level, the narrative is typically overextension followed by an earnings shock (sudden unemployment, loss of earnings, medical issues, divorce, etc). At a macro level, debts go from largely uncorrelated across borrowers to extremely correlated across borrowers based on the business cycle. Broad trends in the economy hit many people at once over a period of months or years, leading to a rise in defaults, which plays out over several quarters in a row.
BNPL firms and capital providers say “If the economy goes south, we have multiple countermeasures we can undertake in days or weeks, not in quarters. Credit cards can’t retroactively unmake the loans they were happy to have made a year ago but less happy about now. We structurally don’t have any loans older than about 6 weeks and those will age off our books in, hmm, 6 weeks or less. We also underwrite new purchases per transaction rather than per credit line, and therefore get our pick of where we spend our risk budget. We can adjust underwriting based on geography, business sector, basket size, even the individual items purchased, to stay open for business while temporarily slowing our riskiest transactions. Or we could go to capital partners or businesses and find a price for that risk that makes everyone happy.”
These arguments have not been tested in the unhappy half of a business cycle, but they do not sound as implausible as “makeup and mortgages are both systemic risks to the financial system.”
Broadening New Payment Lanes
New double sided marketplaces (which effectively all payment rails are) are extremely difficult to incubate in the world, which is one reason why platforms are so powerful: snap into them and you can save a lot of the effort on one or both sides of the market. Much of the world is, like my home of Japan, moving towards an equilibrium where there are many accepted payment methods and more coming all the time.
One thing that Stripe can hopefully offer to upstart payment methods is broad reach on the business side. To businesses, we can offer decreased integration cost and technical churn as new methods start to win consumers’ hearts and wallets.
This is good for both businesses and consumers. Credit card networks were and are a stupendous achievement for the world, but a bit of competition helps to keep them, like anyone, on their toes. This keeps prices (net of rewards) down for the rest of the economy and spurs ongoing product innovation happening.
Sometimes those product innovations are genuinely powerful for various parts of society. Sometimes they help people buy sneakers.
Sometimes they do both.
Want more essays in your inbox?
I write about the intersection of tech and finance, approximately weekly. It's free.