Buy-now-pay-later (“BNPL”) has become a massive global category in financial services. At their peak, Affirm and Klarna were both worth north of $45b.
A number of players have emerged in every major geography around the world.
Yet, in recent months the space has seen a pull back. Affirm is presently valued at $7b. Klarna was recently valued at $6.7b, also about 85% off its peak.
So many are asking what the future of the business model is.
One of the reasons I’ve personally been bullish on BNPL is not because of the core product alone, but because of what BNPL can do in emerging markets. In markets with lower financial inclusion, BNPL’s real potential is as an onramp into a broader set of financial products. That was why I was excited to partner with companies like FinAccel in South East Asia and Kueski in Mexico.
But that doesn’t mean the core product should be overlooked. I thought it would be a great opportunity to double click on how the business model actually works in its pure form in the US and Europe. I’m excited to bring another guest post series, this time with my friend Jeremy Solomon, who is a fellow VC and the former Chief Capital Officer at Affirm.
But without further ado…
Thanks for reading [99%Tech]! Subscribe for free to receive new posts and support my work.
While I no longer have to answer the question “what firm?” when describing where I was previously employed, I still find that the general public—and even the consumer finance community—does not grasp how Buy Now Pay Later (BNPL) works. In fact, the market view feels more like a kitchen sink approach to the alphabet soup of players in modern purchase finance. While one could generically say that all BNPL players are letting a consumer (SMB is worth its own piece) purchase an item today in exchange for installments, this fails to recognize the diversity of business models and risk being taken by platforms. (Note, I hesitate to use the word “lender” in BNPL for many reasons, to be explained below.)
Let’s start with the basics. BNPL is nothing more than a rebranding of purchase finance, allowing a consumer to purchase a good today that s/he would prefer to pay for over time. Installments may be motivated by convenience, cash-flow matching, or affordability. As mattress, furniture, appliance, and electronics retailers learned long ago, providing financing tools to increase the target consumer market is a winning proposition. That said, purchase finance has a long history of being predatory toward consumers, loaded with hidden fees or deferred interest. Incumbents have profited most on borrower mistakes, making customers irate and negatively impacting the merchant’s brand and image.
Why do merchants love purchase finance / BNPL? Adding a consumer financing tool has consistently proven to increase funnel conversion (more consumers purchasing items) and cart size (consumers spending more). This is the holy grail of commerce. Merchants spend the majority of their budget on marketing to consumers. Once the consumer is convinced to enter the store or go to the website, merchants are motivated to convert at a high cost. Financing drives top line results, and nearly all merchants end up with positive returns from the incremental cost. Please note that some offerings such as co-branded or white-label credit cards are economically accretive to the merchant; i.e., the merchant is paid for allowing the bank to offer the card to consumers.
I’ll refer to the modern approach to purchase finance as BNPL. There are three categories of products offered in BNPL, and each is worth discussing individually:
Pay-in-4: The only truly new product to come out of BNPL used to finance smaller purchases ($100-$400). Please note that this is offered as a payments product by platforms, not a loan.
Zero Percent Installment Loans: A 0% interest loan for larger purchases. The consumer pays the sticker price of the item, without being subject to interest of any kind. This is basically “free money”, coming at the same cost as cash but allowing for a flexible payment period.
Interest-bearing Installment Loans: Loans offered for larger purchases and bearing an interest rate. Rates typically are similar to credit cards; however, interest-bearing installment loans do not allow for the opportunity to revolve like credit cards. This forces the borrower into a more disciplined repayment schedule.
All three offerings make money through a Merchant Discount Rate, or “MDR”. One could think of this as similar but distinct from interchange. Merchants pay interchange to the networks in order to compensate for payment settlement risk. MDR is charged for services rendered - financing the merchant’s customer for a major purchase. The MDR varies depending on the product and objective, as described below. Interest-bearing loans typically charge an explicit interest rate similar to credit card rates and also charge an MDR to the merchant. Generally, MDR, interest income, and (for some) late fees are the only sources of revenue for pure-play BNPL platforms.
I cannot emphasize enough that Pay-in-4 is a fundamentally different business than installment loans. As mentioned above, Pay-in-4 is designed as a payments product and not legally considered a lending product. This is a big deal. It enables firms that offer Pay-in-4 to make the option available to all customers without triggering the complex disclosure regime associated with Reg Z and TILA. Note also that Pay-in-4 products are not exempt from all federal and state consumer laws. FCRA and ECOA apply, for example, as do a number of state laws. Sidestepping consumer lending regulation is a tremendous advantage in speed to market and ability to offer to a broad audience.
Typical product design is oriented toward smaller dollar purchases. As the name suggests, the consumer makes four payments over a six-week period, with the first payment due at checkout upon selecting the offering. Since Pay-in-4 is not a lending product, consumers are able - and encouraged - to pay with their credit card. Debit cards are allowed, although the platform would prefer credit due to potential capacity beyond the consumer’s bank account balance. The platform will deduct equal payments from that initial payment instrument every two weeks. There is no interest in the offering, although most platforms will charge late fees.
Platforms make money by negotiating a discount, similar to interchange, with the merchant. Typical pricing is a 4% discount, or ~1% more than credit card interchange fees that merchants pay with every swipe. Merchants value the offering because the additional 1% cost produces meaningful funnel conversion lift, making the additional cost inconsequential.
In practice, this product works like a charge card for individual purchases. Consumers decide to use Pay-in-4 to delay paying for an item for a period of time and settle in full on a 45-day period. Charge cards allow for no payments for those 45 days (purchase window and payment grace window), but charge cards have historically been super prime credit offerings.
Pay-in-4 is a “fool me once” offering, meaning if the consumer fails to pay on time, s/he is unable to access the platform again until the previous use is brought current or paid in full. Platforms like Affirm, AfterPay, and Klarna have robust merchant networks where target market consumers like to shop. Most merchants are exclusive with only one BNPL offering, motivating consumers to pay their obligations.
This product design is intentional, as Pay-in-4 has limited to no credit underwriting component. No credit bureau pull, no cash-flow-based underwriting, etc. The platforms rely on a) success of the initial payment, indicating availability of funds; b) previous payment history; and c) threats of banning a customer from the service. Separately, the best platforms have top notch fraud controls to protect against loss severity.
Standalone, this is a tough business. The challenge with Pay-in-4, as is true of most payment businesses, is that it is only profitable at scale. Because the product is "free" to the consumer, the MDR needs to cover all of the costs associated with managing both the merchant and consumer relationships as well as the cost of supporting the platform itself. If the platform charges 4% to the merchant and accepts credit card payments costing ~2.5%, gross margins are ~1.5% per transaction for those who pay on time. (Please note that network rules prohibit using credit cards to make payments on credit products; Pay-in-4 escapes this designation, and therefore credit cards can be accepted.) Removing major expense lines such as losses, fraud, servicing, and capital funding leaves the platform with basis points IF the business is well-run. Late fees do help out unit economics but also feel like a continuation of legacy purchase finance. We’ve ignored the cost of building a merchant network, where millions to tens-of-millions are being thrown around to win exclusivity with top brands.
Zero Percent Loans
As the name implies, Zero Percent Loans charge no interest to borrowers. The product is commonly used for terms from 3-24 months, although some merchants will offer even longer terms (36+ months) for higher margin goods to drive higher funnel conversion.
Since this product form is a loan, it must be offered through a bank origination partnership or on a state-by-state basis. Most platforms work through bank partnerships today, as banks can offer standardized products nationwide, adhering to the regulations of the state where the bank issues loans rather than the home address of the borrower.
More traditional credit risk is involved with Zero Percent Loans, although the “free money” nature of the product leads to a payment plan psychology rather than a traditional loan. Historical performance on Zero Percent Loans is extremely strong across the credit spectrum. Economics are analogous to a zero coupon bond, where the lender will push a payment to the merchant for the price of the item, net of the MDR. The lender is then repaid over the term of the loan, eventually receiving the full price of the item. For example, a sofa selling for $1,200 might be financed over 12 months with a 10% MDR. The lender would send the sofa merchant $1,080 when the consumer purchases the sofa. The consumer then pays $100/mo for 12 months. The resulting income stream equates to a 20% yield on the loan (bond math: MDR / WAL = yield; WAL = weighted average life).
As opposed to Pay-in-4, Zero Percent Loans have the potential to be an incredibly lucrative business line. However, only merchants with sufficient gross margins will be able to tolerate a material discount on the item. Interestingly, I’ve observed Zero Percent Loans providing more sales lift than holiday sales. A borrower’s fixation on monthly payment is likely a key purchasing criteria and often outweighs the absolute cost of an item.
Interest Bearing Loans
Last but not least, Interest Bearing Loans comprise a meaningful percentage of loans offered by BNPL platforms. In this offering, the merchant pays an MDR similar to interchange and the borrower is charged an explicit interest rate by the lender. Economics are derived from explicit (stated) interest and the implied yield of the MDR.
Borrowers utilizing Interest Bearing Loans are likely using the BNPL offering as additional capacity to credit cards. While many consumers prefer to finance specific purchases, the initial thesis that Millennials and Gen Z were abandoning credit cards in favor of BNPL is unfounded. On the positive side, an interest-bearing BNPL offering limits the risk that the borrower goes into severe indebtedness.
Underwriting may be the most misunderstood component of BNPL. Skeptics have declared the products flawed and a ticking time bomb. To their credit (pun intended), losses have ticked up meaningfully for AfterPay and Klarna due to the platform concentration in Pay-in-4. The “fool me once” credit performance should be bad! And with the proliferation of Pay-in-4 offerings, the risk of being kicked out of the ecosystem is less severe than it once was.
Yet, a sophisticated underwriter like Affirm continues to generate exceptional risk-return. Why? And why can’t legacy players match this performance? Four factors stick out:
Fraud/identity: While the general population is aware of identity theft and generic fraud, organized crime has taken merchant fraud to a different level of sophistication. Synthetic identities, account take-over attempts, and other sophisticated schemes bring significantly more risk to BNPL transactions.
SKU and Merchant-level underwriting: At Nyca we’ve long looked for a player who can bring level 3 data to the market. Card networks are not able to capture individual line items; BNPL’s direct merchant connection allows for capturing and utilization of this data. As it turns out, a loan to purchase a kitchen table will outperform a loan to purchase DJ equipment for identical credit bureau files. Additionally, not all brands are created equal. Great merchants attract more creditworthy customers.
Term: Part of the beauty of a short-term installment loan is the lender is solely focused on the ability of the borrower to repay over the term of the loan. Revolving credit is much more complex to underwrite, as the lender is taking a bet on an individual to perform over different economic periods and is exposed to the volatility of life (job loss, illness, etc.). The short-term lens allows for a larger eligible population and more dynamic risk models.
(Lack of) Friction: Good borrowers don’t want to fill out twenty pieces of information to take out a loan. A great BNPL offering is as frictionless as it gets. How the platform knows as much about the borrower as possible without burdening the application flow is a true skill and leads to performance lift.
If a small dollar consumer lender can’t confirm a borrower’s identity with extremely high certainty in short order, their business is toast. There are three practices I’ve seen in the effort to manage fraud: a) rely on third parties to manage fraud risk; b) add in a kitchen sink of tests to discourage fraudsters; and c) build in-house systems customized for the business line.
Before describing these three scenarios, the most critical learning about fraud is that it is a dynamic, unpredictable problem. What exactly does that mean? Credit risk, while ever-changing due to macro factors, is relatively predictable and stable. Historical credit loss curves display relatively consistent patterns through economic booms and busts. Credit also does not change overnight - a borrower that was creditworthy on Monday is predictably creditworthy on Friday. But over time, health, employment, and other factors do influence the borrower’s ability to pay.
Fraud, on the other hand, is run by criminals whose objectives (in the case of BNPL) are to steal as much as possible from a given merchant. Network fraud can be a more challenging effort, and lending in general is more vulnerable to various fraud schemes. Due to the human element, fraud can spike at a moment’s notice. Most sophisticated fraud rings will test strategies in small enough quantities that the perpetrators go undetected. Then, when confident of the strategy, fraudsters utilize velocity to maximize their take prior to the platform or vendor realizing and shutting down the approach. These schemes have become incredibly sophisticated, with the FBI often getting involved as a result of fraudsters exploiting meaningful vulnerabilities in critical infrastructure such as cell phone carriers.
Nyca portfolio company Sentilink was founded by ex-Affirm risk/fraud team members specifically to fight synthetic identity (piecing together pieces of real and fake information to create a new credit profile). Fraud-fighting professionals enjoy the challenge of trying to outflank incredibly sophisticated crime.
While relying on third party systems isn’t a bad thing—especially if the vendor is offering some degree of fraud coverage—it can create confusion internally about what is a credit loss versus a fraud loss. No system is perfect, and fraudsters will always find a way through. The question is whether the platform has visibility of how much of their business is suffering from fraud attacks. Separating fraud and credit risk is critical to optimizing a BNPL business.
An alternative approach is for a platform to require multiple checks be completed by each potential borrower. One arduous example is taking a picture of a driver’s license, the applicant’s face, and both license and face together. Another is knowledge-based authentication questions (KBA). Both photos and KBA lead to significant funnel leakage, as the pictures are unpleasant to perform, and most of us forget the address at which we lived when we were 11 years old. Funnel optimization is key, and that balance often goes out the window for platforms seeking to truly minimize fraud.
Home-built fraud risk is costly and complex to build. There are relatively few individuals with deep knowledge of how to build great fraud prevention systems, so most platforms will avoid such a step. However, when one looks at top payments companies over the past two decades - PayPal, Square, Affirm (I’m biased) - all have built out their own fraud models and teams.
SKU and Merchant-level Underwriting
Contextual data is becoming more and more important in consumer credit. Why does someone need credit? How are they being reached? How dedicated are they to repaying their obligations? While I don’t pretend to understand the full psychology behind being more inclined to repay housewares over fenceable goods, there is strong empirical evidence of such behavior. Purchase finance has also long been available for furniture, mattresses, and appliances. These goods seem to be valued more by the borrower than other goods with a shorter useful life. While there is no possibility of repossession—BNPL is an unsecured loan—sleeping on that mattress or eating at the financed table reminds borrowers of their obligations. Thus, for equivalent FICO scores, platforms can expect better performance from the consumer purchasing the homegood.
Additionally, merchants can provide meaningful signals of the quality of the applicant. Great merchants build great brands that naturally attract more creditworthy customers. One example is Peloton vs Bowflex. Home stationary bikes were not sexy before Peloton. As Peloton became a social phenomenon and a must-have during the pandemic, it attracted creditworthy customers who were willing to pay a premium for their unique experience. Nothing against Bowflex, but the customer base looking for a bargain or next best alternative to a premium offering may be less creditworthy.
Separate but related, we saw divergent credit performance in the early days at Affirm for mattress merchants selling equivalent products but with different marketing strategies. Sloppy sourcing from Facebook or other social media sources can lead to attracting customers that can’t afford the merchant’s product and are not well suited for the BNPL offering.
Loan term is the most misunderstood component of BNPL. Specifically, many of the old-guard naysayers have ignored a very simple but fundamental component of credit risk theory. Bear with me as I try to explain….
If I lend a dollar to anyone (meaning the US adult population) today, that individual is extremely likely to be able to repay a dollar tomorrow. If I lend that same dollar for a week, there is a sliver of the population that may not be able to pay on that timeline, but it’s once again close to the whole population. A month out, the likelihood of repayment hasn’t really diminished. Six months incorporates incremental risk, 12 months even more. As we go out 2, 3, 5 years, the probability of repayment of that dollar diminishes. Thus, over time, the likelihood of default increases somewhat exponentially. I won’t get into the exact curve, but the point here is the shorter the term, the more confidence the lender has in being repaid by the population at large.
Incorporating revolving credit (credit card) makes performance prediction even more difficult. Not only does time affect the creditworthiness of an individual - positively or negatively - but severity of loss also increases. If a borrower is in financial distress, s/he will most likely utilize available credit to the maximum extent. Installment credit is different, with the loan being repaid linearly (amortizing), thereby reducing the severity of loss with each payment.
Two important factors come out of the BNPL product design.
First, BNPL platforms get performance feedback much faster than conventional consumer credit products. The “fool me once” element and early payment default allow lenders to iterate and adjust underwriting models much more rapidly than, say, credit cards or auto loans. BNPL credit performance will likely deteriorate much more rapidly than credit cards or personal loans during the next downturn. While on the face of it, the quick impact seems like a bad thing, this is actually quite a favorable characteristic. Performance deterioration will cause the BNPL lender to adjust models quickly and be better prepared for the evolving credit landscape. Given the amortizing loan structure, loss severity is limited. All lenders take incremental loan losses in credit downturns, but one could argue a high performing BNPL team should experience a smaller spike overall.
Second, BNPL can take the artificial form of revolving credit by continuing to provide installment loans, subject to an internal credit limit, on an ongoing basis to a borrower. Many customers of Affirm, AfterPay, and Klarna have multiple obligations outstanding at the same time. However, each purchase is amortized on the installment schedule and paid off in short order. The BNPL platform assesses the borrower each time s/he requests financing and has the opportunity to cut the individual off if risk has increased. This once again limits severity of loss versus a credit card offering and allows the BNPL platform to address a broader population.
An anecdote around the origins of Affirm helps crystalize this thesis. Max Levchin was initially focused on building a better credit score than FICO. FICO clearly has its flaws (but still is a good risk metric), and having been harmed by the process in the past, Max sought to build a better score. After assembling what he thought was a great model, he went out to pitch the new score. The universal response was, “If your score is so good, why don’t you lend your own money?” With lenders not wanting to be Affirm’s guinea pigs, the team set about offering a product that would provide as much feedback as possible in the shortest amount of time—digital purchase finance! One important note on building credit models: for models to function best, data needs to include both “bads” and “goods”. If the credit model always picks good risk, the lender has no idea how it’s actually performing. By starting with small dollar lending, Max and team found a relatively cheap source of R&D. The result over time was a sophisticated and nimble approach to consumer underwriting that leverages both traditional and modern data sets to instantly determine creditworthiness for a broader swath of the population.
(Lack of) Friction
As noted above regarding fraud controls, arduous application processes severely affect funnel conversion. The product objective for ALL consumer lending should be to optimize knowledge of the applicant with the smallest amount of data requested. Good borrowers want a great experience, and providing that easy, almost enjoyable, application process is critical to driving full funnel conversion. Most lenders only use more arduous processes, like cash flow underwriting via Plaid, if absolutely necessary. By no means am I endorsing subpar underwriting data; to the contrary, staging information collection to optimize conversion simply brings higher quality loan volume.
Don’t buy it?
Don’t take my word for it; you can read all about Affirm’s credit approach and performance in the latest DBRS securitization ratings report here. An extraction below on credit loss curves shows continued improvement in performance over time. More data, faster turnaround help build better credit models.
A view on consumer credit…
So why did I go through the trouble of writing out what makes BNPL unique? I have nothing riding on this—I hold an inconsequential amount of Affirm and Square stock, mostly out of my desire to see the sector succeed. Public markets will likely continue to trash BNPL leaders for months to years to come. I’m also very bearish on copycat BNPL lenders globally. While the business model is now established, equity value creation will likely remain elusive. As Klarna’s recent 85%+ fall in valuation indicates, a tough road lies ahead for those not on the cutting edge of a newish market.
However, I am very much interested in how startups take different approaches to established industries and leverage modern systems to build better business models. Along with captive SMB financing platforms like Square Capital and Stripe Capital, BNPL stands out as a truly innovative leap forward in providing efficient credit to an underserved population.
I sincerely believe that the next innovations in consumer and SMB credit will come from optimization of context through channel and data. New developments like proactive contextual digital lending offers bring meaningful upgrades to established practices like direct mail. I look forward to meeting the entrepreneurs who seek to understand why something has been done a certain way, then determine how to materially improve the process by starting from scratch.
From a venture perspective, investing in lending businesses is a death trap. And this is coming from someone who built his career in balance sheet-heavy models (SoFi, Lending Club, Affirm, Aven). Capital intensivity is unavoidable. Balance sheet-lite does not exist; those who claim they’ve outsourced all capital needs are likely to fail at the first hiccup. So why even look at it? Truly differentiated business models can generate outsized returns, but these businesses are true outliers, functioning as category-reinventing or category-defining efforts. I look forward to helping support the next generation of credit, but not likely the next BNPL offering!
Thanks for reading [99%Tech]! Subscribe for free to receive new posts and support my work.
Thanks for reading [99%Tech]! Subscribe for free to receive new posts and support my work.