Buy Now, Pay Later: The Products, Economics, Borrowers and Risks Behind the Checkout Button

The central distinction: a payment experience wrapped around several kinds of credit

Buy now, pay later is a convenient shopping label, but it is a poor accounting category. A six-week, interest-free purchase plan, a three-year interest-bearing installment loan and a cash advance from a provider with a BNPL business have different repayment obligations, funding needs and risk. Their shared app or checkout button does not make them economically interchangeable. The central questions are what the customer owes, when cash moves, who owns the receivable and how performance is measured.

The evidence checked for this article through October 4, 2026 supports two conclusions simultaneously: this is a substantial, expanding form of consumer finance, and headline market totals can be misleading without their boundaries. Federal Reserve researchers estimated $156.7 billion of US credit issuance by six providers in 2025, including $78.3 billion of pay-in-four. Their broader total includes Cash App Borrow and other lending outside conventional checkout BNPL. It is an annual flow, not outstanding household debt. [1]

This article examines the market architecture rather than valuing individual companies. Its original calculations are explicitly labeled; illustrative economics are hypothetical rather than estimates of undisclosed merchant contracts. Company figures retain their original reporting periods. Older cohorts remain older evidence, even when the report containing them was released recently. The analysis separates borrower convenience, merchant sales, lender profitability and consumer welfare because none is a sufficient proxy for the others.

Product taxonomy: interest-free does not mean every offer works alike

The most familiar structure divides a purchase into four payments, often with one quarter due immediately and the rest every two weeks. A longer 0% installment offer may look similar at checkout but leave the lender financing the purchase for many months. Interest-bearing installment credit adds an explicit finance cost and often supports larger purchases. Pay-in-30 can involve one future payment rather than four installments. These distinctions determine cash exposure before any model or brand is considered.

Deferred interest is a separate contractual mechanism. Under an actual 0% promotional rate, interest attributable to that zero-rate period is not later imposed retroactively. Under a deferred-interest offer, accrued interest can become payable if the promotional balance is not fully repaid by the specified deadline. The Regulation Z interpretation explicitly distinguishes the two. A deferred payment is also not necessarily deferred interest: postponing the first payment says nothing by itself about whether interest accrues. [6]

An installment conversion on an existing credit card is another adjacent product. It draws on a pre-existing revolving relationship and can use monthly fees or interest. A cash-flow comparison can include it, but a market tally that silently folds it into pay-in-four changes the object being measured. Likewise, a pay-in-full transaction is payment processing, not a loan merely because it passes through the same provider.

Scroll horizontally to see all columns.

One purchase, several institutions and several contracts

A merchant sells the goods or service. A platform displays offers, authenticates the shopper and coordinates the transaction. The legal creditor originates the loan; it may be the platform, a bank partner or another licensed entity. An investor may subsequently acquire the receivable. A servicer collects payments and handles account administration. A card issuer, network and processor can provide the payment rail without owning the customer’s installment credit exposure.

These functions can sit in the same corporate group or in separate firms. The word issuer is therefore ambiguous: issuing a virtual payment card is not necessarily originating the loan that funds the purchase. Sponsor can mean a bank supporting a fintech program or the sponsor of an asset-backed securities transaction. The relevant contract, rather than the logo at checkout, determines which meaning applies.

Consider a hypothetical $400 purchase with $100 paid immediately. The merchant may receive settlement net of its agreed fee, while the customer owes three later $100 installments. The initial $100 is repayment or a down payment, not lender revenue. If the loan is sold, the purchaser receives contractual cash flows under the sale arrangements; the platform may keep servicing income or retained risk. The merchant’s revenue from selling the goods and the lender’s finance revenue are different entries.

Bank partnership does not erase the bank’s obligations. The OCC’s December 2023 guidance addressed national banks and federal savings associations offering the specified short-term product, including through third parties. Its operational concerns included first-payment failures, disputes, fraud and model use. That bank-supervision document is distinct from the CFPB interpretation discussed later. [7]

Scroll horizontally to see all columns.

Sizing the market without changing its definition halfway through

The cleanest starting point is the CFPB’s six-provider series for pay-in-four. Its December 2025 report gives 2023 nominal originations of $43.9 billion across 335.8 million loans. The report’s main tables instead express dollars in 2024 purchasing power: $45.2 billion for the same 2023 activity. Mixing those two figures manufactures a disagreement. The table below retains nominal dollars throughout. [2]

The June 2026 Fed extension is a different exercise: researchers reconstruct 2025 US lending from company disclosures, sometimes allocating global activity through revenue or payment-volume shares. Its $156.7 billion total includes $31.2 billion of other short-term and $47.1 billion of longer-term credit. In particular, $23.7 billion assigned to Block is Cash App Borrow. This makes the total a broad provider-lending estimate, not a measured tally of identical checkout products. Product allocations are estimates, not company-certified segment accounts. [1]

Original calculation: $78.3 billion divided by $43.9 billion minus one equals approximately 78.4% nominal pay-in-four growth from 2023 to the Fed’s 2025 estimate. This is not a growth rate for all household BNPL debt. Nor is the two-year dollar expansion proof that the number of people using the product grew by the same percentage: purchase sizes, frequency, reporting coverage and estimation methods also matter.

Scroll horizontally to see all columns.

Users, transactions and balances: why familiar metrics fail to reconcile

A platform’s annual active users are usually customers who performed a qualifying action within a lookback window. Monthly app users may include people who made no purchase. Registered accounts can include dormant customers. Originations count loans, and payment volume sums transaction dollars. None directly measures the number of distinct people borrowing across the market.

The CFPB’s aggregate collection reports 53.6 million borrower relationships in 2023 and 6.3 annual loans per user per lender. A customer using two firms can appear twice in the first total. By contrast, its January 2025 linked-record study estimated 9.5 loans per borrower in 2022 across the covered firms. Different sampling and aggregation also produced a 2022 scaled loan count of 277.3 million in that study versus 273.8 million in the later aggregate collection. Those are separately sourced estimates, not values to splice into one growth series. [2, 3]

A stock-flow bridge makes the distinction concrete. In an idealized stable book, average principal outstanding equals annual originations multiplied by principal-weighted life in years, provided both use the same origination basis. On a $100 pay-in-four purchase with $25 at origination and $25 on days 14, 28 and 42, post-down-payment balances are $75, $50 and $25 for successive 14-day periods. The area is $2,100 principal-days, or 21 days relative to the original $100 purchase. Relative to the $75 initially financed after the down payment, weighted life is 28 days. Neither figure means the customer’s final due date is day 21.

Under those purely hypothetical assumptions, $78.3 billion of annual purchases would correspond to about $4.50 billion of average unpaid principal: 78.3 × 21 ÷ 365. This is a mechanical illustration, not an estimate of actual industry balances. Seasonality, missed payments, fees, alternative first-payment structures and product mix invalidate a literal extrapolation.

A public-company metric bridge, without a false league table

Affirm’s comparable fiscal-year metrics appear below. Its years end June 30; GMV is transaction value net of refunds, while active consumers and frequency use a 12-month lookback. These definitions support an internal reconciliation, not comparison with another company’s monthly app users. [8]

Original calculation: 27.782 million consumers × 7.0 transactions gives approximately 194.5 million FY 2026 transactions and $258 GMV per transaction. The FY 2025 equivalents are 133.4 million and $275. These use rounded frequency, not exact disclosed loan counts; net refunds and noncredit transactions complicate interpretation.

Klarna’s calendar 2025 global GMV was $128 billion and its active consumer count approximately 118 million. Its platform includes Pay in Full as well as credit products. This is a different geography, fiscal window and product denominator from the Affirm series. Dividing either firm’s annual revenue by platform volume produces a blended monetization ratio, not an APR or a universal merchant discount. Company profiles can analyze each model in depth; a market comparison cannot repair incompatible definitions by placing them in adjacent columns. [10]

Scroll horizontally to see all columns.

Where the revenue comes from

At the simplest level, revenue can come from the merchant, the borrower or another party buying access to the network. Merchant fees compensate a provider for some combination of distribution, credit, payment acceptance and promotion. Borrower interest belongs to interest-bearing products. Some models also have subscription or service fees, late fees, advertising, referral income, card-related revenue, loan-sale gains and servicing income. A particular firm need not have every source.

The economic question is the combined revenue earned for the risks and services supplied, rather than whether the consumer sees 0%. A merchant may subsidize a longer interest-free offer because it helps sell a high-margin durable good. Another merchant may prefer a shorter plan and a lower fee. A provider can also finance a transaction through a general-purpose card route without receiving the same negotiated merchant contribution as an integrated checkout partner.

Loan ownership changes the timing and label of income. Retaining a loan tends to produce interest over time; selling it can produce a gain or loss at sale and subsequent servicing income. A sale premium reflects a purchaser’s expectations about cash flows and risk, not free money independent of repayment. Growth in gains on sale may accompany lower retained balances, but that relationship depends on the actual transaction structure.

For merchants, the relevant comparison includes conversion, basket size, returns, tender substitution and financing cost. Higher average order value among BNPL users does not by itself prove causation: larger purchases may lead customers to select installments. Incremental sales are economically different from existing card sales migrating to a more expensive payment option. This is why a merchant-funded model can be valuable without every reported financed dollar being new commerce.

Funding: a short loan still needs cash before the customer pays

A provider must bridge the time between merchant settlement and consumer repayment. Corporate cash can supply that bridge, but it is scarce equity. A warehouse facility lends against eligible receivables under agreed advance rates and collateral tests. Forward-flow arrangements commit a buyer to purchase qualifying future loans under specified terms. Securitization pools loans and allocates their cash flows to securities with different priorities. A licensed deposit-taking bank can use deposits within its own regulated balance sheet. These routes have different costs and failure points.

Warehouse capacity is not the same as unrestricted cash. A commitment can be unavailable for loans outside eligibility rules, or require equity beneath the lender’s advance. Concentration limits may restrict a merchant, product or borrower segment. Delinquency triggers can reduce availability precisely when the platform needs more liquidity. An apparent pool of unused borrowing capacity therefore cannot automatically finance every new checkout approval.

In securitization, senior investors are typically protected by a combination of subordinated claims, excess collateral, reserves or excess spread. The protection is deal-specific. A short final customer maturity helps cash turn over, but a revolving trust may keep purchasing loans for much longer. Servicing continuity, refund allocation and the timing of collections still matter. Asset-backed borrowing does not necessarily remove assets or liabilities from consolidated financial statements.

Affirm’s FY 2026 filing identifies warehouse facilities, securitizations and forward-flow loan sales among funding sources, and discloses retained interests and loss-sharing structures. Those are concrete examples of the architecture, not evidence that every BNPL provider uses identical financing. [8] Economically, moving loans off balance sheet can transfer much of the credit risk while leaving servicing, representations, retained pieces and future funding-access risk with the originator.

A transparent funding-and-margin example

The following example is original and wholly hypothetical. It does not represent any provider’s contract, accounting margin or observed loss rate. Assume a $100 pay-in-four purchase, a $25 immediate payment, three further $25 payments two weeks apart and a $4 merchant fee. Assume principal-weighted exposure of 21 purchase-dollar days, 8% annual funding cost, $1.20 expected principal loss and $1.10 payment-processing and servicing cost. Ignore settlement lags, taxes, fixed expenses and capital requirements.

Funding cost is 100 × 21 ÷ 365 × 8%, approximately $0.46. Subtracting $1.20 and $1.10 from $4.00 after that funding cost leaves $1.24 transaction contribution. This is not net profit: fraud could exceed the allowance, customer acquisition and technology still cost money, refunds can reverse revenue, and equity must absorb unexpected losses. The example includes the merchant fee as revenue and all scheduled principal collections as return of principal, avoiding a common double count.

If expected loss rises by $1 per $100 purchase, contribution falls by $1. If the annual funding rate rises from 8% to 12%, the additional cost is only about $0.23 at the assumed 21-day weighted exposure. For a hypothetical 12-month equal-principal loan with payments at each month-end and no down payment, average principal-weighted life is 6.5 months; the same 4-point funding shock costs about $2.17 per $100. These calculations illustrate duration sensitivity, not a forecast that losses or rates will move that way.

The short product’s quick turnover is an advantage, but it cannot make a bad cohort profitable automatically. A repeat borrower may generate many small contributions while also accumulating obligations elsewhere. Conversely, a longer loan can support different pricing and customer value. Comparing raw annualized returns without the funding base, operating expenses and loss assumptions obscures the trade-off.

Scroll horizontally to see all columns.

Reported margins and promotional unit economics are not the same thing

Block’s November 2025 investor-day presentation offers an unusually explicit public illustration: forecast 2026 global pay-in-four economics per $100 GMV of roughly $3 gross profit, $1 recurring costs and $2 margin, with 27-day duration. It labels figures rounded forecasts, includes warehouse financing costs and describes directional equity returns using 80% leverage. This is management’s forward-looking framework, not realized 2026 performance or an audited industry benchmark. [9]

The difference between gross profit and the contribution left after credit and funding costs is material. Across firms, some costs appear above gross profit, others below it; a company-defined transaction margin can remove expenses that remain real at the corporate level. A percentage of revenue cannot be compared directly with a percentage of GMV. A gross profit return on average receivables cannot be compared with a profit margin on annual originations.

Provisioning adds a timing problem. Accounting can recognize an expected lifetime credit loss when a loan is originated while interest arrives over future months. A growing long-duration book can therefore show a larger current provision without a matching increase in current-period realized defaults. The reverse is also possible: a mature book can release reserves while new cohorts deteriorate. Loan-sale accounting, fair-value measurement and retained interests can create further differences.

A useful analytical bridge separates reported revenue, transaction-related costs, operating expenses, noncash or unusual items, tax and capital. It then asks whether performance reflects credit quality, funding, product mix, distribution terms or accounting timing. An attractive contribution figure is evidence about that defined layer of the economics; it is not permission to describe the entire company as profitable.

Underwriting: the amount and schedule can matter as much as the approval

Underwriting is not a single yes-or-no score. Identity and fraud checks establish whether the applicant and transaction are credible. A credit model estimates repayment risk. Affordability analysis examines whether the payment schedule fits resources and existing obligations. Policy selects a permissible amount, term, down payment and price. A counteroffer can reduce exposure without outright rejection, but it also changes what an approval-rate comparison means.

A six-week plan can generate repayment feedback rapidly. Successful earlier transactions may support repeat decisions, while a first-time borrower provides less internal history. The August 2026 Fed discussion describes providers beginning with smaller amounts and using repayment experience for later decisions. [5] This is a plausible information advantage for established networks, although historical repayment remains conditional on the loans previously approved and the economic environment in which they matured.

Original analytical example: a customer can repay a $25 installment after payday but cannot safely absorb a $100 debit the day before rent. A model predicting eventual repayment could score both offers similarly while their short-term liquidity effects differ. A small down payment helps only if it settles successfully and is not itself funded by another expensive obligation. Current account balances are a snapshot, not a promise about the funds available at every later due date.

Fast decisions create measurement challenges. Losses among approved customers cannot reveal outcomes for declined applicants. A model comparison can be distorted if merchant mix, loan terms or approval thresholds change at the same time. Higher approvals at unchanged observed losses can reflect a genuinely better model, a safer applicant pool or less mature new loans. Separating those explanations requires matched populations, consistent outcome horizons and an explicit treatment of counteroffers.

Fraud is not one loss bucket

Identity theft, stolen payment credentials, account takeover, synthetic identities, collusive merchants and first-party misrepresentation can all create losses, but their operational signatures differ. A legitimate customer who loses income after purchase has a credit problem; a fraudulent checkout using someone else’s credentials has an authentication problem. A merchant that fails to deliver introduces performance and dispute risk. Combining all three under default can make an underwriting model appear worse or better than it is.

The OCC noted that a first payment may fail when it settles, despite appearing acceptable at purchase, and that instantaneous decisioning and third-party dependence create operational exposure. [7] Analytically, this means approval, authorization, settlement and final collection are separate events. A successful card authorization is not the same as irrevocable money, and an automated debit can later be disputed.

Returns create another channel. A stolen identity can be used to acquire resalable goods; a false return claim can attempt to reverse the debt while retaining the merchandise. Genuine customers can also be harmed by fraud controls that incorrectly block their accounts or delay refunds. A lower fraud-loss ratio achieved through broad declines may carry a substantial customer and merchant cost.

A meaningful fraud trend therefore retains gross attempts, prevented attempts, realized losses, recoveries and false positives separately. It also identifies whether the denominator is approved transaction dollars, all attempted purchases or average receivables. The causal question is whether a control changed fraud outcomes for comparable activity, not merely whether an aggregate percentage declined after the provider stopped serving a difficult merchant segment.

Loan stacking: simultaneous obligations are a visibility problem

The CFPB’s linked-record study found that 63% of covered BNPL borrowers had simultaneous loans at some point in 2022, and 33% did so across multiple firms. About 20% averaged more than one origination per month. These are annual borrower-level measures: they do not mean 63% of loans were delinquent, or that the average borrower constantly carried several plans. The study estimated use by 21% of consumers with a credit record, rather than 21% of all adults. [3]

Overlap can be ordinary. Two affordable purchases a week apart create simultaneous plans. The risk grows when individually modest commitments compete for the same paycheck or when each lender observes only its own piece. A household may owe four providers small amounts on different dates and still face a large clustered debit day. Counting plans without their remaining principal and dates understates the scheduling problem.

Original hypothetical example: four $200 purchases made in the same fortnight each require $50 down and three later $50 payments. The total purchase amount is $800 and the initial cash demand is $200. If payment dates coincide, each later debit day requires another $200. Each provider may observe a $150 remaining exposure after origination; the household faces $600 across the four. The loans need not be delinquent to materially reduce the next month’s flexibility.

This example is not an estimate of typical behavior. It explains why utilization, remaining balances and calendar overlap answer different questions. Repeat use can signal satisfaction, persistent cash-flow smoothing or mounting dependence. The direction cannot be inferred from frequency alone. A lender’s low charge-offs also do not establish that payments to landlords, utilities or other creditors remained unaffected.

Credit reporting has four separate steps

First, the lender must furnish an account or transaction to a reporting company. Second, the reporting company must store and display it in an appropriate product. Third, a scoring model must use it. Fourth, an outside lender must obtain that report or score and use it in a decision. An announcement at one step is not evidence that all four are complete.

FICO announced Score 10 BNPL and 10 T BNPL on June 23, 2025, and its FY 2025 filing says those products were launched. These are distinct versions, not a retroactive change to every existing FICO score. Affirm’s November 25, 2025 explanation said it reported all US loans but that other lenders could not yet see those loans in consumer credit files or scores. That is a dated company statement. The Fed’s August 2026 discussion still described most providers as not reporting and no holistic view of BNPL in credit data. [14, 20, 15, 5]

Good reporting could improve debt visibility and recognize repayment, but the data structure matters. Four short plans in one month should not be naively interpreted as four unrelated long-term credit expansions. Reporting lags can leave a six-week obligation nearly repaid before another lender sees it. Disputed loans, refunds, paid plans and identity errors need accurate status and correction paths.

A reporting requirement can also change demand and selection. New York Fed researchers’ stated-choice experiment, revised February 2026, found strong aversion to interest charges and hard inquiries; simulated changes to the product altered both demand and the composition of interested borrowers. This is survey-experimental and modeled evidence, not a measured nationwide outcome of a new reporting mandate. [16] Better information is valuable, but the effect on access, pricing and customer understanding depends on actual implementation.

Repayment statistics: the denominator is part of the result

The CFPB’s 2023 aggregate data show a 1.83% loan-count charge-off rate and charged-off dollars equal to 0.92% of GMV. Neither figure is an annualized charge-off rate on average receivables. Its 4.1% late-fee incidence applies to loans at the four surveyed firms charging such fees, not every BNPL borrower. The corresponding 2022 figures were 2.63%, 1.71% and 5.2%. [2]

By contrast, the 2025 SHED found 26% of BNPL users reported a late payment during the prior year. These findings can coexist: one late installment on one of many loans puts a person in the annual late-user group without making every loan late or causing a charge-off. Fees can be waived or absent, and a late payment can be cured. The survey also reported 11% of BNPL users had a related overdraft or insufficient-funds fee. [4]

Original arithmetic illustrates why annual borrower incidence can exceed transaction incidence. If a hypothetical customer makes ten independent payments and each has a 2% late probability, the probability of at least one late payment is 1 − 0.98^10, about 18.3%. Actual payments are correlated and BNPL loans contain multiple payments, so this is not an empirical conversion formula. It only demonstrates why apparently conflicting rates can describe different outcomes.

Dollar losses and loan-count losses diverge when larger loans perform differently or borrowers repay part of principal before default. Down payments, recoveries, charge-off timing and product life also matter. A loan charged off 120 days after delinquency belongs to a different reporting timetable from one written off earlier. Comparing raw percentages without the event definition can reverse the apparent ranking of providers.

Scroll horizontally to see all columns.

Cohorts, maturity and the illusion of improvement

A rapidly growing loan book contains many new loans that have not had time to fail. Current delinquencies divided by all outstanding balances can fall even while comparable origination cohorts worsen. Conversely, tightening approvals reduces new balances and leaves a portfolio weighted toward older, troubled accounts; a headline delinquency ratio can rise during genuine credit improvement.

A cohort comparison groups originations by a common period and follows them at the same age. For pay-in-four, useful ages may be measured in weeks; for a multi-year loan, months or years matter. Cumulative principal loss at month 3 in one vintage should not be compared with month 12 in another. A mature cohort’s realized loss is different from a new cohort’s modeled lifetime expectation.

Original hypothetical: vintage A originates $10 million and eventually loses $100,000; vintage B originates $20 million and has lost $80,000 so far. The apparent ratios are 1% and 0.4%. If B is only halfway through its relevant performance window, this is not proof that risk improved 60%. If B’s final loss becomes $260,000, its mature rate is 1.3%. A growth-driven denominator temporarily concealed the deterioration.

Product mix complicates the same problem. Moving toward longer loans delays repayment and loss recognition. Moving toward repeat borrowers can improve observable performance without proving a model improvement. Changing merchant categories can change returns, fraud and seasonality. A coherent account of credit performance therefore distinguishes calendar-period results, origination-vintage outcomes, expected losses and the mix of the population actually served.

Returns and disputes are part of the credit product

The CFPB’s historical five-firm study found that 13.7% of 2021 loans involved at least a partial merchandise return. That is dated evidence of operational importance, not a current market return rate. [12] A purchase and its loan do not always unwind at the same moment: the merchant authorizes a refund, a payment message moves, the servicer adjusts principal, and any money already collected must be allocated.

Affirm’s checked help page illustrates why specific terms matter. It says scheduled payments remain due until the refund is applied; store credit does not reduce the loan; and refunds generally reduce the final remaining payments first, with paid interest treated separately. This is a provider-specific description, not a universal rule for all BNPL offers. A formal dispute process may have different treatment from an ordinary return. [13]

Original example: a customer finances a $400 basket and returns a $100 item after paying $200. If the merchant issues $100 in store credit, the loan may still have $200 due even though the customer no longer has that item. If instead $100 cash is refunded to the lender, remaining principal may fall to $100. Timing determines whether another scheduled debit occurs first. A partial refund need not reduce each remaining installment proportionately.

Travel, home improvement and elective procedures can introduce longer gaps between payment and delivery than ordinary retail. Cancellation rules, merchant insolvency and unfinished services create disputes that outlast a short repayment schedule. The financing provider’s relationship with the merchant, contractual recourse and dispute obligations become central. The customer’s eventual recovery, the merchant’s refund expense and the investor’s principal loss are separate outcomes, even when they begin with the same failed purchase.

Consumer outcomes: useful liquidity and financial strain can coexist

The 2025 SHED reported BNPL use by 16% of adults. Among users, 31% identified spreading payments as their main reason, 29% said it was the only way to afford the purchase and 16% selected avoiding interest. One in five used it for groceries or food delivery. The survey changed its main-reason question, so these figures should not be compared uncritically with earlier multiple-response convenience percentages. [4]

The August 2026 Fed analysis adds a clear financial-buffer gradient: use was 31% among adults whose savings could cover less than $100 of emergency expense, compared with 8% among those able to cover $2,000 or more. [5] These are group associations. They do not prove BNPL created the shortfall, and they do not mean every user is financially distressed.

A borrower with predictable income may value a fixed repayment schedule and avoid revolving card interest. Another may use the same plan because no cash remains for essentials. A third may overestimate future room in the budget. These are economically different situations even if all repay the lender. The welfare comparison depends on the alternative: paying cash, delaying the purchase, using a card, overdrafting, taking another loan or going without.

Low provider losses therefore cannot settle the consumer-protection debate. Automated repayment might reduce missed BNPL payments while increasing pressure on another bill. Equally, identifying financially vulnerable users does not demonstrate that removing the product improves their options. An informative evaluation examines total fees, consumption, hardship, other debt and the value of the financed purchase over time, rather than treating either approval or default as a complete measure of wellbeing.

What the research can establish, and what remains unsettled

Boston Fed research using 2021–2023 payment surveys found users had lower liquid balances and were more likely to revolve credit-card debt: 71% of BNPL users versus 40% of nonusers in 2023. Its use measure concerned the 30 days before the October survey, unlike SHED’s 12-month question. The findings identify selection and financial context, not a causal effect of BNPL access. [17]

The CFPB linked-record analysis explicitly does not determine whether BNPL causes higher non-BNPL balances or instead responds to declining available credit. [3] This distinction is fundamental: measuring customers after adoption without an appropriate counterfactual can mistake pre-existing strain for a product effect. Merchant adoption studies, lending-threshold designs and randomized offers can provide stronger identification, but their results remain tied to a studied product, market and period.

Norges Bank Working Paper 2/2025 examines how private BNPL information can improve consumer-bank lending decisions. Its contribution is evidence about information and credit allocation in a specific banking setting, rather than proof that furnishing every US pay-in-four transaction necessarily raises every borrower’s score. [18] The New York Fed preference experiment similarly illuminates product demand rather than realized lifetime welfare. [16]

The evidence is most persuasive when outcome and research design match. A study of spending can establish a purchasing response without showing long-term wellbeing. A repayment study can establish predictive value without showing fair access. A survey can reveal misunderstanding without quantifying default. The open question is not whether BNPL is universally good or bad, but which structures, customers and circumstances produce useful liquidity at a tolerable overall cost.

Competition extends beyond the BNPL button

Competition takes place at several layers: acquiring merchants, winning checkout placement, attracting consumers directly, funding receivables and managing repayment. A large wallet can distribute financing to an existing user base. A specialist can concentrate on higher-ticket categories. A bank can combine credit information and funding with an established account relationship. A commerce platform can make one provider particularly convenient to integrate. These are mechanisms, not a ranking of current winners.

Integration can reduce acquisition friction but create dependence on a few platforms or merchants. Direct-to-consumer cards broaden acceptance but can alter fee economics. A provider that supplies both payment and lending may earn revenue even when a transaction uses no credit, making platform growth different from loan growth. Merchants may offer several installment choices, and a preferred partnership need not be exclusive.

Consumer loyalty can arise from clear terms, reliable refunds and a familiar approval experience. It can also reflect merchant availability rather than an enduring preference for one creditor. Switching may be easy for a small purchase yet difficult after a dispute, when the customer must coordinate several systems. Competition on headline 0% therefore does not eliminate differences in servicing, fees, repayment flexibility or data use.

Concentration creates both benefits and risks. A larger network can spread technology costs, observe repeat repayment and negotiate funding. It can also amplify an operational outage or make a merchant dependent on one financing channel. The economically relevant market can vary by ticket size, geography, channel and product duration; a broad annual volume share does not establish market power within every segment.

Stress transmission: credit deterioration is only one channel

A household-income shock can raise missed payments, but the provider’s response depends on duration and funding. A short book can reprice or reduce new approvals quickly, yet doing so shrinks merchant financing at the same time customers are under pressure. A longer fixed-rate book retains exposure while funding costs or investor risk appetite change. Liquidity and solvency are related but distinct: a business can face a cash shortfall before lifetime loan losses consume its capital.

BIS researchers’ 2023 cross-country analysis highlighted profitability pressures and links between BNPL platforms and the wider financial system. Richmond Fed’s February 2026 assessment described limited financial-stability impact at the then-observed scale, while acknowledging possible spillovers and mixed welfare evidence. Those are dated research assessments, not assurances about future stress. [11, 19]

Original stress framework: weaker repayments can reduce warehouse availability; a more restrictive advance rate requires additional equity per loan; a weaker sale price reduces gains and cash proceeds; slower refunds increase customer-service demand; and merchant failures increase disputes. These mechanisms can compound without every borrower defaulting. The same cash buffer cannot simultaneously cover all contingent demands unless its size and availability support them.

The relevant financial-system exposure is the chain of retained risks. Bank facilities, securitization investors, deposit funding and corporate creditors can absorb different pieces. Annual purchase volume alone exaggerates balance-sheet exposure for fast-amortizing products, while an average balance alone can understate concentration in vulnerable households or counterparties. A careful assessment keeps both the aggregate scale and the distribution of risk visible.

US federal law: the withdrawn interpretation is not the current rule

The CFPB announced its digital-user-account BNPL interpretation on May 22, 2024 and published it on May 31, with July 30, 2024 applicability. It treated qualifying reusable digital accounts as credit cards for specified Regulation Z purposes, bringing billing, dispute and refund protections into the covered model. It did not generally import all of the separate card rules, including subpart G’s ability-to-repay and penalty-fee provisions. [21]

The next events must be distinguished. On May 6, 2025, the agency said enforcement based on that interpretation would not be prioritized. On May 12, 2025, an official notice expressly withdrew it. As of this article’s October 4, 2026 check, describing the interpretation as merely awaiting possible rescission is outdated. Describing withdrawal as a blanket exemption from consumer law would also be wrong. Product-specific federal and state obligations remain relevant, and interest-bearing or longer installment loans may already fall within traditional Truth in Lending coverage. [22, 23, 24]

The Financial Technology Association challenged the interpretation in federal court. The verified withdrawal is an agency action; it should not be retold as a judicial holding that BNPL is outside consumer-credit law. This review did not retrieve an official-hosted final dismissal document and therefore does not assign a dismissal date, prejudice status or merits ruling. That evidentiary limitation does not make the published withdrawal uncertain.

Congressional proposals remain separate. H.R. 9275, introduced June 11, 2026, would expressly address BNPL in the credit-card framework; the official action record checked here shows referral, not enactment. S.J.Res. 134 sought disapproval of the withdrawal, but the Senate rejected its motion to proceed on May 13, 2026. Neither event restored the 2024 interpretation. A proposed bill, an enforcement-priority announcement, withdrawn guidance and enacted law have different legal effects. [25, 26]

Scroll horizontally to see all columns.

New York: an enacted framework with implementation still proposed

New York enacted its Buy-Now-Pay-Later Act on May 9, 2025 as Chapter 58, Part Y. Its substantive commencement is tied to the 180th day after the Department of Financial Services promulgates implementing rules; authority to make those rules began immediately. The enactment date alone is therefore not the start date for the new licensing regime. [30]

The DFS page checked October 4, 2026 still lists its July 15, 2026 package, including proposed 3 NYCRR Part 423, under proposed regulations. The listed comment deadline was September 14. The proposed text specifies effectiveness 180 days after publication of a Notice of Adoption. This review found no BNPL final adoption on that official page and does not assign a fixed operative date. [31, 32]

This state example illustrates the layered US framework rather than providing a fifty-state survey. An enacted statute may delegate detailed implementation; an agency proposal can explain intended requirements without yet creating them. Existing licensing, credit, privacy and unfair-practices obligations may remain applicable independently. The identity of the creditor and the exact product structure still matter.

International comparison: three different implementation clocks

The UK, Australia and European Union illustrate different ways to bring short-term, often interest-free credit inside consumer-credit oversight. They do not establish a single global BNPL rulebook. Definitions, exemptions, creditor licensing, affordability requirements, dispute rights and treatment of merchants depend on the jurisdiction and the agreement.

In the UK, FCA regulation of covered third-party Deferred Payment Credit began July 15, 2026. The framework includes proportionate affordability assessment and access to the Financial Ombudsman for covered agreements, with authorization or transitional permission arrangements. Earlier agreements and specified merchant activity have different treatment. It would now be inaccurate to describe the UK regime simply as a future plan. [27]

Australia’s regime commenced June 10, 2025. Licensing and consumer-credit obligations apply, with modified responsible-lending arrangements available for qualifying low-cost credit contracts and conditions on transitional licensing treatment. That is not a declaration that every installment offer faces identical requirements. [28]

EU Directive 2023/2225 is adopted legislation, but its application date is November 20, 2026, after this article’s cutoff; the national transposition deadline was November 20, 2025. It expands coverage while preserving limited exceptions for supplier-deferred payments. This review does not assert that every member state implemented it identically or on time. The directive’s existence is distinct from the national rules applicable to a particular contract on a particular day. [29]

Economic comparisons need the same discipline as legal comparisons. Bank funding, merchant practices, debit usage, consumer reporting and enforcement systems differ across markets. A foreign default rate or loan limit cannot be transplanted into a US underwriting model without adjusting for those conditions. The useful comparison is the mechanism and its measured consequences, rather than the assumption that one country’s label captures another country’s product.

Scroll horizontally to see all columns.

What would materially change the assessment

New loan-level reporting could improve estimates of distinct borrowers, simultaneous balances and repayment calendars. Mature, comparable cohorts could show whether recent growth preserved credit performance. Evidence on refunds and complaints could establish whether servicing kept pace with distribution. None requires assuming that every new product announcement represents sustainable adoption.

For economic performance, the important distinction is between changes in merchant pricing, borrower yield, funding, losses and operating efficiency. For consumer outcomes, it is whether affordable purchases and avoided costs outweigh additional spending, fees and hardship for the populations actually using the product. For regulation, it is the operative text, effective date and judicial disposition, rather than an old headline or proposed bill.

The enduring analytical conclusion is straightforward: BNPL connects commerce, payments and credit, but those functions remain measurable separately. Product-specific cash flows explain its appeal; incomplete visibility and incomparable metrics explain much of the controversy. The most credible account preserves both sides: genuine consumer and merchant utility, alongside obligations and risks that do not disappear because the checkout experience feels simple.

Sources

1. Federal Reserve FEDS Notes, June 5, 2026: 2025 US provider-lending estimate and methodologyOfficial sourceBack to text: ↑1↑2
2. CFPB, December 10, 2025: The Buy Now, Pay Later Market; 2019–2023 six-provider dataOfficial source · PDFBack to text: ↑1↑2↑3
3. CFPB, January 13, 2025: Consumer Use of BNPL and Other Unsecured Debt; linked 2022 recordsOfficial source · PDFBack to text: ↑1↑2↑3
4. Federal Reserve, May 2026: Economic Well-Being of US Households in 2025, credit chapterOfficial sourceBack to text: ↑1↑2
5. Federal Reserve, August 19, 2026: Consumer & Community Context, BNPLOfficial sourceBack to text: ↑1↑2↑3
6. CFPB Regulation Z official interpretation: advertising and deferred interestOfficial textBack to text: ↑
7. OCC Bulletin 2023-37, December 6, 2023: Risk Management of BNPL LendingOfficial sourceBack to text: ↑1↑2
8. Affirm FY2026 Form 10-K, filed August 27, 2026, year ended June 30, 2026Filing / reportBack to text: ↑1↑2
9. Block Investor Day, November 19, 2025: Financial Outlook; 2026 forecast unit economicsSource · PDFBack to text: ↑
10. Klarna 2025 Form 20-F: global payment volume, product definitions and financial resultsFiling / reportBack to text: ↑
11. BIS Quarterly Review, December 2023: Buy now, pay later, a cross-country analysisSourceBack to text: ↑
12. CFPB, September 15, 2022: historical five-provider market study and consumer impactsOfficial sourceBack to text: ↑
13. Affirm help center: Returns and cancellations; checked October 4, 2026SourceBack to text: ↑
14. FICO, June 23, 2025: announcement of FICO Score 10 BNPL and 10 T BNPLSourceBack to text: ↑
15. Affirm, November 25, 2025: credit reporting, lender visibility and score distinctionSourceBack to text: ↑
16. Federal Reserve Bank of New York Staff Report 1167, October 2025, revised February 2026Official source · PDFBack to text: ↑1↑2
17. Federal Reserve Bank of Boston, May 23, 2024: BNPL use and financial characteristicsOfficial sourceBack to text: ↑
18. Norges Bank Working Paper 2/2025, dated January 30, posted February 7, 2025: private BNPL data in consumer bankingTechnical reportBack to text: ↑
19. Richmond Fed Economic Brief 26-05, February 2026: BNPL developments and financial-stability contextSourceBack to text: ↑
20. FICO FY2025 Form 10-K, November 7, 2025: BNPL scoring product launchFiling / reportBack to text: ↑
21. Federal Register, May 31 2024: original CFPB BNPL interpretation,89 FR 47068Official source · PDFBack to text: ↑
22. CFPB, May 6 2025: enforcement-priority announcementOfficial releaseBack to text: ↑
23. Federal Register, May 12 2025: CFPB withdrawal notice,90 FR 20084Official source · PDFBack to text: ↑
24. Congressional Research Service R48858: BNPL policy issues and optionsOfficial sourceBack to text: ↑
25. Congress.gov: H.R. 9275,119th Congress, all actionsOfficial sourceBack to text: ↑
26. Congress.gov: S.J.Res. 134,119th Congress, all actionsOfficial sourceBack to text: ↑
27. FCA: Buy Now Pay Later consumer protections and commencement, checked October 4, 2026SourceBack to text: ↑
28. ASIC: BNPL credit contracts and credit licensing, checked October 4 2026SourceBack to text: ↑
29. EU Directive 2023/2225, especially recitals 16–17 and Article 48SourceBack to text: ↑
30. New York enacted S3008C, Chapter 58 Part Y, May 9 2025; section 13 commencementOfficial sourceBack to text: ↑
31. New York DFS banking regulatory activity, proposed versus adopted rules, checked October 4 2026Official sourceBack to text: ↑
32. New York DFS proposed 3 NYCRR 423 text, July 15 2026, commencement clauseOfficial sourceBack to text: ↑