Buy Now, Pay Later does not need another definition. Pay in 4, longer-term loans, a provider that pays the merchant (minus a fee) and collects from the customer — that machine is familiar.
What has changed is where it sits. It is no longer only a checkout app. It is a pricing of risk and a payment rail that is showing up inside wallets and agent-commerce experiments, while U.S. rules remain a patchwork after federal BNPL guidance was withdrawn and states began writing their own.
The Congressional Research Service has tracked the scale: Pay-in-4 originations rose from $2.2 billion in 2019 to $43.9 billion in 2023, with CRS estimating further growth to about $63.3 billion in 2025 (the Federal Reserve Bank of Richmond has published a similar 2025 ballpark near $70 billion). That is enough volume to matter operationally. It is not a reason to treat BNPL as free conversion.
Contribution Margin, Not Take Rate Theater
Provider fees run higher than card costs. The business case is conversion and basket on tickets where cash-up-front hurts — furniture, equipment, higher-priced apparel — net of fees, returns, and fraud.
Returns are the quiet leak. The merchant refunds; the installment plan has to unwind; OMS states get messy if partial shipments and restocks are not mapped. Fraud and friendly fraud ride installment offers the same way they ride cards, with a different dispute path. If those costs are not in the model, "AOV lift" is a vanity.
High-AOV B2C (furniture, equipment) is where the fee is easiest to justify if returns are controlled. Fashion with high return rates can erase the lift in a season. Segment the tender by category, not by enthusiasm.
Placement still matters. Revealing "as low as" only at payment is late. Showing it on the PDP without the fee in the margin model is early in the wrong way. Neither is a provider-logo decision. It is an offer-architecture decision.
B2B should not inherit consumer BNPL by default. Contract terms, credit, and invoicing already finance the account. Dropping Pay in 4 on a portal can conflict with credit holds and tax. If a distributor wants installments, it is a commercial-policy project, not a checkout app.
Do not run a four-way bake-off in the article the customer reads. Klarna, Affirm, PayPal Pay Later, and the rest differ by ticket size, geography, and where they already sit in your checkout. Start with one, measure net margin, then add. Multiple widgets that confuse the pay step can erase the lift.
Checkout Is Leaving the Page
Google's Universal Commerce Protocol work has put BNPL inside a broader agent and wallet path — including installment brands inside Google Pay as those surfaces expand. You do not have to bet the company on one protocol. You do have to assume some customers will choose installments somewhere that is not your hosted checkout.
That makes OMS, refund, and merchant-of-record questions part of BNPL, not a later IT story. Who authorized the plan? What if the agent-placed order is returned? What does customer service see?
Partial capture, split shipments, and restocks have to map to the installment object or the provider and the merchant will tell the customer two stories. Test those paths before you celebrate PDP conversion. The first ugly return will teach you what the widget did not.
Fraud and agent identity are the risk siblings. BNPL is the tender and the consumer-credit wrapper. Do not merge the three into one "AI payments" essay.
Regulation Is Becoming an Operating Constraint
In May 2025 the CFPB withdrew its 2024 interpretive rule that would have treated certain BNPL products more like open-end credit under TILA. There is currently no federal BNPL disclosure mandate of that kind in force. New York has moved the other way: BNPL legislation and proposed DFS rules covering licensing, disclosures, fees, disputes, and data use, with a long implementation fuse after adoption.
Operators selling into New York — or planning as if a large state's rule becomes a practical floor — should treat compliance and customer-communication duties as part of the payment integration, not as the provider's problem alone. Disclosures, servicing, and complaint handling will show up in your CX whether or not you hold the loan.
Service Has to Speak Installments
A customer who misses a payment is not only the provider's problem. They will call you. Scripts, order status, and "where is my refund" have to know the tender. If CS only sees "paid," you will promise a refund the provider has not processed.
Fraud teams should see BNPL as a distinct rail: new-account velocity, shipping to a first-time address on a high ticket, stacking plans. Do not copy-paste card rules and hope. Do not ignore the rail because "the provider underwrites." You still lost the goods.
Decide Like Finance, Integrate Like OMS
The leadership questions are dull on purpose. After fees, returns, and fraud, does this tender improve contribution on the SKUs that need it? Can refunds and partials complete without a person in a spreadsheet? If checkout happens off-site, do we still have the order object? Are we ready for the state rules that actually apply to us?
If those are unanswered, adding a widget is not a payments strategy. It is another button. Payment integration has to include the provider, the gateway, and the order states — not only the logo on the pay step.
Finance should own the go-live gate, with ecommerce and risk in the room. If marketing owns it because "conversion," you will discover the fee and the refund path after peak. That is a predictable way to buy a widget you cannot operationally afford.
Drop the 2021 adoption trivia. The product is mainstream. The remaining work is whether it earns its fee inside a real order, a real return, and a checkout that may not be yours.
Run the numbers on a month of actual tickets in the AOV band you care about, with fees and returns included, before expanding providers. If the math only works on a slide, keep cards. If it works on a SKU set, enable those SKUs and keep BNPL off the rest. Selective tender is allowed. A logo on every PDP is not a strategy.