The Bank Behind the Buy Button

A small clause in a US financial filing deserves to be read as a new map of the bank's place in the economy. On February 14, 2025, Affirm signed a global agreement with Shopify to provide installments within online checkout. The filing with the US Securities and Exchange Commission says a merchant using Shop Pay Installments generally pays Affirm a fee consisting of a percentage of the sale and a fixed amount per transaction. Then comes the striking sentence: Affirm pays Shopify a fee for every transaction through the service. The filing discloses neither that fee nor either party's net profit. It does disclose the direction of payment: the finance company that might once have expected customers to come looking for money pays the company that owns the route to the moment in which the money is needed.1

This is more than a marketing agreement. In traditional banking, a customer first approached a financial institution for a card, loan, or account and then carried that purchasing power into the market. In the new structure, the customer reaches the market after choosing a shoe, device, or piece of furniture. The question is not which bank to visit, but whether to pay the displayed total now or in installments. Within those seconds, a door opens onto pricing, verification, credit assessment, capital allocation, settlement, collection, and risk. Most of it never appears on the screen. The paradox is not that the bank disappears, but that the order in which institutions appear differs from the order in which they bear obligations.

The purchase is not merely an excuse to sell a loan; the loan becomes a tool for completing a purchase whose intent already exists. This causal difference matters. The company controlling the storefront does not need a separate campaign to persuade someone that financing is necessary. It already has a person standing at the threshold of a specific financial act: a known item, price, time, and outcome that can readily be measured. Commercial language calls this “high purchase intent.” In banking language, it resembles a credit application arriving late, after someone else has completed the expensive work of finding the customer and awakening the need.

It would be a mistake, however, to reduce the arrangement to a technology company taking the customer while leaving the bank with the risk. The first lesson of this sector is that “the funder” may describe four different functions. A merchant platform may control checkout and the relationship with the buyer. A financial-technology company may own the credit-assessment model. A licensed bank may formally originate the loan under its legal and regulatory authority. Immediately or soon afterward, the loan may be sold to the technology company, then held by another investor or included in asset-backed securities. The name beside the buy button need not be the name on the agreement, or the name of the party that bears the economic loss at the first default.

Affirm's annual report for the year ended June 30, 2026, separates these roles numerically. It says Celtic Bank and Lead Bank originate most loans through the platform. Affirm applies its own risk model, but, in its description, the originating bank retains final authority over origination and requires compliance with its credit policies. The bank may initially fund a loan and then offer to sell it to Affirm. Most notably, Affirm said that through the report date it had purchased every partner-bank loan originated through its platform and offered to it. During the same year, it bought $40.2 billion in partner-bank loans, originated about $9.5 billion directly through licensed entities, and sold loans with $21.9 billion in unpaid principal to investors or through securitization arrangements.2

These facts demand precision. The bank is not legal scenery; its origination and compliance authority is material. Nor does it automatically retain the loan's risk through the final installment. Credit moves across balance sheets, contracts, and parties, carrying revenue and loss with it. Even “the bank in the background” describes its place in the interface, not the full reality of its work. Inferring the disappearance of its economic role from the disappearance of its logo replaces an intricate industry with a cartoon.

The merchant, meanwhile, is not always buying “cheap money.” It is buying a measurable commercial outcome: more completed checkouts, a larger average basket, or access to a new group for an expensive product. Shopify says that, after deducting the service fee, it generally pays the merchant the full captured order amount within one to three business days, while Affirm collects installments from the consumer. The merchant is not free of every burden; a purchase dispute or return creates service and refund pathways, and there are limits on editing an order after financing. But the timing difference is clear: the merchant does not wait for the buyer's installments to receive the sale proceeds.3

That is why Affirm's fee to Shopify says more than dozens of slogans about the future of finance. Three values are being priced implicitly: the merchant's value from closing the sale; the creditor's value from a financing opportunity whose quality and profitability can be tested; and the platform's value from bringing the other two together at the right moment. Public filings do not reveal each party's share of margin or prove that Shopify secures better terms than Affirm. But a finance company paying a commerce platform establishes that access to the customer is a negotiated economic input, not a free, neutral channel.

In economic terms, this resembles more than a conventional distribution channel. It forms part of a multi-sided platform linking seller, buyer, payment, and credit. A platform may charge one side less to expand its participation and recover value from another side that benefits from greater transaction volume. “Price” is therefore not only the cost of a loan but a map of cross-subsidies, fees, and incentives, some public and some embedded in contracts the buyer never sees. Since 2019, the Bank for International Settlements has described how the combination of data, network effects, and multiple activities can produce a reinforcing loop: more activity generates more data, improves assessment and marketing, and attracts still more activity.4

Platform analysis has a limit. Information is not the absolute property of whoever owns the button, nor is every piece of merchant data automatically available to a lender. Contracts, consent, privacy law, and credit rules determine what may be collected and shared. Technical integration alone does not show that the parties share one database or control credit assessment in the same way. BIS research using data from Mercado Libre and other fintech firms found that some e-commerce data predicted borrower losses better than traditional credit scores in those settings. That does not mean every commerce platform will reproduce the result or that additional data grant an unlimited right of use. It helps explain why a company's position in the chain of economic activity has become a potential credit asset rather than merely an advertising outlet.5

The power of the party controlling the flow appears in one of the driest places in a report: concentration disclosure. Affirm defined its fiscal-2026 gross merchandise volume as $50.2 billion. It said 22% of that volume involved Amazon, and its five largest merchants and platform partners together represented 44%. This does not mean Amazon received 22% of Affirm's income, or that five partners owned half its outstanding loans. The measure is transaction volume, not revenue, credit balance, or profit. It reveals something more exact: a small number of distribution gates can influence a large share of new demand even for a finance company with its own brand and application.2

This raises the question usually hidden by “strategic partnership”: which party can do without the other at lower cost? The finance company has technological capital in pricing, fraud prevention, integration, loan servicing, and a network of funders and investors. It has also built direct relationships: Affirm disclosed that 25% of its fiscal-2026 transaction count began through its application, website, and card. The commerce platform has buyer and merchant flow and the place of choice. The licensed bank has its license, balance sheet, regulatory experience, and authority to refuse credit that does not meet its standards. Negotiation does not end in one party's permanent victory. Each sells something the others cannot quickly copy, and replaceability varies with product, country, and stage of growth.2

The Shopify–Affirm agreement puts concrete boundaries around that power. The February 2025 filing granted Affirm exclusivity for certain comparable deferred-payment products in covered territories, initially led by the United States and Canada. It set an initial term ending June 8, 2028, with renewal and conditional early-termination rights. Affirm's disclosures also describe equity-related arrangements and incentives involving major partners. Exclusivity demonstrates the importance of securing a channel, not permanent ownership of its consumers. Termination and renewal provisions show that renegotiation belongs to the economics of the gate. Treating exclusivity as perpetual customer ownership means reading wishes rather than the contract.12

Apple offers a view of the interface from three directions, provided we do not conflate its products. In March 2023, it announced Apple Pay Later as a US installment product whose loans were issued by its subsidiary Apple Financing LLC. The company was moving from the screen into some lending functions. It later stopped making new loans through the program and, according to its official support page, removed its card from Wallet with iOS 18. Installments did not disappear from Apple Pay. In October 2024, Apple announced US installment options from Affirm and other providers depending on market. This is neither documented “defeat” nor proof that one financial model is always best. It is a clear shift between owning the credit product directly and retaining the payment experience while partners provide financing.6

In January 2026, Apple and Chase announced that Chase would replace Goldman Sachs as the issuer of Apple Card, in a transition expected to take about 24 months. As of October 2026, Apple pages continued to identify Goldman Sachs as the current issuer, so the transition cannot be described as complete. Its significance lies in what did not change names: the card remained Apple Card and its experience remained tied to Apple's wallet and services while the banking party was set to change. This credit-card product is distinct from Apple Pay Later; it is not a replacement of the same installment lender. In July 2026, Apple also introduced Apple Upgrade, leasing devices through Klarna; its official terms emphasize that it is a consumer lease, not a purchase or loan. Legal forms change while discovery remains with the company selling the device or owning the wallet.7

A reader may rush to declare the end of banks. Yet a bank is more than an application that delivers pleasant notifications. Credit requires risk pricing that interface appeal cannot measure, capital to absorb shocks, payment and settlement systems, fair-lending rules, and responsibility to customers and regulators. The Financial Stability Institute's October 2024 report A Two-Sided Affair did describe partnerships in which a bank supplies infrastructure while a technology company connects directly with consumers. It did not say the bank had become valueless. It described a changing value chain and dependence risks involving technology firms of significant size and bargaining power. A triumphalist narrative loses that balance.8

The other face of the system appears when we ask whether it reduced customer-acquisition cost or made borrowing easier precisely when hesitation might be useful. A US Consumer Financial Protection Bureau report covered six companies from 2019 through 2023. In that sample, originations rose from 19.8 million loans to 335.8 million, and 2023 originations were about $45.2 billion in inflation-adjusted dollars. This is not a census of the entire credit market, and user counts across the six companies cannot be added as unique people because a person may use more than one provider. It does show that a small button became a major channel for repeated small-debt decisions, turning tens of millions of purchase moments into later repayment relationships.9

Not every installment is free, and embedded finance does not describe a single kind of loan. Some transactions offer zero-interest installments; others carry interest. Under applicable market law, consumers are entitled to know the lender, terms, fees, and obligations before agreeing. Shopify directs buyers to Affirm for repayment and financing disputes. Apple's current guidance expressly says Apple does not provide the loans in its installment service and directs users to the financing provider for repayment or problems. A potential gap therefore separates the place where the decision is made from the place where the problem is solved. A user may remember the store or phone when seeking help, while the loan is administered by an entity visible only in the acceptance terms.310

Checkout speed cannot by itself test the quality of this experience. An institution should measure what happens afterward. How many buyers understood the installments? How many returns became disputes between merchant and lender? Did customers find the right entity before a payment deadline? How do default rates differ between comparable users offered finance at checkout and those not offered it? That last question requires an appropriate causal design, regulatory approval, and privacy protection; higher sales do not answer it. A deal may succeed for merchant and lender while worsening outcomes for some consumers, or widen credit access for people traditional data would exclude. Loan count alone cannot judge fairness or efficiency.

One of the most promising and dangerous parts of the story is that the company benefiting from finance does not always provide it. A commerce platform may know facts about its small merchants that a conventional lender cannot see: repeat orders, seasonality, returns, sales regularity, and customer mix. Used lawfully and tested properly, these signals can improve assessment of a business's repayment capacity. BIS platform-data research associated credit use with improved product variety and sales for some small firms, but it cannot show that the loan caused every later difference; selection and prior business quality may explain part of the result. Theoretical work in Big Techs vs Banks also suggests that platform information may confer power beyond cheaper finance, including an ability to deny a defaulting merchant services within the ecosystem on which it depends. Using commerce information to discipline a financing relationship is especially sensitive.511

In Saudi Arabia, separating names and roles is particularly important. The Saudi Central Bank permitted Tamara to provide buy-now-pay-later services in July 2023 and issued rules for the sector's companies that December. In March 2025, it announced a license for Tamara Finance to conduct consumer finance and BNPL activities. It likewise authorized Tabby in July 2023 and announced a BNPL license for Tabby Finance in November 2025. The sequence shows that the market cannot always be reduced to “a front-end platform with a conventional bank behind it”: the financial party may itself be a licensed finance company operating under the relevant regime. Saudi Arabia's 2024 Financial Stability Report also emphasized credit checks, reporting to SIMAH, and responsible-finance principles in risk control. The legal framework is not a copy of US arrangements, so bank names and loan-purchase mechanisms from the Affirm file should not be imported into a Saudi product without evidence.12

The region's greater commercial opportunity may lie not in reproducing another “pay in four” interface but in identifying where financial need originates. A small merchant may need inventory finance as orders rise; a shipping operator may need liquidity while awaiting collection; a student may need a payment plan during enrollment; a traveler may consider credit when seeing the final ticket price. These are hypothetical opportunities worth testing, not claims that a particular product exists or has succeeded. Each case should ask whether finance meets an independent customer need or was designed principally to turn purchase hesitation into debt. The distinction is ethical, but it also bears on portfolio quality and business sustainability.

I propose a practical “interface-that-remains” test. Instead of asking who owns the logo or has the most registered users, observe what happens when one of three parties changes while the other two remain. If the lender behind a merchant interface changes and installment selection holds steady under equivalent credit terms and decision quality, that is limited evidence of the merchant interface's power to generate demand. If the purchase interface changes and the funder loses users despite an unchanged product, dependence on distribution becomes visible. If the originating bank changes and the product cannot continue because of licensing or risk requirements, the strength of the financial infrastructure unseen by customers becomes visible. Such a test cannot always be arranged, is not costless, and must not be conducted on customers without required disclosure and controls. It makes “who owns the customer?” testable instead of leaving it as a conference story.

The test needs an economic scorecard that does not conflate parties' outcomes. For the merchant, measure the net causal increase in profit margin after payment and return costs, not merely higher order value. For the funder, measure the return adjusted for losses and cost of capital after platform, collection, and fraud costs, not total loan principal. For the originating bank, compare relationship fees and services with compliance, operational, reputational, and capital costs; the contract may also assign it some performance risk. For the consumer, measure cost transparency, comparability, the cost of reaching a remedy, and the longer-term welfare effect of credit beyond the button press. Each figure has a different denominator. Combining them produces the kind of illusion common to polished investment presentations.

A finer measure of economic sovereignty is the right to change the other party. Who can move a store among funders without losing buyers? Who can move the loan asset among funding sources without disrupting repayment? Who can change the originating bank while preserving terms and compliance? In a dispute, who can compel the parties to show a unified record of what the customer was promised, accepted, and owed? These rights are determined not by user-experience language but by contracts, data architecture, integration cost, legal duties, and market size. Sometimes the strongest party can theoretically be replaced, yet replacement requires a year of approvals and tests. There is no single “owner of the customer.” There are disaggregated rights, unequal bargaining capacities, and responsibilities difficult to divide legally even when divided commercially.

The counterargument is serious: this account may exaggerate platform power. Large banks have relatively inexpensive deposits, card networks, accumulated trust, direct credit relationships, and the ability to distribute products through their applications. Affirm is trying to reverse dependence on other companies' stores by building its application and card; its disclosed direct transaction share shows that a relationship can be built beyond checkout. Socially, embedded finance may increase competition, reduce financing cost, and make credit available where it is needed instead of forcing a separate bureaucratic journey. A BIS theoretical study concluded that platform information advantages and banks' cheaper funding advantages can complement each other rather than inevitably driving one party out.11

That argument corrects rather than destroys the thesis. The contest is not between a bank destined to win and a technology company destined to lose. It is over the right to assemble a financial service in the context that creates demand. A bank may acquire that capacity inside its application, a major merchant inside its store, or a credit company through a partner network that is difficult to replace. One party may retain the flow, another risk assessment, and a third funding. Counting users or loan volume without examining distribution and recontracting rights cannot distinguish a necessary partner from a supplier that only looks necessary until renewal.

One human consideration remains absent from ledgers of “value added.” The price is explicit at purchase, while the consequences of time are less visible. The button lowers friction now; it cannot by itself lower what happens in month three when obligations collide, a returned item's loan balance is not adjusted as expected, or a consumer asks a merchant about a lender's problem and asks the lender about an invoice owned by the merchant. Credit distribution through applications is therefore not justified solely by fee collection. The composite arrangement must provide a clear place for accountability and correction. Whoever benefits from hiding process steps from the user is responsible for ensuring that routes to remedy do not disappear with them.

Return to the sentence in the Shopify–Affirm filing: the finance company pays a fee to the owner of the purchase interface. It may look like an accounting detail, but it announces that banking has entered a phase in which presence at the point of decision can be priced as an asset separate from a lending license and loan balance sheet. The point is not that the bank has become invisible or technology has swallowed money. The person no longer has to seek a bank before the bank—or an intermediary working through it—seeks the person's need. The order of encounter has changed. In economics, a changed order of encounter often begins a new negotiation over returns, power, and responsibility.

1. Affirm Holdings, Form 8-K: global Shopify agreement, US Securities and Exchange Commission, February 14, 2025.
2. Affirm Holdings, Form 10-K for the year ended June 30, 2026, US Securities and Exchange Commission, 2026.
3. Shopify, Shop Pay Installments: Frequently asked questions.
4. Bank for International Settlements, Big tech in finance: opportunities and risks, Annual Economic Report, 2019.
5. Bank for International Settlements, BigTech and the changing structure of financial intermediation, Working Paper 779, 2019.
6. Apple, Introducing Apple Pay Later, 2023; Ten years of Apple Pay, 2024; Apple Pay Later support page.
7. Apple and Chase, Chase to become new issuer of Apple Card, January 7, 2026; Apple, Apple Upgrade launches in the United States, July 28, 2026.
8. Financial Stability Institute, Bank for International Settlements, A two-sided affair: banks and tech firms in banking, FSI Insights 60, October 24, 2024.
9. US Consumer Financial Protection Bureau, The Buy Now, Pay Later Market, December 2025.
10. Apple, Paying with installments using Apple Pay, updated 2026.
11. Bank for International Settlements, Big techs vs banks, Working Paper 1037, 2022.
12. Saudi Central Bank, Tamara and Tabby permitted to provide BNPL, 2023; Rules for regulating BNPL companies, 2023; Tamara Finance licensed, 2025; Tabby Finance licensed, 2025; Financial Stability Report 2024.