Credit used to be a product people went looking for. Now it finds them: at checkout, inside a marketplace, in the dashboard where a merchant checks yesterday's sales. US consumers took out about $157 billion in buy now, pay later credit in 2025. In Poland, more than half of BNPL transactions were under PLN 100.
Most of these loans are still originated and funded by licensed lenders. The customer just never visits them. We call this embedded lending, and it changes what banks compete on.
Our 2026 digital lending report, Mobile-first credit products, by Miquido's Dominika Będkowska and Sebastian Malczyk, is the basis for this article. Below, we compare embedded lending with point of sale lending and digital lending, explain how it works, and show three ways in for banks.
Embedded lending vs. point of sale lending vs. digital lending
The three terms overlap, but they describe different things. Embedded lending is about where the offer appears. BNPL and other checkout credit are one kind of embedded lending, tied to a single purchase. Digital lending describes the channel and the process, and it can happen entirely inside a lender's own app.
A working definition: embedded lending is credit offered inside a non-financial company's product at the moment of need. A licensed lender originates and funds the loan in the background. BNPL is the best-known form. Embedded finance also covers business credit: Shopify Capital, Stripe Capital and Adyen Capital lend to small businesses based on their sales data.

The amounts explain why banks struggle to copy the model: a pay-in-4 loan averages about $135, while a prime personal loan is closer to $10,000. As our report puts it, “Credit is a payment option, not a separate product in the menu.”
How embedded lending works
One embedded loan is split across three parties. Each does what it is best at.
- The bank holds the licence, runs KYC/AML and fair lending compliance, sets the credit policy and funds the loan from its balance sheet.
- The embedded lending platform provides the API, a decision engine that applies the bank's policy in real time, and servicing: repayments, collections and reporting.
- The distributing platform (a merchant, marketplace or SaaS product) owns the UX, the moment of need, the transaction data and the customer relationship.
Affirm is the clearest US example. According to its 2026 annual report, substantially all of its US loans are originated through Celtic Bank, with Lead Bank as a second partner. Affirm's model decides within seconds, under the banks' underwriting policies.
The roles can also merge. Adyen Capital scores merchants on their payment data, pays out the funds and collects a fixed share of each sale. In Poland, PKO Bank Polski and Allegro signed an embedded finance deal in 2025 that puts PKO business loans inside the marketplace.
A partnership does not remove accountability. US regulators expect sponsor banks to oversee their partners' compliance under the 2023 interagency guidance on third-party relationships. The 2024 Synapse collapse showed what happens when that oversight fails.
Why speed no longer sets banks apart in digital lending
Fast decisions used to be a selling point. Now they are the minimum. Poland's mBank won an innovation award for a 30-second loan back in 2014. By Q1 2026, 91% of Upstart's loans were fully automated.
Slow lenders lose customers. Fast ones no longer stand out for it.
J.D. Power's 2026 U.S. Consumer Lending Satisfaction Study shows that satisfaction drops 41 points when approval takes more than an hour. It falls 47 points when funding takes more than a day. Non-bank lenders fund 68% of customers within a day, against 58% at banks.
“Instant application processing and fast decisions have stopped being a source of advantage for banks,” our report concludes. “It's now simply the entry price for playing the game at all.”
Embedding credit goes further: a pre-qualified offer has no application process left to speed up. The real work moves to what happens before the purchase and after the decision.
“The biggest problem is no longer what happens inside the decision engine. It's what happens around it, on both sides.”
Sebastian Malczyk Fintech Solutions Lead at Miquido
Benefits of embedded lending for platforms, customers and banks
Each party gains something it could not reach alone. Platforms and merchants get conversion and a new revenue line. A Stripe A/B test across more than 150,000 checkout sessions showed BNPL lifting revenue by up to 14%. Over two-thirds of BNPL volume was net-new sales.
Customers get credit at the moment of need, often at no cost. According to the Fed, $98.7 billion of US BNPL credit issued in 2025 carried a 0% APR. The catch: 11% of BNPL users paid an overdraft or NSF fee because of a BNPL payment.
Lenders get repeat business at a lower acquisition cost. 96% of Affirm's transactions in fiscal 2026 came from repeat customers.
Banks get early access to future borrowers. Poland's credit bureau BIK found that 54% of consumers who started with BNPL later took out another loan. A bank absent at checkout loses that first contact.
The new dividing line in point of sale lending: Distribution, not technology
At checkout, the scoring engine is no longer the moat. What wins is access to the moment of purchase.
“The market has stopped rewarding the scoring engine on its own; it now rewards access to the moment a customer decides to buy. Distribution, not technology alone, decides who wins today.”
Dominika Będkowska Market Researcher at Miquido
Walmart's OnePay made Klarna its exclusive BNPL provider in March 2025, replacing Affirm. Walmart had accounted for about 5% of Affirm's volume. Both companies had mature technology, so the distribution contract decided the outcome.
Market shares tell the same story. Afterpay (Block) and PayPal led US pay-in-4 lending in 2025, with 38% and 32% of volume. Each is backed by its own merchant and consumer network.
For banks, the shift is good news. They rarely beat fintechs on technology. As regulated financial institutions, they hold what distribution partners need: a licence, cheap funding and millions of customers who have already passed KYC.
What embedded lending means for banks
Moving credit to where customers buy changes four things about how a bank works. None of them is solved by a faster decision engine alone.
Small tickets need a separate, cheaper decision path
Ticket sizes in embedded lending are tiny by banking standards. 52.3% of BNPL transactions in Poland in 2025 were under PLN 100, while the average cash loan was about PLN 26,000 (BIK). The average US pay-in-4 loan was about $135. A risk process built for cash loans does not scale down to purchases this small.
For the same reason, SebastianMalczyk argues in our report that a bank cannot transplant its existing risk model. It needs a parallel, cheaper decision path built into the place where the customer buys, without pulling bureau credit scores for every purchase. WeBank in China shows how low unit costs can go. It reports annual IT costs of under 2 yuan (about €0.25) per account, less than a tenth of its peers'.
Rejection becomes a product
When most applications come through partners, how a bank says no matters as much as how it says yes. From 20 November 2026, Article 18 of the EU's revised Consumer Credit Directive (CCD2) applies. It gives consumers the right to human intervention, a clear explanation and a review of automated credit decisions.
The CJEU set a similar bar in Dun & Bradstreet Austria (C-203/22): people must get an understandable explanation of how their data was used. Regulation B already requires US lenders to give specific reasons for every denial.
Reputation raises the stakes. Gen Z is twice as likely as baby boomers to post feedback on review sites and social media. Our report's conclusion is that rejection “becomes a separate product that has to be designed in its own right: with its own UX, its own communication, and its own appeals mechanism.”
Getting rejection right is an architecture decision more than a copywriting one. Sebastian Malczyk's advice is to keep risk-assessment logic separate from customer-communication logic.
“When those two layers are tangled together, every regulatory change forces a rebuild of the whole system instead of just one module.”
Sebastian Malczyk Fintech Solutions Lead at Miquido
Retention has to happen between loans
A Polish cash loan runs for about four years, and a credit-only app gives customers little reason to open it in between. “If a credit app only lives through the moment a loan is granted, it becomes a dead icon on the customer's phone,” the report warns.
The leaders fill the gap with everyday use. Kaspi.kz customers make 77 transactions a month, and Klarna's card has 6.5 million active users. Revolut grew its loan book 120% in 2025 on a base built through payments.
Retention tools alone do not fix the problem. Sebastian Malczyk puts the risk foundation first and builds the retention layer on top of it.
“Loyalty mechanisms, points, programmes like RevPoints or Rewards Club, sit in the engagement layer and solve a real problem of contact frequency, but they don't change an institution's actual ability to make better credit decisions.”
Sebastian Malczyk Fintech Solutions Lead at Miquido
AI pays off most when markets turn
Using AI no longer sets anyone apart. According to the Bank of England and FCA, 86% of UK financial services firms now use AI, up from 75% in 2024. As the report puts it, AI in scoring “has stopped being a differentiator and has become table stakes.”
What matters is how a model behaves when conditions change. A BIS study of a Chinese fintech lender found that machine learning with non-traditional data predicted defaults better than traditional models during a credit-supply shock. The caveat: a better model still needs stable funding behind it.
Three ways banks can enter embedded lending
There is no single right model. A bank with an established customer base starts from a different place than a fintech building distribution from zero. Most banks will choose one of three routes, or combine them.
1. Lender behind a platform (lending-as-a-service). The bank originates and funds loans that a fintech or merchant distributes. It gains scale fast but gives up most of the customer relationship. Celtic Bank, Lead Bank and Cross River Bank behind Affirm are the US template.
The contract decides how much the bank keeps. PKO Bank Polski secured visible branding in its Allegro deal. Goldman Sachs, by contrast, is handing Apple Card to JPMorgan after costly years as a background issuer.
“The first clue is whether the platform handles the payment itself or only sends the customer to buy elsewhere. A price comparison site can have heavy traffic, but the payment decision happens outside it. Then look at the data: how much of its volume the platform already finances and how often customers return. Allegro Pay finances around 17% of Allegro's GMV. Grab's loan book is growing 65% year on year, helped by daily use of its app.”
Dominika Będkowska Market Researcher at Miquido
2. Own BNPL at the point of sale. The bank runs credit under its own brand and keeps the customer relationship. Alior Pay lets Alior Bank customers defer almost any transaction in its mobile app for 30 days, up to PLN 3,000.
PKO Bank Polski went a step further and connected its pay-later product to BLIK Pay Later, which works in more than 20,000 online stores. The trade-off is distribution: the bank has to win merchants itself or join a shared network.
3. Value-added services in the bank's own mobile app. The aim is to stay in the customer's daily routine between loans. PKO Bank Polski's IKO app sells about 3.5 million extra services a month, from phone top-ups to motorway tickets. Each of those moments is a natural place for a credit offer.
“Of the three routes, being the lender behind someone else's platform puts the least customer acquisition cost on the bank. In the US, partner banks such as Cross River Bank fund Affirm's loans, while the customer at checkout usually sees only the Affirm brand. The price can be a weaker relationship and less control over margin and interface, depending on how the bank negotiates data access. For some banks, combining this model with their own BNPL or in-app services makes more sense than picking one route.”
Dominika Będkowska Market Researcher at Miquido
The routes reinforce each other: services bring daily traffic, own BNPL turns it into credit, and partnerships add reach. We build both value-added services for banking apps and lending solutions such as AI credit scoring. Our engine for Nextbank has handled more than 500 million loan applications across 7 banks.
Want the six challenges and six trends behind these numbers? Download our mobile-first credit report.
What is the difference between embedded lending and point of sale lending?
Point of sale lending is a subset of embedded lending, tied to a single purchase at checkout, as with BNPL and other POS financing. The broader category covers any credit offered inside a non-financial platform or product, including working capital for merchants on platforms such as Shopify or Stripe.
What is digital lending and how is it different from embedded lending?
Digital lending describes the channel and the process: online applications, automated decisions and digital servicing, often in a lender's own app. Embedded credit is defined by where the offer appears, inside someone else's product. Most embedded loans are digital, but most online loans are not embedded.
How does embedded lending improve customer retention?
It keeps credit present in everyday activity instead of only when a customer needs a loan. 96% of Affirm's transactions come from repeat customers. In Poland, 54% of consumers who started with BNPL later took out another loan (BIK).
Is embedded lending regulated?
Yes, and the licensed lender carries most of the duties. CCD2 covers most BNPL in the EU from 20 November 2026, and the UK's FCA has regulated it since 15 July 2026. The US has no federal BNPL rule, but sponsor banks are supervised on their fintech partnerships. New York also has its own BNPL law.
How do you choose the right embedded lending provider?
When comparing embedded lending platform vendors, check approval rates by risk segment, not blended averages. Ask who issues rejection notices and whether they meet CCD2 or Regulation B. Get full API documentation before signing, and agree data ownership and exit terms up front.


![[header] embedded lending
what it means for banks (1)](https://www.miquido.com/wp-content/uploads/2026/11/header-embedded-lending_-
what-it-means-for-banks-1-1920x1280.jpg)



![[header] embedded lending
what it means for banks (1)](https://www.miquido.com/wp-content/uploads/2026/11/header-embedded-lending_-
what-it-means-for-banks-1-432x288.jpg)
![[header] technical feasibility assessment – process & checklist](https://www.miquido.com/wp-content/uploads/2026/11/header-technical-feasibility-assessment-–-process-checklist-432x288.jpg)
![[header] code is getting cheaper. the hard part of software development is moving elsewhere](https://www.miquido.com/wp-content/uploads/2026/10/header-code-is-getting-cheaper.-the-hard-part-of-software-development-is-moving-elsewhere-432x288.jpg)
