Banks' Blind Spot: Marketplaces Are Winning Lending — The Counterplay Starts With AI
by Yannis Larios
Distribution has moved to the point of purchase; banks win it back with instant, explainable decisions on a regulated balance sheet.
On a Tuesday evening in Warsaw, a shopper on Allegro, Poland’s largest marketplace, taps “Add to cart” on a new laptop. At checkout she sees the usual options, plus one more that’s become strangely familiar: split the payment, or get an instant line of credit. She doesn’t leave the site, doesn’t upload a tax return, doesn’t wait days. She’s approved in seconds, buys the laptop, and goes on with her week. The lender isn’t her bank. It’s the marketplace she already uses twice a month.
If you are in the payments and banking market I am sure you’ve heard the joke: "airlines are just credit card companies that happen to own planes" — shorthand for how co‑branded cards and loyalty programs drive outsized profits via interchange, fees, and breakage, with the flight acting as customer acquisition. Nowadays, marketplaces apply the same economics at checkout: card issuing and loyalty tied to the basket, plus point‑of‑need financing, create high‑margin financial revenue on top of thin‑margin commerce. In Europe, the punchline is migrating to retail: marketplaces and vertical platforms are becoming lenders that happen to run shops. At the point of need they see exactly what the customer wants, in what basket, with what history; while banks see a generic applicant in a queue for a loan.
That tiny moment is the shift that should keep bank boards awake. Retailers and platforms now sit on the richest behavioral signals in the economy: what we browse, what we buy, when we return items, how fast we pay. They are using that proximity and data to sell credit at the precise moment it’s most relevant: the moment of purchase. The result is a quiet -yet fast growing- re-routing of lending volume from bank channels into embedded ones. This is not a fintech sideshow; it’s a structural change in distribution of lending products.
European examples are everywhere: Poland’s Allegro, one of Europe’s largest marketplaces, reports that its consumer-credit arm Allegro Pay now finances a meaningful share of transactions and is expanding issuance, even adding a payment card so users can take their Allegro Pay limit everywhere. According to public data, in 2024 it financed about 14 percent of GMV in a quarter and kept leaning in through 2025, with management repeatedly citing consumer loans as a revenue driver.
On the seller side, the pattern is the same. From 2020 to June 2025, Amazon Germany routed eligible marketplace sellers into ING’s lending flow, pre-qualifying them inside Seller Central and sending them to an ING page to complete the application for loans up to €750,000. The pitch is brutally simple: we already see your sales; let’s turn that into capital (Note: Since July 2025 ING stopped offering the service to Amazon, which has now switched to YouLend).
Niche verticals show the same momentum. Berlin’s pharmacy marketplace Onfy embedded lending with Banxware so pharmacies can secure funds digitally in minutes and receive cash within days, without leaving their operating flow. In Poland, ecommerce platform Shoper teamed with PragmaGO so SMB merchants can draw working capital based on their Shoper Payments turnover, all online.
Even legacy captives are moving like fintechs. Volkswagen Bank rolled out a 15-minute instant loan product across dealers with credi2, shrinking the consumer wait from “come back tomorrow” to “you’re approved, let’s go.” This isn’t just for cars; dealers can offer quick financing for accessories, service, or micromobility too!
Why this threatens banks
If you’re a bank, the danger isn’t just lost transactions. It’s losing three compounding advantages at once.
First, distribution. The loan is presented exactly where the need arises, not in a separate banking app. That context outperforms any generic “apply for credit” banner a bank can show.
Second, data. Platforms know the customer’s real behavior. Allegro sees cart size, frequency, returns; Shoper sees merchant turnover; Amazon sees seasonality by SKU. Those are better default predictors than a static bureau score for many use cases.
Third, speed. Embedded lenders collapse time to "yes". VW Bank’s 15-minute promise is now a public benchmark. Once customers experience instant credit, next-day feels like denial.
Add Europe’s regulatory evolution and the moat looks even wider: The EU’s open-finance push seeks to standardize access to broader financial data with customer consent, while payments reform continues to sharpen API obligations for banks and PSPs. Meanwhile, the EU AI Act explicitly treats credit-scoring AI as high-risk, demanding documentation, testing, and human oversight. Net effect: platforms keep their proximity and data; banks inherit heavier governance unless they modernize the stack and partner smartly.
The countermove: AI credit scoring at the edge
Here’s the twist that’s both contrarian and practical: banks can win embedded credit back by scoring faster and fairer than the platforms, using richer data than they do today. And there is one more reality check - marketplaces aren’t balance sheets. Most rely on bank partners, warehouse lines or securitisations to fund credit, and they face capital, risk and regulatory limits. They can originate at the edge, but durable scale still needs regulated capital and disciplined risk management. That’s the strategic opening for banks: bring the balance sheet and governance to the point of need, not just a portal link.
So the countermove that Banks can strategically enforce is not a prettier front end; it’s an underwriting brain that works at point-of-need! Think about the data a European bank already touches or could access with consent:
Card-acquiring flows for their merchant clients (sales velocity, seasonality, refunds, chargebacks)
Open-banking account histories (balances, inflows, outgoing commitments)
Lightweight third-party signals (business registry data, ecommerce reviews, sector indices)
A more extreme frontier - Some lenders test smartphone-derived behavioural signals as credit proxies: accelerometer patterns, typing cadence, GPS stability, even battery/charging habits (!) and app-graph and time-of-day usage. The promise is reach and speed for customers with limited credit history; the risk however is opacity, bias and sensitive inference. A word of caution is due here: European debates on algorithmic transparency remind us that if users can’t see how features drive outcomes, trust erodes fast. If banks go near this space, they should draw clear red lines (no health, ethnicity, or location profiling), require explicit opt‑in, log which features were used, and return plain‑English reason codes tied to policy. Done right, you expand access without drifting into surveillance.
Feed that into an AI underwriting model tuned for specific verticals, and you get instant, contextual decisions that feel like the marketplace experience but carry bank-grade guardrails. You can pre-approve a restaurant for a working-capital line every Friday morning based on the last 26 weeks’ flow; you can approve a laptop purchase in three clicks because the account has stable payroll and low commitments.
A European acquirer’s internal plan would break it down crisply:
Product scope: merchant lending and embedded financing for SMBs with at least six months of processing history, initially in retail and hospitality.
Mechanism: combine acquiring data, open banking via a provider like Tink, and selective external datasets; predict default probabilities and recommend loan size and terms per merchant.
Expected lift: increase origination volume ~20 percent with controlled loss rates; net revenue up roughly 15 percent on the lending line; credit write-offs down ~10 percent via better risk discrimination; merchants with financing churn ~5–8 bps less over a year.
Translate that into board language: more approvals, faster; losses at or below baseline; and a stickier merchant base. That is exactly the lever banks need to stop embedded lenders from eating their lunch.
How to build it without breaking things
Start with a tight use case. Pick one vertical where your bank already has data gravity. If you run card acquiring for 15,000 restaurants, start there. Design a small catalogue: cash-advance-style products with revenue-based repayments; 6- to 12-month working-capital loans; and an overdraft-like line for inventory swings.
Assemble the data spine. Wire in your acquiring data. Add open-banking feeds with explicit merchant consent. Normalize it into a single decision lake. You don’t need perfection; you need coherence and recency.
Build the model and the rails. Use an ML platform that supports reason codes and bias testing. Treat the score as a policy-bound engine: it must output the decision, the amount, a simple explanation to instill trust (“approved due to stable weekend revenue and low chargebacks”), and a confidence band. Expose it via an API so you can drop it into partner checkouts.
Add the human loop where it matters. For borderline cases or high amounts, route to a human underwriter with the model’s features visible. That keeps you on the right side of the AI Act and is good discipline regardless.
Partner to distribute. Co-brand with platforms your customers already use. Your model, their checkout. The platform earns a referral fee; you get high-intent flow without buying search ads. That’s the Amazon-ING logic, applied locally and repeatedly.
Instrument it like a payments product. Decision time in seconds and minutes. Take-up rates by funnel. Delinquency by cohort and vertical. Pre-approved line utilization. Underwriter touches per 100 loans. Close the loop fast: when delinquency ticks up in a sub-sector, pull back offers automatically.
Stories from the front lines
The vertical vignettes are already there, writing themselves.
Marketplaces: Allegro turned its embedded credit into a growth lever, then extended it with a card so Allegro Pay limits can be used off-platform. Result: more financed baskets, more frequency, and a longer relationship with the buyer. A bank could be the balance-sheet partner behind that card or compete with an equivalent limit inside partner checkouts elsewhere.
Seller financing: Amazon’s program, formerly with ING and now with YouLend, proves the broker-lender model works at national scale. The marketplace points the hose; the bank underwrites and services. There’s no rule that says only big tech can do this. Regional marketplaces and national retailers can replicate it with local banks that bring capital, compliance, and an AI brain tuned to the sector.
Vertical software: Shoper and Onfy show that embedded credit fits not only mass retail but also SMB platforms and even regulated niches like pharmacies, provided the underwriting is fast, digital, and respectful of the workflow. Banks can plug into these flows with API-first credit—no pop-ups, no PDFs, just decisions.
Captives reimagined: VW Bank’s 15-minute credit is a reminder that regulated lenders can be just as fast when they reorganize around the decision rather than the process. That’s a cultural cue for universal banks: speed and safety are not opposites if your model is conservative, monitored, and explainable.
The governance reality
There’s of course apprehension to early-bolt AI onto old processes. And it is fully justified. In Europe, credit scoring is a high-risk AI use case. That means documentation, testing, monitoring, and a human fallback. The upside is that banks are built for this; you already run model-risk committees, internal audit, stress tests. Treat the underwriting model like market risk: version it, back-test it, and install a kill switch when drift appears. Pair that with open-finance hygiene: explicit consent capture, purpose limitation, and clean revocation flows. These aren’t compliance chores; they’re how you scale trust.
Scorecard — What the Bank Should Track
KPIs should be tight and decision‑useful. One page, same format every quarter.
Decision time (median / 95th): 5 minutes / 15 minutes. If we slip, we fix it next sprint.
Approvals at the same risk: lift in "yes" decisions without raising expected losses; show by product and segment.
Cost of risk: expected loss and 90+ day arrears; pre‑agreed guardrails with automatic tightening when breached.
Risk‑adjusted margin: income after funding and losses (by product and by partner). No growth without margin.
Partner contribution: share of new lending from marketplaces/vertical SaaS; include concentration limits per partner.
Line use & repeat: % of approved limit used within 90 days and repeat‑borrow rate; early read on product‑market fit.
Retention & value: tenure and cross‑sell uplift for financed vs. non‑financed customers.
Boardroom Action Plan: Three Decisions in 90 Days
Approve the shift and budget. Vote to make instant, explainable underwriting a priority in one vertical this quarter. Allocate a pilot budget, name an exec owner, and set a firm go‑live date within 90 days.
Authorize one distribution partner. Green‑light a co‑branded checkout offer with a named marketplace or vertical platform. Approve a simple revenue‑share, a decision‑time SLA measured in minutes, and consent‑based data‑sharing terms.
Require a quarterly scorecard. Demand a one‑page pack: approvals up, losses flat, partner‑originations share, and customer retention. If any metric breaches the guardrail, pause new originations, adjust policy, and review at the next board.
The hard truth for banks is that distribution moved without asking permission. Marketplaces and retailers now meet customers at the moment of need, and they’re getting good at converting that moment into credit. However, the hard truth for marketplaces is that they still need regulated balance sheets and robust risk management if they want this to endure. The opportunity sits exactly in the middle: AI-first underwriting, delivered at the edge, with bank-grade discipline.
Do that well, and the next time our Warsaw shopper checks out, the instant credit will still feel invisible and effortless. The difference is that a bank will have made that decision in seconds, with better data and better guardrails, and earned the right to compete where the action now happens. From there, the model scales: the same underwriting brain can sit behind dozens of partners and millions of checkouts with consistent controls, turning distribution back into an asset rather than a leak. Miss this shift and the opposite happens — distribution keeps sliding to marketplaces, acquisition costs rise, and loan growth depends on someone else’s storefront. Seize it and you compound reach and margin at the edge while staying within bank‑grade risk limits.
If this resonates, please consider subscribing to "The Next Agenda". For briefings or board-level discussions, feel free to reach out; INED dialogues welcomed where my expertise adds value.
Endnotes - My Sources:
Allegro.eu, Q4/FY 2024 Results (presentation). "Allegro Pay helped finance almost 14% of Allegro's GMV last year; Allegro Pay Visa Card launched." 13 March 2025. https://about.allegro.eu/static-files/ac249de6-03d1-4e23-a9fa-bd89ef56524c and https://about.allegro.eu/static-files/227ecd13-14d4-4b42-8002-ce9cb9dde0cb
Yahoo Finance, Allegro.eu SA (ALEGF) Q4 2024 Earnings Call Highlights. 21 April 2025. https://finance.yahoo.com/news/allegro-eu-sa-alegf-q4-210245864.html
ING Newsroom, ING in Germany and Amazon join forces in SME lending. 30 June 2020. https://www.ing.com/Newsroom/News/ING-in-Germany-and-Amazon-join-forces-in-SME-lending.htm ; and Embedded Finance Review, Why the ING & Amazon Germany partnership ended. 24 June 2025. https://www.embeddedfinancereview.com/p/bad-week-for-embedded-finance-ing-amazon-split-intergiro-license-loss-explained-d3bc43808d071395
Banxware blog, How embedded lending helps pharmacies move faster (Onfy x Banxware). 2025. https://www.banxware.com/blog/banxware-x-onfy-how-embedded-lending-helps-pharmacies-move-faster-in-a-changing-market
Shoper Investor Relations, PragmaPay in Shoper (press office). 20 May 2025. https://investors.shoper.pl/en/press-office/pragmapay-w-shoper ; PragmaGO, Collaboration story: Shoper. https://pragmago.com/collaboration-stories/for-partners-collaboration-stories-shoper/
Volkswagen Financial Services, Instant loan from Volkswagen Bank — fully digital, 15 minutes. 12 October 2022. https://www.vwfs.com/en/media/press-releases/2022/instant-loan-vw-bank.html ; credi2 press, New solution for VWFS. 12 October 2022. https://www.credi2.com/company/press/press-release/new-solution-for-vwfs
EUR‑Lex, Regulation (EU) 2024/1689 — Artificial Intelligence Act (classification of high‑risk AI systems incl. credit scoring). 12 July 2024. https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng
European Commission, Framework for Financial Data Access (FIDA). Accessed August 2025. https://finance.ec.europa.eu/digital-finance/framework-financial-data-access_en
European Commission, Payment Services Package (PSD3/PSR). 2023–2025 updates. https://finance.ec.europa.eu/consumer-finance-and-payments/payment-services/payment-services_en
Tanya Goodin, Credit rating and algorithmic transparency. 22 August 2022. https://tanyagoodin.com/2022/08/credit-rating-algorithmic-transparency/
Credolab, Need a credit score? All you need is a smartphone. Accessed August 2025. https://www.credolab.com/news/need-a-credit-score-all-you-need-is-a-smartphone
