Buy Now, Pay Later sounds harmless until you look at who actually can’t pay later

--

The global BNPL market reached approximately $560 billion in 2025, growing 13.7% year over year. In the UK alone, more than 10 million people are expected to use BNPL services this year, with total spending forecast at $38.48 billion. Globally, 380 million people used BNPL in 2024, a number projected to hit 670 million by 2028.

Press enter or click to view image in full size

The growth is real. The convenience is real. And on the surface, everything looks fine.

But here is the number that does not make bold headlines:

In 2025, 41% of BNPL users reported making at least one late payment. In 2024, that figure was 34%. The year before, lower still. Every single year, the number goes up.

The average unsecured debt per UK adult, including BNPL, has risen to £4,308, up from £3,891 in 2022. In the US, 40% of BNPL users have missed more than one payment, and 72% of them have seen their credit scores decline as a result. Klarna, one of the world’s largest BNPL providers, posted a $99 million net loss in Q1 2025 alongside a 17% year-over-year increase in customer credit losses.

These are not abstract statistics. They are real people, in real financial distress. And the pattern behind them is something I have spent a significant amount of time studying.

I spent months going deep into credit default data because I genuinely wanted to understand what the pattern looks like before someone hits rock bottom.

What I found was uncomfortable. Not because it was complex but because it was simple. The same journey, repeating across thousands of cases. Predictable. Preventable. Almost always ignored until it was too late.

The data I worked with was not recent. The numbers have changed. The scale is bigger. But the pattern, the same stages, is playing out right now, in 2026, at a far larger scale.

The defaulter profile I consistently encountered was a combination of two types of people:

The first — low-income households who could not qualify for traditional credit. Their credit limit was already tight, utilisation was high, and there was no financial buffer. BNPL felt accessible to them because no other option was available.

The second — impulsive behaviour that did not account for consequences. The “pay in 3” option appears at checkout. The purchase feels manageable in the moment. The full cost is calculated later, if at all.

BNPL approved both. Without asking. Without checking.

This does not happen overnight. Across thousands of cases, the same four-stage pattern repeated:

Stage 1 — Already stretched before anything starts The user is financially pressured before a single BNPL purchase is made. Low credit limit, high utilisation, no headroom. This is the person BNPL markets itself to and it is also the person most at risk.

Stage 2 — The spending creep Bills begin to grow each month. Payments stay flat minimum or near-minimum. The gap between what is owed and what is being paid widens silently. Within three to four months, the bill can double while the payment remains exactly the same.

Stage 3 — The first missed signal The first late payment arrives. In the data I analysed, a single payment delay of two or more months was the strongest single predictor of default, stronger than income, age, or credit limit. And that first late payment never happens out of nowhere. It is the outcome of months of accumulated financial stress that was already visible in the data, just never flagged.

Stage 4 — The spiral locks in Once the delay pattern starts, it almost never self-corrects. Each subsequent month shows the same or worsening delinquency. Default is not an event. It is a destination that had been set three to six months earlier.

The current numbers confirm this pattern exactly, just at a much larger scale.

This is the fundamental problem with BNPL. And it is not a theory it is visible in how the industry operates.

For a platform, for a brand, what matters is volume. How many approvals. How many transactions. How much GMV growth to report this quarter.

Nobody is watching what happens to that user in month two, month three, month four. Nobody is flagging that bills are creeping up while payments stay flat. Nobody is raising an alert when the gap starts to widen.

Research shows that 44% of frequent BNPL users were already over-indebted before any regulatory intervention began. BNPL loans are largely not reported to credit bureaus they become what analysts call “phantom debt” that introduces systemic risk. Lenders do not know. Regulators do not know. Only the user knows when the bill arrives.

The official default rate sits at around 1.8% to 2%. But 34% to 41% of users report late payments. That gap exists for a reason. “Default” is counted when the provider takes a formal loss. Distress begins the moment a user first pays late. That distress does not appear in anyone’s spreadsheet and so no one fixes it.

The platform grows. The user spirals. And both things happen at the same time.

BNPL providers already have the data. Spending patterns, payment behaviour, purchase frequency, existing obligations all of it is available. Machine learning models can identify the early warning signs of default weeks before it happens. The pattern is detectable. The technology exists.

The reason it is not being used is not technical. It is commercial. Platforms fear that adding any friction will push users to a competitor. So the data sits there. The warning signs go unread. And the spiral continues.

But this thinking is short-term and ultimately self-defeating.

The right approach is to integrate three things simultaneously not as separate features, but as a unified system:

Data-driven affordability checks running silently in the background. The user experience stays seamless. Checkout remains fast. But behind the scenes, the model is assessing whether this user can genuinely afford this purchase based on spending patterns, payment history, and existing BNPL commitments. If the risk is elevated, the limit adjusts automatically. The user does not see friction. The platform avoids a future default.

Real-time spending visibility for the user. A single line — “Your current monthly BNPL commitment is £340” shown at the point of checkout is enough to make someone pause and reconsider. This is not paternalistic. It is transparent. Users who can see their total obligations make better decisions. And users who make better decisions do not default.

Predictive approval decisions at the moment of purchase. When a user reaches checkout, the model already knows their risk profile. The decision to approve, limit, or suggest a lower amount can be automated without delay, without human review, without disrupting the checkout flow. The experience feels identical. The outcome is fundamentally different.

This is not speculative. The data infrastructure for this exists today. The models can be built. The question is whether there is any will to build them.

In the UK, FCA regulation comes into effect in July 2026 bringing stricter creditworthiness assessments, clearer disclosures, and formal consumer protections. Until those rules land, consumers still have no right to affordability checks and no access to the Financial Ombudsman if something goes wrong.

Across Europe, the revised Consumer Credit Directive is being implemented, formally bringing BNPL under regulated credit frameworks. Australia has ended prior exemptions, requiring BNPL providers to meet lending standards under the National Consumer Credit Protection Act.

This is all necessary. And it is long overdue.

But regulation is reactive by nature. It arrives after the harm has already happened, and it sets a floor, not a ceiling. A platform that treats regulation as a compliance checklist box ticked, form filed, move on will not improve outcomes for its users. It will simply be compliant while the same pattern continues underneath.

The platforms that will actually make a difference are the ones that understand something simple: a user who defaults is a user they have lost forever. Long-term user health is long-term business health. Data-driven decisions do not slow growth they make growth sustainable.

This problem does not belong to any single party. And neither does the solution.

Platforms: The data you have should not only serve conversion. It should serve protection. A first missed payment is a signal. Your model can already see it. Do something with it before the spiral locks in.

Regulators: The 2026 UK regulation is a start. But disclosure requirements alone are not enough. Affordability checks must be mandatory, verifiable, and standardised — not self-reported by the same providers who benefit from approving everyone.

Users: Financial literacy matters here. “Pay in 3” is a loan. Calculate the total cost before you buy. Track your monthly BNPL commitment across every platform you use. BNPL is convenient, but the cost of that convenience becomes visible when all the instalments mature at once.

BNPL is a useful tool. I genuinely believe that.

For the right person, for the right purchase, at the right time, it offers real financial flexibility that traditional credit cannot. It has a legitimate place in how people manage their money.

But what is happening right now is not that.

What is happening right now is a tool reaching the people who are already financially stretched, approving them without checks, and then watching silently as they move through a predictable, data-visible spiral while the platform reports another quarter of transaction growth.

“The FCA published its final rules in February 2026 — you can read the full policy statement here.”

Regulation has arrived. The FCA’s new rules are now in effect in 2026. Affordability checks are no longer optional. Consumer protections are now a legal requirement.

But a law being passed and a problem being solved are two entirely different things.

Compliance means doing the minimum the law requires. Responsibility means using the data you already have to genuinely protect the people on your platform not because you have to, but because a user who defaults is a user you have permanently lost.

The rules are here. The data has always been here. The only question now is what the industry chooses to do with both.

BNPL can be fixed. The tools exist. The law has made a start. Whether the industry goes beyond the minimum that is the real test.

Written by Ikram Ul Haq — Fintech BA and Product professional writing about payments, BNPL, credit risk, and the intersection of data and product thinking.