The Credit System Is Racing to See Buy-Now-Pay-Later. It Can Only See the Healthiest Slice.

In the span of a year, three different arms of the American credit machine decided that buy-now-pay-later (BNPL) could no longer stay invisible. FICO, the company behind the score most US lenders rely on, built two new credit scores to read it (FICO, 23 June 2025). The Department of Housing and Urban Development (HUD) asked whether hidden BNPL debt is quietly inflating the mortgages it insures (HUD request for information, 24 June 2025). And on 4 May 2026, four Democratic senators pressed Experian, Equifax and TransUnion to explain how they handle it (Senate Banking Committee letters, 4 May 2026).

These are not one news event; they are a convergence. FICO's new scores became available from late 2025 and are working their way to lenders through 2026, so this is the stretch when the plumbing actually gets connected. All three efforts rest on a single assumption: that the credit file can see BNPL. So we asked one narrow, falsifiable question, and the answer is not the one the system is counting on.

As the system rushes to fold pay-in-4 into credit reports and mortgage math, how much of it can the file actually see today, and is that visible share moving toward household stress or away from it?

The short answer: the gauge reads only a minority of US pay-in-4pay-in-4The most common form of BNPL: a purchase split into four equal, interest-free payments spread over roughly six weeks., and the visible part is the healthy part. The slice growing fastest, and showing the most strain in the limited data we have, is the slice that stays dark. We will walk through exactly how we reached that, and where the comparison has to be made carefully rather than overstated.

BNPL's "pay-in-4" splits a purchase into four interest-free payments over about six weeks. Because there is no interest and often no hard credit check, for years it left no trace on a credit report. The loan was real, but to a bureau it did not exist.

A loan appears on your file only if the lender chooses to send the data, an act the industry calls "furnishing." For pay-in-4, furnishing is voluntary, and most of the largest originators decline. So the first thing to measure is not how the loans behave, but simply who reports them at all.

We built the table above by reading each provider's own furnishing disclosures and the bureaus' announcements. The pattern is stark. Exactly one major US player, Affirm, furnishes its full pay-in-4 book, and even Affirm reaches only two of the three bureaus: Experian (for loans issued from 1 April 2025, announced 19 March 2025) and TransUnion (for loans issued from 1 May 2025, announced 22 April 2025), but not Equifax.

Everyone else either reports nothing by default or reports something other than pay-in-4. Klarna furnishes only its longer 'pay over time' financing plans, not its interest-free pay-in-4. Afterpay (owned by Block), PayPal's US Pay in 4, and Zip's standard pay-in-4 do not furnish to the bureaus at all. Sezzle reports through a free feature called 'Sezzle Up,' and it does reach all three bureaus, but only for borrowers who actively sign up, so it is an opt-in rather than a default.

It is tempting to read FICO's two new scores, FICO Score 10 BNPLFICO Score 10 BNPLA 2025 version of the FICO score designed to take buy-now-pay-later loans into account, but only for loans a lender has already put in the credit file. and FICO ScoreFICO ScoreThe most widely used credit score in the US, a three-digit number lenders use to gauge how risky a borrower is. 10 T BNPL, as the moment the lights come on. They are not. A score is a model: a recipe for weighing data that is already in the file. FICO does not furnishfurnishThe industry term for a lender actively sending loan data to a credit bureau. A bureau can only show a loan if the lender chooses to furnish it. anything itself. If a lender never sends a loan to a bureau, no FICO model, however clever, can see it.

FICO's genuine innovation is narrower: it teaches the score to recognize when a borrower has opened several BNPL loans in a short window, a behavior older models mishandled. Useful, but it only sharpens the picture of loans already furnished. It cannot conjure the dark book into view.

If the invisible book were the safest part of the market, none of this would matter much. It is the opposite. The clearest signs of strain sit precisely where the file cannot see.

The share of US BNPL users who admit paying late has climbed from 34% in 2024 to 41% in 2025 to 47% in 2026, on LendingTree's recurring survey of roughly 2,000 adults. This covers all BNPL users, including Affirm's prime, furnished borrowers, so if anything it understates the strain in the unfurnished, lower-quality book.

And what people are financing has shifted toward essentials. The share using BNPL to buy groceries has roughly doubled, from 14% two years ago to 29% in 2026 (and about 38% among Gen Z). Splitting a sofa into four payments is discretionary; splitting the weekly food shop is a sign of a stretched budget.

Then there is loan stackingloan stackingHolding several BNPL loans at once, often across different apps, so no single lender sees the borrower's full obligations.. In the CFPBCFPBThe Consumer Financial Protection Bureau, the US federal agency that oversees consumer lending.'s most recent loan-level data (2022, published January 2025), 63% of BNPL borrowers held more than one BNPL loan at the same time, and 33% held loans across different firms at once. This is the single pattern furnishing is supposed to expose, and the single pattern voluntary, partial furnishing structurally cannot: if borrower X has one Affirm loan and three unfurnished loans elsewhere, the file shows one. The credit report is not just incomplete; it is systematically biased toward looking calmer than reality.

Here is the honest difficulty at the center of this story. For the visible book we have hard data: realized dollar losses. For the dark book we have only soft data: people telling a survey they paid late. These are not the same kind of number, and we refuse to draw them as one trend line, because that would fake a precision we do not have.

On the hard-data side, the picture is reassuring. The industry's charge-offcharge-offThe point at which a lender gives up on collecting a loan and writes it off as a loss. The charge-off rate is the share of dollars lost this way. rate, the share of dollars lenders give up on, actually fell from 2.63% in 2022 to 1.83% in 2023 (CFPB, December 2025 report). Affirm does not break out its pay-in-4 loss rate, but every signal points to it running well under 1% of volume, consistent with the CFPB's roughly 0.92% industry pay-in-4 charge-off for 2023. The visible book is genuinely well-behaved.

So the finding is not 'a hidden crisis is showing up in the losses.' It is subtler and, we think, more important: the one book with trustworthy loss data is the healthiest one, while the books showing rising late payments, grocery financing and stacking are the ones with no comparable loss data at all. The incomparability is the point. The system is wiring itself to the gauge that reads clean, and is blind to the gauge that reads stressed.

How big is the thing we are arguing about? Even that is unsettled. Three serious attempts to size US pay-in-4 originationsoriginationsThe total value of new loans made over a period, as opposed to the balance still outstanding at a single point in time. span from about $45bn (CFPB, 2023 actual) to $63bn (Congressional Research Service, estimate for 2025) to $70bn (Richmond Fed, estimate for 2025), a 1.5x spread. Widen the lens to all BNPL products and a June 2026 Federal Reserve note puts 2025 originations near $160bn. The base year itself is disputed: the CFPB's own December 2025 report says $45.2bn for 2023, while the CRSCRSThe Congressional Research Service, a nonpartisan body that writes briefing reports for members of the US Congress. and the Fed cite an earlier CFPB figure of $43.9bn. When the experts cannot agree on the size to within a third, the case for measuring it properly, rather than guessing, gets stronger, not weaker.

A popular worry holds that private creditprivate creditLoans made by investment funds rather than deposit-taking banks, a fast-growing and lightly-disclosed corner of finance., loans funded by investment funds rather than banks, is secretly inflating BNPL risk. The marquee example is KKR's deal to buy BNPL receivablesreceivablesMoney a company is owed by its customers; here, the outstanding balances on BNPL loans.. But that pipe is European: the 2023 KKR agreement covers PayPal 'Pay Later' loans originated in France, Germany, Italy, Spain and the UK, not the US. It tells us nothing about American credit files.

And the logic runs backwards anyway. Affirm is heavily fund- and market-financed, through warehouse lines and securitizations, yet it is the one player that furnishes its whole book. Funding structure does not determine visibility. The only lever that determines whether a loan shows up is whether the originator chooses to furnish it. That is the whole game, and it is a choice, not a market force.

Honesty requires holding two things at once. The first half of this piece argues the file is dangerously blind. The second half admits that blindness may not be a disaster, and that the cure has its own side effects.

Pay-in-4 is small, fast-clearing and low-loss. A six-week, interest-free loan that pays itself off rarely festers into the kind of balance that sinks a household. So the economic stakes of invisibility are probably modest today. And forcing visibility is not free. Reporting a tiny, short-lived loan can actually lower the score of a thin-file borrowerthin-file borrowerSomeone with little credit history on record, so a single new loan moves their score a lot. Often the young or first-time borrowers BNPL attracts., someone with little credit history, because a brand-new account and a fresh credit check ding exactly the young, first-time borrowers BNPL most attracts. HUD's own concern with FHA mortgages is the mirror image: there, invisible BNPL debt may let buyers look more affordable than they are, inflating the mortgages the government insures.

So the policy question is not simply 'turn the lights on.' It is 'light up the right book, in the right way, without punishing the people the system says it wants to protect.' What our analysis shows is narrower and, we hope, durable: as currently wired, the credit file sees the healthiest slice of pay-in-4 and misses the stressed one, and it does so by design, because furnishing is a voluntary choice the safest lender happens to be making and the rest are not.