"Four interest-free payments" is neither free nor simple.

BNPL underwrites per-purchase, but portfolio health is read in vintages: all loans originated in a given month, tracked as cumulative losses over their life. Pay-in-4 vintages mature brutally fast — weeks, not years — which is BNPL's secret risk advantage: when the economy turns, a BNPL book reprices its risk in one quarter while a credit-card book drags years of seasoned balances. Charge-off typically happens after ~120–180 days of delinquency (accounting policy varies); before that, collections escalate from app nudges to agencies. The metrics that matter: first-installment default rate (the fraud-adjacent one — installment one bouncing usually means stolen payment method or no intent), cohort loss curves against plan, and the share of users on their fifth-plus concurrent plan — 'loan stacking', the blind spot bureaus are only now closing.

Every BNPL balance is lent money that must itself be borrowed. The funding ladder, cheapest to priciest: deposits (Klarna holds an EU banking licence partly for this), warehouse credit lines from banks, forward-flow agreements (sell the receivables to an investor as they originate — Affirm's deals are the public example), and securitization (bundle receivables into ABS notes). Unit economics live or die on this: a ~5% merchant fee funds maybe weeks of receivable life, so the spread over funding cost is thin, and every 100bp of central-bank rate hikes in 2022–23 fell straight through BNPL P&Ls — the sector's pivot from growth to profitability was a funding-cost story wearing a strategy costume. Diligence question one for any BNPL: what's your blended cost of funds, and what happens to it if rates move 200bp?

BNPL's early growth exploited a reporting hole: pay-in-4 plans mostly weren't furnished to credit bureaus, so each provider underwrote blind to the others — and disciplined-looking consumers quietly stacked five plans across five apps. That hole is closing: bureaus began ingesting BNPL data via specialty programs, US furnishing expanded through 2024–25, score builders announced BNPL-inclusive models, and regulators (UK FCA rules live 15 Jul 2026; EU CCD2 folding BNPL into consumer credit) increasingly expect affordability checks that see total exposure. The strategic effect cuts both ways: reporting builds thin-file consumers real credit history (the emerging-market promise) — and it ends the era when BNPL risk models could pretend their applicant had no other debts.

Repayments ride card rails or bank debits, so a failed installment is a decline, not a default — yet. The provider's machine: smart retries (timed to paydays, using account updater for reissued cards), method switching (try the backup card), then late fees where permitted (capped in many markets; the UK regime and several EU states restrict them), then plan freezing (no new purchases — the real behavioral lever), then collections. Disputes are two-layered and genuinely confusing: a merchant problem (item never arrived) is a BNPL-provider dispute — there's no card chargeback against the merchant, the BNPL absorbed that risk; but the consumer's repayment can itself be charged back against the BNPL if it rode a card. BNPL replaced the issuer in the liability chain, and support tickets everywhere prove consumers haven't been told.

The merchant fee only looks irrational until you see what it buys: measured conversion lift and larger baskets (numbers vary by category and study — treat all vendor-quoted lifts as marketing until A/B'd on your own funnel), zero credit risk (BNPL pays the merchant in full, T+1-ish, and owns the default), and increasingly the lead-gen channel: BNPL apps became shopping destinations, and 'sponsored placement' in them is straight retail-media spend. That last one explains fee tolerance best — merchants compare BNPL fees not to card fees but to customer-acquisition cost. It also explains regulator interest in the app-store dynamics: when the payment method is also the marketplace, whose customer is it?