The article considers credit risk in the buy now — pay later (BNPL) model from the perspective of portfolio profitability. It is shown that the key to sustainable growth is behavioral scoring at the checkout, a separate antifraud circuit, and limit “dosing” based on the actual payment discipline. The economic goal is formulated as maximizing the contribution margin taking into account expected losses (PD/LGD/EAD), funding costs, returns, and operating costs. The data architecture, calibration, and reject correction, rules for merchant categories, and penalties for returns are described. Based on pilots and simulation modeling, it is shown that such a combination increases conversion without deteriorating the quality of risk, reduces the share of disputed transactions, and stabilizes cash flows in stress scenarios. Practical recommendations cover approval “handrails”, growth limits, collection strategies, and requirements for decision transparency. Control metrics are discussed separately - vintage analysis, transitions by expiration, model drift and data quality.